A Comparison of TTF and RTF UHF RFID Protocols | Henri Marais

September 13, 2018 | Author: Anonymous | Category: Documents
Share Embed


Short Description

12 RFID Systems 1 0.8 Read Rate (Tag Counts/Queries) q = 0, f = 0 0.6 q = 30, f = 0 q .... The reader transmits at a pow...

Description

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/229907323

A Comparison of TTF and RTF UHF RFID Protocols Chapter · July 2010 DOI: 10.1002/9780470665251.ch9

CITATIONS

READS

0

821

3 authors, including: Johann Erich Wolfgang Holm

Henri Marais

North West University South…

North West University South…

15 PUBLICATIONS 49 CITATIONS

8 PUBLICATIONS 4 CITATIONS

SEE PROFILE

SEE PROFILE

Some of the authors of this publication are also working on these related projects: Applicability of energy-based fault detection for transient systems View project

All content following this page was uploaded by Johann Erich Wolfgang Holm on 31 December The user has requested enhancement of the downloaded file.

RFID SYSTEMS RESEARCH TRENDS AND CHALLENGES Edited by Miodrag Boli´c University of Ottawa, Canada

David Simplot-Ryl INRIA, France and University of Lille, France

Ivan Stojmenovi´c University of Ottawa, Canada

A John Wiley and Sons, Ltd., Publication

RFID SYSTEMS

RFID SYSTEMS RESEARCH TRENDS AND CHALLENGES Edited by Miodrag Boli´c University of Ottawa, Canada

David Simplot-Ryl INRIA, France and University of Lille, France

Ivan Stojmenovi´c University of Ottawa, Canada

A John Wiley and Sons, Ltd., Publication

This edition first published 2010  2010 John Wiley & Sons Ltd. Except for: Chapter 5, ‘Design of Passive Tag RFID Readers’  2010 Intel Corporation Registered office John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, United Kingdom For details of our global editorial offices, for customer services and for information about how to apply for permission to reuse the copyright material in this book please see our website at www.wiley.com. The right of the author to be identified as the author of this work has been asserted in accordance with the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by the UK Copyright, Designs and Patents Act 1988, without the prior permission of the publisher. Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic books. Designations used by companies to distinguish their products are often claimed as trademarks. All brand names and product names used in this book are trade names, service marks, trademarks or registered trademarks of their respective owners. The publisher is not associated with any product or vendor mentioned in this book. This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold on the understanding that the publisher is not engaged in rendering professional services. If professional advice or other expert assistance is required, the services of a competent professional should be sought.

Library of Congress Cataloging-in-Publication Data RFID systems : research trends and challenges / edited by Miodrag Bolic, David Simplot-Ryl, and Ivan Stojmenovic. p. cm. Includes index. ISBN 978-0-470-74602-8 (cloth) 1. Radio frequency identification systems. I. Bolic, Miodrag. II. Simplot-Ryl, David. III. Stojmenovic, Ivan. TK6570.I34R4868 2010 658.7 87–dc22 2010003318 A catalogue record for this book is available from the British Library. ISBN 978-0-470-74602-8 (H/B) Set in 10/12 Times by Laserwords Private Limited, Chennai, India Printed and Bound in Singapore by Markono

To my wife Andjelka and children Marija, Natasa and Katarina. Miodrag Boli´c

To Isabelle, my wife. David Simplot-Ryl

To my wife Natasa and children Milos and Milica. Ivan Stojmenovi´c

Contents About the Editors

xvii

Preface

xix

Acknowledgements

xxi

Part I

1 1.1

1.2 1.3

1.4 1.5

2 2.1 2.2

COMPONENTS OF RFID SYSTEMS AND PERFORMANCE METRICS Performance of Passive UHF RFID Systems in Practice Miodrag Boli´c, Akshay Athalye, and Tzu Hao Li Introduction 1.1.1 Overview 1.1.2 Background Ideal RFID System Practical RFID Systems 1.3.1 Complexity of RFID Systems 1.3.2 Single Reader, Single Tag 1.3.3 Single Reader, Multiple Tags 1.3.4 Multiple Readers, Single or Multiple Tags 1.3.5 Mobile Readers and/or Mobile Tags 1.3.6 Large Deployments Including Many Readers and Tags 1.3.7 Other Desired Features of Practical RFID Systems Overview of the Book Conclusion References Performance Metrics and Operational Parameters of RFID Systems Raj Bridelall and Abhiman Hande Overview Key Operational Parameters 2.2.1 Operating Distance 2.2.2 System Throughput 2.2.3 Localization 2.2.4 Impact of Materials 2.2.5 Other Factors Considered

3 3 3 4 5 7 7 7 12 15 16 17 18 19 21 21 23 23 24 26 32 39 43 44

viii

2.3

2.4

3 3.1

3.2

3.3 3.4

3.5

4 4.1

4.2

4.3

Contents

Classification of Commercially Available Products 2.3.1 Near-Field Coupled Systems 2.3.2 Far-Field Propagating Systems 2.3.3 Ultra Wide-Band 2.3.4 Passive Solutions 2.3.5 Semi-Passive Architectures 2.3.6 Far-Field Solutions 2.3.7 Near-Field Solutions 2.3.8 Active Architectures Conclusion Problems References

47 48 50 51 52 52 53 53 53 54 55 55

UHF RFID Antennas Daniel Deavours Dipoles and Relatives 3.1.1 Dipole 3.1.2 Radiation 3.1.3 Impedance and Bandwidth 3.1.4 Radiating Resistance 3.1.5 Polarization T-Match and Relatives 3.2.1 The Classic T-Match 3.2.2 The Modified T-Match Putting it Together: Building an RFID Tag The Environment 3.4.1 Dielectric Constant 3.4.2 Dielectric Loss 3.4.3 Metals 3.4.4 Propagation 3.4.5 Practical Steps to Overcome Environmental Challenges Conclusions, Trends, and Challenges References

57

RFID Tag Chip Design Na Yan, Wenyi Che, Yuqing Yang, and Qiang Li Tag Architecture Systems 4.1.1 Tag Architecture 4.1.2 Design of High Efficiency Frontend Circuits Memory in Standard CMOS Processes 4.2.1 Why Have a Standard CMOS eNVM? 4.2.2 Basic Cell Structures and Operation Mechanisms 4.2.3 Memory Architecture and Peripheral Circuits 4.2.4 Future Challenges Baseband of RFID Tag

58 59 60 61 65 67 69 69 71 74 81 81 83 84 86 87 97 98 99 99 99 100 109 109 110 113 115 115

Contents

4.4

4.5

5 5.1 5.2

5.3

5.4

5.5

6 6.1 6.2

6.3

6.4

6.5

ix

4.3.1 Introduction 4.3.2 Low Power Baseband Design 4.3.3 Clock Rate 4.3.4 Clock-Related Low-Power Techniques 4.3.5 Sub-Threshold Digital Circuit 4.3.6 Adiabatic Circuit RFID Tag Performance Optimization 4.4.1 Low Power 4.4.2 Low Cost Conclusion Problems References

115 116 117 119 121 121 122 123 123 125 125 126

Design of Passive Tag RFID Readers Scott Chiu Overview Basics of Passive RFID Operation 5.2.1 An Introduction to ISO 18000-6C Air Interface 5.2.2 Tag Singulation and Access Passive RFID Reader Designs 5.3.1 RFID Reader Read Range and Transmitted Power 5.3.2 RFID Reader Implementation Advanced Topics on RFID Reader Design 5.4.1 Integrated Transceiver 5.4.2 Cancellation of Transmitted Carrier Leakage 5.4.3 Dense Reader Operations Conclusion Problems References

129

RFID Middleware: Concepts and Architecture Nathalie Mitton, Lo¨ıc Schmidt, and David Simplot-Ryl Introduction Overview of an RFID Middleware Architecture 6.2.1 The Need for a Middleware 6.2.2 Architecture Readers Management 6.3.1 Reader Protocol/Interface 6.3.2 Manage and Monitor Data Management and Application-Level Events 6.4.1 Data Management and ALE Functionalities 6.4.2 Specs and Reports 6.4.3 Research Challenges Store and Share Data 6.5.1 EPC Information Services

129 130 131 134 136 137 139 146 146 147 148 150 151 151 155 155 156 156 157 160 160 162 164 165 166 170 171 171

x

6.6 6.7

Part II 7 7.1 7.2 7.3

7.4

Contents

6.5.2 Object Naming Service 6.5.3 Discovery Services Example Conclusion Problems References

173 174 174 176 176 176

TAG IDENTIFICATION PROTOCOLS Aloha-Based Protocols Kwan-Wu Chin and Dheeraj Klair Pure Aloha Slotted Aloha 7.2.1 Pure versus Slotted Aloha Variants Framed Slotted Aloha 7.3.1 Basic 7.3.2 Dynamic 7.3.3 Enhanced/Hybrid Conclusion Problems References

181 182 184 185 187 188 189 193 199 200 201

8

Tree-Based Anti-Collision Protocols for RFID Tags Petar Popovski

203

8.1 8.2

Introduction Principles of Tree-Based Anti-Collision Protocols 8.2.1 System Model 8.2.2 Basic Tree Protocols 8.2.3 Improvements to the Basic Tree Protocol 8.2.4 General Arbitration Framework for Tree-Based Protocols 8.2.5 Numerical Illustration Tree Protocols in the Existing RFID Specifications 8.3.1 Tree Protocol for EPCglobal Class 0 8.3.2 Tree Protocol for EPCglobal Class 1 Practical Issues and Transmission Errors 8.4.1 Token Generation 8.4.2 Transmission Errors 8.4.3 Dealing with Moving Tags Cooperative Readers and Generalized Arbitration Spaces 8.5.1 Two-Dimensional Arbitration Space 8.5.2 Further Remarks and Multi-Dimensional Arbitration Conclusion Problems References

203 205 205 207 209

8.3

8.4

8.5

8.6

210 214 214 215 216 217 217 217 221 222 223 226 227 228 228

Contents

xi

9

A Comparison of TTF and RTF UHF RFID Protocols Alwyn Hoffman, Johann Holm, and Henri-Jean Marais

231

9.1 9.2

Introduction Requirements for RFID Protocols 9.2.1 Categories of RFID Technology 9.2.2 Requirements for Passive UHF RFID Different Approaches Used in UHF Protocols 9.3.1 Deterministic versus Stochastic 9.3.2 RTF versus TTF Description of Stochastic TTF Protocols 9.4.1 Supertag 9.4.2 IP-X 9.4.3 TOTAL 9.4.4 Comparison between Different TTF Protocols 9.4.5 TTF Performance with Additional Data Pages Comparison between ISO18000-6C and TTF Protocols 9.5.1 Areas of Comparison 9.5.2 The Impact of Progress on Technology 9.5.3 A Comparison between RTF and TTF for Fast Moving Tags Conclusion Problems References

231 232 232 236 238 239 240 241 242 244 246 248 253 255 255 258 261 265 266 267

9.3

9.4

9.5

9.6

Part III

READER INFRASTRUCTURE NETWORKING

10

Integrating RFID Readers in Enterprise IT Christian Floerkemeier and Sanjay Sarma

271

10.1 10.2 10.3 10.4 10.5

Related Work RFID System Services Reader Capabilities RFID System Architecture Taxonomy EPCglobal Standards 10.5.1 Discovery, Configuration and Initialization (DCI) and Reader Management (RM) 10.5.2 Low Level Reader Protocol (LLRP) 10.5.3 Reader Protocol (RP) 10.5.4 Application Level Event (ALE) 10.5.5 EPC Information Service (EPCIS) 10.5.6 Tag Data Translation Specification (TDT) Adoption of High-Level Reader Protocols Potential Future Standardization Activities Conclusion Problems References

272 272 277 278 280

10.6 10.7 10.8

282 282 284 285 289 290 290 292 293 294 294

xii

11 11.1 11.2 11.3

11.4 11.5

12

12.1 12.2 12.3 12.4 12.5 12.6 12.7

12.8 12.9

12.10 12.11

12.12

12.13

Contents

Reducing Interference in RFID Reader Networks Sung Won Kim and Gyanendra Prasad Joshi Introduction Interference Problem in RFID Reader Networks Access Mechanism, Regulations, Standards and Algorithms 11.3.1 Regulations 11.3.2 Standards 11.3.3 Reader Anti-Collision Algorithms Comparison Conclusion Problems References

297

Optimal Tag Coverage and Tag Report Elimination Bogdan Carbunar, Murali Krishna Ramanathan, Mehmet Koyuturk, Suresh Jagannathan, and Ananth Grama Introduction Overview of RFID Systems Tree Walking: An Algorithm for Detecting Tags in the Presence of Collisions Reader Collision Avoidance 12.4.1 Implementation Coverage Redundancy in RFID Systems: Comparison with Sensor Networks Network Model Optimal Tag Coverage and Tag Reporting 12.7.1 Problem Definition 12.7.2 Problem Complexity Redundant Reader Elimination Algorithms: A Centralized Heuristic 12.8.1 Analysis RRE: A Distributed Solution 12.9.1 RRE 12.9.2 RRE-HC 12.9.3 Analysis 12.9.4 Dependency on RCA Adapting to Topological Changes 12.10.1 Tag Count Resetting The Layered Elimination Optimization (LEO) 12.11.1 Implementation 12.11.2 Analysis Related Work 12.12.1 Coverage Problems in WSNs 12.12.2 Collisions in RFID Systems Conclusion Problems References

321

297 298 300 301 302 303 314 316 317 317

321 324 326 326 327 328 330 331 331 332 334 335 335 336 338 338 339 340 341 342 342 343 343 343 344 344 345 345

Contents

13

13.1 13.2 13.3 13.4 13.5 13.6

Delay/Disruption-Tolerant Mobile RFID Networks: Challenges and Opportunities Hongyi Wu and Zhipeng Yang Motivation Overview of FINDERS General Feasibility Study Unique Challenges and Tactics Related Work Conclusion Problems References

xiii

349 349 350 351 355 358 359 359 360

Part IV ADDRESSING OTHER CHALLENGES IN RFID SYSTEMS 14

14.1 14.2

14.3 14.4 14.5 14.6

15 15.1 15.2

15.3

15.4

Improving Read Ranges and Read Rates for Passive RFID Systems Zhiguang Fan, Fazhong Shen, Jianhua Shen, and Lixin Ran Introduction Signal Descriptions and Formulations for Passive Backscatter RFID Systems 14.2.1 Signal Descriptions 14.2.2 SNR and Read Range Formulation Improving the Read Range of a Passive RFID System Improving the Read Rate of a Passive RFID System Two Design Examples for RFID System Conclusion Problems References Principles and Techniques of RFID Positioning Yimin Zhang, Xin Li, and Moeness Amin Introduction Tag Range Estimation Techniques 15.2.1 RSS-Based Techniques 15.2.2 Phase-Based Techniques 15.2.3 Time-Based Techniques DOA Estimation Techniques 15.3.1 Directional Antenna 15.3.2 Phased Array 15.3.3 Smart Antenna RFID Positioning Techniques 15.4.1 Trilateration/Multilateration 15.4.2 Triangulation 15.4.3 Hybrid Direction/Range Methods 15.4.4 Radio Map Matching Methods

365 365 366 367 369 374 379 381 386 386 387 389 389 392 392 394 396 397 398 398 398 399 399 401 403 405

xiv

15.5 15.6

16 16.1 16.2 16.3 16.4

16.5

16.6

17

17.1 17.2

17.3

Contents

15.4.5 Proximity Improving Positioning Accuracy Conclusion Problems References

408 409 411 411 412

Towards Secure and Privacy-Enhanced RFID Systems Heiko Knospe and Kerstin Lemke-Rust Introduction Security and Privacy Classification of RFID Systems Attacks on RFID Systems and Appropriate Countermeasures 16.4.1 Eavesdropping of Messages 16.4.2 Denial-of-Service 16.4.3 Manipulation of Messages 16.4.4 Generation of Messages 16.4.5 Relay of Messages 16.4.6 Tracking and Hotlisting 16.4.7 Cloning of Transponders 16.4.8 Cryptanalytic Attacks 16.4.9 Physical Implementation Attacks Lightweight Cryptography for RFID 16.5.1 Random Number Generators 16.5.2 Block Ciphers 16.5.3 Stream Ciphers 16.5.4 Hash Functions 16.5.5 Public-Key Cryptography Conclusion Problems References

417

Cryptographic Approaches for Improving Security and Privacy Issues of RFID Systems Miyako Ohkubo, Koutarou Suzuki, and Shingo Kinoshita Introduction Threats against the RFID System 17.2.1 Passive Reading Attack 17.2.2 Active Reading Attack 17.2.3 Rewriting Attack 17.2.4 Cloning Attack 17.2.5 Destruction/DoS Attack 17.2.6 Scanning/Tracking Attack 17.2.7 Side-Channel Attack 17.2.8 Attack against Overall System Security Required Properties 17.3.1 Identification

417 417 418 420 421 422 423 423 423 424 425 425 427 431 432 434 437 439 440 443 443 444

447 448 449 450 450 451 451 451 452 452 452 452 453

Contents

17.4 17.5 17.6

17.7 17.8

17.9

18

18.1 18.2

18.3

18.4

18.5

17.3.2 Authentication 17.3.3 Privacy 17.3.4 Indistinguishability 17.3.5 Forward Security 17.3.6 Delegation and Restriction 17.3.7 Proof of Existence 17.3.8 Distance Bounding 17.3.9 Synchronization Cryptographic Protocols for Identification with Privacy Cryptographic Protocols for Authentication without Privacy Cryptographic Protocols for Privacy and Other Requirements 17.6.1 Approaches with Hash Functions 17.6.2 Approaches for Forward Security with Hash Chain 17.6.3 Approaches with Binary Tree 17.6.4 Approaches with Block Ciphers 17.6.5 Approaches with Lightweight Methods 17.6.6 Approaches with Public-Key Methods 17.6.7 Approaches for Proof of Existences 17.6.8 Mutual Authentication 17.6.9 Approaches without Cryptography Implementation Real Systems and Attacks 17.8.1 e-Passport 17.8.2 MiFare Card 17.8.3 KeeLoq 17.8.4 Approach to Strengthen EPC Conclusion Problems References Novel RFID Technologies: Energy Harvesting for Self-Powered Autonomous RFID Systems Raj Bridelall and Abhiman Hande Introduction Novel Low Power Architectures 18.2.1 Dual-Active Standards 18.2.2 Micro-Wireless RFID 18.2.3 Semi-Active Energy Harvesting Optimized for RFID 18.3.1 Solar Cells 18.3.2 Thermoelectric Transducers 18.3.3 Vibration Energy Scavenging Solutions Future Trends in Energy Harvesting 18.4.1 Thin-Film MEMS Piezoelectric Cantilevers 18.4.2 Integrated Power Management with Load Balancing Conclusion

xv

453 454 455 455 456 456 457 457 457 459 460 460 461 462 462 462 463 463 463 464 464 466 466 466 467 467 468 468 468

473 473 475 475 476 477 478 480 482 483 488 489 491 493

xvi

Contents

Problems References 19

19.1

19.2

19.3

19.4

19.5

19.6

Index

Simulators and Emulators for Different Abstraction Layers of UHF RFID Systems Christian Steger, Alex Janek, Reinhold Weiß, Vojtech Derbek, Manfred Jantscher, Josef Preishuber-Pfluegl, and Markus Pistauer Introduction 19.1.1 Motivation 19.1.2 Goal of the Simulation/Emulation Platform 19.1.3 Model-Based Design and Verification of UHF RFID Systems 19.1.4 Higher Class RFID Tags and Energy Harvesting Devices 19.1.5 Basics on Conformance, Performance and Interoperability Testing The Simulation/Emulation Platforms 19.2.1 Layers of the Modeling and Verification Framework 19.2.2 Implementation Languages UHF RFID Simulation Platform 19.3.1 Multi-Layer Optimization 19.3.2 Modeling and Simulation Techniques 19.3.3 Model for the Simulation of the UHF RFID System 19.3.4 Use Case: UHF RFID Systems 19.3.5 RFID Application and System Design Kit+Library Real-Time HIL-Verification and Emulation Platform 19.4.1 Timing Analysis 19.4.2 Use Case: Multi UHF Tag Emulator 19.4.3 RFID Tag Emulator Higher Class Tag Architecture Based on Energy Harvesting 19.5.1 Proposed Mapping of Functional Blocks to Tag ASIC Architecture 19.5.2 Cosimulation for Functional Verification: The Partitioning of the UHF RFID System Simulation Model 19.5.3 Two-Level Simulation Method for Verification and Improvements Evaluation 19.5.4 Use Case Logistics: A Container Transport Conclusion Problems References

493 493

497

497 497 498 499 500 502 505 506 509 511 512 514 520 520 524 525 526 528 530 531 531 532 535 536 539 539 540 543

About the Editors Miodrag Boli´c, [email protected], www.site.uottawa.ca/∼mbolic Miodrag Boli´c received his B.S. and M.S. degrees in electrical engineering from the University of Belgrade, Serbia in 1996 and 2001, respectively, and his Ph.D. degree in electrical engineering from Stony Brook University, NY, in 2004. Since 2004, he has been with the University of Ottawa, Canada, where he is an associate Professor at the School of Information Technology and Engineering. His current research interests include computer architectures, biomedical signal processing and RFID. He has eight years of industrial experience from the US and Serbia related to digital signal processing and embedded system design. He is a co-founder of a start-up Astraion Inc., NY, that develops novel RFID systems. He is a founder and director of Computer architecture research group and RFID research group at the University of Ottawa. He has been a principal investigator on a number of projects funded by NSERC, Canada, Ontario Centres of Excellences and industry. Dr. Boli´c has been involved in a number of research service activities including: chair of the joint chapter of signal processing, oceanic engineering, geosciences and remote sensing for the IEEE Ottawa section, and associate editor of Telecommunication Systems journal, Springer. David Simplot-Ryl, [email protected], http://www.lifl.fr/∼simplot David Simplot-Ryl received the Graduate Engineer degree in computer science, automation, electronic and electrical engineering, and M.Sc. and Ph.D. degrees in computer science from the University of Lille, France, in 1993 and 1997, respectively. In 1998, he joined the Fundamental Computer Science Laboratory of Lille (LIFL), France, where he is currently professor. He received the Habilitation degree from the University of Lille, France, in 2003. His research interests include sensor and mobile ad hoc networks, mobile and distributed computing, embedded operating systems, smart objects and RFID technologies. Recently, his main occupation is contributing to international standardization on RFID tag identification protocols in partnership with Gemplus and TagSys companies. He has written scientific papers, book chapters and patents and received the Best Paper award at the 9th International Conference on Personal Wireless Communications (PWC 2004) and at the 2nd International Conference on Mobile Ad-hoc and Sensor Networks (MSN 2006). He is an associate editor of Ad Hoc and Sensor Wireless Networks: An International Journal (Old City Publishing) and a member of the editorial board of International Journal of Computers and Applications (Acta Press), the International Journal of Wireless

xviii

About the Editors

and Mobile Computing (Inderscience), and International Journal of Parallel, Emergent and Distributed Systems (Taylor & Francis). Ivan Stojmenovi´c, [email protected], www.site.uottawa.ca/∼ivan Ivan Stojmenovi´c received the Ph.D. degree in mathematics. He has held regular and visiting positions in Serbia, Japan, the USA, Canada, France, Mexico, Spain, the UK (as Chair in Applied Computing at the University of Birmingham), Hong Kong, and Brazil, and is a Professor at the University of Ottawa, Canada. He has published over 250 different papers, and has edited four books on wireless, ad hoc and sensor networks and applied algorithms with Wiley/IEEE. He is the editor of over a dozen journals, is editor-in-chief of IEEE Transactions on Parallel and Distributed Systems (from January 2010), and founder and editor-in-chief of three journals (Multiple-Valued Logic and Soft Computing; Parallel, Emergent and Distributed Systems; and Ad Hoc and Sensor Wireless Networks). Dr. Stojmenovi´c has h-index 35 and >6000 citations. One of his articles was recognized as the Fast Breaking Paper, for October 2003 (the only one for the whole of computer science), by Thomson ISI Essential Science Indicators. He is the recipient of the Royal Society Research Merit Award, UK. He was elected to IEEE Fellow status (Communications Society, class of 2008), and is a recipient of Excellence in Research Award of the University of Ottawa, 2008–2009. He has chaired and/or organized >50 workshops and conferences, and served on over 100 program committees. Among others, he was/is program co/vice-chair at IEEE PIMRC 2008, IEEE AINA-07, IEEE MASS-04&07, EUC-05&08, WONS-05, MSN-05&06, ISPA-05&07, has founded workshop series at IEEE MASS, IEEE ICDCS and IEEE DCOSS, and been Workshop Chair at IEEE MASS-09, ACM Mobicom/Mobihoc-07 and Mobihoc-08.

Preface RFID networks are currently recognized as one a research area of priority. Research activities related to RFID technology have been booming recently. A number of ongoing projects are being funded in Europe, Asia, and North America. According to leading market analysts, the development of the RFID market is projected to increase from approximately $3 billion in 2005 to $25 billion in 2015. Several countries have dedicated innovation programs to support and develop RFID systems and related technologies: the RFID initiative in Taiwan, Ubiquitous Japan and the NSF SBIR program in the USA. The EU has recently advertised its Strategic Research Roadmap concerning the Internet of Things, which first of all refers to the RFID technology before being extended to communicating devices as in M2M (Machine to Machine). In this roadmap, several application domains have been identified: • • • • • • • • • • • • • • • • • •

Aerospace and aviation Automotive Telecommunications Intelligent buildings Medical technology, healthcare Independent living Pharmaceutical Retail, logistics, supply chain management Manufacturing, product lifecycle management Oil and gas Safety, security and privacy Environment monitoring People and goods transportation Food traceability Agriculture and breeding Media, entertainment and ticketing Insurance Recycling

The potential of RFID technology is huge. Contrary to popular belief, RFID technology is not recent and the delay in its deployment in commercial applications is not only due to its excessive cost. Ten years ago, standardization activities were insufficiently developed to allow the emergence of one standard which guarantees interoperability. In the meantime, ISO and worldwide organizations such as GS1 have proposed solutions, but new problems have arisen such as privacy issues and reading accuracy in proximity of certain materials

xx

Preface

such as water. The integration of RFID data in information systems is also a non-trivial problem. In the vision of the Internet of Things, future applications bring scalability and programmability issues. The book is intended to cover a wide range of recognized problems in RFID protocols and low-level research challenges, striking a balance between theoretical and practical coverage. The theoretical contributions are limited to the scenarios and solutions that are believed to have some practical relevance. This book is unique in addressing RFID protocols and communication issues in comprehensive manner. The book is divided into four parts. Part I provides an introduction and describes architectures of both passive UHF readers and tags. In addition, it defines performance metrics and introduces different classifications of RFID systems. Part II is related to networking protocols that involve one reader and multiple tags with the goal of resolving tag-to-tag interference. Tag identification protocols are covered in a systematic way. They include Aloha-based and tree-based protocols, which are the most popular. In addition tag-talks-first and tag-talks-only protocols are discussed and compared with reader-talksfirst protocols. Part III provides coverage of networking protocols that involve a host and multiple readers. First, the interface between the host and the readers is considered. Next, MAC layer solutions for reducing reader-to-tag interference are discussed. In addition, the redundant reader elimination problem and delay-tolerant networks are covered. In Part IV, several major research challenges in the RFID field are presented, such unsatisfactory read accuracy even in the most favorable RF environments, low read ranges, security problems, localization of tags, energy harvesting and simulators and emulators for RFID systems. Some of these challenges are so serious that they are preventing the widespread use of RFID technology (e.g. low read accuracy and security). Therefore, a number of these challenges and potential solutions are analyzed in this part of the book. At the end of most chapters, problems are presented and the solutions to some of the problems are provided on the book’s website http://www.wiley.com/go/bolic rfid. We believe that this book is an appropriate and timely forum, where industry, and academics from several different areas can learn more about the current trends in RFID networking and become aware of the protocols and current issues in RFID networks. It is well recognized that RFID technology will become a part of everyday life soon. Additionally, we believe that, given the huge interest in this topic shown by the industrial and academic worlds, this book can become a standard guide to modern RFID systems. Miodrag Boli´c University of Ottawa, Canada David Simplot-Ryl INRIA, France and University of Lille, France Ivan Stojmenovi´c University of Ottawa, Canada

Acknowledgements We would like to express our gratitude to the authors of book chapters who not only contributed a book chapter but also reviewed one additional chapter. In addition, we would like to thank a number of people who helped us review this book as shown below (the reviewers are not listed in any specific order). Gustaw Mazurek (Warsaw University of Technology), Daniel M. Dobkin (Enigmatics), Timo Kasper (Ruhr-Universit¨at Bochum), Venkatesh Sarangan (Oklahoma State University), Carlisle Adams (University of Ottawa), Zhang Xiong (Beihang University), Jeffrey S. Fu (Chang Gung University), Masahiro Miyakawa (Tsukuba University of Technology), Justin Wenck (University of California, Davis), Pradeep Shah (Texas MicroPower Inc.), Christoph Angerer (Vienna University of Technology), Seok Joong Hwang (Korea University), Ilker Onat (University of Ottawa), Md. Suruz Miah (University of Ottawa), Lin Wang (University of Pittsburgh), Fusheng Wang (Emory University), Junho Yeo (Daegu University), Stevan Preradovic (Monash University), Nicolas Pauvre (GS1 France), Mustapha Yagoub (University of Ottawa), Rony Amaya (Carleton University), Francesca Lonetti (ISTI-CNR), Francesca Martelli (Universit`a di Pisa), Gaetano Marrocco (Universit`a di Roma), Michael E. Knox (Mode1corp), Pankaj Mishra (University of Ottawa), Nemai Chandra Karmakar (Monash University), Zhou Yuan (Nanyang Technological University), Guan Yong Liang (Nanyang Technological University), Petar M. Djuric (Stony Brook University), Ali Miri (University of Ottawa), Mohamad Forouzanfar (University of Ottawa), Daniel Shapiro (University of Ottawa), Qinghan Xiao (Defence Research and Development Canada), Qiang Guan (Chinese Academy of Sciences), Bela Stantic (Griffith University), Xianjin Zhu (Stony Brook University) and many others. We greatly appreciate the support, guidance and encouragement given by Wiley’s team including Sarah Tilley, Anna Smart and Tiina Ruonamaa.

Part One Components of RFID Systems and Performance Metrics

1 Performance of Passive UHF RFID Systems in Practice Miodrag Boli´c1 , Akshay Athalye2 , and Tzu Hao Li1 1

School of Information Technology and Engineering, University of Ottawa, Canada

2 Astraion

LLC, NY, US

1.1 Introduction 1.1.1 Overview Radio Frequency Identification (RFID) is a technology that has risen to prominence over the past decade. The clear advantages of this technology over traditional identification methods, along with mandates from supply chain giants like Wal-Mart and the Department of Defense, led to a large number of research and commercialization efforts in the early 2000s. However, almost a decade on, the early promise of widespread, ubiquitous adoption of RFID is yet to materialize. This is due to a combination of several technical and commercial factors. The technical imperfections and shortcomings existing in presentday RFID systems pose a very significant obstacle to the widespread adoption of RFID. Overcoming some of these challenges would amount to a very significant step forward towards realizing the tremendous potential of RFID technology. This book describes the ongoing efforts of some of the leading researchers in the field towards tackling the most challenging issues in today’s RFID systems. With this in mind, the aim of this chapter is to clearly demonstrate, through experimentation, some of these technical challenges faced by RFID systems in practice. This chapter will enable the reader to better recognize the shortcomings of today’s RFID systems and will allow for a better understanding and appreciation of the research efforts described in the rest of the book. In this chapter, we focus on passive RFID systems operating in the Ultra High Frequency (UHF) band and adhering to the popular EPC Global Class 1 Generation 2 (Gen 2) RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

4

RFID Systems

standard [1]. We begin with the characterization of a hypothetical “ideal” RFID system. We then proceed to examine the performance of practical RFID systems through simple experiments and point out the non-idealities and problems that arise in practical systems. We begin this examination by considering a simple system involving a single stationary reader and a single stationary tag in free space. We then examine systems with increasing degrees of complexity with multiple (possibly mobile) readers and tags in more challenging deployment environments. As complexity of RFID systems increase, more problems (non-idealities) are observed in the performance while problems identified with simpler systems remain. We believe that the approach of analyzing RFID systems with an increasing degree of complexity and identifying challenges as they appear will give the reader a sound understanding of the challenges facing real-world RFID systems. Please note that this book chapter represents our viewpoint on imperfections of RFID systems. We have tried to point out some of the major issues in existing UHF RFID systems. This is not meant to be an exhaustive listing of all the possible challenges in practical UHF RFID systems, and there may be some problems and issues that have not been addressed here.

1.1.2 Background RFID is a wireless technology that allows for automated remote identification of objects [2]. The major components of an RFID system are tags or transponders that are affixed to objects of interest and readers or interrogators that communicate remotely with the tags to enable identification. RFID systems exist in various flavors that can be classified based on the frequency of operation, power source of the tag and the method of communication between the reader and the tags. A detailed classification of the commercial RFID systems based on the above criteria is presented in Chapter 2. In addition, the overview of RFID technology is presented in a number of publications including [3, 4]. In this introductory chapter, we focus on passive RFID systems operating in the 860–960 MHz band. Passive RFID tags draw the power required for operation from the radio wave transmitted to them by the reader and communicate with the reader by controlled reflection of a portion of this incident wave. This technique of communication by controlled reflection is referred to as backscatter modulation. Although this technique was used as early as World War II, RFID transponders were expensive, large devices that remained confined to military applications. However, the tremendous progress in VLSI technology along with the establishment of standards in the early 2000s, enabled RFID tags to be manufactured in high volumes resulting in a price point that initiated numerous commercial applications. The main goal of commercial RFID systems is to automate and enhance asset management by providing global asset visibility. This ability of RFID systems finds various applications in diverse fields such as supply chain management, indoor asset and personnel tracking, access control, robotics and many more. The immense commercial potential of RFID is mainly due to the numerous advantages that the technology possesses over traditional identification mechanisms such as barcodes. Some of these advantages are: (i) passive RFID tags can be read at much greater distances than barcodes; (ii) there is no need for a line of sight between the reader and tag; (iii) multiple tags can be read at much higher rates than barcodes; (iv) RFID tags have much larger memory than barcodes which allows storage of a lot more information than just the ID; and (v) the information contained in the RFID tag can be modified dynamically using the interrogator.

Performance of Passive UHF RFID Systems in Practice

5

As mentioned earlier, in order to harness the advantages of RFID technology to build viable commercial solutions, a number of technical challenges needs to be overcome. Some of these challenges are common to other wireless technologies while others are unique to the RFID system to hand. Each RFID technology, including passive, semipassive and active RFID systems operating at different frequencies, poses a unique set of challenges to obtain the desired performance. In addition, design requirements, performance specifications and protocols for active, passive and semi-passive systems are also very different. Therefore, in this chapter, we will limit our discussion to long-range passive backscatter-based UHF RFID systems operating in the 860–960 MHz band. In our opinion, this type of RFID has the most potential for significant commercial impact. As a result, it has seen the most research and standardization activity in recent years, more than other types of RFID systems. Today commercial systems of this type adhere to the EPC Global Class 1 Generation 2 (Gen 2) standard that has been in effect since 2005 [1]. Gen 2 compliant readers and tags are readily available in today’s market from several vendors all over the world including Alien Technology, Impinj Inc., Motorola and others. There have been several approaches in characterizing RFID systems. They are mainly based on (1) experimental characterization; and (2) mathematical modeling and simulation-based analysis. Experimentation is either performed in a controlled environment such as anechoic chamber, in the laboratory environment [5], [6], [7], [8] or in the application-specific setups such as conveyer applications [8]. An example of modeling and simulation-based analysis is performed in [9] where tag characteristics, propagation environment, and RFID reader parameters have been modeled and simulated. In this chapter, RFID systems are characterized through experimentation in an anechoic chamber and laboratory environment.

1.2 Ideal RFID System We begin our analysis of practical RFID systems by presenting the characteristics of a hypothetical ideal RFID system. Of course, like most other ideal systems, this RFID system would be unrealizable in practice. However, formulating such a system will give us a better understanding of the problems faced by real-world RFID systems. Once again, we point out that this ideal system is formulated in the context of UHF passive RFID systems. Since passive tags do not have a battery, they need to receive enough energy to turn on the tags’ integrated circuit. Therefore, in order for a passive RFID system to operate, the tag needs to receive enough power to wake up, and its backscattered response needs to be correctly received and decoded by the reader. In addition to this basic functionality, an RFID system has several other requirements for efficient operation that will be described later as desired features. The characteristics of an ideal RFID system, that mainly correspond to the basic functionality, can be summarized as follows: 1. There exists a well-defined, controllable read zone for each reader. For every tag within its read zone, each reader has a 100% read rate or read accuracy and for tags outside its read zone, each reader has a 0% read rate.1 1

For this chapter, we define read rate as the fraction of the number of times the reader is able to read a tag over the number of queries it sends to a tag. Please note that in RFID literature, the term read rate often refers to the speed at which the reader and tag communicate.

6

RFID Systems

2. 3. 4. 5.

Performance is insensitive to the physical orientation of tags. Performance is insensitive to the nature of the object on which the tag is placed. Performance is insensitive to the environment in which the system is deployed. Multiple tags communicate with the reader in a collision-free manner and the time for reading a fixed number of tags is a deterministic function of the number of tags while utilizing the maximum allowable bandwidth. 6. Performance is unaffected by the presence of multiple readers with overlapping read zones or of multiple tags within a read zone. 7. Performance is unaffected by relative motion between the readers and tags as long as the tags remain within the read zone of the reader. In the context of above characterization, read rate or read accuracy for each tag is defined as the percentage of times a reader is able to correctly read the tag’s ID. Hence for the ideal system if a reader sends out N queries to a population of tags that are all within its well-defined read zone, it receives N responses from each of the tags containing their respective IDs. Similarly if a reader sends out N queries to a population of tags that are all outside its read zone, it will receive no response. In the ideal system, this holds true for multiple readers with overlapping read zones, that is, if multiple readers send out N queries to a population of tags that are simultaneously in the read zone of all the querying readers, then each of the readers will receive N responses from each of the tags containing the respective tag IDs. As mentioned in the above characterization, the behavior of the ideal system is unaffected by factors such as orientation, environment and relative motion. Besides these ideal characteristics, there are several other desired features that researchers are trying to bring to practical UHF RFID systems. Some of these include: 1. 2. 3. 4. 5. 6.

High level of security of an RFID system. Localization of each tag within the read zone with high level of accuracy. Low cost of RFID components and high return of investment. Easy integration of RFID software into existing application software. Simple deployment and networking of multiple readers. Simple synchronization of multiple readers.

Items 4-6 are related mainly to complex RFID systems with a large number of readers. Although the ideal RFID system is unrealizable in practice, the above characterization provides a useful reference against which to measure various performance metrics of practical systems. Moreover, one can view most of the research efforts described in the rest of the book, whether at the physical, protocol or software level, as attempts to bring practical systems as close as possible to the aforementioned ideal or desirable RFID system. We now proceed with the analysis of practical systems by considering deployments with increasing levels of complexity, and pointing out the divergences from the ideal system that occur in practice.

Performance of Passive UHF RFID Systems in Practice

7

1.3 Practical RFID Systems 1.3.1 Complexity of RFID Systems Figure 1.1 shows simple block diagrams of several RFID systems. The complexity of the systems increases from left to right. Systems consisting of stationary readers and tags are shown in Figure 1.1(a), (b), (c), while the mobile systems are shown in Figure 1.1(d), (e). Stationary readers can be attached to walls, portal constructions or ceilings resulting in fixed deployments, while mobile readers are attached to moving objects such as forklifts, hand carts or are carried by people (handheld readers) and robots. Stationary tags are attached to the objects that usually do not move during the query round, for example, tags attached to shelved items. Mobile tags are attached to mobile objects or people and they move relative to the reader antenna during the query round. Large RFID systems like the one outlined in Figure 1.1(e) require the consideration of several other issues such as networking, synchronization, data processing software and middleware in order to enable an end user to reap the benefits of a deployed RFID system.

1.3.2 Single Reader, Single Tag The simplest practical RFID system from an analysis viewpoint is the one that consists of a single stationary reader and a single stationary tag (Figure 1.1(a)). Our setup for examining the performance of such a system consists of a Gen 2 compliant tag and a Increasing complexity

T T

R

T

R

T

Rd

Ri • Lack of well-defined read zone • Sensitive to • physical orientation of tags • nature of the object the tag is placed on • the environment

(a)

T

R

T R

T

R

T

T T

• Previously introduced non-idealities • Effect of collisions among tags • Sensitive to proximity of other tags

• Previously introduced non-idealities • Reader-to-reader interference • Reader-to-tag interference • Unwanted reads

• Previously introduced non-idealities • Missed tags • Increased level of interference

(b)

(c)

(d)

Stationary systems

T

T

T

R

R

T

T

T T

T

Complexity issues • Software integration • Networking • Deployment • Synchronization (e)

Mobile/Stationary systems

Figure 1.1 RFID systems of increasing complexity together with non-idealities encountered in each system with (a) single reader R single tag T system, (b) single reader multiple tag system, (c) multiple readers single tag system which includes interfering reader Ri and reader that communicates with the tag Rd , (d) mobile reader and multiple tags, (e) complex system that includes many readers and tags where readers are connected to the router and the computer.

8

RFID Systems

Gen 2 compliant reader. The Gen 2 protocol uses a dynamic frame slotted Aloha-based anticollision protocol that enables multiple readers to communicate with a single reader. In this protocol, the reader requests tags to reply to the reader commands in defined time slots. The reader specifies a fixed number of slots in a so-called Inventory Round or Query Round . An inventory round is defined as a single cycle of an algorithm by which a reader attempts to singulate the tags within its environment. Singulation is defined as a process of identifying a single tag and reading its ID number. A Query Round begins with a Query command which species a so-called Q parameter which indicates the number of slots in a query round. Each tag in a population then selects one random slot out of these slots to communicate with the reader. The reader then sends out successive Query Rep commands which designate the start of each slot. A reader may also send Query Adjust commands that dynamically increase or decrease the number of slots in the round. In its chosen slot, the tag replies with a 16-bit random number (RN16) using backscatter modulation. Upon successful reception of this RN16, the reader sends out an Acknowledge command with the same RN16 back to the tag. If this number matches the number that the tag originally sent out, the tag backscatters a Protocol Control (PC) header, followed by its EPC ID and a 16 bit CRC. A more detailed explanation of the Gen 2 standard is given in Chapter 5. The description of the Aloha-based anticollision algorithm used is presented in Chapter 7 of this book. In our experimental setup, the Gen 2 reader is attached to a host computer that is capable of monitoring the read rate of a detected tag, that is, the ratio of the number of times a tag responded successfully to the number of Query Rounds sent out by the reader. We are using commercial UHF Gen 2 passive dipole tag. The reader transmits at a power of 30 dBm over a 6 dBil gain circularly polarized antenna. In addition we deploy a sniffer device in the proximity of the tag so as to examine the actual communication happening over the wireless channel. This sniffer device is connected to an oscilloscope that stores the snapshots of the baseband signals going between the tag and the reader. Figure 1.2 shows the captured waveform for single reader, single tag scenario. Figure 1.2 shows the Query Rep commands that a reader sends out designating the tag communication slots. In its selected slot, the tag backscatters an RN16 as shown in the Figure 1.2. This is then followed by transmission of the Acknowledge command by the reader and the subsequent backscattering of the EPC ID by the tag. Upon examination of the performance, we see that even this simple system differs from the ideal system described in Section 1.2 in several ways. The deviations are presented in Figure 1.1(a) and analyzed below. Lack of Well-defined Read Zone: The practical single reader-single tag system does not have a specific read zone with respect to the reader antenna wherein the tag exhibits a 100% read rate and outside which the tag exhibits a 0% read rate. It has been shown in several independent works that, in the case of passive RFID systems, such as the one currently under consideration, the read range depends mostly on the power in the forward link which is needed in order to power tags IC [4, 5, 8, 9]. In free space, as a tag moves away from a reader, the mean value of the power it receives drops off as per the Friis equation describing the wireless link. As this power drops off, so does the tag’s read rate until the tag reaches a place where it is unable to receive sufficient power and the read rate drops to zero. Figure 1.3 shows the read rate of a single tag as a function of the

Performance of Passive UHF RFID Systems in Practice

Reader Transmission :

Query Adjust

Query Repeat

Tag Backscatter:

Figure 1.2

9

ACK + Tag RN16

RN16

Query Repeat

PC + EPC + CRC16

Communication between a single Gen 2 reader and single Gen 2 tag.

1

Read Rate (Tag Counts/Queries)

0.8

0.6

Trail #1 Trail #2

0.4

0.2

0 0

0.5

1

1.5

2

2.5

3

3.5

4

Distance (m)

Figure 1.3 Read rate for a single tag in an anechoic chamber.

4.5

5

10

RFID Systems

distance from the reader. Two curves on the graph are two different trials with the same setup. The experiment is carried out in an anechoic chamber (dimensions of chamber are: length 5 m, width 5 m and height 2 m). For this experiment, the tag is placed in the best possible orientation with respect to the reader antenna. By performing the experiment in the anechoic chamber and by fixing the orientation of the tag and reader antennas, we wanted to avoid influence of other parameters to read rate besides the distance. The experiment was started by placing the tag at a distance of 1 meter from the reader antenna. A total of 500 Query rounds were sent by the reader, and the number of responses from the tag was noted. The steps were repeated while increasing the distance between the tag and the reader antenna in 0.5 meter steps in the range between 2 m and 4 m where the read rate was maximum, and in 0.1 m steps in the zone when read rates start to drop. As seen from Figure 1.3, even in an anechoic chamber, there exists a gray area around the reader antenna wherein a tag may or may not be read in a particular query round. Thus, in the practical system it is not possible to define a clear read zone as described for the ideal system. This is because of the inherent properties of electromagnetic radiation which is the basis of the communication between the reader and the tag. Sensitivity to tag orientation: The relative orientations of the tag and the reader significantly affect the performance of a practical system. Figure 1.4 shows the effect that the tag orientation has on the read performance even when the tag is within the read range of the reader antenna. In order to collect the data we performed the experiments in an anechoic chamber with the similar setup as presented above. In this experiment,

1

0.8 q = 0, f = 0 Read Rate (Tag Count/Queries)

q = 30, f = 0 q = 60, f = 0

0.6

X

0.4

f Y

0.2 q

Z

0

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

Distance (m)

Figure 1.4 Read rate for a single tag with different orientations in an anechoic chamber.

Performance of Passive UHF RFID Systems in Practice

11

the reader is fixed at the position (x, y, z) = (0, 0, 0) and tag is moved in 0.5 m steps in y-direction. Measurements are repeated three times: each time a tag was oriented differently relative to the orientation of the reader antenna. Read rates are recorded in the same way as in the previous experiment and presented in Figure 1.4. From Figure 1.4, it is obvious that the read range drops significantly when the orientation becomes less favorable for the tag because the tag is not able to collect enough energy to turn on the IC. In addition, Figure 1.4 demonstrates that the read rates depend on the relative orientation of the tag and reader antennas and on the distance between tag and reader. The problem of orientation sensitivity can be handled by innovative tag antenna designs involving multiple dipoles or monopoles. In fact, ensuring orientation insensitivity is one of the most important goals in designing antennas for tags. The latest research efforts in designing efficient tag antennas are presented in Chapter 3. Experimentation with different tag orientation are published in several different works including [10]. Please note that experimentation of dependency of orientation to read rate in [10] is performed with EPCGlobal Class 1 tags (generation before Gen 2). Sensitivity to deployment environment: The performance of a practical system is highly dependent upon the environment in which the system is deployed. Like any other wireless system, the nature of the environment affects the multipath and fading properties of the channel. This effect is even more pronounced in RFID systems due to the passive nature of the tag operation and the inherently low signal to noise ratio (SNR) of the weak backscatter signal. In order to examine the effect of environment on performance, we repeated the experiments mentioned previously in the significantly cluttered environment of a normal computer lab room (dimensions of lab room are: length 10 m, width 6 m and height 4 m). Figure 1.5 shows the read rate performance in the cluttered environment of a computer lab. As we can see, the deployment environment hampers the read performance of the system and also introduces some blind spots/null spots due to multipath interference and channel fading. In comparison to the read rates in the anechoic chamber, it is obvious that even for a fixed relative orientation of the reader and tags antenna it is difficult to specify the range in which the read rate is maximum. It is clear from the above-mentioned experiments, that the read rate and read range of a tag are not solely dependent on the distance between the reader and tag, but are also affected by factors such as orientation and environment. A similar inference has been drawn in [9] and [11] wherein the authors suggested defining read range as the range in which a pre-defined read rate or accuracy of reading a tag can be achieved. The reading accuracy problem is presented in Chapter 14 of this book together with the directions for improving reading accuracy. Sensitivity to the nature of the object on which the tag is placed: It is known that UHF RFID systems do not perform well when attached to objects that contain metals and fluids. These materials not only attenuate the signal when placed between the tag and the reader, but also result in detuning of the tag antenna. Detuning occurs if the materials are in close proximity to the tag. The effect of metal and water on read range and read accuracy was analyzed in several publications including [5]. Read ranges were reduced up to three times in proximity of water and metal in comparison with the read range in free space.

12

RFID Systems

1

Read Rate (Tag Counts/Queries)

0.8

q = 0, f = 0 q = 30, f = 0 q = 60, f = 0

0.6

X

0.4

f Y

0.2

q

Z

0 0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

Distance (m)

Figure 1.5

Read rate for a single tag with different orientations in a computer lab room.

Approaches to handling this problem are based on modification of tag antennas to work specifically in the proximity of metals and liquids [12]. Design of such tags is described in Chapter 3 as well. We will now move on to more complicated practical scenarios involving multiple readers and tags. Note again, that all the non-idealities and problems identified with the simpler systems are carried forward to the more complicated systems.

1.3.3 Single Reader, Multiple Tags As mentioned earlier, Gen 2 tag anticollision protocol is based on a slotted Aloha wherein multiple tags can communicate with the reader in separate time slots. The following nonidealities can be identified in the practical situations based on an examining performance. Collision between tags: In the practical system, unlike the ideal system, collisions can occur between tags that are trying to communicate with the reader within the same query round. In the context of Gen 2, a collision occurs when two tags select the same slot number to backscatter. In this case, the reader is often unable to decipher tag transmissions and the communication attempt is unsuccessful. Figure 1.6 shows a Query round involving 2 tags where frequent collisions occur.

Performance of Passive UHF RFID Systems in Practice

13

Collision

Query

Query Adjust Query Rep

Figure 1.6

Query Adjust

Query Adjust

Query Rep

Collisions in Gen 2 query round.

Tag collision adds a significant overhead to the time needed to read a population of tags. This non-ideality is further exacerbated in scenarios involving a large number of tags simultaneously in the field of view of the reader as is the case in many commercial applications of RFID. However, these collisions are result of the air protocol selected in Gen 2 standard which is based on the Aloha paradigm. The standard does not specify implementation of the algorithm for anticollision and only gives recommendations. Chapter 7 describes different slotted Aloha anticollision algorithms with the emphasis on the methods for dynamic estimation of the number of tags to be singulated. The time to query population of tags can be minimized if the number of tags in the population is known or if it is correctly estimated. Effect of tags in proximity to each other: In an ideal RFID system, neighboring tags will have no effect on each other’s performance. In reality, however, this is not the case. In most practical applications, RFID tags are placed on objects that are densely co-located. Hence it is very important to understand the effect that the tags have on each other. The effect of proximity of the other tags to the read range and read accuracy is experimentally analyzed in [11] and [4].

14

RFID Systems

(a) Case 1 Z

#5

#4

#3

#2

Y

#1

Reader Antenna

Cardboard X

2m (b) Case 2 Z

#1 #2 #3 #4 #5

Y

Reader Antenna X Cardboard

Figure 1.7 Experimental setup for examining effect of tag proximity on performance. Tags are spaced 1 cm apart and parallel to one another and (a) orthogonal to the reader antenna, (b) parallel with the reader antenna.

The presence of multiple tag antennas in close range alters the current distribution, the radiation pattern and introduces mutual impedance. As a result, tags in close proximity tend to cast a shadowing effects on the neighboring tags [4]. This shadowing effect increases as the number of tags increases and as the inter-tag distance decreases. The effect causes tags to receive less power than when no other tags are present in the proximity. This leads to a drop in the read rate or read accuracy of the tag. In order to examine the effects of tag proximity, we use an experimental setup consisting of a Gen 2 reader with a circularly polarized patch antenna with a 6 dBil gain and five Gen 2 tags with dipole antennas. The tags are placed 2 m away from reader antenna and the reader output power is set to 23 dBm. All tags are placed in the same plane on a single cardboard platform with the best possible orientation angle to the reader. This is done so as to eliminate the influence of orientation sensitivity on the measurements. The experimental setup is shown in Figure 1.7. We consider three cases; in Case 1, tags are spaced 1 cm apart and parallel to one another. The cardboard platform is placed along the X –Y plane and the reader antenna is placed along the X –Z plane. For Case 2, we place the cardboard platform along the X –Z plane while keeping the position of the reader antenna unchanged. Case 3 is similar to Case 1, but we increase the inter-tag spacing to 30 cm. The read rate results are shown in Table 1.1. Case 3 shows that when the tags

Performance of Passive UHF RFID Systems in Practice

15

Table 1.1 Effect of tags in proximity on tag read rate. Read Rate (#reads / #queries) Tag Tag Tag Tag Tag

1 2 3 4 5

Case 1

Case 2

Case 3

0.98 0.98 0.56 0.38 0.11

0.45 0.096 0.02 0.178 0.87

0.98 0.98 0.97 0.98 0.98

are spaced far enough apart, all of them have a pretty good read rate. Case 1 and Case 2 results clearly demonstrate the significant impact of tag proximity on the read rates of a tag population. This problem becomes very important in practical applications involving a large number of densely packed objects, each having a separate RFID tag.

1.3.4 Multiple Readers, Single or Multiple Tags In this subsection we will consider a situation in which there are multiple stationary readers and one or more stationary tags. Having multiple readers introduces diversity and redundancy which help in solving some of the previously recognized issues. If multiple readers are used to read the same tag, the chance that the tag will be in the blind spot for all the readers is reduced. Redundant readers or one reader with multiple antennas are commonly used in industry to improve the accuracy of reading. Localization of tags might also be improved by using multiple readers. Coarse-grained localization information, based on association of tags with the reader that detected it, will be improved if tags are read by two or more readers placed at known location. The tag would then be located in the space that represents the intersection of the read ranges of these readers. However, the presence of multiple readers in an environment gives rise to the very serious problem of interference caused by their simultaneous operation. We examine the effect of this problem on system performance below: Interference: A system with two readers and one tag is shown in Figure 1.1(c). In this figure, communication between desired reader Rd and tag is affected by the signal sent by the interfering reader Ri . If multiple readers attempt to singulate tags simultaneously, two types of interferences might occur: reader-to-reader and reader-to-tag interference. Readerto-reader interference or reader jamming occurs when the interfering reader affects the reception of the tag signal by the desired reader. Reader-to-tag interference or tag jamming can occur when the interfering reader affect tags reception of the signal from the desired reader. We examine the performance of interfering readers through the following experiment. Two Gen 2 readers from different vendors were programmed to work in a dense reader mode. Dense reader mode is proposed in Gen 2 standard [1] to support large-scale enterprise applications with many readers. Spectral allocation is defined so that reader-to-reader interference is reduced. The reader transmits at a power of 30 dBm over a 6 dBil gain

16

RFID Systems

Table 1.2

Effect of multiple reader interference on tag read rate.

Reader 1 Mode M= OFF M= M= M=

4, BLF = 256 kHz 4, BLF = 256 8, BLF = 256 0, BLF = 640

Reader 2 Mode OFF M= M= M= M=

4, 4, 4, 4,

BLF BLF BLF BLF

= = = =

200 kHz 200 200 200

Reader 1 read rate

Reader 2 read rate

1 – 0.355 0.372 0.261

– 1 0.39 0.38 0.35

circularly polarized antenna. We are using a Gen 2 dual-monopole tag that is orientation insensitive. The tag is placed in between the readers at a distance of 1.5 m from each reader. Read rates of each reader are recorded when the other reader is off as well as when both readers operate simultaneously. As we see from Table 1.2, the read rate of both readers dropped significantly when both of them are querying simultaneously. When both readers send their commands at same time, the tag cannot decode the collided reader signals. The more readers broadcast at same time, the harder it is to read passive tag. There are multiple challenges in resolving interference problems that are unique to RFID systems [13]. Passive UHF tags have limited tuning capacity and their reception will be affected by signals from multiple readers even if the readers operate at different frequency channels. In addition, readers transfer high amounts of power that is needed to power up the tag. This high power worsens the interference problems. Reader-to-reader interference problem is addressed in the Gen 2 standard by introducing the dense reader mode. In this mode, lower data rates are specified, and Miller Subcarrier encoding is used so that the channel for reader and tag transmission can be well defined and separated. Dense reader mode is also analyzed in Chapter 5 of this book. For regions with only narrow frequency band available in UHF band such as Europe, the listen before talk approach together with applying spectral mask constraints are viable solutions [1, 14, 15]. Even though readers transmit in separate frequency channels, they can still cause readerto-tag interference. In [16], it has been shown that the tag might function even if there are interfering readers. If desired and interfering readers transmit simultaneously, tags will be able to detect the signal from the desired reader if that signal is much stronger than the signal from the interfering reader. Experimentation was performed at the frequency of 866 MHz. Tags can detect the signal from the desired reader if the following conditions are met: (1) the signal power from the desired reader needs to be 6 dB higher than the signal power from the interfering reader when the difference in reader’s carrier frequencies is more than 800 kHz; and (2) the signal power from the desired reader needs to be 13–20 dB higher than the signal power from the interfering reader when the difference in readers carrier frequencies is less than 800 kHz [16].

1.3.5 Mobile Readers and/or Mobile Tags Both reader and tags can be mobile. Figure 1.1(d) shows an example in which the reader is mobile and the tags are stationary. This corresponds, for example, to the situation when

Performance of Passive UHF RFID Systems in Practice

17

the reader is attached to a forklift and the tags are placed on items that are on the shelves. In mobile systems, it will not happen that the tag remains in the blind spot. Due to the mobility of tags and/or readers, the reader will eventually be able to read the tag. While mobility can aid in handling some of the identified issues, it also introduces new problems into the system. Centralized algorithms for reducing interference might fail if readers are mobile and some tags might not be read because they appear in front of the reader for the limited amount of time. Missed tags: Let us consider the case in which tags are mobile and the reader is stationary. As pointed out in [17], RFID tag antennas designed for stationary applications operate reliably when tags move at low speed. For example, a speed of about 16 km/h is considered low enough for reliable operation. In [18] the effect of accuracy is examined when ten single-dipole tags are moved at the speed of 1 m/s and were kept at 1 m from the reader. Different relative orientations between tags are explored. It was determined that the minimum distance between tags should be 4 cm in order to be able to detect all the tags. Other experimentation is described in [19], where an RFID reader is placed on the side of the conveyer belt whose speed is 2 m/s and tags are attached on boxes placed on the conveyer belt. The system is first calibrated to avoid unwanted reads. Excellent read rate is achieved even in cases when 50 tags are placed on the same box and the speed of the conveyer belt is 2 m/s and when the reader is 44 cm away from the tags. However, the high reading rate in both papers might be due to the fact that the reader was placed very close to the tags. Simulations with detection of high speed objects are described in Chapter 9 of this book. It has been shown that it is possible to read four tags at a distance of 4 m at speeds of 250 km/h with appropriate setting of parameters of Gen 2 protocol. It also shows that much better performance for high speed tags can be achieved using tag talk first protocols. The experimentation with mobile tags attached to a vehicle is performed in [20]. In the experiment, six tags are attached to the windshield and two antenna are placed at the height of 4.2m with an angle of 45 degrees. High read rate is achieved at the speed of 10 km/h, while very low read rates are achieved for speeds above 70 km/h. The method for predicting the read rate is presented in [20] and it relies on the support vector machine model. Mobile readers are used in combination with stationary tags in several different applications for localization, mapping and navigation of robots and people. In almost all applications stationary tags are used as landmark tags to assist in positioning of the mobile readers and their carriers. In [21] it was also confirmed that the major problem is how one can cope with low reading accuracy. If the landmark tag is not read, the position cannot be determined. One possible solution, self-localization of robots that takes into account unreliable RFID readings, is proposed in [21]. There, the robot first collects information about the read landmark tags and the number of their detection during the training phase. This information is stored and used to compare with the real measurement results during operation of the robot.

1.3.6 Large Deployments Including Many Readers and Tags Figure 1.1(e) shows an RFID system consisting of a large number of readers and tags. The figure shows RFID system with four readers of which three are stationary and connected

18

RFID Systems

using Ethernet to the switch. One reader is mobile and it communicates with the switch using a wireless link. The switch is connected to the host that runs middleware software. The role of the middleware is to aggregate the data from the deployed readers, filter and process it as per case-specific requirements and present it to the higher level software. In addition to the physical non-idealities that we have described, this large deployment of multiple readers and tags gives rise to a whole new set of challenges related to the middleware, application software and network management. Synchronization of many readers: In large deployments, implementing the synchronization scheme is a significant challenge. Many algorithms have been developed to reduce interference in large-scale deployments (Chapter 11 of this book). Centralized synchronization is impractical for large-scale solutions and should be avoided [22]. In addition, in multivendor environments, such as shopping centers, receiving dock doors might belong to different organizations and synchronization is again impractical [15]. In these situations, readers will need to be able to operate asynchronously of each other and be truly event driven [15]. Easy integration of RFID software into existing application software: Full benefit of an RFID system can only be utilized if data collected from the RFID system is used for decision-making [23]. Therefore, the RFID reader system needs to be integrated into the existing application software. RFID middleware software is used to provide, besides other functions, easy integration with legacy applications. If the applications have proprietary interface, then the integration becomes complicated [23]. RFID Middleware as a part of EPCGlobal architecture is described in Chapter 6. The architecture of RFID system is presented in Chapter 10. Installation and tuning: In deployments with a large number of readers, installation of the RFID components can be complicated. It requires installation of power connections and network cabling. In addition, readers and antennas have to deployed properly to reduce interference among the readers and other wireless devices or machineries to the RFID networks. Improper deployment can result in some areas being not properly covered by RFID readers. Therefore, the RFID system needs to be tuned by the experienced hardware engineer [23] to assure that required performance is achieved.

1.3.7 Other Desired Features of Practical RFID Systems As mentioned earlier, in addition to the basic operational characteristics embodied by the ideal RFID system, practical systems have some other features that are necessary in enabling commercial applications. We describe a couple of these below. Privacy and security in Gen 2: Privacy and security aspects are extremely important for successful deployment and application of RFID systems. A large number of publications have recently appeared that deal with improving or developing security and authentication frameworks for RFID applications [24]. RFID tags based on Gen 2 standard are considered to belong to low-end systems based on their capacity to implement schemes for security. Current security and privacy

Performance of Passive UHF RFID Systems in Practice

19

mechanisms in the Gen 2 standard, although better than previous standards, are still considered weak (Chapter 16 of this book). In the Gen 2 standard, the reader does not transmit the EPC code, parts of the tag memory can be locked, a kill command can permanently disable the tag. Functions such as access to special memory, ability to modify tags and the ability to kill tags require a 32-bit access password. However, the 32-bit access password can be eavesdropped and then easily computed. More advanced security and privacy methods are described in Chapters 16 and 17 of this book. As is pointed out in Chapter 2 of this book, adding security features to passive tags would increase the complexity of the tag’s hardware and then likely reduce the tag’s range and throughput performance. This tradeoff makes the task of designing security mechanisms for passive RFID systems all the more challenging. Reducing ambiguity in tag location: Unlike an ideal system, the practical reader-tag RFID system is unable to determine the precise location of the tag within the read zone of the reader. As mentioned earlier, one of the main goals of an RFID system is to provide global asset visibility. In most practical applications, precise asset location is an important attribute of asset visibility in order to automate the asset management process. The application of traditional ranging techniques such as those based on received signal strength (RSS), time of arrivals (TOA), time difference of arrivals (TDOA) and phase of arrivals (POA) to passive RFID is very challenging due to the weak backscatter signal used to communicate back to the reader. Localization techniques such as trilateration rely on precision of ranging techniques. This is the reason why many RFID localization solutions are based on using landmark tags or some kind of map matching. Precise localization using passive RFID is an important problem that is the focus of a number of research efforts today. The overview of localization algorithms is presented in Chapter 2 and detailed analysis of algorithms for ranging and localization is presented in Chapter 15. Reducing the unwanted reads: The fact that the reading zone is not properly defined causes tags to be read by multiple readers. In some applications, this is not desired. For example, in dock door application, each reader covers a different door. It is important to detect which door the pallet went through. If the tags on the pallets are read by multiple readers, this will introduce ambiguity in associating pallets with corresponding dock doors. Reading of the tags by the readers that cover neighboring doors are called unwanted reads. Other names used in practice are unintended reads and cross reads [25]. Solutions against unwanted reads are based on both hardware (shielding) and software (filtering of data).

1.4 Overview of the Book This chapter presented problems and non-idealities in current passive UHF RFID systems. The rest of the book describes approaches to deal with the defined problems. In addition, the book tackles a number of other issues that have not been described in this chapter. Therefore, we summarize major problems, methods and approaches covered in the book. Chapters 3–5 deal with design of tag antennas, tag and reader circuit for passive RFID systems operating in UHF band. Chapter 3 introduces the design of antennas for passive UHF tags. The chapter also describes the effects of environmental factors including proximity of dielectric materials and metals on performance of the antennas. Then, several

20

RFID Systems

approaches for designing antennas that operate in proximity of metals have been proposed. In Chapter 4, tag’s architecture is presented. RF, baseband and memory design are detailed. Low power design issues are pointed out. Chapter 5 focuses on the reader design issues that are unique to RFID systems including handling transmitter leakage during tag reception, frequency generation, transmit linearity and transmit AM noise. Chapters 6 and 10 cover EPCGlobal architecture framework and protocols. Chapter 10 analyzes services that the RFID system supports and defines the system architectures. The chapter then analyzes a number of EPCGlobal specifications that standardize interfaces to configuration, monitoring and data processing services. Chapter 6 describes RFID middleware in details. It focuses on the following aspects: reader management, data management and storing and sharing data in RFID networks. Non-idealities of the RFID systems are tackled throughout the book. The problem of read range is considered in Chapters 2 and 14. The parameters that affect read range in a RFID system are pointed out and ways to achieve longer reading ranges are introduced. Parameters that affect the throughput of RFID systems are considered in Chapter 2. Reading accuracy is explained in more detail in Chapter 14. Localization of passive and active tags is the application that is attracting increasing interest. The localization problem is introduced in Chapter 2 and elaborated in Chapter 15. Privacy and security aspects of RFID networks are considered in Chapters 16 and 17. Chapter 16 focuses on low level approaches and covers attacks against common RFID components and lightweight cryptography. Chapter 17 describes higher level aspects of security and privacy including description of the cryptographic protocols for privacy, identification with privacy, and authentication without privacy. Even though the book mainly deals with passive RFID system, active RFID systems are also described in Chapter 2. Some localization solutions based on active RFID systems are mentioned in Chapter 15. Dual frequency active RFID systems are presented in Chapter 18. Chapter 18 is dedicated to novel technologies and it is especially focused on energy harvesting. Energy harvesting is also briefly discussed in Chapter 19. In Chapter 19, a methodology that allows translation of the models of RFID system automatically onto a hardware platform to enable real-time verification in the target operating environment is presented. A detailed description of the simulator and the emulator is provided. Tag identification protocols are presented in Chapters 7, 8 and 9. Reader talk first protocols including Aloha and tree-based protocols are described in Chapters 7 and 8 respectively, while tag-talk-first approach is described in Chapter 9. Combinations of tree-based and Aloha-based algorithms are described in Chapter 7. In general, the number of tags to be singulated is not known. For both tree- and Aloha-based protocol, knowledge of the number of tags would reduce the average time it takes for the protocol to singulate all the tags. Therefore, in Chapters 7 and 8 ways to estimate number of tags while the algorithm is running are described. Tag talk first protocols are analyzed in Chapter 9. Detailed comparisons with Aloha-based protocols have been performed regarding throughput and speed of reading. In the near future, RFID deployments with large number of readers will be common. In these applications, reducing interference among the readers and performing synchronization of the readers will become a major problem. Detailed comparison of reader anticollision algorithms is presented in Chapter 11. Extension of the tree based protocols

Performance of Passive UHF RFID Systems in Practice

21

for multiple collaborative readers is described in Chapter 8. An example of the reader anticollision algorithm is also presented in Chapter 12. With improving technology and dropping price of RFID readers, dense reader deployments will soon become reality. In addition to improving accuracy, densely deployed readers can be used, for example, to estimate more accurately the location of the tagged items. On the other hand, in order to reduce overall power consumption in the network and reduce amount of interference, it is important to turn off redundant readers when they are not needed. A redundant reader can be safely turned off or removed from the network without affecting the number of tags covered. Detecting redundant readers is a complex problem that is described in Chapter 12. Several other networking problems are considered in the book. They include a problem when readers that are connected wirelessly and communicate using multi-hop communication need to optimally report detected tags. The solution to eliminate redundant tag reports generated by multiple readers is provided in Chapter 12. Delay-tolerant mobile networks of RFID readers are introduced in Chapter 13. In these networks information among stationary readers is transferred using mobile tags. The many challenges that this type of network introduces are presented in Chapter 13.

1.5 Conclusion Although radio frequency identification is a rapidly emerging technology, several technical challenges need be to overcome to enable its ubiquitous adoption. In this chapter, we have examined some of these important technical challenges in present-day UHF passive RFID systems. We did this by formulating a hypothetical ideal RFID system which exhibits optimal performance in various aspects. We then examined the performance characteristics of practical RFID systems and how they diverge from the ideal system. Overcoming these non-idealities will be the key in enabling RFID technology to achieve its immense potential. We hope that this chapter has given the reader a better understanding of what ails the RFID systems of today. With this understanding, the reader can better appreciate the research efforts being described in the rest of the book and how each of these efforts fits into the bigger picture of enabling ubiquitous adoption of UHF RFID systems.

Acknowledgements We would like to thank Alexey Borisenko for performing some of the experiments. We would also like to thank Dr. Michael Knox and Dr. Mustapha Yagoub for giving constructive comments that helped improve this manuscript.

References [1] EPCglobal Inc. (2008) EPC Radio Frequency Identification protocols class-1 generation-2 UHF RFID protocol for communications at 860 mhz–960 mhz, Standard Specification version 1.2.0. [2] Wyld, D. C. RFID (2005): The right frequency for the government. A research monograph from the IBM Center for the Business of Government. [3] Want, R. (2006) An introduction to RFID technology, IEEE Pervasive Computing, 5(9): 25–33. [4] Dobkin, D. M. (2007) The RF in RFID: Passive UHF RFID in Practice. Oxford: Elsevier-Newnes.

22

RFID Systems

[5] Derbek, V., Steger, C. Weiss, R. Preishuber-Pflugl, J. and Pistauer, M. (2007) A UHF RFID measurement and evaluation test system, Electrotechnic and Informationstechnik , 124(11), 384– 390. [6] Buettner, M. and Wetherall, D. (2008) “An emperical study of UHF RFID systems,” in MobiCom, San Francisco, California, USA. [7] Muhlmann, U. and Witschnig, H. (2007) “Hard to read tags”: an application-specific experimental study in passive UHF RFID systems, Electrotechnic and Informationstechnik , 124(11). 391– 396. [8] Ramakrishnan, K. N. (2005) Performance benchmarks for passive UHF RFID tags, M.S. thesis, University of Kansas. [9] Nikitin, P. and Rao, V. (2006), Performance limitations of UHF RFID systems, in IEEE Antennas and Propagation Symposium, pp. 1011– 1014. [10] D’Mello, S., Mathews, E., McCauley, L. and Markham, J. (2008) Impact of position and orientation of RFID tags on real time asset tracking in a supply chain, Journal of Theoretical and Applied Electronic Commerce Research, 3(1), 1–12. [11] Currie, I. A. and Marina, M. K. (2008) Experimental evaluation of read performance for RFID-based mobile sensor data gathering applications, in Proceedings of the 7th International Conference on Mobile and Ubiquitous Multimedia, Umea, Sweden, pp. 92–95. [12] Mitsugi, J. and Hada, H. (2006) Experimental study on UHF passive RFID readibility degradation, in Proceedings of the International Symposium on Applications and the Internet. [13] Impinj. Inc, (2007) RFID communication and interference, White Paper, Grand Prix Application Series. [14] Leong, K. S., Ng, M. L. and Cole, P. H. (2006) Positioning analysis of multiple antennas in dense reader environment, in Applications and the Internet Workshops. [15] Tuner, C. The dense reader problem in Europe, White Paper, November 2005. [16] Martinez, R. Interference in RFID systems, Presentation for SG3. [17] Rao, K. V. S., Nikitin, P. V. and Lam, S. F. M. Antenna design for UHF RFID tags: A review and a practical application, IEEE Transactions on Antennas and Propagation, 53(12). [18] Rahmati, A., Zhong, L., Hiltunen, M. and Jana, R. (2005) Reliability techniques for RFID-based object tracking applications, in 37th Annual IEEE/IFIP International Conference on Dependable Systems and Networks, pp. 113– 118. [19] Ren, Z., Tan, C. C., Wang, D. and Li, Q. (2009) Experimental study of mobile RFID performance, in International Conference on Wireless Algorithms Systems and Applications, Boston, MA, pp. 12–20. [20] Jo, M., Youn, H. Y., Cha, S.-H. and Choo, H. (2009) Mobile RFID tag detection influence factors and prediction of tag detectability, IEEE Sensor Journal , 9(2): 112–119. [21] Schneegans, S. Vorst, P. and Zell, A. (2007) Using RFID snapshots for mobile robot self-localization, in Proceedings of the 3rd European Conference on Mobile Robots (ECMR 2007), Freiburg, Germany, pp. 241–246. [22] Tanaka, T. and Sasase, I. (2007) Interference avoidance algorithms for passive RFID systems using contention-based transmit abortion, IEICE Transactions on Communications, E90-B(11): 3170– 3180. [23] Bhuptani, M. and Moradpour, S. (2005) RFID Field Guide: Deploying Radio Frequency Identification Systems, Sun Microsystems Press, Prentice Hall. [24] Ahson, S. A. and Ilyas, M. Eds., (2008) RFID Handbook: Applications, Technology, Security and Privacy. Boca Raton, FL: CRC Press. [25] Krishna, P. and Husalc, D. (2007) RFID infrastructure, IEEE Communications Magazine, 45(9): 4–10.

2 Performance Metrics and Operational Parameters of RFID Systems Raj Bridelall1 and Abhiman Hande2 1

Axcess International, Inc.

2 Texas

Micropower, Inc.

2.1 Overview Automatic Identification and Data Capture (AIDC) technologies come in a wide variety of functionality, all targeted towards quickly linking a physical item with associated data contained within an information-technology (IT) system. Unlike other forms of AIDC technology such as magnetic stripe and barcodes, end-users often select RFID technology for its transparent ability to automatically locate and monitor the condition of physical assets and personnel using minimal or no manual intervention. Tagged items are generally associated with high value, high liability, or both. Examples of high value assets include laboratory equipment, medical instrumentation, pre-fabricated construction material, heavy-duty tools, and controlled medical substances. Personnel tracking applications include unattended contractor time and attendance logging for automatic invoicing, locating miners and first responders for safety, and the automatic accounting of chemical and nuclear plant employee whereabouts during an emergency. Organizations spend a large sum of money to deploy RFID technology because they expect significant improvements in operational efficiency, personnel safety, and exposure to liability. It is widely understood that a supplier’s failure to map the true capabilities of the technology to actual application requirements can lead to failed pilots and significant economic losses. Over a period of several decades, innovators have introduced numerous RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

24

RFID Systems

types of homogeneous RFID technologies to address a broad range of requirements for unattended data capture across various physical asset tracking and personnel location applications. However, it has become evident after a period of trial and error that not all RFID technology types can be successfully configured for a given application, nor does a single technology type exist that can be applied across all application categories. This chapter examines key operational parameters that make up the various technology categories and how these map to both performance and application requirements. The second section examines the technical trade-off required to achieve the desired range, throughput, omnidirectionality, localization, environmental compatibility, security, and standards compliance. Parameters that affect operating distance are covered in detail, including the amount of power that the interrogator radiates, interrogator and tag sensitivity and others. Next, system throughput is evaluated by analyzing relevant parameters such as the bandwidth occupied, data rate, modulation scheme, signal to noise ratio and channel sharing mechanisms. In addition, interrogator and tag interferences are discussed briefly. The major application of RFID systems is to detect the presence of tagged objects and/or people. Another important application is to provide the location of the tagged objects or people. Discussion on localization accuracy starts with the basic algorithms for estimating positions followed by the impact of multi-path and omni-directionality on localization accuracy. Impact of materials such as water or metal within proximity of interrogators and tags are examined. Other factors considered include reliability of RFID systems, size and thickness of RFID tags, health and safety aspects of RFID systems, security, and total cost of ownership. In the third section, several classifications of RFID systems are discussed. Based on the type of EM link formed with the interrogator, we distinguish between near-field versus far-field RFID systems. Commercial near-field RFID systems that operate at low, high and ultra-high frequencies are described and compared. Then, basic features of narrow-band, wide-band and ultra wide-band far-field RFID systems are presented. Next, classification is considered based on the way the tags obtain their power. We then describe basic characteristics of active, passive and semi-passive RFID systems. The final section presents concluding remarks and some research directions.

2.2 Key Operational Parameters This section will identify the key operational parameters that affect RFID performance in terms of range and throughput. Later sub-sections will also touch on factors that influence design choices for size, scalability and security. The key parameters that affect range and data rate are interdependent as shown in the dependency graph of Figure 2.1. In this figure, the parameters and decision choices that the designers/users of RFID system have control over are shown within the hexagons. These parameters include operating frequency, transmit power, bandwidth, digital modulation encoding, and maximum tolerable bit error rate (BER). The key operational parameters that are affected by those design parameters include operating distance (range) and system throughput (proportional to data rate). These operational parameters are shown within the rectangles of Figure 2.1. The oval objects in Figure 2.1 are intermediate factors that are influenced by one or more design parameters. They include signal-to-noise ratio (SNR), noise level, receiver sensitivity, and noise figure (NF). For example, the amount of “noise”

Performance Metrics and Operational Parameters of RFID Systems

Digital Modulation

25

BER

Data Rate SNR

Range

TX Power & Frequency

Noise

Bandwidth

RX Sensitivity

NF

Semiconductor Technology

Figure 2.1 Key operational parameters and design trade-off.

at the input of a receiver is strongly dependent on the “bandwidth” of the receiver. Noise figure is a measure of degradation of the signal to noise ratio, caused by components in the RF signal chain. NF depends on the type of semiconductor technology and determines how much the noise at the input of the receiver is amplified relative to the signal before arriving at the demodulator. We would like to point out that some parameters and some possible dependencies are omitted in Figure 2.1 for reasons of clarity. The specific parameter combination and their values ultimately determine the type of tag. Although there are a large number of possible parameter permutations, the type of tag can be categorized by the type of communications link (near-field or far-field), the method of tag transmission (emission or reflection), and the type of power supply (battery or energy harvesting). The third section covers the benefits and deficiencies of each tag type. Two of the most widely deployed tag types for long-distance operation are passive and active operating in the ultra-high frequency (UHF) bands. The impact from adjusting any of the key operational parameters of just these two tag types adequately illustrates RFID system behaviour without much loss of generality. The far-field begins where the wave impedance quickly settles towards the free-space impedance value of 377 ohms [1]. For electrically small antennas, this distance is λ (2.1) 2π This distance is essentially the radius of the near-field region around an antenna, and is less than one foot for popular commercial UHF RFID systems that operate in the 915 MHz band. Passive tags transmit data by reflecting power from the interrogator. This is also referred to as backscatter modulation for systems that operate in the far-field, and load modulation for systems that operate in the near-field. Passive tags harvest energy from the interrogator instead of using a battery. Active UHF tags, on the other hand, utilize an on-board battery for the power needed to emit a modulated RF signal that encodes the data. RNF =

26

RFID Systems

The most commonly utilized frequencies for both active and passive RFID operation are at or near 125 kHz, 13.65 MHz, 315 MHz, 433.92 MHz, 915 MHz, 2.45 GHz, and 5.8 GHz. Systems operating in bands below 70 MHz tend to operate in the near-field with inductively coupled coil wire antennas because far-field antennas at those frequencies would be impractically large. The majority of standards prescribe passive UHF RFID operation in the region between 860 MHz and 960 MHz because of its favorable output power and RF energy harvesting characteristics [2]. Unfortunately, this is the only spectral region without uniformly allocated frequency bands around the world. Consequently, international standards are divided between the European allocations from 865–868 MHz and the North American allocations from 902–928 MHz [3, 4]. Regulations outside of Europe and the Americas have also opted for a blend of the European and American regulatory allowances. Figure 2.2 shows the distribution of countries that follow the FCC allocations for Effective Isotropically Radiated Power (EIRP), and countries that follow the EU allocations for Effective Radiated Power (ERP). Countries that adopt a blend of the FCC and EU regulations are shown in detail, and the EIRP amount is shown unless otherwise indicated as an ERP amount. The difference between EIRP and ERP amounts relates to the type of antenna assumed, and this is further addressed in a later section. Figure 2.2 also indicates those countries that use Frequency Hopping Spread Spectrum (FHSS), and whether the specification is for indoor (InDr) or outdoor (OutDr) environment, and if the band is licensed (Lic). These allocations are still evolving as of the time of this writing. Nevertheless, regardless of the operating frequency selected, performance will be constrained by the amount of signal power allowed, the amount of noise or interference in the system, and the available bandwidth.

2.2.1 Operating Distance Operating distance is defined as a maximum distance between the interrogator’s antenna and the tag at which the interrogator can reliably read from or write to the tag. In passive RFID systems, maximum operating distances for reading and writing operations are not always the same. The reason is that when writing to its memory, most tags require more power and, therefore, the operating distance is reduced. The maximum distance for reading operation is called read range while the corresponding distance for the writing operation is called write range. All things being equal, the greater the power that the interrogator is allowed to radiate, the greater the distance at which a tag will be able to receive and interpret commands. Reciprocally, the more power that a tag can direct towards the interrogator, the greater the distance at which the interrogator will be able to receive and interpret tag responses. Power is often listed in decibels (dB) of milliwatts (mW) or dBm, and generally includes the power gain of the antenna supplied with the unit. Receiver sensitivity and interference are the next dominant parameters that affect range. However, these tend to be a strong function of bandwidth and another section addresses this relationship. 2.2.1.1 Radiated Power in the Americas Most interrogators designed for sale in FCC regulated environments can radiate up to four Watts of power in the 915 MHz band. This includes the antenna gain and any losses

928.00 867.00 920.00 869.00 928.00 925.00 869.00 925.00 869.00 955.00 925.00

8%

37%

8%

29%

EN 302-208 865.6–867.6 MHz 2W ERP

1.00 ERP FHSS InDr 0.50 ERP 0.50 ERP 0.50 ERP 0.50 ERP FHSS OutDr 0.50 ERP FHSS-NoLic 0.50 ERP 0.50 No Lic 0.05 0.02 LBT 0.00

9%

9%

India Israel Belgium China-A China-B Hong Kong A Iran Macedonia Malaysia B Singapore B Vietnam B

Australia Brazil-A Brazil-B Hong Kong B Japan Lic. Korea A Korea B New Zealand South Africa B South Africa C Thailand A 867.00 917.00 867.60 844.50 924.50 868.00 868.00 867.60 923.00 925.00 925.00

End 926.00 907.50 928.00 925.00 954.00 910.00 914.00 868.00 919.00 921.00 925.00

Similar Range to Europe

865.00 915.00 865.60 840.50 920.50 865.00 865.00 865.60 919.00 920.00 920.00

Start 920.00 902.00 915.00 920.00 952.00 908.50 910.00 864.00 915.40 919.20 920.00

Similar Range to FCC

4.00 ERP 2.00 2.00 ERP 2.00 ERP 2.00 ERP 2.00 ERP 2.00 ERP 2.00 ERP 2.00 ERP 2.00 ERP 2.00 ERP Lic.

Watts Sharing 4.00 4.00 4.00 4.00 4.00 LBT 4.00 LBT 4.00 FHSS 4.00 4.00 FHSS 4.00 CW 4.00 FHSS-Lic.

Worldwide regulatory allowances for UHF RFID operation as of January 2009.

Lowest Performance Regions

922.00 865.60 918.00 866.00 922.00 920.00 866.00 920.00 866.00 952.00 923.00

Figure 2.2

Taiwan A Armenia Philippines Singapore A Taiwan B Thailand B Vietnam A Vietnam C Malaysia A Japan UnLic. Indonesia

Rest of World Allocations Pending

FCC CFR 47 Part 15.247 902-928 MHz, 4W, EIRP

Performance Metrics and Operational Parameters of RFID Systems 27

28

RFID Systems

in the system. Specifically, the FCC limits the power delivered into the antenna to 1 Watt and the antenna gain to 6 dBi. The norm for antenna gain specification is with respect to an Effective Isotropic Radiated Power (EIRP) as denoted by the small ‘i’ next to the ‘dB’ unit. The isotropic radiator is a theoretical antenna that radiates with zero dB gain in all directions. Such an antenna is, however, not physically possible. In practice, adding antenna gain adds directionality to the radiated pattern. For example, a theoretical halfwave dipole antenna has a gain of about 2.15 dBi [1]. The FCC also allows for antenna gains beyond 6 dBi in the 915 MHz band, but with a corresponding reduction of power to the antenna so that the radiated power in the direction of maximum gain does not exceed 4 Watts. If a circularly polarized antenna is used, then the antenna gain can actually double. For example, commercially available systems deliver either one Watt into a 6 dBi linearly polarized antenna or 1 Watt into a 9 dBi circularly polarized antenna. This is possible because the FCC measures antenna gain with respect to a linearly polarized receiving element that effectively captures half (−3 dB) of the incident circularly polarized power. 2.2.1.2 Radiated Power in Europe The European specifications cite the maximum output power in terms of the Effective Radiated Power (ERP), which is different from EIRP. The ERP specification is with reference to a theoretical half-wave dipole antenna. As an example, the European Union (EU) standard EN302-208 establishes the maximum radiated power at 2 Watts, ERP [4]. System manufacturers interpret this as 2 Watts radiated by a theoretical half-wave dipole in the direction of highest gain. Figure 2.3 illustrates the difference between these two interpretations. European regulations for UHF RFID specify 15 channels of 200 kHz bandwidth each, but only ten of those allow a maximum of 2 Watts ERP [4]. With everything else being equal, this power level limits the maximum unobstructed theoretical read range to nearly two-thirds that of systems operating within FCC regulations. A lower UHF operating frequency, nevertheless, is preferred because of the more favorable energy capture characteristics of a longer wavelength. The Friis far-field transmission formula for unobstructed

Dipole Orientation

Theoretical Isotropic Radiator Theoretical Half-Wave Dipole

Figure 2.3

Radiated power relative to EIRP and ERP.

Performance Metrics and Operational Parameters of RFID Systems

narrow-band propagation shows this effect [5].   λ 2 T ψri Pi_tag = Ps 4πr

29

(2.2)

That is, the amount of signal power that the tag receives, Pi_tag is proportional to the wavelength λ squared. Ps is the power supplied by the interrogator, r is the operating distance, T is the interrogator antenna gain, and ψri is the tag’s antenna sub-system gain when it is absorbing energy. 2.2.1.3 Radiated Power in the Rest of the World Adoption trends show that most countries are either following UHF RFID regulations similar to the Americas, the European Union, or a blend of both. Countries that cannot allow as much power in the UHF frequency bands may achieve at least equivalent or better performance with active RFID even when the permitted power levels are only a few microwatts. For example, technology providers have successfully deployed approved products that transmit no more than 2 microwatts in the 315 MHz and 433 MHz bands to achieve performance significantly exceeding that of passive systems operating at the maximum power levels in the 915 MHz FCC bands [6]. 2.2.1.4 Interrogator and Tag Sensitivity Receiver sensitivity is the minimum signal power that is required to produce a specified output signal. The interrogator and tag receivers will decode a command if the average required signal power is greater than their respective receiver sensitivity levels. The amount by which this signal power must exceed the receiver input noise power is the minimum signal-to-noise ratio (SNR) required to recover data of a specified digital modulation scheme and maximum tolerable bit error rate (BER). Digital data communications systems typically accommodate a BER of one bit per million received (10−6 ). Applications such as digitized voice transmissions can tolerate a higher BER. In general, a higher SNR reduces the BER. However, increasing the SNR for a specified range implies increasing the transmitted signal power or reducing the receiver input noise power. Figure 2.1 illustrates the various dependencies on SNR. Constraints on Passive Tag Sensitivity The lower the power a tag requires for activation and signal decoding, the greater its operating distance from the interrogator. Figure 2.4 illustrates this relationship for a selected sub-set of system types that operate under different regulatory allocations. Figure 2.4 shows, for example, that an EU compliant tag operating at 868 MHz that requires 50 microwatts to energize and decode a signal will respond from an unobstructed distance of about 3.25 meters, under ideal conditions. The same tag can respond from an unobstructed distance of about 5 meters if the tag sensitivity improves to 20 microwatts. Signal power diminishes as the inverse square of the distance (1/r2 ) in the far-field, and as the inverse sixth power of the distance (1/r6 ) in the near-field [5]. Therefore, near-field tags will operate at a significantly reduced distance from far-field tags of the same sensitivity.

30

RFID Systems

1 106

Passive Tag Sensitivity (µW)

1 105 1 104

EU: 869 MHz

FCC: 915 MHz

1 103 50

100 10 1

EU: 2.45 GHz

EU: 13.56 MHz 0.1 0.01

0

1

2

3

4

5

Operating Range (Meters)

Figure 2.4

Effect of sensitivity on range.

Constraints on Active Tag Sensitivity Technology providers optimize passive tags for RF energy harvesting. Therefore, their receivers are inherently less sensitive than those of active tags. Passive tag receivers typically consist of signal envelope detectors constructed from passive diodes or an equivalent construction [7]. They detect a signal only when the input signal is sufficiently strong to overcome the passive diode detector forward bias threshold. Active tags utilize the onboard battery to bias active rectification circuits, which substantially lowers this forward bias threshold. Consequently, the sensitivity difference can be significantly greater than 100 dB [7]. Constraints on Interrogator Sensitivity for Passive UHF Tags The interrogator signal received from a backscattering passive UHF tag decreases with free-space distance as follows [5]:   λ 4 2 PB = Ps T ψro (2.3) 4πr That is, the amount of signal power PB received by the interrogator decreases at the rate of 1/r4 for passive UHF systems. Ps is the power generated from the interrogator and delivered to the antenna, r is the operating range, T is the interrogator antenna gain, and ψro is the tag antenna subsystem gain when it is reflecting energy, including any transmission losses. In an optimized design, the interrogator decoding sensitivity should be at least PB when the tag is at its maximum operating distance. Solving for r in Equation 2.2 and setting Ptag = Ptag_sens gives the maximum powering range rmax as a function of the tag’s decoding sensitivity,  λ Ps T ψri rmax = (2.4) 4π Ptag_sens

Performance Metrics and Operational Parameters of RFID Systems

31

Substituting Equation 2.4 into Equation 2.3 and setting PB = PI_sens to maintain maximum range gives the required interrogator sensitivity as a function of the tag sensitivity, PI _sens

2 Ptag_sens ψro = Ps ψri2

(2.5)

Intuitively, for fixed tag sensitivity, the greater the power (Ps ψri ) transferred to the passive tag, the further out it can respond with a valid signal, and the weaker the signal that the interrogator must accommodate. In addition, a lower tag antenna gain in reflection mode, ψro will weaken the reflected signal that the interrogator must accommodate. This relationship is independent of the interrogator antenna gain (T ) as anticipated from the RF reciprocity principle that balances the outbound gain with an identical inbound gain. As a final insight, the required interrogator sensitivity under ideal propagation circumstances is at most twice that of the passive tag’s sensitivity on a logarithmic scale because of the round trip that the same signal must travel. Constraints on Interrogator Sensitivity for Active UHF Tags Since active UHF tags transmit by generating a signal rather than backscattering, the interrogator will receive signal power,   λ 2 PI = Ps_tag  T ψr (2.6) 4πr Even though interrogators can radiate a relatively large signal for powering passive tags, the available backscatter signal (PB ) from passive tags is still orders of magnitude less than a low-power signal (PI ) arriving from a typical active tag from the same distance. For example, from equation 2.2, a four Watt EIRP (∼36 dBm) signal transmitted from an FCC compliant interrogator operating at 915 MHz will become approximately 3 microwatts (−25.27 dBm) after traveling a distance of 30 meters in free space. A battery-powered backscatter tag will reflect a portion of the three microwatts towards the interrogator. In comparison, a low power active tag in the same band will typically transmit 1 milliwatt (0 dBm) of signal, which is over 300 times more powerful than the signal backscattered from a passive tag. This means that an interrogator with the same sensitivity can decode the √ Active tag’s response from a distance that is a factor of approximately 17 (that is 300) further away. With all other parameters unchanged, increasing range means either increasing the average signal power transmitted or increasing the receiver sensitivity. However, regulatory rules limit the transmitted power, and the receiver noise figure bounds its sensitivity. The theoretical noise floor is a function of bandwidth and temperature. For example, at room temperature, the theoretical minimum noise spectral density is −174 dBm/Hz [8]. Other Design Considerations The operating distance depends on many other factors including: the environment, the orientation of the antennas, the directionality of the antennas, the modulation scheme, the type of RFID system used, the antenna gains, and many other factors. Some of the key environmental factors include the proximity of metal to the tag or reader antennas, the presence of in-band noise sources, and operation indoors or outdoors. Relative antenna

32

RFID Systems

orientations will affect the amount of received power incident at the tag and interrogator antennas. For a given radiated power, greater operating distance can be achieved with directional antennas; however, this would limit readable locations. The type of RFID system used significantly affects operating distance. For instance, mobile interrogators are designed to have relatively small form factor and long battery life. Therefore, large operating distance is not generally available with hand-held and mobile devices. The modulation scheme can affect the operating distance since different modulation types provide different levels of noise resistance. For practical implementations, transmission power, noise figure, and bandwidth affect key application level considerations such as power consumption, cost, and size. Increasing antenna gain in order to increase the received signal strength tends to increase the size of the antenna. Higher antenna gain also favors transmission or reception in one direction over others. On the other hand, lower gain antennas must be used if the application requires omni-directional performance. A lower noise figure amplifier increases sensitivity but also tends to demand more power or require a more expensive semi-conductor technology [9]. A higher output power interrogator amplifier uses more power and tends to require additional heat sinking elements that also increase size and bulk. This trade-off between range, bandwidth, power consumption, cost, and size are some of the most important RFID systems design considerations.

2.2.2 System Throughput Given a data exchange protocol, the amount of bandwidth per channel that an interrogator can legally transmit is directly proportional to the maximum achievable data rate of a specified modulation scheme. Interrogators select an available channel from the sub-divided frequency band by utilizing traditional narrow-band multiple access techniques such as frequency hopping or carrier sensing. Therefore system throughput is evaluated by analyzing relevant parameters such as the bandwidth occupied, data rate, modulation scheme, signal to noise ratio and channel sharing mechanisms. Although interrogators can simultaneously operate in separate channels, increasing the number of available channels does not necessarily lead to an overall increase in system throughput. Interrogator interference and Tag interference are two phenomena that account for this barrier as described in Sections 2.2.2.6 and 2.2.2.7 respectively. 2.2.2.1 Occupied Bandwidth Once the interrogator selects a channel, it may modulate its carrier in order to send data to the tag. Carrier modulation generates a power spectral density (PSD) that the regulatory bodies measure. The PSD is essentially an electromagnetic (EM) energy footprint that the interrogator creates across the occupied channel. The interrogator generally centers the carrier within the channel and the data energy is distributed on either side of the carrier. For a selected modulation scheme, higher data rates result in spreading the data energy further away from the carrier and potentially into adjacent channels. Therefore, the specified channel bandwidth ultimately constrains the data rate. The majority of low-power RFID systems utilize a form of ASK modulation for both the forward and return links. Wireless network standards such as IEEE 802.11b (Wi-Fi)

Performance Metrics and Operational Parameters of RFID Systems

200 kHz

33

200 kHz

0 dBc 500 kHz FCC 15.247 −20 dBc

−30 dBc

200 kHz EN 302 – 208

−36 dBm

Fcarrier − 200 kHz

fcarrier

Fcarrier + 200 kHz

Figure 2.5 Comparison of FCC and EU spectral masks.

and IEEE 802.15.4 use other digital modulation schemes such as FSK, PSK, and QPSK to transmit a higher data rate within the same channel bandwidth. In general, more complex modulation schemes require higher SNR, as well as increased power demands from the transceiver architecture. The simplicity of ASK modulation is often the primary driver for minimizing cost, power consumption, battery size and overall tag footprint. Each standard specifies a spectral mask within which the transmitted signal must be contained. Figure 2.5 compares the UHF RFID spectral masks of the FCC and EU regulatory domains. The main lobe of the EU mask is not only 200 kHz narrower, but also imposes a more stringent limit on sidebands and spurious emissions outside of the band. Unlike the FCC, the EU standards also regulate the backscatter modulation from tags. Figure 2.6 shows the backscatter spectral mask prescribed by the EN 302-208 standard. The center frequency of the interrogator carrier f c and its value depend on the channel that the interrogator selects for transmitting. The modulated backscatter energy is constrained relative to this center frequency. Some exceptions are possible, as noted in Figure 2.6, where the standard limits any spectral response above 870 MHz to −36 dBm regardless of the channel the interrogator selects. 2.2.2.2 Protocol Throughput Throughput in terms of the number of tags read per second depends on many factors including: • • • •

the the the the

data rate of the forward link (interrogator commands to the tag); data rate of the return link (tag responses to the interrogator); Media Access Control (MAC) protocol; operating environment.

The air interface and MAC protocols of the selected standard determines the number of bits transferred in each direction before a single tag read can be completed. Some protocols

34

RFID Systems

200 kHz

−15 dBm −27 dBm −36 dBm −47 dBm −54 dBm

863 MHz 862 MHz

fc − 300 kHz fc fc + 300 kHz fc − 600 kHz fc + 600 kHz

870 MHz

Figure 2.6 EN 302-208 backscatter spectral mask.

do not require a forward link message in order to read tags and result in significantly faster throughput. However, the trade-off is an inability to command tags to configure their behavior in real-time before they automatically respond, or to write data to their on-board memories. Standards often refer to these unidirectional protocols as Tag-talkfirst (TTF). Reader-talk-first (RTF) protocols are more prevalent, however, because of the added degree of control over the link characteristics. For example, in the popular EPC Class I Generation 2 (EPC C1G2) or the ISO 18000-6c standards, the interrogator initiates the data transfer protocol and configures tags to respond at data rates between 40 kbps and 640 kbps [10]. This provides for some degree of adaptability depending on the EM properties of the application environment. The MAC protocol used affects how collisions in tag-to-reader communications are handled. The RF environment can also affect the protocol throughput since the number of retransmissions is partially dependent on the environmental conditions. 2.2.2.3 Modulation Scheme The achievable bit rate is a direct function of bandwidth occupied, bit encoding scheme, and SNR [11]. Once allotted a fixed bandwidth, the maximum bit rate achievable depends on the spectral efficiency of the selected bit-encoding scheme. For example, non-coherent ASK modulation requires a bandwidth that is twice the bit rate and is often described as having a spectral efficiency of 0.5 bits-per-second-per Hertz (bits/sec/Hz). Similarly, binary phase shift keyed modulation (BPSK) theoretically provides up to 1 bit/sec/Hz, whereas quadrature phase shift keyed modulation theoretically provides up to 2 bits/sec/Hz [11]. The actual data rate and BER achieved depends on the available SNR. Nearly all passive and most active RFID technologies utilize non-coherent ASK modulation and demodulation because their implementation inherently requires low-power and is low-cost. It is also possible to utilize a sub-carrier modulation type such as FSK and PSK

Performance Metrics and Operational Parameters of RFID Systems

35

1

Bit Error Probability

0.1 0.01 1 10−3 1 10−4 1 10−5 1 10−6 1 10−7 1 10−8 0

2

4

6

8

10

12

14

16

Eb/No(dB)

Figure 2.7

Signal quality for non-coherent ASK demodulation.

along with ASK. For example, FSK sub-carrier modulation is achieved by manipulating the ASK modulation speed in a data stream to encode binary digits. The bit error probability (pe ) for non-coherent ASK demodulation is 



1 1 Eb (2.7) pe = e 2 N 0 2 E b is the energy per bit in Joules, and N o is the single-sided noise power spectral density in Watts per Hertz [11]. This is plotted in Figure 2.7 where it is observed that slightly more than 14 dB of bit energy per unit of noise power spectral density is required to achieve a lower bit error rate than one bit per million received. In general, the greater the spectral efficiency of the modulation scheme, the greater the SNR required for an acceptable bit error probability. 2.2.2.4 Signal to Noise Ratio The maximum achievable bit rate for a UHF backscatter system is     λ 4 1 Rbit (r, pe ) = Ps T2 ro × × × M(r, Ro , NB ) × 4πr kB T0 fr

1 

1 2 ln 2pe

 (2.8)

The parameters kB and To are the reference temperature and Boltzmann constants respectively [5]. The signal attenuation factor M (r, Ro , NB ) accounts for non-line-of-sight conditions using a model for signal propagation where M(r, Ro , NB ) = 

1 2(NB −2)

r 1+ R0

(2.9)

The parameters Ro and NB are the free-space equivalent breakpoint distance and the environmental attenuation factors respectively. For free space, NB = 2 and the factor

36

RFID Systems

becomes unity. A receiver manufacturer typically specifies the amplifier’s noise factor, which is fr =

SNRin SNRout

(2.10)

This is the same as a noise figure which is typically specified with a dB value. From Equations 2.3 and 2.8, the SNR can be written as     PB M(r, Ro ) 1 SNR = = Rbit × 2 ln (2.11) kB T0 fr 2pe Solving for Eb /No in Equation 2.7 and substituting the result into Equation 2.11 gives the minimum required SNR as  Eb  SNRmin = Rbit × (2.12) N0 pe That is, the minimum SNR needed to decode a backscattered signal with maximum tolerated probability of bit error (pe ) is directly proportional to the bit rate. Equation 2.8 proves that a higher SNR is required if the application requires more range (r), a lower bit error probability (pe ), or a higher throughput from increasing the bit rate (Rbit ). 2.2.2.5 Channel Sharing Interrogators designed for use with passive tags must transmit sufficient power to energize tags and consequently do not utilize multiple access techniques such as Direct-SequenceSpread-Spectrum (DSSS) or Ultra-Wide-Band (UWB) that spread energy across a wide spectral region using relatively short time period signals. Rather, multiple access techniques that require channel hopping are used because the tag can collect energy from a non-modulated carrier using a simple diode rectifier front-end. Consequently, interrogators for passive tags are limited to time-sharing or activity-based media access (MAC) protocols. They either randomly select a channel and occupy it for a fixed duration before moving on to another, or wait until a channel becomes vacant before transmitting. Standards have dubbed the former method as frequency hopping (FH) and the latter a Listen-Before-Talk (LBT). FH with a randomized hopping pattern is a requirement for UHF transmission levels greater than 125 milliwatts under FCC regulated domains. The intention is to spread the power averaged across the band over a specified duration. FCC compliant UHF interrogators that transmit the maximum allowed power must frequency hop randomly across 50 of the 52 channels allocated within the 902 MHz to 928 MHz band. Each channel is 500 kHz wide with spectral mask shown in Figure 2.5. Interrogators must also hop at a rate such that the average time transmitting in any one channel is less than 400 milliseconds over any 20-second period. The European regulations prescribe an LBT/FH channel sharing method since there are far fewer channels for FHSS to be effective. Figure 2.8 illustrates the European channelization for UHF RFID. The EN 302-208 standard currently defines channel occupancy as any carrier signal that can be detected above −96 dBm. The link margin for an EU compliant interrogator transmitting a signal at full power (2 Watts ERP) is

Performance Metrics and Operational Parameters of RFID Systems

37

2W

500 mW 100 mW

865.0 MHz

865.6 MHz

Figure 2.8

867.6 868.0 MHz MHz

EU channelization under EN 302-208.

33 dBm – (−96 dBm) = 129 dB. From the Friis equation, this equates to several miles. Therefore, it is possible for EU compliant systems to become sluggish as the number of interrogators within a radius of several miles exceeds the number of available channels. A typical warehouse can utilize dozens of interrogators, including those fixed at dock doors, conveyor belts, and storage rooms, as well as mobile and hand-held interrogators brought within close proximity to other interrogators. Therefore, it is likely that the EU regulations will continue to undergo modifications to mitigate such constraints. 2.2.2.6 Interrogator Interference When interrogators operate simultaneously within a few channels of each other, they produce interfering signals that affect the operation of both tags and interrogators within range. The impact on broadband backscatter tags is more significant than for narrow-band active tags. Another section of this chapter covers the performance impact from this type of tag interference phenomenon, which is different from MAC-related tag interference also known as tag collision. The mixing of a carrier from the desired interrogator with a carrier from another nearby interrogator causes interference. Nonlinearities from practical RF amplifiers and mixers produce signal components called inter-modulation products (IM) that appear within the receiver bandwidth and can distort the tag data signal. The relative frequency position of the individual spectral energy components received are illustrated in Figure 2.9, namely the spectrum of the carrier signal plus the backscattered data (grey arrow and lines), the

Carrier

Backscatter Data

CW Interferer

IM Products

Figure 2.9 Interrogator interference.

38

RFID Systems

continuous waveform (CW) interferer (dotted arrow), and multiple IM products components (black arrows). Demodulators can recover the data signal contained within the upper side band or the lower sideband of the carrier, but not when unwanted signal components overlap the data sidebands as shown. The C1G2 standard defines a “dense reader environment” as one where as many interrogators are operating as there are channels allocated in the band [10]. The net effect from any interference is reduced system throughput. That is, adding a second interrogator does not necessarily double the throughput of tags identified per unit interval of time. 2.2.2.7 Tag Interference Increasing the number of available channels in a band may increase aggregate throughput for active tags but not necessarily for passive tags. The reason for this is that passive tags have fixed broadband receivers that listen to the entire band. Unlike active tags, passive tags cannot dynamically hop to another vacant channel to establish an isolated link with an interrogator. They simultaneously receive and integrate signals within range from all interrogators transmitting anywhere in the band. Figure 2.10 illustrates how multiple interrogator signals can add destructively at the tag to distort the received signal. The figure shows interference between the modulated and unmodulated signals in two different channels. Other equipment that shares the band such as cordless phones and wireless local area networks can also add to this distortion. Under certain circumstances, adding a low pass filter to the tag’s baseband receiver circuit may partially mitigate this issue [12]. Demodulated data signal

(a) Carrier modulated signal received without interferer

(b) Interfering carrier in another channel Demodulated data signal

(c) Corrupted data signal

Figure 2.10

Signal from tag interference.

Performance Metrics and Operational Parameters of RFID Systems

39

Table 2.1 Common RTLS techniques for indoor positioning. Number of antennas 1 2 3 or more

Location method

Positional Information

Proximity (near-field) Proximity (far-field) Triangulation Trilateration

In or out of a magnetic field zone Radial proximity to antenna A 2D position between antennas A position in 3D space

2.2.3 Localization In addition to automatic identification, advanced RFID implementations can also provide information about a tagged item’s position, speed, and direction of travel. These Real-Time Location System (RTLS) techniques range from simple proximity to computational positioning using a local or a global positioning system (GPS). Table 2.1 summarizes common techniques for indoor positioning solutions. Localization accuracy depends on operational parameters as well as the effectiveness of the signal processing algorithms used. In this section, we first describe several basic algorithms for localization. They include proximity detection (which only determines that the tag within the reading range of the reader), as well as algorithms for determining 2D and 3D position of the tagged objects. These algorithms rely on estimating the range between the interrogator’s antenna and the tag. Range estimation algorithms calculate the range by correlating the distance with either the time of flight (TOF) of a signal, the Received Signal Strength (RSS) or with the ratio of phase versus frequency variations. The ranging accuracy is affected by many factors. Multi-path conditions depend on the indoor layout structure, line-of-sight conditions, mobility of objects, height of the tag’s antenna and others. Next, better time resolution (or proportionally larger available bandwidth) results in better ranging accuracy. Hence, in the UHF range, higher localization accuracy is expected in North America than in Europe due to wider available bandwidth. The level of SNR is important since low levels of SNR can affect the accuracy of the signal processing algorithm applied for range estimation. Tag antenna types as well as their orientation are important factors in determining localization accuracy. RTLS usually requires tags that are orientation insensitive. Proximity of metals and liquids can cause detuning which will degrade either powering range for passive RFID systems or SNR for active RFID systems, which then affects RSS and Angle-of-Arrival (AoA) estimates. 2.2.3.1 Proximity Presence determination at specified spatial regions may be the lowest cost RTLS method because fewer antennas are required, and the amount of computation is limited to determining whether or not the tag is within range of the antenna coverage pattern. Near-field antennas provide the highest degree of control when the coverage area must be relatively small and non-overlapping. These include transitional and restricted areas such as doorways and work stations. Far-field antennas cover large areas with less precision because of signal reflections. They are adequate when the zones are large and far enough apart from each other, for example, detached buildings across a campus.

40

RFID Systems

The near-field zone is also called a control point when the presence of a tag in the near-field is utilized to control an apparatus such as an automatic gate motor, a door latch, a camera, a buzzer, or some other actuating device. Control points are less reliable with far-field signaling because of signal reflections from multi-path propagation. 2.2.3.2 Positional Computation Triangulation and trilateration techniques are used to determine 2D or 3D position of a stationary tag using stationary interrogators. These techniques require additional signal processing, computation, and data exchange between interrogators in order to determine the velocity of a tagged item. Trilateration algorithms rely on an estimate of the radial distance between the tag and the interrogator. RSS and TOF algorithms can provide a distance estimate with varying degrees of accuracy depending on multi-path conditions, bandwidth, and SNR [13]. Triangulation algorithms determine a tag’s velocity by successively computing the AoA from the average RSS. Triangulation Method Given two antennas positioned a known distance D apart as shown in Figure 2.11, with each having sub-systems capable of estimating the tag’s AoA, the perpendicular distance to the tag r can be estimated from trigonometry using the law of sines [14], rˆ = D

ˆ sin(α) ˆ sin(β) ˆ sin(αˆ + β)

(2.13)

The accuracy of this estimate depends upon the SNR, multi-path signal propagation conditions, and the accuracy with which each interrogator is capable of estimating the two angles of arrival, α and β. interrogators can also conceivably estimate the AoA with phased-array antennas capable of adaptively steering the beam to search for the maximum likelihood source direction [16].

Tag

Region Of Uncertainty

rms Angle error

r

a RX1

b D

Figure 2.11 Triangulation.

RX2

Performance Metrics and Operational Parameters of RFID Systems

41

(x1,y1) (x2,y2)

d1 d2 Region Of Uncertainty

(x,y)

d3 rms

(x3,y3)

range error

Figure 2.12

Trilateration.

Trilateration Method With three or more antennas, it is possible to approximate the 3D coordinates of the tagged item. Each antenna sub-system first estimates the tag’s radial distance and then sends this data to a common computer that can solve for the intersecting area illustrated in Figure 2.12. This technique requires that every antenna cover the same target area. The position estimate of (x, y, z) is determined from estimates of each radial distance {d1 , d2 , d3 } and the known coordinates of each fixed position antenna (xn , yn , zn ). The uncertainty of each distance estimate translates into a tag position uncertainty. As an example, by assigning (x1 , y1 ) as the origin in the layout shown in Figure 2.13, and simultaneously solving the set of circle equations for a common solution, the position coordinate estimate becomes, dˆ12 − dˆ22 + x22 2x2

(2.14)

dˆ12 − dˆ32 + x32 + y32 x3 − xˆ 2y3 y3

(2.15)

xˆ = and yˆ =

A 3D solution is determined similarly by simultaneously solving the three equivalent equations for a solid geometric sphere.

42

RFID Systems

Ceiling

Obstruction Signal Peak Interrogator Antenna

Tag Floor Signal Null

Figure 2.13

Multi-path interference.

2.2.3.3 Multi-Path Impact The signal arriving at the tag or the interrogator is a combination of signals reflected from materials in the environment plus the direct path transmission. The theoretical impulse response for a multi-path signal is, h(t) =

L 

αi ej φi δ(t − τi )

(2.16)

i=1

The amplitude and phase of each signal component i are αi and φi respectively. The TOF for the ith component is τi . L is the total number of received signal components. Since time of flight is directly proportional to the distance and the direct path signal is the first to arrive, the goal of the signal processing algorithms is to determine the direct path signal from the received composite signal. The RSS is the vector sum of all signal components. For non-line-of-sight conditions, the direct path signal may be much weaker than it would have otherwise been. The reflected signal components will combine constructively or destructively as illustrated in Figure 2.13. Therefore, RTLS engines average the RSS over several transmissions in order to improve the radial distance estimate. This implies that tags must transmit more frequently at each position in order to maintain the same velocity and positional update latency. Therefore, the trade-off for higher speed and accuracy is higher power consumption. In general, better localization accuracy can be achieved by collecting more diversified information that includes time, frequency, space, and polarization. For example, a DirectSequence Spread Spectrum (DSSS) technique can be used to increase the time resolution of the signal in order to estimate TOF more accurately. With such an approach, the

Performance Metrics and Operational Parameters of RFID Systems

43

receiver correlates a known pseudo-random number (PN) sequence with the composite signal to resolve each signal component. The algorithm measures the number of chips between the transmitted and the backscattered signal. The TOF estimate and the ability to resolve the first arrival signal improve with PN sequence length. However, the chipping sequence rate must also increase in order to maintain the same data rate [17]. A faster chipping rate also requires more bandwidth, and this may not be available. Increasing the bandwidth raises the noise floor and consequently reduces the receiver sensitivity, which ultimately reduces link margin and range. TOF of the first arrival signal can also be determined from UWB transmissions, which consist of nanosecond wide pulses. The time duration between pulses encodes the data. Signal absorption and multi-path conditions can cause inter-symbol interference making it nearly impossible to distinguish between pulses. Shortening the pulses and improving the SNR can improve the TOF estimate, but shorter pulses also occupy more bandwidth. UWB transmissions occupy several gigahertz of bandwidth and few regulatory bodies currently allow them at power levels below that of non-intentional radiators [3]. 2.2.3.4 Omni-Directionality Impact RTLS assumes that each interrogator can simultaneously read the tag throughout the common coverage area regardless of its orientation with respect to the interrogator. However, it is likely that an RF opaque material that the tag is attached to will block its signal from reaching more than one interrogator. The degree of RF transparency depends on the type of material and the operating frequency as described in Section 2.2.4. While omnidirectional antennas and RF transparent materials are likely to provide the best results, RF transparency is not a guarantee.

2.2.4 Impact of Materials RF reflective materials in the environment contribute to multipath conditions that reduce localization accuracy, read rates and operating distance of RFID systems. When these types of materials come into close proximity with either the tag or the interrogator, they could cause an adverse change in the receiver’s impedance characteristics that reduce the amount of energy coupled. The type of material and the selected operating frequency are the two most important parameters that determine the impact on receiver sensitivity.

Near Field

Far Field

Metals

Detuning

Reflections & Diffractions

Water

Some Detuning

Absorption

Near Full Performance

Figure 2.14

Reduced Performance

Effect of materials.

Poor or No Functionality

44

RFID Systems

Figure 2.14 summarizes the performance impact of materials on each system type. Antenna detuning or signal absorption occurs when material with dielectric properties substantially different from air comes within proximity of either the interrogator or tag antennas to cause an impedance mismatch. Any change in the impedance from a conjugate match with the propagating medium will decrease the amount of power transferred to the tag. A passive tag will receive less energy for activation, and a battery-based tag will experience a lower SNR. The net result is a decrease in operating distance and read rates for either type of tag. The degree to which these materials will degrade performance depends on the type of electromagnetic link and the material as shown. 2.2.4.1 Metals When attached to material with high metallic or water content, the electrical coupling between the tag’s antenna and the material causes an impedance change that reduces the power coupled into the tag. Manufacturers seemingly optimize planar or printed antennas either for free-space impedance characteristics or for the type of material intended for tagging. The electric field intensity near the antenna-material boundary appears to dominate a designer’s ability to practically implement the required impedance matching conditions [18]. Therefore, designers often add dielectric materials to the antenna to alleviate a trade-off between free-space and direct attach performance. However, dielectrics form an antenna spacing that forces non-planar and thicker antenna designs. 2.2.4.2 Water Microwave ovens work because water molecules in the food absorb the microwave energy. The H2 O molecules behave like tiny dipole antennas that attempt to align themselves with the oscillating EM fields. The water molecules generate heat in the process, much in the same way that a mechanical piston does when it absorbs energy to smooth out the vibrations. Any UHF tag antenna that is within close enough proximity to water will lose energy to the vibrating H2 O molecules because the molecules effectively form a lower impedance path for absorbing the EM energy. Water also acts like a dielectric that could detune the antenna. The degree of impact depends on whether the tag is completely immersed or attached to a water container.

2.2.5 Other Factors Considered 2.2.5.1 Read Rates One of the most important problems facing the RFID industry nowadays is that RFID systems are not 100% reliable. This means that if 100 tags are in the operating distance of the interrogator, very rarely will the interrogator be able to read all 100 tags. The percentage of tags that are in the operating distance of the interrogator and that are read (identified) by the interrogator relative to the total number of tags that are within the operating distance of the same interrogator is called the read rate. Low read rate can occur due to poor positioning of RFID tags so that their orientation relative to the reader antenna is unfavorable, close proximity of tags to one another, proximity of tags to water

Performance Metrics and Operational Parameters of RFID Systems

45

Antenna ASIC Package Battery

Figure 2.15

Packaged active tag. Courtesy Axcess International, Inc.

or metal, the number of tags within the operating distance, relative speed between the tagged objects and the interrogator, and many other factors. 2.2.5.2 Limitations on Size and Thickness The most basic passive tag consists of a substrate containing the electronics and antenna. An active tag also includes a battery and may incorporate additional components such as actuators, sensors, and indicators requiring additional electronics. Figure 2.15 shows an active tag packaged into a credit card form factor. Single chip implementations for active tags are now common [19]. Given the ability to reduce the size of the electronics, the size of the antenna and the battery remain the two dominant factors that constrain the tag’s form factor. Size Impact from Antennas The antenna size is a function of wavelength, gain, radiation pattern, bandwidth, and the required electrical impedance characteristics. Without the ability to further increase source power or sensitivity for a selected operating frequency, enhanced range is still possible by increasing the receive antenna gain if the required impedance matching characteristics are maintainable. The concept of an effective aperture (Aeff ) relates the wavelength (λ) and antenna gain () for far-field antennas where λ2  (2.17) 4π The ratio of the effective aperture to the physical antenna size Aphy is the aperture efficiency, which is typically about 80% for the best designs. The actual value depends on the type of antenna construction [1]. The interrogator’s antenna size is comparable to the wavelength in the far-field region. The antenna size decreases with increasing frequency and, therefore, there is a tendency to increase the operating frequency of RFID systems. In the near-field region, interrogator antennas are considered electrically small even though they can be several meters in diameter [20]. Near field RFID tag and interrogator antennas are basically loops of conductive material such as copper wires. Aeff =

46

RFID Systems

Size Impact from Batteries The battery size depends on the volumetric energy density of the battery type. Lithium coin-cell batteries store near 90 µW-Year of energy per cubic centimeter. Therefore, the size of the battery depends on both the tag’s power consumption and the desired operational life before needing a recharge or a replacement. A ‘low-power’ narrow-band active tag consumes an average of 30 microwatts under typical usage scenarios [1]. Therefore, a two cubic centimeter size Lithium battery will last, 90 µW · year/cm3 × 2 · cm3 ×

1 = 6 years 30 µW

(2.18)

The actual battery capacity should be degraded based on its leakage current or shelf-life rating. Active tags use different solutions to preserve battery life including: • low-power motion sensors integrated with tags so that tags do not report when they are stationary; • incorporating short-range radio in the long-range active tag and using short-range communication as an activation (waking-up) point so that the tag is awakened after it enters the area of interest [6]. By using these techniques, tags can reduce their beaconing frequency and reduce power consumption. For example, consider an application that requires detecting the presence or condition of tagged assets in storage. When using dual-frequency active tags, for example, those from Access International, Inc., the presence of these assets can be detected at the activation point that wakes up the tags as they enter the storage area. After that, tags will periodically report their condition and status to the interrogator. A very low beaconing frequency such as once per hour suffices for this typical scenario. In this way, the battery life can be significantly prolonged as pointed out above (about six years for this scenario). 2.2.5.3 Health and Safety In 1995, the European Telecommunications Standardization Institute (ETSI) subcommittee, European Committee for Electro Technical Standardization (CENELEC) published the pre-standard ENV50166-2 advising that uncontrolled emissions should not exceed 10 watts per square meter when averaged over any six-minute interval, and within about 20 centimeters of the antenna. However, for a controlled environment, the exposure limits are frequency dependent. Exposure limits are much more stringent for microwave frequencies starting at 1 GHz [21]. The implication is that designers should consider the frequency dependent Maximum Permitted Exposure (MPE) when installing antennas in people-occupied facilities. 2.2.5.4 Security Secure tags are capable of disguising their over-the-air data exchange. The level of security is proportional to the difficulty or cost to infer, reveal, or capture the actual data exchanged. The ease or cost with which the exchanged data is recoverable and applied to expose

Performance Metrics and Operational Parameters of RFID Systems

47

a person’s identity or activity relates to the privacy factor. Cryptography schemes can encrypt and decrypt data using secret digital keys and passwords. The strength of the cryptographic scheme is highly correlated to cost and power consumption. Therefore, adding security features will likely reduce the tag’s range and throughput performance. 2.2.5.5 Total Cost of Ownership Tag cost is often a key point of comparison among the various technology types. However, a complete RFID system also requires interrogators, label printers, software, and networking technologies to provide the value sought. Other cost components also include the manual labor to install, integrate, apply, and use the technology. The sum of all related RFID expenditures is the total cost of ownership (TCO). Tag price tends to be more sensitive in applications such as the retail supply chain where the tag is likely to be disposed of once read. Conversely, reusable tag embodiments and higher order functionality such as RTLS, tamper alert, and conditional sensing, tend to increase tag price. Therefore, return-on-investment (ROI) analysis for active tags often considers amortizing the tag price over its useful life.

2.3 Classification of Commercially Available Products Three RFID characteristics form the orthogonal bases for their categorization. These are: the way they obtain power, the way they transmit data, and the type of EM link formed with the interrogator. The four near-field and four far-field combinations form the eight categories shown in Figure 2.16 together with the major applications of RFID systems for each combination. The type of EM link impacts performance near liquids and metals as described in Section 2.2.4. The two most popular categories are active

No Batteries

Batteries

Energy Harvesting

Active

Solar

Vibration

Semi-Passive

Supply Chain Labels

Toll Collection

Reflected Energy

Passive

Figure 2.16

Tag technology categories.

Far-Field –EM Radiation

Asset Tamper and Seal Tags

Near-Field –Magnetic Coupling

Emitted Energy

(Semi Active)

48

RFID Systems

and passive, with their key performance attributes described earlier. Conceptually, it is convenient to consider semi-passive as passive with a battery, and semi-active as active without a battery. Batteries can improve both the sensitivity and functionality of a passive architecture. The absence of a battery forces reliance on other energy sources, for example, solar, thermal, and vibration, which may not be readily available. Figure 2.17 summarizes the trade-off between each of the remaining categories focused on the link between the tag and the interrogator. The highlighted portions of Figure 2.17 imply that products that integrate near-field and far-field capabilities can provide the combined benefits shown. Some of these architectures will be covered in Chapter 18. Figure 2.18 is a qualitative categorization of the two key performance requirements of range and throughput with respect to representative applications and tag types.

2.3.1 Near-Field Coupled Systems The most significant feature of near-field coupled systems is their resilience to impedance perturbation from the proximity of liquids and metals as described in Section 2.2.4. This near-field characteristic allows for covert antenna installations, such as burying them in the ground beneath control points such as gates and doorways. Control point boundaries can also be sharper because the near-field power declines three times faster on a logarithmic scale than in the far-field. However, this characteristic becomes a deficiency when the application also calls for long-range communications and higher throughput. By extension, both capabilities are obtainable by producing a combined near-field and far-field tag. 2.3.1.1 Low-Frequency Tightly wound low frequency antennas operating in the 130 kHz region have made it possible to construct capsule-size tags. Prevalent applications are animal tracking in areas such as wildlife management, animal rearing, and pet safety. In general, systems that must operate in environments with high humidity, fluids, and metallic content tend to use low-frequency near-field communications. A low frequency near-field antenna typically requires several dozen turns of a copper wire to provide sufficient operating range. However, these types of antennas often break from bending when implemented in a planar form factor such as within proximity access control cards. 2.3.1.2 High-Frequency In the higher frequency band such as 13.56 MHz, the antenna requires fewer turns for equivalent form-factor LF tags that can operate within the same coupling distance. Fewer turns make them less expensive to print or etch antennas than their LF counterparts. The higher frequency also facilitates more complex modulation schemes that can increase data rate and throughput. HF designs have become the preferred approach for contactless smart card, electronic transit pass, and electronic passport applications. Their accommodating near-field characteristics near liquids and metals combined with the higher data rates as compared with LF have increased their utility in short range supply chain operations such as automatic item identification in the pharmaceutical supply chain.

Emitted Energy

Reflected Energy

• Multi-tag arbitration speed limited by data rate

• Magnetic field zone control • Simple narrow-band protocols maximize battery life

Figure 2.17

• Robust media penetration

• Robust near-field energy harvesting for passive HF/LF RFID

Benefits • Excellent zone control

Deficiencies • Some bands require spread spectrum and complex multiple access protocols; leads to higher power consumption • Poor zone control

Benefits • Tens of meters of range for passive tags • Longer range for semi-passive tags; limited primarily by reader sensitivity • High multi-tag arbitration rate • Longer battery life

• Highly orientation sensitivity due to weaker backscatter and multi-path propagation

Deficiencies • Poor zone control • Poor RF media penetration

e.g. UHF RFID (ISO18000-6 & EPC)

• High multi-tag arbitration rates possible due to larger bandwidth & data-rate

Benefits • Long range due to RF propagation and higher transmit power

e.g. Wi-Fi (IEEE 802.11), UHF RFID (ISO18000-7), UWB

Far-Field

Performance and feature trade-off between technology categories.

• Multi-tag arbitration limited by bandwidth and data rate

Deficiencies • Backscatter reader sensitivity and loop antenna diameter limits practical range to within one meter

e.g. HF RFID (ISO 14443), LF RFID (ISO 14223-1)

Deficiencies • Range limited to antenna loop diameter

Benefits • Robust link around dense RF media

e.g. RuBee (IEEE P1902.1), NFC (ISO 18092)

Near-Field

Performance Metrics and Operational Parameters of RFID Systems 49

External Power Source (e.g. RF, Vibration, Light) Internal Power Source (e.g. Batteries)

RFID Systems

Range

50

Long

Vehicle Tracking

Personnel Tracking

Short

Shelved Items

Conveyor Belts

Low

High Throughput

Semi-Active

Far-Field Active

Near-Field Passive

Far-Field Passive & Semi-Passive

Figure 2.18 Application vs. technology.

2.3.1.3 Ultra High Frequency Near-field UHF RFID technology was introduced to provide an alternative to HF but at a much higher data rate. Tags can be read within short distances of about 1 ft with UHF antennas designed for near-field operation. High read rates have been demonstrated with near-field UHF tags placed on metals, on bottles containing liquids, as well as immersed in liquids. The advantages in comparison with LF and HF tags is that near-field UHF technology provides significantly higher data transmission rates, and that tag antennas can be relatively smaller since they require only one loop turn of a conductive trace or wire. The tags use loop-like antenna structures of about 1 centimeter in diameter, which may also be coupled to dipole-like structures for better sensitivity in the far-field region [22]. Tags that support both near-field and far-field UHF antennas can communicate with the interrogator using either near-field or far-field or both, depending on the type of interrogator antenna, and the power transmitted from the interrogator antenna.

2.3.2 Far-Field Propagating Systems Early far-field RFID systems utilized narrow-band communications in the lower UHF bands. However, the proliferation of commercial wireless standards in the 2.45 GHz Industrial, Scientific, and Medical (ISM) band, for example, Wi-Fi and Bluetooth, drove significant electronic integration and cost reduction. This fueled the emergence of wideband RFID tags that are compliant with those short range wireless technology standards, but not without a power consumption penalty. 2.3.2.1 Narrow-Band Regulations in the lower UHF bands such as 315 MHz and 433 MHz allocate a few tens of kilohertz of bandwidth. This allocation suffices for a majority of traditional RFID applications. Implementations near 433 MHz are widely deployed for container security and military logistics. This application was partially a catalyst for the ratification of international narrow-band standards such as ISO18000-7 and ISO18185 [23, 24]. Regulatory regions that have no allocations near 433 MHz tend to allow operations near 315 MHz.

Performance Metrics and Operational Parameters of RFID Systems

51

FCC CFR47 Part 15.231 limits the output power in these bands to approximately 6 microwatts [3]. Therefore, solutions in these bands are generally limited to active systems. 2.3.2.2 Wide-Band Although RFID implementations require very little bandwidth to transmit their unique identification number, the wider bandwidth ISM bands at 915 MHz, 2.45 GHz, and 5.8 GHz have attracted many wireless technologies, including RFID. Wi-Fi, ZigBee, and Bluetooth, are the most popular examples in this category [17, 25]. The number of active tag implementations using Wi-Fi, ZigBee, and Bluetooth are growing in the overall RFID market To share the band, these devices utilize spread spectrum with a relatively high data rate of at least one MBPS. In addition, active tags that operate in the ISM bands transmit at power levels that are at least an order of magnitude higher than narrow-band active tags. Consequently, they utilize proportionally larger batteries for equivalent longevity. Wi-Fi and Bluetooth are now widely available in consumer products such as personal digital assistants and cellular phones. ZigBee specifies a suite of protocols that comply with the IEEE 802.15.4 standards [25]. This IEEE standard defines an ad-hoc mesh network where nodes within range of each other can exchange data. Nodes initiate communications by either continuously listening for requests from other nodes in the network, or listen only at pre-determined time-slot intervals. The former requires that their receivers remain continuously activated. This increases energy consumption and shortens battery life. Waking up at pre-determined time-slots requires clock synchronization. This tends to add both cost and power consumption to the architecture. Therefore, a key trade-off in mesh network architectures is power consumption for responsiveness. RFID tag implementations that are compliant with these standards will, therefore, suffer from these same constraints on power consumption and battery life. Consequently, commercially available tag products that are compliant with these short range wireless network standards typically provide for user replaceable batteries.

2.3.3 Ultra Wide-Band The fundamental carrier signal for UWB systems is a nanosecond scale pulse width instead of a continuous wave (CW) frequency. The relative time separation between pulses encodes the data. While a relatively short time pulse improves the TOF estimate, it also spreads the occupied frequency across more than 500 MHz of bandwidth. Therefore, regulatory bodies severely restrict the output power to avoid interference with other equipment. For example, FCC CFR47 Part 15.517.c limits indoor UWB system power levels to −41.3 dBm EIRP, which is equivalent to about 74 nanowatts per MHz of resolution bandwidth [3]. This is less than 2% of the power density allowed for narrowband 433 MHz systems of equivalent bandwidth. UWB spectral allocation across the world is currently sparse. The link budget, which is the difference between the transmitted power and the achievable receiver sensitivity at the required data bandwidth, provides a good indication of the fundamental range difference between a narrow-band RFID system and a RFID implementations based on UWB technologies of equivalent data rate.

52

RFID Systems

2.3.4 Passive Solutions Nearly all of the passive tag performance deficiencies are due to its reliance on the interrogator for energy and a carrier signal to reflect during backscatter communications. The tag must receive reliable power for sufficient time duration in order to maintain a stable logic state during the collision arbitration process. Regardless of the algorithm used, tags must either keep track of a time-slot, a bit position in their serialized identification number, or an acknowledged response. However, multi-path propagation, electromagnetic interference, and electrical impedance perturbation from material proximity often hamper a tag’s ability to receive power for a sufficiently long period. Even when conditions allow the tag to obtain power from the interrogator for a sufficient period, or if a battery is available for powering the electronics, the significant loss in signal strength from backscattering becomes the next significant performance constraint as described in Section 2.2.2.4. 2.3.4.1 Silicon Based The traditional passive RFID tag, regardless of operating frequency, utilizes a single silicon chip. A simple protocol such as ISO18000-6a will require a few thousand transistors, while a more complex protocol such as C1G2 will require tens of thousands of transistors [12]. 2.3.4.2 Chipless RFID Researchers continuously demonstrate low functionality tags implemented with surface acoustic wave (SAW) resonators, MEMS cantilever finger reflectors, and RF reflective materials. These technologies create a unique backscatter signature from the incident RF energy [26]. The signature is usually a function of the mechanical construction methodology so this restricts their use to applications where a unique identifier without any prescribed format suffices. Other limitations are typically an inability to remotely address a specific tag, to write data from a distance, and to implement a collision arbitration MAC protocol. Printed transistor and organic circuit technologies capable of implementing low complexity RFID tags are also now emerging [27]. However, practical implementations are still constrained by their relatively large transistors, high operating voltage, and low charge carrier mobility [27]. Printed transistor technology cannot yet achieve the density or electron mobility required to implement complex RFID protocols in practical form factors [28].

2.3.5 Semi-Passive Architectures Also known as battery-assisted passive (BAP), semi-passive RFID is based on a hybrid architecture that reflects energy from the interrogator to transfer its data in the same manner as passive RFID, but utilize a battery for operating power. Therefore, their performance limitations stem from the same backscatter link constraints as passive tags. Although the battery facilitates improved receiver sensitivity through lower threshold voltage-biased rectifiers, the weak backscatter signal imposes greater sensitivity challenges on the interrogator. Equation 2.5 highlights that this challenge is exponential with improvements in

Performance Metrics and Operational Parameters of RFID Systems

53

tag sensitivity. Increasing the interrogator sensitivity will impose bandwidth restrictions, which ultimately reduces system throughput.

2.3.6 Far-Field Solutions Semi-passive far-field tags utilize the same backscatter modulation schemes as passive far-field tags. Aside from power transfer, they will impose similar constraints on link performance. Compelling reasons for semi-passive far-field implementations are backwards compatibility with existing passive tag standards, the potential for longer operating range than passive tags, and support for sensors and actuators.

2.3.7 Near-Field Solutions Semi-passive near-field architectures create controlled magnetic field disturbances that the interrogator senses and decodes. The benefits of semi-passive near-field solutions are similar to those of semi-passive far field solutions in that they are backwards compatible with passive systems but offer some performance and feature enhancements. Commercial examples are devices that comply with the RuBee [29] and NFC [30] standards. At lower RF carrier frequencies, magnetically coupled near-field systems operate more reliably near liquids and metals and provide better zone control accuracy than far-field systems.

2.3.8 Active Architectures Active tags use a battery for communications as well as other tag operations. Much higher operating distance can be achieved because of the higher receiver sensitivity possible with voltage biased front-end detectors. Several classifications of active tags can be made depending on the type of the standards used, their predominant application, wake-up mechanism, and frequency of operation. Active systems that utilize narrow-band standards such as ISO18000-7 or similar legacy protocols, are distinguished from those that are based on existing short range wireless network standards such as Wi-Fi. Active RFID solutions span the same frequency bands and electromagnetic link types available for passive tags, plus additional frequency bands where passive RFID do not operate due to regulatory imposed transmit power limitations. For example, narrow band active RFID tags operate within the 315 MHz and 433.92 MHz bands for which passive RFID tags are not commercially available. The 433.92 MHz UHF band is more uniformly allocated world-wide, whereas UHF allocations for passive RFID are more fragmented. Although more uniformly allocated, regulatory bodies such as the FCC restrict transmitted power within the 433.92 MHz band to about 6 micro-watts. The interrogator transmission periodicity is also limited to one transmission every 10 seconds or less often for regular reporting. Active RFID architectures span single frequency (LF, HF, UHF, microwave, UWB) implementations as well as multi-frequency implementations that include LF wakeup and UHF transmission [19]. Architectures compatible with WLAN infrastructures tend to be deployed for RTLS applications that track high value assets. The tags are typically designed to use replaceable batteries because of higher power consumption demands for operational compliance

54

RFID Systems

with wireless network infrastructures. Tag architectures designed for compliance with ZigBee or BlueTooth tend to support some form of wireless mesh networking so that data can be moved through intermediate nodes and eventually to a network connected hub. This architecture requires that each node awaken often enough to maintain synchronization with the network timing parameters, and to relay data packets that are part of its route. Since each node can uniquely identify a tagged asset, these types of network compatible technologies are also considered a type of RFID solution. However, the trade-off is usually higher power consumption than low power, narrow-band implementations such as protocols that comply with or are similar to the ISO18000-7 standard.

2.4 Conclusion In this chapter, the key operational parameters of an RFID system have been defined and the influence of other parameters, factors and design choices that affect them is elaborated. The key operational parameters are operating distance, system throughput and localization accuracy. This chapter provided detailed parameter and decision choices that the designers have control over including operating frequency, transmit power, bandwidth, digital modulation encoding, and maximum tolerable bit error rate. Factors that are influenced by one or more design choices include environmental characteristics based on deployment decisions such as signal-to-noise ratio (SNR), noise level, receiver sensitivity, and noise figure. The chapter also covers performance impact from materials such as water and metal on design considerations for read rates, size and thickness of RFID tags, health and safety aspects of RFID systems, security, and total cost of ownership. The available RFID technologies are then classified and qualitatively compared. Although RFID performance is critically important for successful system implementation, technology selection guides that systematically address key operational parameters and performance metrics for proper deployment have not been sufficiently covered in available literature. Most of the previously published works on parameters of RFID systems address mainly performance of passive HF [31] or passive UHF systems [32], and as such they cannot be used to select the appropriate technology for a given application anywhere in the world. In [32], read rate and read range are defined as main parameters of passive UHF systems. In [31], RFID systems are specified by their cost, size and performance where performance is determined by read range, speed, integrity of communication and compatibility among the systems from different vendors. A second line of research is related towards quantifying performance of RFID systems experimentally [33]. A set of performance benchmarks is proposed mainly to determine read rates and ranges versus different parameters for passive UHF RFID systems. The key parameters that we selected in this chapter are different from the previous publications because the focus of this chapter is beyond passive and far-field RFID systems. This chapter represents a comprehensive survey of performance metrics and operational parameters of overall RFID systems. Due to limited space, the factors that are also important but have not been treated in this book include the cost of RFID tags and readers, the effect of environmental factors such as temperature and humidity, the effect of the placement of the tags next to each other, unwanted reads, and several others. In addition, the analysis is restricted to RF and physical level parameters and do not include the effects of reader to tag protocols on performance and other high-level factors.

Performance Metrics and Operational Parameters of RFID Systems

55

In this chapter mathematical expressions for quantifying the performance of several parameters are given. This represents a step towards addressing larger research challenges in which all the parameters can be quantified and cost functions defined when comparing different technologies for various applications. Possible research directions include modeling and RFID system, designing cost function and developing an application software that would compare different design choices for a specific set of applications. In conclusion, we foresee that this chapter can be useful for the classification of RFID systems and quantification of their performance.

Problems 1. A tag receives 1 microwatt of ASK-modulated signal at 1 kbps data rate. What receiver sensitivity is required to demodulate the signal with a BER no greater than 10−6 ? 2. Consider a receiver with the sensitivity level solved from problem 1, and a corresponding FCC Part 15.231.e compliant transmitter operating in the 433.92 MHz band. If each transceiver in the system uses a 2.2 dB gain antenna, what is the link margin? 3. Given the same receiver sensitivity as problem 1, what is the link margin for an UWB system providing identical data rate and BER as the system in problem 2, but compliant with FCC 15.517.c using the 3.1 GHz to 10.6 GHz band? 4. Compare the theoretical unobstructed far-field range for each system of problems 2 and 3.

References [1] Krause, J. (1950) Antennas, 2nd edn. New York: McGraw-Hill. [2] Bridgelall, R. (1999) UHF Tags – the answer to the retail supply chain’s prayers? RF Innovations Magazine, August. [3] Federal Communications Commission (FCC). Code of Federal Regulations (CFR) Title 47, Part 15. [4] European Telecommunications Standards Institute (ETSI). EN 302 208: Electromagnetic compatibility and radio spectrum matters (ERM) – Radio-frequency identification equipment operating in the band 865 MHz to 868 MHz with power levels up to 2 W, Part 2 – Harmonized EN under Article 3.2 of the R&TTE Directive. [5] Bridgelall, R. (2002) Bluetooth/802.11 protocol adaptation for RFID tags, in Proceedings of the 4th European Wireless Conference, Feb. 28. [6] Bridgelall, R. (2008) Introducing a micro-wireless architecture for business activity sensing, paper presented at IEEE International Conference on RFID, April 16. [7] Mandal, S. and Sarpeshkar, R. (2007) Low-power CMOS rectifier design for RFID applications, IEEE Transactions on Circuits and Systems, 54(6): 1177– 1188. [8] Smith, AA. Jr. (1998) Radio Frequency Principles and Applications. New York: I.E.E.E. Press. [9] Gray, P.R., Hurst, P.J., Lewis, S.H., and Meyer, R.G. (2001) Analysis and Design of Analog Integrated Circuits, 4th edn. New York: John Wiley & Sons, Ltd, pp. 127–128. [10] EPCglobal Inc. (2005) EPC Radio-Frequency Identification protocols Class-1 Generation-2 UHF RFID protocol for communications at 860 MHz–960 MHz. Ver. 1.1.0. Dec. 17. [11] Skalar, B. (1988) Digital Communications Fundamentals and Applications. New Jersey: Prentice Hall. [12] Karthaus, U. and Fischer, M. (2003) Fully integrated passive UHF RFID transponder IC with 16.7-µW minimum RF input power, IEEE Journal of Solid-State Circuits, 38(10): 1602– 1608. [13] Assad, M.A. (2007) Master’s thesis: A real-time laboratory test bed for evaluating localization performance of Wi-Fi RFID technologies, Worchester Polytechnic Institute, May 4.

56

RFID Systems

[14] Thomas, G.B. and Finney, R.L. (1984) Calculus and Analytic Geometry, 6th edn. Reading, MA: AddisonWesley Publishing Co., pp. 151– 152. [15] ISO/IEC 18000-6 (2004) Information technology automatic identification and data capture techniques – Radio frequency identification for item management air interface – Part 6: Parameters for air interface communications at 860–960 MHz. [16] Tseng, J.D., Ko, R.J., and Wang, W.D. (2007) Switched beam antenna array for UHF band RFID system, IEEE International Workshop on Anti-counterfeiting, Security, Identification. Issue 16–18. April: 92–95. [17] IEEE Std. 802.11. Specification is a basis for Wi-Fi. [18] Dobkin, D.M. and Weigand, S.M. (2005) Environmental effects on RFID tag antennas, in IEEE International Microwave Symposium, June. [19] Axcess International Inc. (2007) Dot Micro-wireless technology for business activity monitoring. Press release, November. [20] Sears, F.W., Zemansky, M.W., and Young, H.D. (1984) University Physics – Part II . 6th edn. Reading, MA: Addison-Wesley Publishing Co., pp. 616– 617. [21] IEEE C95.1-1991. Standard Safety Levels with Respect to Human Exposure to Radio Frequency Electromagnetic Fields (3 KHz–300 GHz). [22] Nikitin, P.V., Rao, K.V.S., and Lazar, S. (2007) An overview of near field UHF RFID, IEEE RFID 2007 conference, Grapevine, TX, March 2007. [23] ISO 18000-7 (2004) Parameters for active air interface communications at 433 MHz. [24] ISO 18185-1 (2007) Freight Containers – Electronic Seals. [25] IEEE Std. 802.15.4. Specification is a basis for ZigBee. [26] Preradovic, S., Balbin, I., Karmakar, N.C., and Swiegers, G. (2008) A novel chipless RFID system based on planar multiresonators for barcode replacement, in IEEE International Conference on RFID, April 16, pp. 289–296. [27] Subramanian, V. (2009) Printed electronic tags and sensors for smart packaging applications, in Proceedings of the 8th Annual Conference on Flexible Electronics and Displays, Feb. 2–5. [28] Gowrisanker, S. (2009) Low temperature integration of CMOS devices on flexible substrates, in Proceedings of the 8th Annual Conference on Flexible Electronics and Displays, Feb. 2–5. [29] IEEE Std. 1902.1. Specification is a basis for Rubee. [30] ISO 18092. Specification is a basis for Near-Field Communications at 13.56 MHz. [31] Scharfeld, T.A. (2001) An analysis of the fundamental constraints on low cost passive radio-frequency identification system design, Master’s thesis, Massachusetts Institute of Technology, Cambridge, MA, August. [32] Nikitin, P.V. and Rao, K.V.S. (2006) Performance limitations of passive UHF RFID systems, in Proceedings of IEEE Antennas and Propagation Symposium, Albuquerque, NM, July, pp. 1011– 1014. [33] Ramakrishnan, K. and Deavours, D. (2006) Performance benchmarks for passive UHF RFID tags, in 13th GI/ITG Conference Measurement, Modeling, and Evaluation of Comp. and Comm. Systems, N¨urnberg, Germany, March 27–29, pp. 137– 154.

3 UHF RFID Antennas Daniel Deavours University of Kansas, USA

Passive UHF RFID antennas are primarily based on a “printed” dipole. Figure 3.1 shows several commercial tags. At first pass, one can see that they are a combination of long and skinny dipoles, though some of them are fat (those tend to be more expensive and thus less popular commercially). They are certainly not always straight ribbon dipoles and include a number of wiggles, slots, curves, and other interesting geometries. The RFID IC seems to always be attached to some loop within the antenna structure. Some of these features are designed for aesthetics, but clearly some are functional, that is, they affect behavior of the antenna in some intentional way. What do all of these features do? How does size impact performance? How does the environment affect tag performance? What makes an antenna a “good” antenna? These are many of the topics that we will address in this chapter. The chapter is set out as follows. In the first section, we review the behavior of a dipole antenna. Those who are familiar with antenna theory can safely skip this section, and those who are vaguely familiar or unfamiliar will find the most useful concepts and expressions there. The second section will focus on the T-match in its variants [1, 2]. The T-match is essentially the loop-like structure that we see in many of the RFID tags. For many of these analyses, we attempt to reduce the physical structure to circuit-equivalent approximations. Usually, some detail is lost in such simplifications, but they are also invaluable to providing intuition, insight, and design principles. In Section 3.3, we walk through a design process using a combination of equations and CAD tools to design an antenna. In Section 3.4, we leave the dipole-based antennas and focus on microstrip antennas. Though the literature is now full of microstrip-based RFID antenna designs, we present here some theory and examples of how one can build microstrip RFID antennas using the same T-match used for dipoles. We also present some recent research that shows an interesting “hybrid” dipole/microstrip solution. We conclude with some remarks and references. RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

58

RFID Systems

Figure 3.1 Variety of commercially available UHF RFID tags.

After reading this chapter, we hope you will have a better appreciation of the antenna designs. You should be able to look at a commercial antenna and identify the important aspects of the antenna. You should understand the fundamental limits that govern all printed antennas. We won’t be able to tell you which tag is the right one for a particular application, but if you’re ambitious, you might be able to design one yourself.

3.1 Dipoles and Relatives The classic dipole antenna consists of two wires of equal length placed in line and attached to an AC source in the middle. To simplify the analysis, we often assume that the wires are hollow cylindrical tubes. RFID antennas are usually not made out of straight hollow wire tubes with the chip placed in the middle. However, understanding something about the basics of the simple dipole will translate well to the printed dipole. By “printed” we do not necessarily mean that it was literally printed (by a printing press or an ink jet printer, e.g.), though it certainly may be printed using conductive silver or other metallic inks. What we mean is that it is thin, flat, and all the metal residing in one plane. Cylinders are three-dimensional; printed antennas are (primarily) two-dimensional structures. We will begin this section by reviewing the classic dipole antenna, then move on to printed dipoles, meandering dipoles, tip-loaded dipoles, T-match, bandwidth, gain, power transfer efficiency, and other important metrics.

UHF RFID Antennas

59

3.1.1 Dipole The dipole antenna is fed differentially, that is, one side is attached to a positive voltage and the other to the negative voltage. At DC, with such a structure, nothing interesting will happen. The structure will have some DC capacitance, that is, some positive charges will pile up in one of the poles, and an equal number of negative charges on the other end, and that’s it. At AC, things begin to get interesting. Capacitors take time to charge, and moving currents along a wire create magnetic fields. Those moving currents also launch electromagnetic waves that propagate into the far field, which acts as a resistor. Conceptually, one can think of the dipole as two inductive wire segments that are capacitively coupled to each other. Capacitance accumulates charge, which creates a force that opposes more charge accumulation. As charges accumulate in one of the poles, like charges repel and unlike charges attract, so the capacitance resists currents “piling up” in one pole or the other. Mathematically, this kind of force is similar to a spring that resists motion. As current moves through one of the poles, it creates a magnetic field according to Ampere’s law. The magnetic field serves as a kind of momentum, a force that resists changes in current: electrons in motion tend to stay in motion and electrons at rest tend to stay at rest. (This is partly due to physical momentum since electrons have mass, but the energy stored in the magnetic field is much stronger than the energy stored in electron momentum, so we usually ignore the momentum due to mass.) At some frequency, the two forces are equal and tend to cancel, causing resonance (Figure 3.2). Mathematically, this is no different than a child on a swing or a guitar string vibrating. A common analogy is a mass on a frictionless surface connected to a spring. At some frequency, the energy stored in the spring and in the momentum is the same and the two forces cancel themselves out in a kind of resonance, which is observed in a sinusoidal oscillation of the mass.

Electric Fields

Magnetic Fields

Figure 3.2 Electric and magnetic fields around a dipole antenna.

60

RFID Systems

A very important concept with dipoles is that of impedance, which simply stated, is the ratio of voltage to current. Because the voltage and current of an AC system may not be in phase, the impedance may be a complex number, that is, having a magnitude and phase, or a real and imaginary part. Resonance can be defined as when the current and voltage is in phase, that is, the impedance has no imaginary part. This happens when the dipole is approximately one half of a wavelength in length. Predicting the impedance looking into a dipole is a very fundamental problem, but also a complex problem. While a dipole has been around since Heinrich Hertz experimented with them in 1886, much has been learned (e.g. [3]), but unfortunately very little is simple. However, over a relatively short range of frequencies near resonance, there have been several circuit models that can moderately accurately predict the dipole impedance. Even those tend to be more complex than we would like to work with, so often we simplify the equivalent circuit to a series RLC circuit shown in Figure 3.3. It’s important to note that this is only an engineering approximation, and only works moderately well over a short range of frequencies. The reactance does tend to follow the series LC circuit, so the model works very well when describing the reactance. However, the resistance is not constant. We will develop the resistance in more detail below. Again, over reasonably small ranges of frequencies, such as 3% or 5%, these assumptions hold fairly well.

3.1.2 Radiation As with many things pertaining to a dipole, an accurate model is quite complex, but a reasonably good approximation is simple. Let’s begin by looking at what happens with one very small segment of a dipole. Imagine chopping up the dipole into short segments, and on each segment assume the current is uniform. (These little dipoles are known as Hertzian dipoles.) Let I0 be the current in the dipole, d the distance from the antenna (assumed to be more than a wavelength away), and θ the angle formed from the X axis as illustrated in Figure 3.4. Then the intensity of the electric field is given in Equation 3.1. E=

−j I0 sin θ L j (ωt−kd) e . 2ε0 cd λ

(3.1)

Here, j 2 = −1, ε0 is the free space permittivity, c is the speed of light, k = 2π/λ, and ω = 2πf . The important point is that: (1) the power intensity is the square of the

RA

CA

LA

Voc

Figure 3.3 Simple circuit model of dipole antenna near resonance.

61

Y

UHF RFID Antennas

q

L

d

X

Figure 3.4 Radiation pattern around a dipole antenna.

E-field intensity, and the E-field drops off as 1/d, so power drops off as 1/d 2 ; and (2) the E-field intensity falls off with θ as sin θ . The radiation pattern is thus a toroid-like or “donut-shape” radiation pattern. However, a half-wave dipole does not have a uniform current distribution over its length. At resonance, it is actually a half sine wave, so the current is larger in the center and falls off to zero at the ends. The resulting electric field intensity follows a pattern that is given in Equation 3.2. cos( π2 cos θ ) ≈ sin θ sin θ

(3.2)

For many applications, the approximation of sin θ is good enough. The bottom line is that a dipole radiation is at a maximum in the broadside direction, and goes to essentially zero in the direction of the poles. We’ll see what that looks like when we consider a specific example in Section 3.3.

3.1.3 Impedance and Bandwidth The first question we will ask is: what is the effect of the diameter of a wire to a wire dipole? The answer is that it affects a number of things, but most importantly, bandwidth. Informally, bandwidth is the range of frequencies over which the antenna provides a given level of performance. Usually, the limiting factor in antenna bandwidth is the antenna

62

RFID Systems

impedance, that is, the input impedance of the antenna changes (relatively) rapidly with frequency. For most antennas, such as the antenna you might have on a television, the intention is for the antenna to match the characteristic impedance of the transmission line connecting the antenna to the television. The characteristic impedance is essentially real, for example, 75 or 300 Ohms. This is also the case with RFID reader antennas that are commonly connected to the reader through a 50 Ohm coaxial cable. With RFID tag antennas, the antenna is directly coupled to the RFID integrated circuit (IC). The IC usually has some kind of circuit that converts small voltages into usably large DC voltage to power the IC, usually using diodes or transistors in diode configuration. What that basically means is the IC impedance is not resistive, but contains a large reactive component. In this case, because diodes are used, the impedance tends to have a large capacitive component. For example, an IC may have a series resistive component of 10 Ohms and a reactive component of −j 150 Ohms. If the IC has an impedance of 10 −j 150 Ohms, then to maximize the power delivered to the IC, the antenna should present an impedance of 10 +j 150 Ohms, that is, a conjugate pair. This follows from the maximum power theorem. To measure the quality of the impedance match, we can use something called the power transfer efficiency, or power transfer coefficient, typically denoted in the literature by the Greek letter τ . If τ = 1, then we know that the IC and antenna form a conjugate pair and the maximum power is transferred between the two. The astute reader might notice that only half of the power is transferred to the load; the other half is scattered (re-radiated) back into space by the antenna. This can’t be helped. The power transfer efficiency can be expressed in terms of the impedances [4]. τ =

4RA RC |ZA + ZC |2

(3.3)

In Equation 3.3, the subscript A represents the antenna impedance and resistance, and C the IC impedance and resistance. If you substitute RA = RC and ZA = ZC∗ , you’ll see that τ = 1 as expected. Directly related to power transfer efficiency is return loss. Return loss is the fraction of available power not delivered to the load. There is a simple identity: power transfer efficiency + return loss = 1. To be specific, we can define the scattering wave s as shown in Equation 3.4. s=

ZA − ZC∗ ZA + ZC

(3.4)

Then the return loss is |s|2 , and the identity τ + |s|2 = 1. Return loss is usually specified on a decibel scale, that is, RL = 20 log10 |s|. There are some convenient conversions to remember: • a 10 dB return loss = 90% power transfer efficiency; • a 20 dB return loss = 99% power transfer efficiency; • a 3 dB return loss = 50% power transfer efficiency etc. The IC impedance does change with frequency. We’ll return to the IC impedance changing with frequency later on, for now, let’s focus on the antenna. For the antenna, especially

UHF RFID Antennas

63

for small, short printed dipole antennas that are commonly used for RFID, the impedance changes rather rapidly with changes in frequency. As the impedance changes, τ also changes. One way to quantify the change in impedance is through measuring the quality factor or Q of the antenna. The name may be rather misleading because a larger quality factor indicates an antenna with a small bandwidth, which is generally regarded as a lower quality antenna. The term “quality factor” was historically used to describe the quality of reactive elements such as inductors and capacitors. Hence, a high quality inductor is one with a large quality factor or Q. However, the concepts still apply to antennas, except that a large Q is often inversely proportional to the quality of the antenna. Generally, Q = 2π

Energy Stored 1 ≈ Average Power Dissipated Bandwidth

(3.5)

So what does that mean? Remember that the dipole stores energy in the magnetic and electric fields. The more energy stored in those fields as a fraction of the power radiated, the larger the Q of the antenna. Power is dissipated by an antenna through radiation. (It can also be dissipated through resistive or Ohmic losses, dielectric (material) losses, but that negatively impacts efficiency. We’ll return to this below.) So a very large Q antenna stores a lot of energy and radiates relatively little of it, and thus will have a very small impedance bandwidth. An antenna that has a large bandwidth stores relatively little energy and dissipates much of it. How can you decrease the energy stored in a dipole? Simple: use thicker poles. A thin dipole stores a lot of its magnetic energy very close to the thin wire, while a thicker cylinder just stores a lot less magnetic energy. Interestingly, the thicker dipole has a larger surface area, which stores less capacitive energy, and as the diameter gets larger, the amount of energy is decreased in both the magnetic and capacitive fields in very close to the same amount. That means that a dipole basically resonates at the same frequency regardless of how thin it is. That assumption starts to break down when the dipole becomes fairly fat, but for thin wires, it holds remarkably well. Thus, we talk about the resonant frequency as (mostly) a function of its length, and the Q as a function of the length-to-radius ratio. 3.1.3.1 Printed Dipoles Until now, we have considered a simple, cylindrical dipole, but nearly all RFID antennas are some variant of a printed dipole. A printed dipole isn’t necessarily printed using, for example, an ink jet printer (though it could be); it simply means that the antenna is thin and flat. The term comes from the use of printed circuits such as a printed circuit board, which has been turned into the acronym PCB. This term comes from the old printing processes used to print a layer of resist before etching the pattern away using a corrosive solution. Some RFID tags are literally printed using silver inks. Regardless, the end result is an antenna that is fundamentally flat, not round. What are the consequences of a flat dipole versus a round dipole? Bottom line, we see a loss of bandwidth. There is a rough equivalence between a wire dipole with radius r and a printed dipole with width W :W = 4r, which is shown in Figure 3.5.

64

RFID Systems

W t r

Figure 3.5 Relationship between cylindrical and ribbon dipoles.

Not surprisingly, wider printed dipoles have larger bandwidths. Unfortunately, wider dipoles take up a lot more space, and if the dipole is made out of silver ink, printing a wide dipole uses much more ink than a narrow dipole. If the antenna is etched from copper or some other subtractive process, then it makes sense to leave as much metal on the antenna as possible, but if the antenna is intended to be printed using expensive silver ink, then you may see designers trading off bandwidth for decreased cost. Because IC sensitivity (discussed below) has improved so much in recent years, it is now common for designers to trade off performance for reduced cost. 3.1.3.2 Meandering Dipoles Unfortunately, much of the previous discussion has little to do with practical RFID antennas because most RFID tags are designed to comfortably fit within 50.8 mm by 101.6 mm label (2 by 4 inches). A large number of tags fall within the general size of 8 mm by 94 mm. Recall that we said that a resonant dipole is a little short of half a wavelength long, and at 915 MHz, that’s about 150 mm, or more about six inches. So somehow we have to shrink the size of the dipole. We basically have two options: (1) we can take the dipole and make it meander so that the full 150 mm of length will fit within 94 mm of space, or (2) we can just not use a resonant dipole. Most commercial tags use some combination of the two approaches. What is the impact of shrinking the dipole from 150 mm to 90 mm? To demonstrate, we consider two antennas that are both 1 mm wide. The first is resonate length (151.8 mm). The second is a 1 mm line that is meandered within a 90 mm by 10 mm “box” so that it also resonates at approximately 915 MHz. See Figure 3.6 for an illustration of the tested antennas. The long, straight dipole was found to have Q = 6.7, which gives a

Figure 3.6 Straight vs. meandering dipole used to evaluate changes in Q for different dipole geometries.

UHF RFID Antennas

65

3 dB bandwidth of approximately 130 MHz, which will cover the world-wide RFID band comfortably. The meandering dipole was found to have a Q = 15.7, or a 3 dB bandwidth of approximately 58 MHz. What if the dipole is too short to resonate? That’s actually OK, because we don’t need the dipole impedance to be resistive. We actually want the dipole impedance to be inductive (positive reactance) because the IC typically has a negative reactance. We’ll get into the details of the matching network in the next section. However, for now, we can say that a “short” dipole, or one operating below resonance, has a capacitive impedance (negative reactance). Somehow, we have to make that impedance inductivelooking (positive reactance), and we usually have to add some kind of inductor or inductorlike circuit. Remember, inductors store magnetic fields to operate. So, basically, we have to add reactance to the antenna, and remember that adding reactance increases the antenna Q. We threw out a few examples about antennas, Q, and bandwidth, but it is very natural to ask a question about the fundamentals. Let’s assume for a moment that we are not particularly clever about how we are building antennas. Perhaps there is something else out there, other than a dipole, which might have much better bandwidth. How can we quantify just how good these printed dipoles perform in terms of bandwidth? This has been a topic of theoretical antenna work for a long time. Today, there is a generally accepted lower bound on the Q of an antenna that fits within a sphere of radius a. This is called the Chu limit, after Andrew Chu, who first published the concept in 1948 [5]. The limit is given here in Equation 3.6.   1 1 Qlb = ηr (3.6) + (ka)3 ka Here, ηr is the relative free space impedance, which for most purposes, is unity. How is our dipole doing? Well, for one thing, we know that the printed dipole doesn’t use space as efficiently as a cylindrical dipole, so we’re giving up some bandwidth there. For a device that is 150 mm long, the theoretical smallest Q is 1.03. A Q of 6.7 is quite a bit worse. The 90 by 10 mm dipole? Its minimum Q is about 2.7, which would have more than 300 MHz of bandwidth, so a Q of 15.7 and bandwidth of 58 MHz is, again, considerably worse. So the bottom line is that printed dipoles are not at all bandwidthefficient. However, they are very cheap to make and simple to design, and within the field of RFID antennas, performance almost always takes a back seat to cost. Why are dipoles so bandwidth-inefficient? It’s fairly simple. If you consider the bounding sphere, the dipole, and especially the printed dipole, uses very little of that volume. Other antennas that fill that volume more effectively will be able to store less energy, just as we saw thicker dipoles have smaller Q. A dipole built out of two hemispheres would have a very small Q and approach the Chu limit. But who wants to buy RFID tags out of two hemispheres?

3.1.4 Radiating Resistance In the previous section, we focused a lot on the antenna reactance, Q, bandwidth, and resonance. Next, we will look at the radiating resistance. Simply put, if you were to break up the dipole into little segments of a dipole (perhaps “dipolettes”), each one would induce an electric field at some distance. If we integrated the square of the electric field (power)

66

RFID Systems

over all solid angles, we would find the total radiated power. That power lost to radiation is what we call the radiating power, and shows up in our circuit model as resistance, that is, the radiating resistance [3]. To cut to the chase and simplify, we’ve put the radiating resistance of a dipole in a convenient form:  2 2 πL RA = 80α (3.7) λ Here, L is the total length of the dipole, and α is a term that is dependent on the current distribution along the dipole. If the current is uniform, then α = 1. If the current is triangular, maximum at the center and zero at the ends, then α = 0.5. If the current is distributed as a half sine wave, which is the case for resonant dipoles, then α = 0.62. So, for example, our 150 mm dipole with a half sine wave distribution should have a radiating resistance of about 64 Ohms; we found 67 Ohms, so it was a pretty good estimate. The meandering 90 mm dipole would have a radiating resistance of about 23 Ohms (we saw 27.5 Ohms). If the antenna is very short, then the current tends to be more triangular, so the radiating resistance of a 90 mm straight (not meandering) dipole would fall to about 15 Ohms (about what we saw); another reason why short dipoles aren’t as effective. However, armed with this knowledge, we know that α 2 could be much larger than 0.38, which is the case for a resonant antenna with a half-sine wave current distribution, so there have been several “tricks” used to try to increase α. Remember that Q is a ratio of energy stored to power dissipated, and if you can increase the power dissipated without increasing the energy stored, then you can increase the bandwidth-efficiency of the dipole. The main concept for most of these techniques is to try to make the currents as uniform as possible across as much length of the dipole as possible. Recall the 90 mm meandering dipole. Before, we had the meandering distributed evenly throughout the length of the antenna. Now, consider pushing all those meanderings to the edge of the antenna, and having a large section in the middle that is straight. Sometimes, instead of latitudinal meandering sections, people use longitudinal meandering, or what looks like a hat at the ends of the antenna. Another way to increase α is to place a large capacitor at the ends. Obviously, adding a real capacitor would not be practical, but adding a very large metalized area, with enough area to store a significant amount of charges, can sometimes work. This is what we call capacitive tip loading. In some of our experiments, we found that the size of the capacitor needed to be impractically large for it to be an effective technique. However, on some of the smaller, wider tags, like the one shown below, the large metal area at the ends might be large enough to make a difference. Figure 3.7 shows some examples of commerciallyavailable tags that use either capacitive or inductive (“hat”) tip-loading technique. The best way we found to increase α is to use a spiral inductor. Because of the reinforcing magnetic fields, the spiral inductor can create the same amount of inductance in a much smaller space than meandering lines [6]. Using a spiral inductor, we were able to get α close to 0.8, which is quite good. Unfortunately, the spiral inductor used to create that large α also stored a lot of magnetic as well as capacitive energy, and we ended up increasing the Q of the antenna, so tip-loading with a spiral inductor is definitely not a good design choice. Figure 3.8 shows a prototype of a spiral-loaded dipole that we designed.

UHF RFID Antennas

Figure 3.7

67

Examples of commercial tip-loaded dipole tags.

Figure 3.8 Example of a spiral-loaded tag. This tag gives excellent radiating resistance for its size, but poor Q.

From our experience, the simplest and most effective way to increase α and decrease Q is to use a meandering dipole where the meandering sections are pushed towards the end of the dipole. In doing that, we were able to decrease the Q from 15.7 to 13.8. Not a big difference, but it also didn’t cost us anything in terms of extra size or material to get it. Another way to decrease Q is to use fatter traces, but there is an obvious limit to how fat you can make the traces and still fit within a confined area, plus there is the additional cost if one uses expensive metallic ink-based antennas. Bottom line, printed dipoles have adequate, but not great bandwidth. It is very difficult (but possible) for an antenna to have such large Q and operate over a world-wide frequency range. However, below, we will explore some of the details of how one can use the antenna Q together with the reactance of the IC to further improve the bandwidth of the tag. Before we can do that, we have to understand one more part of the antenna: the matching circuit. But before we dive into the matching circuit, there is one more component to the antenna that is worth covering.

3.1.5 Polarization A propagating electromagnetic wave is called a transverse electromagnetic (TEM) mode. In the TEM mode, the electric and magnetic fields form a 90 degree angle, and both are perpendicular to the direction of travel. Other modes exist, but they don’t propagate in free space; only the TEM mode propagates efficiently through space. The right hand rule applies here: point your fingers of your right hand in the direction of the electric field vector, curl your fingers towards the direction of the magnetic field vector, and if you stick your thumb out, it will point in the direction of travel. To keep matters simple, we consider only the electric field component of the TEM wave. Notice that for a particular direction of propagation, let’s say in the Z direction, the E-field can be anywhere in the X–Y plane. Let’s say you’re standing and facing a tag

68

RFID Systems

10 meters away. Let the X axis extend to your right out of the tag (the tag’s left), the Y axis extends up from the tag, and the Z axis is pointing right at you. If the tag is a dipole antenna and the long dimension of the dipole is horizontal, that is, parallel to the ground, then we say that the dipole is horizontally polarized . The electric field will be oriented completely in the horizontal plane. If the dipole is perpendicular to the ground (straight up and down), then any transmitted or reflected wave will be vertically polarized , that is, the electric field will be oriented vertically. Now consider a transmitting dipole antenna (let’s say the tag) and a receiving dipole antenna that you’re holding in your hand. What happens if the transmitting antenna is horizontally polarized and the receiving antenna is also horizontally polarized? The antennas work as expected, and power is transferred between the antennas as expected. But what happens if you rotate the receiving antenna by 90 degrees and hold the dipole vertically? The received power becomes essentially zero; there is a polarization mismatch between the transmit and receive antenna. So far, the polarization we have considered is in dipoles, which are linearly polarized (LP), that is, the E-field is aligned linearly in one direction. Conceptually, imagine if you could rotate the antenna really, really fast, like 915 million rotations every second? (Which corresponds to one rotation per complete cycle of a wave.) This would result in an electric field vector that would corkscrew the air, which yields an electric field that we say is circularly polarized (CP). While it may not be realistically possible to spin the dipole at 915 million rotations per second, it is possible to rotate the electric field electrically through some other means, usually using some antenna other than a dipole. In fact, it’s quite common for RFID reader antennas to be circularly polarized. It’s possible for a wave to be partially circularly polarized and partially linearly polarized; to understand this, just imagine the superposition of the two. The resulting wave, we say, is elliptically polarized . What happens when a LP antenna, such as a dipole, and a CP antenna, such as a CP patch antenna commonly used with RFID readers, meet? What happens is that there is a partial polarization mismatch, and exactly one half of the power is lost (compared to the case where there is a polarization match). This is commonly tolerated for a couple of reasons. First, it is common for tags to be oriented both vertically and horizontally. If we choose a vertically polarized reader antenna and some of the tags are oriented horizontally (or the reverse), the system performance will suffer severely. With a CP reader antenna, it doesn’t matter if the tags are oriented horizontally or vertically or any angle in between – the performance will be the same, though degraded by 3 dB. Note that polarization is distinct from the radiation pattern discussed in Section 3.1.2. There is a special class of RFID tag antennas that are called dual dipole antennas. (See Figure 3.9 for an example.) You can think of this antenna as two orthogonal dipoles. The dual-dipole antenna requires the use of a special, three-terminal IC, with the terminals labeled RF1, RF2, and GND (ground). The chip has a rectifier circuit so that it, basically, automatically selects which dipole it should use, and when helpful, can combine the signal from both dipoles. This allows the dual-dipole antenna to be polarization-matched to both linear and circularly polarized radiation, which is very convenient when using CP reader antennas. This is one reason why dual dipole antennas tend to perform so well. Another is because of the large size, the antenna bandwidth is excellent, and thus tends to be less

UHF RFID Antennas

69

Figure 3.9 A dual dipole RFID tag.

affected by environmental parameters (discussed more in Section 3.4). The downside is the large size tends to significantly increase the cost of the tag.

3.2 T-Match and Relatives 3.2.1 The Classic T-Match Remember that the IC has a significant (capacitive) reactive component to its impedance, and to have an efficient transfer of power between the antenna and the IC, we need to present an antenna with a conjugate impedance. We used an example of 10 −j 150 Ohms for a typical IC impedance. Thus, we would like the antenna to present 10 +j 150 Ohms. How do we do that? One way is to make the antenna short enough so that the radiating resistance is about 10 Ohms, and then meander the antenna so that it is electrically long and presents a large (inductive) reactance. While that will work, it’s not a good, general design tool. A variant of that design is to start with a meandering dipole, and make it sufficiently meandering so as to provide enough inductance. This may provide too large of a resistance, especially for antennas that are longer than resonant-length (for reasons we won’t go into here), so one proposal is to use a “loading bar,” which is a second, parallel conductive strip (dipole) nearby, which serves to effectively reduce α, which in turn reduces the radiating resistance. We don’t recommend this approach because it requires lots of extra space for both the electrically long antenna and the loading bar, plus it intentionally decreases α, meaning the Q of the antenna is significantly increased in the process. The resulting antenna will of course work, but in our opinion, it does not make good use of space.

70

RFID Systems

S

W2 Vin W1

Figure 3.10

Dipole with T-match.

In practice, most commercial RFID antenna uses some kind of modification to the T-match. The T-match was a structure developed by Uda in the 1930s. The structure of the T-match is shown in Figure 3.10. You can see that there is a dipole component and a loop-like structure. If you extend that loop-like structure to the ends of the dipole, the T-match basically becomes the folded dipole. Hopefully, this starts to look a little bit familiar after examining the commercial antennas in Figure 3.1. The T-match basically has three critical parameters: the length of the T-match S, the width of the secondary trace W1 , and the distance between the two traces, see Figure 3.10. A number of textbooks cover the T-match, so we will skip the analysis here and get to the results. The analysis follows a minor variation of the even/odd mode analysis with a second port added in the middle of the dipole (top trace). The equivalent circuit model is given below, where Zd is the impedance (which is almost entirely reactive) seen from the ports when driving the two ports differentially, and Zc is the impedance seen when driving the ports in common, or the common mode impedance, which is about that of the dipole impedance. Notice that Zd is basically a short circuit transformed by a short length of transmission line, so if the line is relatively short and we consider a relatively narrow range of frequencies, we can model Zd with an inductor. The α term here is a current splitting factor in the common mode, which acts as a transformer, and for equal diameter conductors, α = 1. (When the loop becomes a half wavelength long, Zd becomes an open circuit, and all we see is the dipole impedance scaled by (1 + α)2 . This is the “folded dipole.”) The equation that governs the input impedance is given in Equation 3.8. This is the same as the circuit diagram shown in Figure 3.11. Zin =

Zin

(3.8)

(1+a)2:1

Zd

Figure 3.11

(1 + α)2 Zc Zd (1 + α)2 Zc + Zd

Zc

Uda’s circuit model of T-match.

UHF RFID Antennas

71

(1+a)2Zc||Zd

(1+a)2Zc

Zc

Figure 3.12

Smith chart view of impedance matching using the T-match.

To illustrate what the T-match does, we consider a dipole before and after a T-match. We use a Smith chart in Figure 3.12 to illustrate. If you’re not familiar with a Smith chart, unfortunately there is not enough space to give a tutorial here. Basically, the Smith chart captures the positive complex plane through a Moebius transform of Z: zˆ =

Z0 − Z , Z0 + Z

(3.9)

where Z0 is commonly taken to be 50 Ohms for convenience, as it is here. The circles on the Smith chart represent lines of constant resistance; the radial arcs represent constant reactance; the horizontal line in the center is the real axis; the top region is inductive (positive reactance) and the bottom is capacitive (negative reactance); the left-most point is a short circuit or zero Ohms, and the right-most point is an open circuit. The common mode is approximately that of a center-fed dipole of comparable size, and is plotted as Zc in Figure 3.12. The splitting factor a provides an impedance scaling factor by (1 + a)2 . Finally, Zd provides a shunt reactance. A short, shorted transmission line provides an inductive reactance, so following a line of constant susceptance, results in the final input impedance Zin . Hopefully, if we designed the T-match correctly, then ∗ Zin = Zic (or whatever the desired criteria).

3.2.2 The Modified T-Match If you look at a modern RFID tag, you will probably see something that looks like a T-match, but doesn’t exactly fit the model of a T-match. In fact, only one of the antennas shown in Figure 3.1 uses a classic T-match. Two others use something we call the embedded T-match (the tag in the upper right corner of Figure 3.1 is an excellent example), where the T-match is inscribed into the dipole section, and behaves substantially like the classical T-match. The rest are some modification to the basic principle. Not surprisingly, we call these structures the modified T-match. In Figure 3.13, we show a

72

RFID Systems

Lh/2

Lh/2 Le/8

Le/8

Le/8

Le/8

Le/8

Le/8

Le/8

Le/8

Figure 3.13 Intuitive inductor-based circuit model for the T-match. Both the series and shunt inductor are shown as distributed inductors.

commercial antenna’s matching network, and interposed on the image is one way we propose that we think about the matching network. (The following is based roughly on the work of [7].) Imagine that each short trace acts like a small inductor [6]. Thus, we see distributed inductance all along that matching network. (Note that determining the actual inductance of straight wire segments in the presence of other conductors with currents causes mutual inductances, making exact calculations exceedingly difficult. This approach requires a little imagination, but it’s not really that much of a stretch.) Again, imagine the inductances are distributed along that loop. For reasons that will become clear in a moment, we’ll consider two parts of the loop: the “series” section, which we will call Le , and the “shunt” portion, which we will call Lh . We must emphasize that this model does not consider the coupling between components. When metal strips are in close proximity, as is shown in Figure 3.13, there is substantial coupling between the conductors, which is difficult to predict accurately without detailed computer models. However, as we will see later, while you may not be able to easily directly map the physical structure to literal inductor values, the qualitative circuit can be quite useful in guiding antenna design. So one can use a qualitative process and either experimentation or computer simulation to guide the design process. One alternative is to simplify the circuit and structure the matching circuit into simpler and largely uncoupled segments, which is not necessarily space-efficient, and therefore rarely used in practice. Until now, we have neglected to describe the RFID IC’s equivalent circuit. The circuit model that people use is determined by its purpose. We should mention that the RFID IC impedance is non-liner, that is, the impedance varies (sometimes considerably) with the input power. It is common to characterize the impedance of the IC at the smallest power setting, that is, the turn-on power, which will cause the IC to turn on and operate, that is, the turn-on impedance. It is reasonable to match the antenna at the IC turn-on impedance because that’s where the IC needs the most efficient transfer of power; when there is excess power collected by the antenna, we can afford to lose efficiency in the power transfer. In fact, if there is too much power transferred to the IC, the sensitive circuits can be damaged, and so IC-makers build special overload circuits. IC design is outside the scope of this chapter, but it is important for us to understand how to model the IC impedance. Sometimes, you may see the IC impedance specified as a complex number, that is, 20 −j 167 Ohms. This number should be specified at a particular frequency, that is, 915 MHz,

UHF RFID Antennas

73

because to complicate matters, the IC impedance also changes with frequency. If you’re only interested in designing a tag for a particular frequency or small range of frequencies, this might be fine, but if you’re trying to design a tag for world-wide operation, then it’s important to know the IC impedance over the world-wide band. Fortunately, most ICs can be modeled as a parallel RC circuit. This is not a perfect model, but it’s accurate enough for most applications. Putting together the dipole, which is approximately a series RLC circuit, the T-match, which is approximately a shunt-series inductor circuit, and the IC, which is approximately a parallel RC circuit, we end up with the circuit shown in Figure 3.14. Probably the worst assumption in this circuit is that the resistance of the dipole is constant. We know, for example, from [3] that the resistance is inversely proportional to the square of the frequency. So we know it’s not perfect, but it is a good approximation, and it allows us to translate something as tricky as currents and fields into something as simple as lumped element circuits. However, there is one more “trick” that we can use that will provide a lot more insight. Let’s introduce a parameter β = Lh /(Le + Lh ). Then it can be shown that the circuit shown in Figure 3.14 is identical to the figure shown in Figure 3.15. Here, Lse = Le Lh /(Le + Lh ), that is, the parallel combination of Le and Lh . The proof [7] is not too difficult and can be performed by someone with a modest background in circuit theory. There are a number of important results of the circuit in Figure 3.15. First, we can see how the ratio β can effectively scale the antenna (or IC) impedance. Also, note that β acts much like α in the T-match. In fact, one can draw an equivalence: β 2 = 1/(1 + α)2 . The classic T-match includes the inductors in the common mode impedance Zc , while the distributed inductor model tries to separate the matching circuit from the antenna. This separation may be OK if the inductors are relatively small, but if they become too large, they become a part of the antenna and need to be explicitly included. CA

RA

LA

Le

Lh

VOC

CIC

RIC

Figure 3.14 New inductor-based circuit model of the T-match dipole.

RA

CA

LA

Lse

b2(Le+Lh)

VOC

Figure 3.15

CIC /b2

Transformed circuit model of RFID tag.

b2RIC

74

RFID Systems

Second, an electrical engineer familiar with filters can spot this circuit and may recognize that the circuit in Figure 3.15 looks like a band pass filter, and a band pass filter can be used to extend the bandwidth of a circuit. Briefly, it works like this. Let the series and parallel RLC circuits have the same resonant frequency. At resonance, the series RLC circuit has a reactance with a positive slope. At resonance, the parallel RLC circuit has a reactance with a negative slope. Then, over some range of frequencies near resonance, the reactance will nicely cancel. If you’re lucky (or clever, or diligent) enough so that the slopes are just right, you can actually increase the bandwidth out of your system by about a factor of three. For several practical examples, we’ve been able to achieve this three-fold increase in bandwidth, which is significant. Unfortunately, this technique relies on the Q of the IC to be what you want it to be. Using an equal ripple (Tchebychev) filter, the Q of the antenna and IC should be identical. As we saw before, the Q of the antenna is largely determined by the size of the antenna, and the Q of the IC is usually set by the chip-maker. Fortunately, many antennas and chips have a Q around 10 to 15, which means that this technique can be used. For really small antennas where the antenna Q is very large, like 100, the bandwidth will be entirely dominated by the dipole Q and this technique won’t offer any benefit. Also unfortunate is that this technique requires both the series and parallel RLC circuits to resonate at the same frequency, which is also the (geometric) center frequency of the frequency range of interest. Recall from earlier, most commercial RFID antennas are not resonant antennas. In fact, it requires quite a bit of meandering to make an antenna resonate at around 900 MHz. (Look at the antennas in Figure 3.1 and guess which ones are resonant and which ones are not. We think only two of them are actually resonant antennas.)

3.3 Putting it Together: Building an RFID Tag Next, let’s walk through the process of designing an actual RFID tag using the lessons we’ve learned so far. We are using Ansoft Designer to do most of the “heavy lifting” for us, although Designer is by no means the only tool available for doing this kind of work. In general, any good method of moments (MoM) tool would be suitable. Let’s assume we’re given the task of designing an RFID antenna to work on RF-friendly material (we assume air). The antenna must be cheap and therefore as small as possible. Initially, the antenna will be etched from copper but in the long term will be printed with silver, so it must be a thin-wire dipole. After consulting with a contract manufacturer, you learn that the manufacturer can reduce costs by US$0.003 if the maximum antenna length is 86 mm, and the antenna can be as wide as 12 mm. Any wider, and the antenna will cost a fraction of a penny more. The antenna should be designed to be used with a new, superlow-cost RFID chip that has a parallel resistance of 2200 Ohms and a parallel capacitance of 1.15 PF. The tag must give as much performance as possible in the European ETSI band and the FCC band in North America, that is, 865 to 930 MHz, and we can expect minimal detuning from the environment. An 86 mm design is going to have an impact on the antenna Q, but we can make up for some of it with a 12 mm width. The line width may be a concern. 1 mm lines might be too costly, so we will try 0.5 mm lines and see if we can meet the objectives with that, and if not, try 0.75 mm lines.

UHF RFID Antennas

75

12 mm

86 mm

Port1

Figure 3.16 Meandering dipole used in the first cut.

First, we do a first rough pass at the antenna and see what kind of Q we can get out of it. The geometric mean of 865 and 930 MHz is 897 MHz, so we’ll target that as the resonant frequency. We’ll target an antenna with inductively loaded ends to get the smallest Q with the least amount of metal. Figure 3.16 shows a rough first cut. We adjusted the number of meanders so that it will resonate at about 897 MHz. It doesn’t have to be exact; we’ll tune it later. All we need now is a good estimate. We use the Equation 3.10 as a simple, reasonable estimate Q, which works well for a series or parallel resonant RLC circuit with resistance R: Q=

ω dX(ω) , 2R dω

(3.10)

Estimating the Q, we see that it’s about 17, which is probably good enough. We don’t actually have to cover the entire band from 865 to 930 MHz; the region between 868 through 900 MHz is unused, and we may be able to use that to our advantage. We also notice that the resistance is about 35 Ohms, which is good for an 86 mm antenna; the extra width allows us to make longer and therefore fewer meandering segments, increasing α. From this, can estimate a circuit shown in Figure 3.17. We know that that (Le + Lh ) must resonate with the IC capacitance, so (Le + Lh ) = 27.3 nH. What remains is choosing β, that is, the ratio of Le to Lh . One very useful rule of thumb is as shown in Equation 3.11:  Ra β= 2 . (3.11) Rc This estimate works well for √ choosing a 10 dB return loss (90% efficiency) over a broad band. So for this case, β = 2 · 35/2200 = 0.1784 is a good estimate. We decided to write a little computer program to vary β and see what happened. We found that we could 35 Ω

0.298 pF

VOC

Lh

105 nH

Le

Figure 3.17

1.15 pF

Circuit model of proposed antenna.

2200 Ω

76

RFID Systems

−6 −8 −10

-Return Loss (dB)

−12 −14 −16 −18 −20 −22 −24 −26 860

870

880

890

900

910

920

930

940

F (MHz)

Figure 3.18 Return loss for circuit with f0 = 897 MHz, β = 0.165.

actually give up some bandwidth to get a better return loss when we used β = 0.165. The resulting return loss is shown in Figure 3.18. We see that we have a very nice result in which the return loss stays under 12 dB over the entire range from 865 to 930 MHz. However, we can note that the performance is best at about 875 MHz (with a very nice 99.5% efficiency), not at 866 MHz where we would like it to be, and again at 925 MHz, not 915 MHz. So let’s try changing the resonant frequency. Recall that we set the center frequency to be 897 MHz. Next, we tried reducing the resonant frequency to 890 MHz, and then increased β a little bit so that β = 0.17. The resulting return loss is shown in Figure 3.19. The overall worst-case performance over band has little difference. We see that instead of 12 dB return loss at 930 MHz, it is only 10 dB, or still at 90% efficiency. However, the antenna operates at well over 99% efficiency at 865–868 MHz, and again has better than 99% efficiency at around 917 MHz and is better centered within the operating band. What we basically have done is chosen the geometric mean of the center of the two bands, 866 and 915 MHz. Then we adjusted β until we achieved a minimum return loss at those two frequencies (Figure 3.19). Now that we have a return loss curve, the next step is to go back to the antenna and fit it with a T-match that will (hopefully) produce an antenna with the same return loss. The way we do this will require a few iterations. We know, for example, we can control the resonant frequency of the parallel RLC circuit by the size of the loop formed by the T-match, and we can control β by the ratio of Le to Lh . The resonant frequency of the dipole antenna can be modified by the length of one of the meandering elements. So, armed with that intuition, we can start to construct a T-match to our desire. Because β is

UHF RFID Antennas

77

0

Return Loss (dB)

−5

−10

−15

−20

−25 860

870

880

890

900

910

920

930

940

F (MHz)

Figure 3.19 Return loss for circuit with f0 = 890 MHz, β = 0.17.

small, we know that Lh ≈ 6Le , so that gives a good starting point on the proportions. For convenience, we plotted the impedance looking into the circuit-equivalent of the antenna from the IC in Figure 3.20. This is the impedance behavior that we want the simulated physical antenna results to duplicate. It took us about 10–12 iterations, for maybe 20 minutes of work to get it just right, and we finally arrived at the modified T-match shown in Figure 3.21. The tag shown in its entirety is shown in Figure 3.22. As you can see, the design looks like a conventional RFID antenna design, except for the rather blocky design. (It’s at this point in the design process where management will probably introduce a new requirement regarding aesthetics.) Next, let’s look at the return loss of the simulated antenna versus the circuit model, which is plotted in Figure 3.23. We see that we are able to achieve remarkably close agreement between the simulated result and the circuit-based result. So the circuit is a very good guide for determining what is achievable. The reason for using a circuit is because it is much easier and faster to write a program or script to manipulate the circuit to achieve a desired level of performance than it is to manually change the physical layout of the antenna and run a full-wave simulation model over a range of frequencies. It also provides a great deal of insight and intuition as to the frequency response of tags. We see that there is pretty good agreement, though it isn’t perfect. Again, several of the assumptions that went into the circuit model, especially the assumption that the radiating resistance is constant over frequencies, are not accurate assumptions. But we can conclude that the circuit model was able to guide us to a solution that was able to meet the predicted performance remarkably well and with relatively little effort given the assumptions.

78

RFID Systems

180 160 140 120 Ohms

Rin 100

Xin Ric

80

−Xic 60 40 20 0 860

870

Figure 3.20

880

890

900 F (MHz)

910

920

930

940

Circuit predicted impedance looking into antenna.

2 mm

4.2 mm

2.5 mm

Port1 18 mm

Figure 3.21

Geometry of matching circuit.

Next, we plotted the three-dimensional radiation pattern in Figure 3.24 using a dB scale. The long dimension of the antenna is oriented along the X axis. As we can see, the antenna exhibits the “normal” dipole radiation pattern where radiation is poor along the X axis and is at a maximum in the Y-Z plane. So are we done with the tag design? Only if we’re lucky! In reality, there are numerous ways in which reality can deviate from our nice circuit and simulation model. For one, the

UHF RFID Antennas

79

Figure 3.22

Geometry of the completed antenna.

0 Circuit Simulation

−5

-Return Loss (dB)

−10 −15 −20 −25 −30 −35 −40 860

870

880

890

900 F (MHz)

910

920

930

940

Figure 3.23 Return loss of circuit and simulated antenna.

data sheet of our fictitious IC may be accurate under assumptions that don’t hold for us. For example, the method of attaching the IC to the antenna will introduce some parasitic reactance. What if we’re using a different adhesive, or a different pressure setting when attaching the chip? Does the data sheet IC impedance assume a particular attachment? In reality, the chip attachment process introduces some “parasitics” that affect the impedance. Some data sheets give the probed IC impedance (with no attachment reactance), and you have to add about 0.1 to 0.15 pF of shunt capacitance to get values that you want to use. Add 0.15 pF of capacitance to the IC that we designed the antenna for and you’ll see that our carefully designed antenna won’t work nearly as well. One IC data-sheet that we had indicated the IC reactance should be about j 194 Ohms, but we found through experimentation that after all the parasitics, we needed to match to j 127 Ohms. That’s one rare and rather extreme example, but a real one. Also, the impedance of these chips comes from analog RF front ends to the IC, and there are natural, random variations within the manufacturing process. Some chips will have different impedances than others. You may need to produce dozens of permutations and test dozens of tags with each permutation to get the design that you want that works well on average. Is this for a strap? Different

80

RFID Systems

Y

Phi Z

Theta

Figure 3.24

X

Radiation pattern of designed antenna.

straps introduce different parasitics. All this work was done putting blind faith into our simulation results to produce accurate results; did we configure the software properly? So don’t be fooled into thinking that you can design to the data sheet and everything will magically work as expected. There’s still a lot of hard work to be done before you can move from a prototype to production. There’s no substitute for experience and good support from the chip vendor. The aim of this chapter isn’t to turn you into an expert in building RFID antennas. The aim is to introduce simple and practical ways of thinking about antennas that will give you insights into how the antennas work. We wanted to give enough information to really understand the mechanics in terms of simple circuit elements, and to give insights into the fundamentals of what can and cannot be achieved. The “trick” of using the tag and matching circuit as a band pass filter is still not well known in the industry, and if you look back to Figure 3.1, you can see some of the antennas on that page do not have enough meanders to be resonant dipoles, and thus cannot take advantage of the band pass filter. That might be fine for antennas that are used only in one region, but makes it hard to believe claims of “world wide operation” – unless the vendor’s definition of operation is fairly loose.

UHF RFID Antennas

81

3.4 The Environment In the previous section, we pretended to be antenna designers for a little while and built an antenna that would work in these very nice, well-defined ways. What happens when we place the tag on a plastic bin? On a cardboard box? A bottle of water? Near metal? Unfortunately, the answers to those are complicated, and usually the results are that the tag degrades its performance one way or another. In this section, we’ll examine the mechanisms of how the environment impacts the tag performance, and steps that can be done to mitigate that impact.

3.4.1 Dielectric Constant To begin with, we can classify materials into three categories: dielectrics, conductors, and magnetic materials. Magnetic materials such as ferrites are relatively rare, so we’ll ignore those. Dielectrics are basically electrical insulators; they have very large resistance to electrical current when a voltage is placed across them. Conductors, on the other hand, conduct electrical current very readily and offer very little resistance. At 900 MHz, no material is a perfect conductor or a perfect dielectric. A dielectric material basically does one thing under the presence of an electric field: it resists the electric field. Under an electric field, a dielectric material produces its own electric field that opposes the applied electric field. Every material is made up of molecules and atoms. Two atoms that are bound in a molecule often do so because one atom is an electron-donor and the other atom is an electron-receiver, and the resulting molecule is slightly polar. Under an electric field, the polar molecule tries to align itself with that field, but the polar molecule produces its own, counteracting electric field. This alignment of polar molecules is the way that the material stores energy. Atoms themselves consist of a positively charged nucleus and negatively charged electrons in a surrounding electron cloud. Normally, the nucleus is in the center of the electron cloud, but when an external electric field is applied, the cloud can shift slightly to one direction and the nucleus the other direction to produce a slight polarization and an opposing electric field. The amount of counteraction to the electric field is called the dielectric constant , or permittivity, and is represented by the symbol ε. Because it has units, we often talk about the relative dielectric constant, εr , which is relative to the dielectric constant of a vacuum, ε0 , that is, ε = εr ε0 . This relative dielectric constant has an important implication to the propagation of electromagnetic fields. We won’t go into the details to derive this here, but we will simply state that the velocity of a propagating wave through a dielectric material is slowed down. Let c be the speed of light in a vacuum. Then the velocity of a propagating wave through a dielectric medium with permittivity εr is c v=√ . εr

(3.12)

What does any of this have to do with RFID antennas? Remember that the antenna resonates when the electric field and magnetic fields essentially cancel. What happens if the electrical energy stored in the antenna is increased? The resonant frequency of the antenna decreases. Or, another way to think about it, if a half-wave dipole (where

82

RFID Systems

half a wavelength is measured in air) is completely embedded a dielectric material, the wavelength is reduced, so the antenna, relative to the wavelength in the dielectric, becomes √ larger, and it becomes larger by a factor of εr . The new resonant frequency is decreased √ by a factor of 1/ εr . If we know an antenna is going to be completely embedded in a √ dielectric material, we need to scale all the dimensions by a factor of 1/ εr . What frequently happens is that tags are placed on some object. If the object is relatively thick (comparable to the size of the antenna), then the tag is surrounded in part by the object (dielectric) and in part by air. See Figure 3.25. We can assume air has a relative dielectric constant of 1. In this case, we can estimate an effective dielectric constant. The effective dielectric constant is the dielectric constant of a (fictitious) homogeneous medium that would produce nearly-identical behavior to the nonhomogeneous environment. In this case, we can simply estimate εeff ≈

εr + 1 . 2

(3.13)

This is only an estimate, which breaks down when εr becomes large, but it’s a good estimate when εr is small, which is the case in a lot of instances. Then we can substitute εeff into the previous equations and estimate the change in resonant frequency, for example. Note: if the dielectric is thin, then you should not assume equal weighting; the actual weighting is really quite complex and involves complex mathematical equations, and all but the simplest structures require complex numerical calculations that are best done by computers. A simple question to ask is then: what kind of materials gives what kind of dielectric constants? That’s actually a hard thing to catalog because materials commonly found in the environment of RFID tags are rarely in their pure form. Cardboard is mostly air, except that it readily takes on water, depending on the humidity of the air, and can have a dielectric constant between 1.2 and 1.4. Plastics like ABS are a mixture of three (or more) materials in varying proportions, but 2.8 is typical. Polyethylene is about 2.2 or 2.3, polypropylene about 2.3, polystyrene between 2.4 or 2.5, and polycarbonate about 2.8. Nylons tend to be around 3.5, but can also vary considerably because nylons readily take on water from the environment. Glass for windshields and wine bottles are a challenge because glass can have a permittivity of 6 or larger. Water, we should mention, is about 80, and thus is one of the most challenging dielectric materials for RFID. Motor oil can vary considerably, but may be around 20 to 30. However, white petroleum jelly is about 2.1.

Figure 3.25 Tag on semi-infinite dielectric. This is a common model for a tag placed on an item.

UHF RFID Antennas

83

3.4.2 Dielectric Loss The other thing about dielectric materials is that in the presence of time-varying electric fields, the material may not align its electric field exactly at the same time as the electric field is being applied. You can imagine, a water molecule may not be able to spin around 915,000,000 times a second and keep up. In fact, it doesn’t; its rotation lags the electric field a little bit. That lag introduces a complex component to the dielectric constant. The ratio of the imaginary to real is called the loss tangent, and is usually written tan δ. The larger that number, the more the material converts energy in the form of the electric field into heat, and it doesn’t have to be very large at to start causing problems. For RFID tags, this is not a good thing. We spent a lot of time looking at how an antenna can efficiently deliver power to the chip, but if there’s a lossy dielectric material near the antenna, the antenna is going to lose some of its power to the dielectric. Just how bad is it? Again, a lot depends on the details. Rarely is a tag completely immersed in a dielectric. So we again use a theoretical homogeneous material with an effective loss tangent, determined in roughly the same way as we would estimate the effective dielectric constant. Let Qd = 1/ tan δeff be the quality factor of the effective dielectric homogeneous medium. Assume our antenna has a quality factor of Qa . Then following the definition given in [5], we can find the efficiency of the antenna as shown in Equation 3.14: η=

Qa Qd Qa + Qd

(3.14)

So it turns out that one way to combat dielectric loss is to have an antenna with a small Qa . Unfortunately, as we saw earlier, a small Qa requires a physically large antenna, which usually means more cost. But otherwise, if you place an antenna on a lossy material, you will expect to see a reduction in antenna performance. Materials that have a lot of dielectric loss tend to be very polar molecules. So arcylates and vinyls tend to be very lossy. One of the most lossy materials that we’ve come across is phenolic resin, which is sometimes used to make hard, plastic dishes that you might have noticed gets very warm in the microwave. Phenolic resins have loss tangents between 0.2 and 0.4. Water is fairly common and pretty bad, and at around 900 MHz has a loss tangent of 0.05, or a Q of 20, which is about the same as acrylic plastic. But if you look at this from a purely efficiency standpoint, the antenna efficiency may only be 50%. The majority of the challenges around water come from the dielectric constant and the resulting change in the antenna impedance, and the modification to the radiation pattern (the antenna tends to radiate more energy into the water), which is why acrylic is a much easier material to work with for RFID than water even though it has a similar loss tangent. Most other plastics are fairly low loss, with the exception of Nylon (when hydrated) and most polyurethanes, which can be close to 0.1. Acrylic, as we stated is about 0.05, polyester or PET is about 0.02, and polycarbonate is about 0.012. The polyethylene, polypropylene, and polystyrene plastics are very low loss with a loss tangent of .001 or less, as is PTFE (Tefon). Now, let’s briefly consider the combined effects of the dielectric constant and loss tangent on the performance of an RFID antenna. The dielectric constant will change the resonant frequency of the antenna, which has a similar affect to operating the antenna at √ higher frequency, higher by a factor of εeff . Making a dipole assumption, this normally means the resistance of the antenna increases a little bit and the reactance of the antenna

84

RFID Systems

√ increases a lot, by a factor of approximately 2 εeff Ra Qa . That will normally result in a considerable impedance mismatch, which is the normal source of lost performance. Again, we see that Q is fundamental to the performance of tags and their sensitivity to dielectrics. In addition, any loss in the material will result in the antenna resistance decreasing and additional loss in performance as described by [15]. Note, however, that decreasing the resistance from dielectric loss might actually be a positive benefit. If reducing the resistance results in better efficiency in delivering power to the load, the antenna may lose some radiating efficiency but gain more power transfer efficiency and have a net increase in performance.

3.4.3 Metals Metals produce a completely different set of challenges for RFID. Metals can very profoundly and negatively impact RFID tag performance [8]. The best way to explain this is using image theory. You know, for example, when you look in a mirror you see a mirror image of yourself. Everything in the mirror image is identical but backward, that is, left is right and right is left. At RF frequencies, a large, flat metal surface acts as a mirror for RF signals. So if an antenna is some distance h above a ground plane, we can replace the ground plane with a mirror image of the antenna a distance 2h away (see Figure 3.26). Recall from earlier, a small dipole segment with a unit of current radiates that energy into space. If there is another small dipole segment very close by with a current exactly opposite, the electromagnetic fields produced by the two will nearly cancel. The only portion that will survive is due to the small offset in phase produced by the distance 2h. So, we can surmise that when the dipole antenna is placed near a metal ground plane, we will see the radiating resistance reduce by a factor of roughly sin2 (2πh/λ), which, for small h, reduces to (2πh/λ)2 . (Note that if 2h = λ/4, you have the condition in which the image significantly adds to the radiation in the direction normal to the ground plane.) If the tag is placed right on metal, h becomes so small that the resistance plunges to nearly zero. So placing a tag right on metal is a good way to prevent the antenna from working at all. The reactance is much more difficult to analyze rigorously, so we’ll take a more qualitative approach. The inductance of the wire antenna is slightly reduced by the presence of the ground plane, but since most of the energy is stored in the magnetic field very close to the wire, the affects are mild until the antenna gets very close. The capacitance of the antenna changes quite a bit. The antenna now stores energy in the electric field between the antenna and the ground plane. Recall from basic circuit theory that a series

Figure 3.26 Tag operating above a ground plane, shown with image.

UHF RFID Antennas

85

RLC circuit has the following resonant frequency. f02 =

1 2πLC

(3.15)

With equal or a little less inductive energy and lot more capacitive energy, we expect the resonant frequency of the antenna to be related as follows. With C increasing and L the same or decreasing only slightly, we expect f0 to decrease. That’s exactly what happens. The resonant length of a dipole will decreases from about 0.46λ to about 0.38λ or less, for example; the actual value is dependent on h, the width of the dipole, and other factors. In Figure 3.27, we show the (simulated) input impedance of a dipole above a 300 mm2 ground plane at varying heights on a Smith chart normalized to 50 Ohms. For heights less than 1/4 of a wavelength, we see both the input resistance and reactance decreasing with decreasing height. At about 6 mm, the effects of the ground plane affect the impedance in a significantly different way, causing both the resistance and reactance to move sharply to zero with decreased height. So a resonant dipole with impedance of about 70 Ohms in air, now with a 6 mm separation has an input impedance of 0.6 −j 20 Ohms – a very significant change in impedance! This is why metal is considered one of the more challenging materials to work with for RFID.

110

100

120

90

80

70 60

1.00

50

130 140

0.50

2.00

40 30

150 160

5.00

54 mm

0.20

20

170

10 0.00 0.0

180

0.20

0.50

1.00

−170

2.00

5.00

−10

100 mm

−160

−0.20

−5.00

−150

−20 −30

−140 −130

6 mm −0.50 −120

Figure 3.27

0

−110

−2.00 −1.00 −100

−90

−80

−70

−40 −50

−60

Impedance of a near-resonant dipole above a 300 mm2 ground plane.

86

RFID Systems

What can be done? Lots of things, but unfortunately, none of them are simple. In Section 3.4.5, we will investigate some alternatives. But before that, we will look at one more environmental issue affecting RFID tag performance.

3.4.4 Propagation Until now, we’ve primarily focused on the impact that a nearby material has on the antenna. What about material interposed between the tag and the reader? Depending on the kind of material and size of material, it will have varying impact on the performance of the tag. For example, consider a tag that is on a pallet of cardboard boxes. The dielectric and metallic material inside the boxes can detune the material, but if the tag is surrounded by boxes, then those interposing boxes (and material inside the boxes) can pose additional obstacles to getting RF energy in and out of the pallet. From the previous section, we can classify materials as dielectric or conductors. Let’s consider conductors first. Conductors, naturally, readily conduct electricity. The phrase commonly used in the RFID industry is that “conductors reflect,” and to a first approximation, that’s correct. However, that’s not quite the case. The term commonly used in the technical community is scattering. There are obvious differences in the connotations of the words reflecting and scattering, and those connotations are appropriate here. We usually think of mirrors as reflecting, that is, light bouncing off the surface of the mirror in a very predictable direction. Scattering is, well, bouncing off in a bunch of different directions at the same time, that is, all over the place. The description that is most apt depends on the size of the metallic object. Recall that with electromagnetism, size is always relative to the wavelength. At 900 MHz, a wavelength is roughly a third of a meter (about 1 ft). If the metallic object is large relative to a third of a meter, like, say, a large metal wall, then the RF energy will reflect off the metal wall in much the same way that light reflects off a mirror. If the metallic object is small relative to a third of a meter, like, say, another RFID antenna, then the RF energy will scatter. In this example, a nearby tag will scatter RF energy in much the same way that it radiates RF energy, according to a radiation pattern (see Section 3.1.2). In fact, two RFID antennas can couple and become something like an antenna array with an entirely different radiation pattern, and is the subject of ongoing work within the research community. So, if there is a large, intervening metallic wall, the wall will reflect RF energy in much the same way that a mirror reflects light. If there is a series of small metallic objects, some RF energy may be able to get through the space between metallic objects, and some of the RF energy may get through by scattering off of the metallic objects. Furthermore, other objects that are not intervening between the tag and reader may reflect or scatter RF energy. These multiple signals come together and may add (constructive interference) and serve to boost the signal, or cancel (destructive interference) and suppress the signal. In a complex environment, all of these factors are happening and the results can be very difficult to predict. This is one of several things that make RF and RFID behave somewhat randomly or magically to the initiated. What about dielectrics? Recall that we can characterize dielectric material through two properties: the dielectric constant, and the loss tangent. Let us first consider the dielectric constant. Recall that one of the effects of the dielectric constant is that it causes the

UHF RFID Antennas

87

velocity of RF energy to slow down. This has two important consequences. First, any time there is an abrupt change in the dielectric constant in a medium, an electromagnetic wave incident on that transition undergoes both reflection and refraction, that is, some of the energy is reflected, and some it is transmitted through the medium but at a different angle. A set of equations called the Fresnel equations precisely describe what happens. Bottom line, at an air-dielectric boundary, the larger the dielectric constant of the material, the more energy is reflected. So, for example, trying to read through a stack of (full) milk cartons (dielectric constant about 80) will be very difficult because most of the RF energy will be reflected. But reading through paper or plastics can be relatively easy, with most of the energy being transmitted and little reflected. Note that this is different from an antenna being placed very close to a dielectric material. One vendor has sort of a gimmicky trick for showing off their product. They place a tag within a fish tank (complete with water and fish) and a reader antenna very close to the tank. Most of the electromagnetic wave actually forms inside the fish tank, and thus does not experience the air–water dielectric boundary. If the vendor were to move the antenna back by half a meter or more, they would have a much more difficult time getting energy into (and out of) the fish tank. Second, the loss tangent of a material causes RF energy to propagate poorly through the material. The loss tangent is a measure of how much energy is lost to the material per wavelength. Since water is common (and particularly problematic), consider a wave that is propagating through pure water with a dielectric constant of 80 and a loss tangent of 0.05. For a relatively small loss tangent, we can approximate the loss (in dB) after propagating a distance x as shown in Equation 3.16. √ AdB (x) = x8.686k0 εr tan δ

(3.16)

Here, k0 is the free space wave number, or about 19.16 m−1 at 915 MHz. For water, this works out to 74.4 dB per meter. We can forget about using long-range RFID under water! 10 cm yields only 7.4 dB of loss, which is significant, but still possible to overcome. However, keep in mind that this component is only the propagation loss. What does this mean? As an example, putting RFID tags on one liter milk cartons is probably not a good idea. The dielectric constant can change the antenna impedance significantly, creating power transfer inefficiencies. The dielectric loss of milk will result in a less efficient antenna. Trying to read through milk cartons will introduce a sizeable reflection loss at the air–milk boundary, and if milk has a larger loss tangent than water (which is very likely), then there will be significant propagation losses. Bottles of wine offer a similar challenge, but since wine bottles are generally cylindrical, there are possibilities of waveguide structures formed by the bottles, especially if foils are used in the bottle labels. A detailed analysis of tagging wine bottles and similar challenging RF product is the subject of ongoing research and the exchange at several research conferences.

3.4.5 Practical Steps to Overcome Environmental Challenges In this section, we focus on overcoming some of the environmental challenges of RFID tags. Note that we do not address propagation challenges here; we focus only on getting the antenna to work well when placed on a challenging item.

88

RFID Systems

The simplest way to overcome the environmental challenges is to physically separate the tag from the offending material. The further away the tag, the less the impact the material has on the tag. Again, far is defined in terms of a wavelength, or in this case, in terms of the length of the antenna. For metals and liquids, the separation has to be significant or else the tag will suffer from significant performance degradation. For example, for a 3.2 mm foam spacer from a metal plate, we observe that the typical tag degrades by about 16 dB. A 4.8 mm spacer degrades performance by about 13 dB, and a 6.4 mm spacer by about 11 dB. Tags vary by a few dB, but nearly all tags degrade pretty much the same. The exception to the rule are the exceptionally large, 90 mm2 tags that perform significantly better, though they still degrade. There are several solutions for solving the metal problem. The easiest thing to try is to provide some spacing from the metal object, usually using some thick foam backing. These are inexpensive, but tend to work poorly. When those don’t work, people usually use a class of tags called “asset tags” or “metal tags” that are usually based on some kind of microstrip antenna, for example, [9]. These usually include a metal ground plane, a hardened case, and tend to be relatively expensive. On the positive side, they work well, sometimes better on metal than in air. On the negative side, they can be considerably more expensive. A second approach that has recently been developed is a hybrid approach: a special dipole antenna that can also work as a microstrip antenna. Below, we will go into more detail. 3.4.5.1 Separation The simplest approach to dealing with metal is to separate the antenna from metal. Recall the results of dipole spacing from metal in Figure 3.27. A dipole near metal will experience very significant changes in input impedance, resulting in a substantial decrease in power transfer efficiency. The larger the separation, the better the performance, though for complex reasons the relationship may not be monotonic for larger separation distances. Usually, people use a thin foam spacer made of polyethylene (PE) or polyurethane (PU). PE is popular because it tends to be the lowest cost, has a very low dielectric constant, and excellent resistance to water. PU is sometimes used because it can provide a nice, open cell “sponge” that can be easily compressed, which simplifies manufacturing, but unfortunately, sponges are also good at absorbing water. Typical thicknesses are 4.8 mm (3/16 inches), though 3.2 and 6.4 mm are sometimes used. Suppose a tag can be read in free space at a distance of 32 feet, and with a 4.8 mm spacer, the tag suffers 18 dB of performance degradation, that is, it operates at 1.6% efficiency. The read distance will be reduced by a factor of 8, meaning the tag will be readable to about 4 feet. For a number of applications, this read distance is sufficient. So sometimes, nothing more need be done. 3.4.5.2 Microstrip Antennas A microstrip antenna, see Figure 3.28, is the class of planar antennas that are designed to operate above a ground plane. Figure 3.28 illustrates a typical microstrip antenna with the associated fringing electric fields. The fringing fields are primarily where the radiation comes from. Energy is stored in the area between the antenna and the ground plane,

UHF RFID Antennas

89

Figure 3.28 Microstrip antenna fed by a microstrip transmission line, shown with fringing electric fields. L

CA

RA

CA

RA

Figure 3.29 Transmission line model of a microstrip antenna.

similar to a parallel plate capacitor. Small, thin antennas will thus have a very large Q, while thick, wide antennas will have a smaller Q. One especially simple but effective way to think of the microstrip antenna is to view it as a pair of RC circuit elements connected by a transmission line, illustrated in Figure 3.29. The resistances at the edges represent the (shunt) radiating resistance of the antenna, and the capacitors represent the energy stored in the fringing fields illustrated in Figure 3.28. Note that the resistance here is a shunt resistance in parallel with the capacitance, not the series resistance as in a dipole. Thus, the resistance represented in Figure 3.29 tends to be quite large, as opposed to small in the dipole case. We’ll draw equivalence between the two views below. The two circuit elements are connected by a microstrip transmission line of length L and has some characteristic impedance determined by the width, height, and dielectric constant of the substrate. Consider viewing the impedance of the antenna at the left radiating edge. The antenna will resonate when the impedance at the right edge, transformed by the transmission line of length L, presents the conjugate impedance to the RC element on

90

RFID Systems

CA

RA

Vs

CA

RA

CA

RA

(a) Vs

CA

RA

(b)

Figure 3.30 “Traditional” unbalanced way to feed the microstrip antenna (a) and “dipole” balanced way to feed the microstrip antenna (b).

the left. This happens when L is slightly less than half a wavelength. Typical resistance values may be 100 Ohms for a very wide, thick antenna, to 30,000 Ohms for a narrow, thin antenna. Note that the circuit model of the microstrip antenna implicitly assumes that the way we are feeding the antenna is with respect to ground, as illustrated in Figure 3.30 top, hence a conceptual bias towards viewing the antenna as an unbalanced device, that is, a signal feed with reference to ground. If we split the transmission line exactly in half and feed one side with respect to the other, as illustrated in Figure 3.30 bottom, then we are back to something that resembles a dipole. In that case, the rather large edge resistance is transformed by a (roughly) quarter wave transmission line, which will result in a small series resistance. In fact, the parallel RC circuit will be transformed to a series RC circuit. The circuit equivalent of the frequency-dependent behavior of a resonant antenna shown in Figure 3.30(a) will be that of a parallel RLC circuit, while same antenna fed as shown in Figure 3.30(b) will behave as a series RLC circuit, or qualitatively the same impedance behavior as a free-space dipole. The only difference is that the series resistance can be small, for example, less than an Ohm. The traditional challenge with microstrip antennas is that microstrip antennas are generally thought of as unbalanced devices, that is, you feed them with a signal (in Figure 3.30, a microstrip transmission line), which is referenced to the ground plane. The question has been: how do you attach the chip to the antenna? One simple way is to attach one terminal of the IC to the signal source and the second to the ground through a “via” or through-hole connection. Drilling a hole or wrapping a conductor is an extra step that can be cumbersome and adds to the cost of the tag. It may be surprising (or not) that the T-match structure, or the modified T-match, can be used to transform the antenna/IC impedance to something that is usable. Recall in

UHF RFID Antennas

Figure 3.31

91

Narrowband microstrip antenna utilizing a modification to the modified T-match.

the modified T-match, the term β is used to transform the IC resistance to the dipole resistance, reducing the IC resistance by a factor of β 2 . There is no theoretical limit to how small we can make β, although there are practical limits. If β = .01, for example, then β 2 = 10−4 , which can transform a 2200 Ohm resistor to 0.22 Ohms if necessary. If that’s not sufficient, we can repeatedly apply the T-match. In Figure 3.31 we show an antenna that is designed to produce an impedance match for an antenna separated from the metal ground plane by 1.6 mm of HDPE. The antenna size is approximately 94 mm long and 30 mm wide. The matching network is can be viewed as two applications of the modified T-match. Because of its size, the bandwidth of this antenna is very small. At another extreme, the antenna shown in Figure 3.32 uses a very simple T-match. It is physically large, measuring 100 mm wide and 138 mm long, and a 15 mm expanded polystyrene substrate. This antenna was designed using the filter theory described above to give 90% power transfer efficiency from 865 to 955 MHz.

Figure 3.32 Large, wideband microstrip antenna utilizing classic T-match.

92

RFID Systems

3.4.5.3 Microstrip Dipole Recall that a convenient perspective on the microstrip antenna is to view the antenna as a center-fed dipole with a small radiating resistance, that is, a small fraction of an Ohm (recall Figure 3.27). Remarkably, the unmodified T-match is adequate to perform an impedance match. Consider a rectangular dipole antenna with a width of about 15 mm. We vary the length of the dipole from 70 mm (well below resonance) to 320 mm (past the anti-resonance) and observe the impedance on a Smith chart. Second, we observe the same dipole placed 3.2 mm from an infinite metal ground plane. The two impedances are plotted on a Smith chart in Figure 3.33. As one can clearly see, the resistance at resonance is drastically reduced to a little more than a tenth of an Ohm. Next, consider the permissible region of the matching circuit for a T-match circuit. While α can take on any positive number in theory, there are practical limits. For simplicity, we assume that α is bound by 10. Recall that the T-match first transforms the common mode impedance by a factor of (1 + α)2 , and second transforms the impedance by a shunt reactance Zd , resulting in the circuit shown in Figure 3.11. Figure 3.34 illustrates the permissible region, that is, the region in the Smith chart in which the common mode impedance can be matched to the conjugate IC impedance using the T-match. Notice that the dipole impedance shown in Figure 3.33 will have considerable overlap within the permissible region. Next, consider the impedance of the microstrip dipole shown in Figure 3.33. Clearly, there is a small region close to the series resonance of the microstrip dipole that crosses into the permissible region. Depending on the particular IC and whether the radiating resistance is large enough, there may be a small region of overlap when the antenna is slightly longer than resonant, but that doesn’t happen for the example antenna and IC we show here.

320 mm 320 mm 70 mm

70 mm

Figure 3.33 Impedance of variable length dipole in free space (left) and 3.2 mm separation from an infinite metal ground plane (right).

UHF RFID Antennas

93

Figure 3.34 Permissible region. If ZC lies in the shaded area, then the T-match can provide a conjugate impedance match.

Figure 3.35 A prototype microstrip antenna using an embedded T-match.

It may be quite surprising that with a modest value of α and ZD , one can find a conjugate impedance match for a microstrip dipole using only the T-match. A prototype of this kind of tag antenna is shown in Figure 3.35. The resulting antenna is 100 mm long and 30 mm wide, and constructed on a 3.2 mm HDPE substrate. The width of the narrow strip region is used to determine the desired value of α, and the length of the slot is used to control ZD . Small conductive losses cause ZD to become slightly resistive, which affects Zin , and must be accounted for. The T-match again can be used as a very simple, systematic means of developing a tag that works on metal. Again, such a small antenna has a large Q and thus a narrow bandwidth. The antenna shown in Figure 3.35 has a Q of approximately 100, and thus a 3 dB bandwidth of approximately 9 MHz. At the edge of the bands, the performance can be as much as 10 dB above the performance in the center of the band. Depending on the application, this may or may not be acceptable. For some applications, it is more important for the tag to have more uniform sensitivity across the band. Because we’re starting with an antenna with a Q of 100, there isn’t

94

RFID Systems

much we can do to improve on the situation without sacrificing performance. One way to achieve broadband performance with a small, microstrip antenna is to purposely add loss to the system. One way to add loss is to use a dielectric with higher loss, or a less conductive material for the antenna. Or, we can add loss in the power transfer efficiency. To purposely decrease the efficiency to a certain level, we can rearrange the expression for power transfer efficiency (3) in terms of the antenna resistance:

 1 1 2−τ RA = RI C ±2 2 − (3.17) τ τ τ Assuming RI C = 20 Ohms and we want to reduce the power transfer efficiency to 0.5, we find that RA = 3.43 Ohms or RA = 116 Ohms. The smaller resistance value decreases the slope of the reactance, which increases the bandwidth, while the larger resistance increases the slope of the reactance, which decreases the bandwidth. Clearly we want to use the smaller resistance. By matching to RA = 3.43 Ohms, we will reduce the slope of the reactance by a factor of about 5.8, which is significant. It is difficult to predict the resulting bandwidth since there are several other factors involved, including the resonance of various circuits. However, as a rule of thumb, by reducing the efficiency by a factor of 2, we lower the Q of a high- Q antenna by a factor of 2 and increase the 3-dB bandwidth by a factor of 2. This is only a rule of thumb and relies on a number of assumptions that may not hold. However, with antennas that have very large Q, the technique of decreasing the Q of the antenna by reducing RA can be effective. 3.4.5.4 Metal-Tolerant Antennas Owing to the popularity of foam spacers, our research group at the University of Kansas started investigating whether there was something that could be done to make tags using foam spacers to work better [10]. Intuitively, we know that there is considerable design freedom to pick a particular set of antenna geometries, electrical lengths, and T-match sections. Given that, it seemed obvious that certain choices would be better than others. We suspect that most commercial antennas were not designed with near-metal performance in mind, and so it seemed likely that there was considerable room for improvement. Recall that a number of things happen when a dipole antenna is placed near metal, namely that the radiating resistance decreases drastically, the resonant frequency decreases modestly, and that the antenna Q increases substantially. Also, keep in mind that both the antenna and the matching network are affected by the presence of metal, and that in the previous section we showed that it is possible to use the T-match to perform impedance matching of a microstrip dipole. It is at least conceivable, then, that one may be able to find an antenna design that behaves as a dipole with a proper T-match in air, and as a microstrip dipole with the appropriate T-match when near metal (with a HDPE foam substrate, e.g.). Figure 3.36 shows such a tag. The tag is mounted on a 3.18 mm HDPE foam separator. We have not yet fully devised a simple design methodology for constructing such a tag, but we can present an analysis of the tag here. Because it uses a loop-based T-match, we use the distributed circuit model to describe the antenna and matching circuit behavior both in air and when on metal with a 3.2 mm HDPE foam substrate (εr = 1.095). The

UHF RFID Antennas

95

(c) U. Kansas 2007

Figure 3.36

Patent Pending

A combined dipole/microstrip antenna.

T-match

Antenna 20.0 Ω

1.152 pF

13.70 nH

5.56 nH

Voc

IC

14.43 nH

1.3 pF

1500 Ω

Figure 3.37 Circuit model of combined dipole/microstrip antenna functioning in air.

T-match

Antenna 0.348 Ω

Voc

1.908 pF

11.20 nH

IC

12.13 nH

3.351 nH

1.3 pF

1500 Ω

Figure 3.38 Circuit model of combined dipole/microstrip antenna functioning on metal.

circuit-equivalent models for the antenna in air and on an infinite ground plane are shown in Figure 3.37 and Figure 3.38 respectively. Note that the resistance in air of 20 Ohms reduces considerably to a fraction of an Ohm on metal. For the matching circuit, we note that both the shunt and series inductor reduces when placed on metal; the inductance per unit length is partially cancelled by the presence of the ground plane, turning the traces into transmission lines. Next, let’s look at the input impedance as a function of frequency. In air, the impedance is given in Figure 3.39(a), along with the conjugate IC impedance. We see that the antenna doesn’t quite achieve a conjugate impedance; the reactance lines cross at about

96

RFID Systems

500

150

100 Rin Xin Ric −Xic

50

Resistance / Reactance (Ohms)

Resistance / Reactance (Ohms)

450 400 350 Rin Xin Ric −Xic

300 250 200 150 100 50

0 860 870 880 890 900 910 920 930 940 950 960

Figure 3.39

0 8 080

900

910

920

F (MHz)

(a)

(b)

930

940

950

Impedance of combined dipole/microstrip antenna in air (a) and on metal (b).

1

0.5 0.45 Power Transfer Efficiency

0.9 Power Transfer Efficiency

890

F (MHz)

0.8 0.7 0.6 0.5 0.4

0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05

860 870 880 890 900 910 920 930 940 950 960 F (MHz) (a)

0 880

890

900

910 920 F (MHz)

930

940

950

(b)

Figure 3.40 Power transfer efficiency of combined dipole/microstrip antenna in air (a) and on metal (b).

910 MHz and the resistances match at 920 MHz, but overall the antenna achieves an excellent impedance match. On metal, it’s a very different story (see Figure 3.39(b)). As expected, the resistance is dramatically reduced, and the reactance changes very rapidly with frequency, especially above resonance. Clearly, we have a very large Q antenna. Indeed, from the circuit values we estimate a Q of about 250. Next, let’s translate the impedance information into power transfer efficiency. The efficiency is plotted in air and on metal in Figure 3.40(a) and (b). We see excellent performance in air, though not perfect, and certainly not wideband. On metal, however, we see the efficiency peak at about 50% efficiency, but the bandwidth is extremely narrow. This should be expected with such a large Q, and is about the best one can do. Achieving a higher peak performance would result in an even narrower bandwidth. (Indeed, performance and bandwidth are often traded off.)

UHF RFID Antennas

97

How does the tag perform? The data sheet for the IC indicates a minimum power requirement of −14 dBm (39.8 milliwatts). For a reader at full power (30 dBm and 6 dBi transmit antenna) and a 3 dB polarization loss (assuming the reader is transmitting a circularly-polarized wave and the tag is linearly polarized), both the reader and tag antenna are optimally aligned, the maximum free-space read distance for a 2 dBi tag is 7.36 meters (24.1 feet). Environmental effects may add or subtract from that number. On metal, the directivity of the antenna is enhanced from 2 dBi in the dipole case to about 8 dBi. The antenna efficiency (primarily due to conductive losses) is reduced by about 6 dBi, so the resulting gain is back to 2 dBi. That is further reduced by about 3 dB by the peak power transfer efficiency, but at the edge of the bandwidth, can be reduced by 8 to 10 dB. So the effective gain of the tag (including power transfer efficiency) ranges between −1 and −11 dBi. Under similar assumptions, the free space read distance will vary between 5.2 meters (17.1 ft) and 1.6 meters (5.4 ft). In the real world, such a narrow bandwidth antenna poses interesting behavior. Recall that the EPC protocol (and most country regulations) require the RFID reader to frequency hop. Every 0.4 seconds, the reader will hop to a new frequency according to a pseudorandom hopping sequence. In the 915 MHz band in North America, there are 50 channels, so it takes 20 seconds for the reader to cycle through all 50 channels. A narrow-band RFID tag will be seen with decreasing frequency at increased distance. The reader may take 7 seconds on average to see the tag at the furthest read distance, may take only one or two seconds at half the furthest read distance, and will always see the tag at about one fourth of the furthest read distance. What is acceptable is entirely a function of the application. For many asset-tracking applications, a read distance of 7 to 10 feet is sufficient, and the reader takes 2–3 seconds to scan the area, so a narrowband tag is quite acceptable, and in fact this hybrid dipole/microstrip antenna may be “overkill.” If the tag is passing through a large portal with other losses (perhaps lossy dielectrics interposed) where the highest tag performance is necessary and at high speeds where the tag remains in the read field for only a fraction of a second, then the narrowband tag won’t prove adequate. For those applications, a larger tag with a wider bandwidth response is necessary. In practice, those applications are relatively rare, and the market seems to readily adopt narrowband RFID tags for asset tracking.

3.5 Conclusions, Trends, and Challenges We’ve covered a lot of topics in this chapter, starting with some of the basics about how antennas work, to advanced topics and leading-edge research. RFID antennas coupling to reactive RFID chips and using the T-match and modified T-match circuit offer a new host of possibilities of what can be done, if they are well understood. For the majority of RFID applications, price is a dominant factor for the tag, so small, short, printed dipoles are the norm. We covered some of the basics about how energy is stored and dissipated in the antenna, and different ways to minimize the antenna Q. Next, we examined the T-match and modified T-match, the circuit equivalent, and how using the two and some simple filter theory we can develop wideband RFID tags. Next, we put all those concepts together and designed our own antenna for a set of hypothetical design requirements. Finally, we looked at the issues involved for RFID tags working in various

98

RFID Systems

environments. We looked at what the environment fundamentally does to those antennas, and what we can do to mitigate the impacts. We’ve seen that the future of RFID is likely to include tags that work much better near metal, and while we are not aware of any tags that work well near water, we see no reason why they could not be developed. Hopefully, the reader has come away with an appreciation for how antennas are designed, what the various design elements are, what trade-offs are involved, how the design elements may be composed, and what some of the fundamental limits are.

References [1] Marrocco, G. (2008) The art of UHF RFID antenna design: impedance-matching and size-reduction techniques, IEEE Antennas and Propagation Magazine, 50(1): 66–79. [2] Rao, K.V.S., Nikitin, P.V., and Lam, S.F. (2005) Antenna design for UHF RFID tags: A review and a practical application, IEEE Transactions on Antennas and Propagation, 53(12): 3870– 3876. [3] Balanis, C. (2005) Antenna Theory, 3rd edn. Chichester: John Wiley & Sons, Ltd. [4] Nikitin, P.V., et al. (2005) Power reflection coefficient analysis for complex impedances in RFID tag design, IEEE Transactions on Microwave Theory and Technique, 53(9): 2712– 2725. [5] Chu, L.J. (1948) Physical limitations of omni-directional antennas, Journal of Applied Physics, 19: 1163– 1175. [6] Wadell, B.C. (1991) Transmission Line Design Handbook . Wilmington, DE: Actech. [7] Deavours, D.D. (2009) Analysis and design of wideband passive UHF RFID tags using a circuit model, in IEEE RFID, Orlando, FL. [8] Aroor, S.R. and Deavours, D.D. (2007) Evaluation of the state of passive UHF RFID: An experimental approach, IEEE Systems Journal , 1(2): 168– 176. [9] Ukkonen, L., et al. (2005) Reliability of passive RFID of multiple objects using folded microstrip patchtype tag antenna, in IEEE Antennas and Propagation Society International Symposium. 2005. [10] Mohammed, N.A., Sivakumar, M., and Deavours, D.D. (2009) An RFID tag capable of free-space and on-metal operation, in IEEE Radio and Wireless Symposium, San Diego, CA. [11] Dobkin, D.M. (2007) The RF in RFID: Passive UHF RFID in Practice. Oxford: Elsevier.

4 RFID Tag Chip Design Na Yan1 , Wenyi Che1 , Yuqing Yang1 , and Qiang Li2 1

Fudan University, China

2

Quanray, China

This chapter will explore the fundamental theory of RFID tag chip design in detail, including RF/analog front end, baseband and non-volatile memories. The other focus of this chapter is to combine layers of a UHF RFID and analyze the energy/signal transmission theory applied to RFID. Some low-power and low-cost design techniques will be characterized.

4.1 Tag Architecture Systems The tag system consists of RF/analog frontend, non-volatile memory and digital baseband sections. The functions of every module and design specifications of the system have been described from the standpoint of system architecture. High efficiency frontend design includes: (1) rectifier; (2) power (voltage) regulator; (3) demodulator; (4) clock extraction or generation; (5) backscattering; (6) power on reset; and (7) voltage (current) reference, etc. The emphasis of Section 4.1.1 is on rectifier design.

4.1.1 Tag Architecture An RFID tag is a low-end wireless communication device, however, it has several features that distinguish it from other low-end wireless devices. The low fabrication cost is mandatory for applications such as logistics, tickets and personal identification purposes. Chip designers have to use a process comparable to the standard CMOS process to reduce extra masks and abolish almost all the complicated circuits to save the chip size. On the other hand, power supply for the tag system is harvested from the reader’s interrogating RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

100

RFID Systems

Matching Network

Antenna

RF PAD

ESD

RF PAD

Demod

Rectifier

Modback

Clock Generator

RNG Vdd

Regulator & Bias

POR Generator

Baseband & EEPROM

RFID Tag IC

Figure 4.1 Tag system architecture.

electromagnetic (EM) field. Such characteristics limit the use of power-dissipative circuit modules, such as power amplifiers and high-speed high-precision A/D converters, frequency synthesizers, and so on. For uplink (R-T) communication, the tag changes its complex input impedance to backscatter the reader’s interrogating power. For downlink (R-T), the envelope of the input RF carrier is extracted and converted to digital signals by an incoherent detector. Figure 4.1 depicts a typical tag system. The matching network is to provide maximum power transmission from the antenna to the tag system. The modback circuit consists of several passive devices such as capacitors and switches. It performs the function of load modulation at the simplest circuit. The rectifier converts the input RF power to a DC voltage to power the whole system. Since the input RF power of the tag will vary a lot along with the change of the distance between reader and tag, a regulator is incorporated in the tag system to provide stable voltage supply. A bias circuit provides voltage and current references for the whole system. The clock signal for the digital baseband is generated by a low power oscillator, such as a ring oscillator. The demod module detects the input R-T signal [1]. A power-on reset (POR) circuit detects the rectified DC voltage level to estimate whether the total power is sufficient to supply the whole tag or not. The POR signal is used to set up the digital baseband which will process the input signal and control the logic operation of the whole tag system. EEPROM is also an essential part of the tag circuit, because it stores the data when the tag is not powered up. The passive operating manner of the tag makes it necessary to add a large storage capacitor CS to smooth the supply voltage in case of the interrogating energy gap.

4.1.2 Design of High Efficiency Frontend Circuits Power consumption is the primary limitation when designing the frontend circuitry of a tag. Let us take the Alien Higgs-3 tag as an example. Its reported sensitivity is −18 dBm

RFID Tag Chip Design

101

[2], which actually means that the overall power consumption during the whole operation period should not exceed 16 µW. Even if the tag’s rectifier reaches 50% PCE (power conversion efficiency), the tag’s operating power needs to be lower than 8 µW. This is really stringent requirement for any IC designer. In order to design a tag with such a small power budget, many design methods have been proposed in recent years [3]. These technical contributions can be divided into two categories: (1) to lower down the power consumption of the tag’s processing circuits; and (2) to increase the rectifier’s PCE. 4.1.2.1 Low Power Frontend Circuit There are two reasons for designing low power circuits: (1) to save the overall energy in a long time span; and (2) to decrease the transient power consumption. Most portable devices use the first reason because their design target is to prolong the battery life. However, low power RFID tag design follows the second reason so that the reader transmits EM energy continuously to power on the tag. Therefore, a sporadic large current is forbidden during the tag’s operating period. Energy used to supply the tag system needs to be scattered over the whole time span. Low voltage design in tag system has stimulated a great deal of interest in recent years. The sub-1 V technology trends [4, 5] have made it possible to enhance the system performance of RFID systems. Unlike the other IC design, there are two different supply voltage levels in the tag’s frontend design. Figure 4.2 shows the supply voltage change during the “frame-sync” period of EPC Gen 2 protocol [6]. When the ASK modulation signal is at a high level, the tag senses the modulated carrier and the supply voltage increases. When the ASK modulation is at a low level, the opposite process takes place. Before time t1 , Vdd is at a level of Vop , which represents the operating voltage. During the period of DELIMITER, Vdd drops to a lower level labeled as Vmin , which should be higher than the tag’s minimum supply voltage requirement. The I–V relationship in this period is: t2 Cs (Vop − Vmin ) = I (t) · dt (4.1) t1

where I (t) is the transient current of the whole tag system. Moreover, Vop is a function of the operating distance: Vop · Iop ∝ 1/d 2

(4.2)

Where, Iop is the operating current and d is the operating distance. Equations (4.1) and (4.2) indicate that a low Vmin not only increases the operating distance, but also decreases the chip size by reducing the storage capacitor. With a systematic view in designing the frontend circuitry, different building blocks have different Vmin according to their operating speed. In order to have a global low voltage performance, multi-level supply voltage generation appears to be a good solution [7]. The essence of this method is to combine building blocks with similar supply voltage requirements together as a group, and to generate different voltage levels for different circuit modules. Each group has its optimal supply voltage level and operates independent of the others. Figure 4.3 presents a typical circuit architecture of a tag system using

102

RFID Systems

Delimiter

Data 0

R=>T Calibration

Modulation signal

ASK Modulated carrier Vop Supply voltage Vmin t1

t2

Figure 4.2 The “frame-sync” sequence of the Gen 2 protocol.

Vdd1 Rec1 CS1

V&I References

Clock Oscillator

Power-On Reset

Vdd2 Rec2 CS2

Baseband

Vdd3 Rec3 CS3

Figure 4.3

EEPROM

Multi-level supply voltage generation.

multi-level supply voltage generation. Three rectifiers named Rec1-3 are employed to drive different building blocks. Analog circuits such as voltage and current references, clock oscillator and power-on reset have the same supply voltage of Vdd1 which is generated by Rec1, digital baseband uses Vdd2 from Rec2 and the EEPROM is driven by third rectifier Rec3. Storage capacitors are also divided into three segments named Cs1−3 for different supply voltages. In this way, each building block of the tag system can be optimized separately and the tag system has the lowest supply voltage requirement as a whole. Another way to reduce power consumption is to use proper power management. Since all the functional blocks are not necessarily working in the whole operating period of the tag system, supply paths of each functional block can be turned on and off properly to have minimum power consumption averaged in time. For instance, when the tag is in the process of demodulating the signals, the modulation blocks might be shut down to save

RFID Tag Chip Design

103

power and when the tag is transmitting signals, the demodulator might be shut down and vice versa. Some functional blocks like RNG (random number generator) can be turned on at the power-on period. This generates several bits of random number seeds before the whole digital baseband is set up. When the whole tag is power on, the RNG is turned off. The power management method saves total power consumption at the cost of more logic control devices. Besides, the power management circuit also consumes power and it has to be turned on all the time. Therefore, trade-offs need to be carried out to gain the optimal complexity of power management logic. 4.1.2.2 High Efficiency Rectifier The rectifier is an essential component for all types of RFID tags. If PCE (η) can be improved from 17% [8] to 23.5% [9] with a constant load power of Pload , the necessary power threshold of tag Ptag,th would be decreased by 38%. That is why increasing the rectifier’s PCE has always been a hot topic in the RFID area. In this section, some common techniques, which enhance the efficiency of the rectifier, are introduced. Ptag,th = Pload /η

(4.3)

The remotely powered characteristics of the RFID tag require the rectifier to be operated with limited input RF signal amplitude, which is often at the level of several hundred mV. In order to generate a 1 V supply voltage for the whole system, almost all the tag designers use voltage-boosting rectifier circuits [9]. An AC–DC charge pump is a common choice. Most of today’s AC–DC charge pump circuits are based on the fundamentals of Dickson’s structure [10]. Figure 4.4 shows the circuit architecture. From Figure 4.5, it is obvious that the PCE of Dickson’s charge pump is determined by all the parameters such as stage number, or process related parameters such as the threshold voltage of the diode (in CMOS circuit design, the diodes are often replaced by diode-connected MOS transistors for the sake of the chip area and the small turn-on voltage), the loading current and the output voltage level.

N

4 3

2

Vout

1

Vin

Figure 4.4

An N-stage Dickson charge pump.

104

RFID Systems

Matching network

Rectifier efficiency Read range Stage number Received energy

Operating bandwidth

Process related parameters Output voltage & current load

Figure 4.5 The factors affecting the power conversion efficiency of the charge pump.

Among all these factors determining PCE, the threshold voltage is found to be the crucial one. Therefore, Schottky diodes and zero thresholds MOS transistors are widely used in conventional charge pump circuit because their threshold voltages are much smaller than those of the common diode or MOS transistor [3, 11]. However, to integrate Schottky diodes in a tag, one needs to have more masks in the fabrication process which upsets the idea of low cost. On the other hand, zero threshold MOS transistors are sensitive to process variations and have the drawback of large leakage current. This situation promoted the invention of threshold compensation techniques with standard CMOS devices. Some recent works on charge pump design are illustrated below to explain the threshold compensation method. Figure 4.6 illustrates an N stage self-threshold compensated charge pump circuit. A detailed circuit diagram of each stage is depicted in the cloud chart [12]. The gate of the NMOS transistor Mn is controlled by Voutdc which is the generated high DC voltage, and the gate of the PMOS transistor Mp is controlled by Vindc which is lower than the output. Such gate connections make both transistors operate in the linear region during the forward biasing period. The voltage drop on each transistor can be far less than its threshold. However, the constantly turned on manner of the transistors has also some drawbacks. Leakage current from Voutdc to Vindc is not negligible during the reverse biasing period of the circuit. Figure 4.7 shows another topology of a threshold compensated charge pump circuit which has a balance in forward voltage drop and reverse leakage current [13]. The basic concept of this circuit is that it employs a constant current source to bias a diode-connected MOS transistor Mb and uses its gate-source voltage VGSb to compensate the threshold of pass transistor Mi . The combined effect of forward voltage drop and reverse leakage current can be minimized by carefully setting the value of VGSb . Another merit of this circuit is that it introduces two transistors, namely Ms1 and Ms2 , at each stage to shift the substrate voltage of Mi and thus eliminates the body effect which degrades the PCE in conventional charge pump circuit. It should be noted that in Figure 4.7, complementary blocks, namely Cell_P and Cell_N are applied for odd and even stages respectively. The

RFID Tag Chip Design

105

VinRF

Vindc

Voutdc Mn

Mp

Stage-N VinRF Voutdc Vindc Stage-2 VinRF Voutdc Vindc Stage-1

Vout

VinRF Voutdc Vindc Vin

Figure 4.6

Charge pump circuit with a self-threshold compensation.

difference of Cell_N and Cell_P is that Cell_N uses NMOS transistor as Mi and Mb which are shown in the cloud chart. Compared to the self-compensated and constantly compensated charge pump circuits which use a static voltage level to bias the gate of the pass transistor, Figure 4.8 shows a charge pump circuit which uses dynamic voltage levels for gate control [9, 14]. The symmetric structure of the circuit makes the differential RF inputs of each stage (denoted by VRF + and VRF − ) have same amplitude and opposite phase from each other. In this way, during the operating period, only one pair of transistors is turned on at one time instant. For example, when VRF + >(Vindc + Voutdc )/2 and VRF − < (Vindc + Voutdc )/2, Mn2 and Mp1 are turned on and the other two transistors are turned off. The complementary on/off behavior of the transistor pairs minimizes the forward voltage drop on pass transistor and the reverse leakage current at the same time. Table 4.1 lists some of the recent works on AC–DC charge pump circuits. Though the reported PCEs differ a lot from one another, almost all of these works utilized the threshold compensation method. It should be noted that all of the circuits with incident power less than −10 dBm used the differential structure, while the singled-ended structure was mostly chosen for incident power larger than −10 dBm. This is because that circuit using differential structure gains twice the incident RF voltage amplitude than those using single-ended structures. Such a method makes the differential charge pump structure a promising candidate for future tag design, which requires a better sensitivity performance than today’s state of the art of −18 dBm.

106

RFID Systems

Ibias Mb Mi

Cp

Vin

Vout

Ms1

Ms2

Stage-N Cell_P Vin Vout Ibias Vout Ibias Cell_N Stage-2 Vin Stage-1 Cell_P Vin Vout Ibias

Vout

Vin

Figure 4.7

Charge pump circuit with constant threshold compensation and substrate shift.

4.1.2.3 Random Number Generator Information security of RFID systems is now becoming a crucial factor in a number of applications such as personal identification and military cargo management. Therefore, a great deal of attention has been given to security-related research on cryptographic circuitries and algorithms. Since random numbers are employed in these algorithms as security keys, their randomicity performance determines the overall security level. There are several special requirements for RNG circuitries, which can be implemented in tag systems. These requirements and some real RNGs are discussed in the following section. For the sake of flexibility and small chip size, pseudo random number generators (PRNGs) are widely used in conventional tag systems. However, PRNGs are mostly digital state machines which have repetitions of outputs with constant cycles. Such a feature makes them susceptible to attacks. For this reason, implementing a low power truly random number generator (TRNG) into a tag has become a new challenge for RFID. According to the random number generation mechanism, there are three main categories of TRNGs hardware: (1) direct amplification of noise using a wideband high gain amplifier; (2) sampling of a high frequency oscillator with a jittered low frequency oscillator; and (3) discrete time chaos systems using analog signal processing technique. Though

RFID Tag Chip Design

107

Mn1

VRF +

Mp1

Vindc

Voutdc

Mn2

VRF -

Mp2

Stage-N

Stage-1 VRF + Vin

Vindc

VRF +

Voutdc VRF -

Figure 4.8

Voutdc

Vindc

Vout

VRF -

Symmetrical charge pump circuit with dynamic threshold compensation.

Table 4.1 Recent work on AC–DC charge pump circuits. Authors [16] [9] [9] [9] [12] [13] [14]

Structure

PCE (%)

Incident power

Process

Differential Differential Differential Differential Single-ended Single-ended Differential

36.6 3.2 23.5 16.7 29 36 66

−10 dBm −12.1 dBm −20.7 dBm −22.2 dBm −9.9 dBm −9.5 dBm −12 dBm

0.35 µm FeRAM CMOS 0.5 µm standard CMOS 0.18 µm standard CMOS 0.18 µm standard CMOS 0.35 µm standard CMOS 0.18 µm standard CMOS 0.18 µm standard CMOS

the first method is easy to realize, it requires a large gain and wideband amplifier which consumes lot of power. In the second category, the amplified noise will affect other parts of the circuit and thus brings the problem of physical isolation. The third method requires complicated analog circuitry with large current consumption [15]. Such drawbacks make these two methods unacceptable for tags. With careful analysis and trade-offs among chip size, the second method of oscillatorbased TRNG is now being accepted by researchers for its power consumption and output bit rate. It consumes less power, shows good quality against 1/f noise and insensitive to process variations [15]. Figure 4.9 is an ideal model to illustrate the operating principle of oscillator-based TRNG. In the circuit, white noise is delibaretely affixed to a low frequency clock to jitter its rising and falling edges. A fast oscillating clock is sampled by the jittery slow clock to generate serial random bits as seeds for post-digital processing. Since the edges of the jittery slow clock are unpredictable, the output of the circuit is truly random. There are several approaches nowadays to implement an oscillator-based TRNG into a tag and main difference is in the way to select the fast clock and noise source.

108

RFID Systems

Fast Clock

Sampling Circuit

Slow Clock Generator

Post Digital Processor

Output Random Bits

Jittery Slow Clock White Noise

Figure 4.9

An oscillator-based TRNG.

Fast Clock According to the basic mechanism of the circuit, the standard deviation of the slow clock’s jitter σ (Tslow ) needs to be several times of the period of fast clock Tfast . The higher the ratio between σ (Tslow ) and Tfast , the higher the output randomicity will be. The optional source for a fast clock in a tag includes the input RF carrier and the system clock for baseband. To exploit the input RF carrier is a straightforward method and it is the signal with highest frequency in tag [15]. The circuit works well for tags operating in HF band but encounters some difficulties for UHF band. The 900 MHz UHF carrier presents difficulties in designing the sampling circuit. Extra power consumption and hardware overhead are required for accurate sampling [17]. Using the system clock as the fast clock is a feasible way for the UHF tag [18]. Since the prevalent system clock frequencies of today’s UHF tags are from 1.28 MHz to 2.56 MHz [19], the sampling circuit can be realized by a T flip-flop, which is simple to be realized and consumes no static power. Noise Source Common noise sources in IC to jitter the slow clock are to use the channel noise of MOS transistors [17] and the thermal noise of resistors [15, 18]. Thermal noise of resistors is a good choice as its noise spectrum density is almost white in the entire frequency span. The ideal noise of resistors excludes the low frequency flicker noise which will contaminate the output randomicity. The drawback of using resistors as a noise source is that resistors occupy a larger chip area and cannot be fabricated in digital CMOS process. On the other hand, MOS transistors are smaller than resistors in size but their noise spectrum is not uniform. Such a feature makes the channel noise of MOS transistors a good option in designing TRNG for anti-collision purpose rather than security. Table 4.2 lists some recent work on TRNG circuits for RFID tags. The level of randomness is represented by low, medium and high as specified by the pass rate performances of the NIST standards [21] and are concluded by randomicity levels of low, medium and high. As can be found in Table 4.2, all designs of TRNG circuits are making trade-offs in power consumption, output bit rate, chip size and randomicity performance. There still remains a big challenge for tag designers to design a low power high randomicity TRNG circuit with compact chip size.

RFID Tag Chip Design

109

Table 4.2 Recent work on TRNG for RFID tags. Author [17] [18] [15] [20] [20]

Power consumption

Bit rate

Chip size

0.528 µW 1.04 µW 2.3 mW 180 µW 2.92 µW

320 kb/s 40 kb/s 10 Mb/s 50 kb/s 0.5 kb/s

0.0056 mm2 0.05 mm2 0.0016 mm2 1.49 mm2 0.031 mm2

Randomicity low medium high high high

4.2 Memory in Standard CMOS Processes In RFID tag chip, the embedded non-volatile memory (eNVM) is used to store the related essential information such as the electronic product code (EPC), the expiration date, the security passwords, the tag manufacturer information, and possibly user data. The memory organization of the tag defined in ISO18000-6C protocol is shown in Figure 4.10, which is redrawn from ISO 18000-6C. The capacity of the eNVM ranges from 96 bits to 2 Kbits. Comparable to the other memory applications, the capacity is very small and the memory area is still limited by peripheral circuits, such as the decoder, the latch, and the high voltage generator for the write/erase operation. In the past couple of years, conventional EEPROM memory has been widely used in RFID technology. However, the conventional EEPROM process requires additional mask steps and a long manufacturing cycle. So the small memory area results in high manufacturing costs. Obviously, it is necessary to find a low-cost, high-performance and reliable solution for embedded non-volatile memory.

4.2.1 Why Have a Standard CMOS eNVM? Several memory solutions have been proposed recently to replace the EEPROM. Ferroelectric memory has been used as RFID tag memory [22, 23], because of the following advantages: no requirement for high voltage for the write operation, fast write cycles and 10-year data retention. Unfortunately, fabrication process modifications and additional mask steps are still required to produce FeRAM. Therefore, if one takes the cost and chip yield into consideration in a low-cost design, FeRAM is not the best solution. One time programmable (OTP) non-volatile memory is compatible to standard CMOS process and does not require an additional mask layer [24, 25]. In the factory, the stored information is written into OTP memory one time, and then locked forever. The user can not rewrite/modify data of memory and therefore, the application of OTP memory in RFID systems is limited. Resistive RAM (RRAM) stores information according to the electrical resistance of certain materials by applying an external voltage or current, instead of storing information in transistors (flash memory) or capacitors (DDRAM) [26]. RRAM is non-volatile, and its simple structure is ideal for future generations of CMOS chips. It requires no extra mask layers, but few process modifications. The type of memory has two states: high resistive and low resistive to be denoted as “1” and “0.” The low resistive state needs more current consumption than the high resistive state, which is still a challenge for low power tag chip design. The related researches and verifications are still under development and not mature.

110

RFID Systems

MSB

LSB ...

10h

TID[15:0]

00h

TID[31:16]

MSB

LSB USER

Bank11

TID

Bank10

EPC

Bank01

RESERVED

Bank00

EPC[15:0] ...

20h

EPC[N:N-15]

10h

PC[15:0]

00h

CRC-16[15:0]

LSB

MSB ...

30h

Access Passwd[15:0]

20h

Access Passwd[31:16]

10h

Kill Passwd [15:0]

00h

Kill Passwd [31:16]

Figure 4.10

h

Logical memory map. Redrawn from ISO 18000-6C.

Along with the development of the standard CMOS process, the characteristic dimension of transistors is shrinking; the thick gate-oxide of transistors is becoming thinner and thinner. Consequently, the needed voltage for tunneling is decreasing. That is the reason why eNVM can be implemented in standard CMOS process and be a better choice in RFID tag chip design. Standard CMOS eNVM brings two advantages: (1) shorter time to market; and (2) cost advantage over traditional eNVM. Now it has already been used in RFID tag chip products (Impinj Monza).

4.2.2 Basic Cell Structures and Operation Mechanisms Aiming at low cost and low power, a single-poly nonvolatile memory (SPNVM) in standard 0.18 µm CMOS processes has been developed [27], which is compatible with the standard CMOS process. A basic single-end SPNVM cell is shown in Figure 4.11. By storing a different amount of charges on a floating gate denoted as FG1, a different voltage can be established. As a

RFID Tag Chip Design

111

V

M1

M1c

M1t FG1

I1

M1s Row V1t

V1c

Figure 4.11 A basic single-end SPNVM cell structure.

result, it is feasible for the sense-amplifier to discern the logic value stored in the memory cell by comparing I1 with reference current Iref . Figure 4.12(a) illustrates the key part of the SPNVM cell. It is composed of four transistors: a tunneling transistor M1t , a coupling transistor M1c , a reading transistor M1s , and a select transistor M1s . There is a common end between tunneling and coupling transistors, which act as a floating gate denoted as FG1. The physical size of coupling transistor M1c is much larger than that of tunneling transistor M1t . M1t and M1c are connected as capacitor transistors. Thanks to the capacitive divider, a large fraction of the voltage, which is enough to induce the FN tunneling at M1t , will be applied. The coupling transistor and the tunneling transistor act as the control capacitor and the tunneling capacitor respectively. Figure 4.12(b) shows the cross-section of the tunneling transistor and the coupling transistor. The source and drain area are connected to N-well. The gates must be connected together by poly-silicon, not by metal. Table 4.3 illustrates a set of operation states and the applied voltages at corresponding nodes for the single-ended SPNVM. During the program operation, V1t is pulled down to the ground while V1c is connected to the high voltage Vhh generated from the charge pump, while the readout circuit is turned off (V = 0V , Row = Vdd ). Due to the high electric field intensity established across the oxide dielectric of M1t , electrons tunnel from M1t ’s bulk to the floating gate and decrease the floating gate voltage. During the read operation, both V1t and V1c are pulled down to the ground, while the readout circuit is power-on (V = Vdd , Row = 0V ). Then the readout current I1 is generated and sensed by the sense-amplifier. Let us compare the cost of a tag chip between the conventional EEPROM process and the standard CMOS process, based on SMIC 0.18 µm EEPROM/Standard process respectively, to exemplify the cost advantage of standard CMOS eNVM, as shown in Table 4.4. Suppose the two tag chips are the same, except the EEPROM circuit and the tag chip area is 0.5 mm2 , including the 512-bit memory cell. In the SMIC 0.18 µm

112

RFID Systems

Vc Vc Mt

Cc

Mc Vt FG FG Ct

Vt

(a) Mt

Mc

Vt

Vc FC

N+

P+

N Well

P+ Oxide

P+ Oxide

P+

Oxide

N+

N Well P-Sub (b)

Figure 4.12 Schematic (a) control capacitor Mc and tunneling capacitor Mt; (b) cross-section.

Table 4.3 Operation states versus applied voltages. Operation

V1t (V)

V1c (V)

V(V)

Row(V)

Program Erase Read

0 Vhh 0

Vhh 0 0

0 0 Vdd

Vdd Vdd 0

EEPROM process, the EEPROM cell is 3.96 µm2 . If we define the proportion of control capacitance to tunneling capacitance as 9:1, then the maximum area of SPNVM cell is about 110 µm2 . The comparative results show that the total costs of the tag chip in the standard CMOS process reduce to 80.9% of the EEPROM process. The limitation of a standard CMOS NVM bit cell is a larger memory cell size compared with the conventional EEPROM bit cell size. However, it is not a severe constraint in RFID technology since the required data storage capacity is small and the capacity of 512-bit is enough in most cases. Other attempts have been reported in this regard [28–31]. Pesavento et al. [28] reported an eNVM used in RFID chips in 0.25 µm standard CMOS processes by applying a

RFID Tag Chip Design

113

Table 4.4 Comparison of two tag chips with different processes. Process (SMIC 0.18 µm)

EEPROM

CMOS

Area of memory bit cell Number of mask layers Number of photolithograph Memory chip/tag in area Cost increase due to area Cost increase due to additional masks Total cost increase of tag chip

3.96 µm2 28 29 0.41% 0 0 0

110 µm2 19 19 11.3% 10.89% −30% −19.1%

differential cell structure. Ohsaki et al. [29] used an nMOS transistor as a tunneling capacitor. However, it has been proved that a pMOS capacitor has better data retention than a nMOS capacitor due to its higher insulator energy barrier. In [30], each bit cell has its own localized sense amplifier and switching circuits. Although it offers better reliability, the bit cell is too complicated and thus is too large and consumes too much power. Kee-Yeol et al. [31] used a MIM capacitor as a coupling capacitor, but this reduces the memory retention time due to leakage from the poly contact and the metal interlayer dielectric.

4.2.3 Memory Architecture and Peripheral Circuits Figure 4.13 presents the general architecture of eNVM. For eNVM memories with small capacity, the area is mainly occupied by peripheral circuits. The memory includes the following modules: (1) input/output buffers, which buffer address/data signals; (2) a sync module, in which the address/data signals are synchronized with the clock; (3) a decoder including a block/word/bit decoder, used to select the operated memory cell; (4) a charge pump, which generates the high voltage required for write operation; and (5) a sense amplifier that senses the state of a bit line and amplifies the signal. Considering the power consumption/read speed requirements for RFID applications and the intrinsic characteristics of the memory cell, eNVM memory is always designed to have parallel input and serial output. Namely there is only one sense-amplifier needed, since the data is read out one bit by one bit. The sense amplifier (SA) is one of the most critical peripheral circuits in memory design and it function is related to the read access time. It is used to retrieve the stored data from memory, by amplifying small signal variations on the bit-lines. In general, increasing the memory capacity will result in increasing parasitic capacitance of the bit-line, consequently lowering the read speed of the sense amplifier and enhancing the average power consumption. There are mainly two kinds of architectures for SA, voltage-mode sense amplifier and current-mode sense amplifier shown in Figure 4.14(a) and Figure 4.14(b), respectively. In the case of low power supply, current-mode architecture is usually used for high speed and low power consumption because the parasitic capacitance on the bit-lines is very large. Most of the current-mode SAs are based on cross-transistors [32, 33]. On the other hand, high write voltage, for example, which is above 14 V in SMIC 0.18 µm EEPROM process, need to be generated on the chip for write operation. Most

114

RFID Systems

Date Latch

Charge Pump

Storage Cell

Word Line

Bit Line

Block Line

Block Line Driver

Decoder

Input Buffer

Sync Module

Word Line

Bit Line

Ctrl

Bit Select Sense Amplifier Output Buffer

Figure 4.13

VDD

General standard CMOS eNVM architecture.

VDD

M3

VDD Precharge

Vbias

Isen

M4

Trans-impedance Amplifier Out

Out M1

Bit Decoder

Bit_Line

Iref

Word Decoder

Bit_Line

SE

(a)

Figure 4.14

Sout

M2

(b)

(a) Voltage-mode sense amplifier; (b) Current-mode sense amplifier.

charge pumps are based on the circuit proposed by Dickson [10]. However, due to the existence of the body effect, the pumping efficiency of Dickson charge pump has been degraded. In recent years, several attempts have been made to minimize the body effect of the charge transfer transistors to improve the pumping efficiency of the charge pump [34, 35]. A high performance ALL-PMOS charge pump for low voltage operations is proposed [36] and a maximum power efficiency of 87.1% has been reached at the load current of 2.5 mA for the power supply voltage of 1.5 V.

RFID Tag Chip Design

115

High voltage switch is another important part, which allows the voltage to shift from low voltage to high voltage. A high voltage device in standard CMOS processes is required here to ensure the reliability, since the write operating voltage is typically much higher than VDD. Several articles and patents have been published to address this issue [37–39].

4.2.4 Future Challenges As one of the leading innovators in developing RFID products, Impinj Inc. has invented AEON/MTP NVM which can be produced in standard CMOS processes. Impinj had already completed the qualification of its AEON Memory. Available in 0.18 µm and 0.25 µm TSMC CMOS processes, AEON Memory has attracted a lot of attention from the NVM industry. Impinj Inc. [40] shows the specifications for a 4 K-bit configuration with integrated high voltage on TSMC’s 90 nm LP process. It seems that standard CMOS eNVM is a perfect solution when only small amount of NVM is needed. However, with the opportunities provided by the standard CMOS eNVM technology, designers still have to face some challenges in order to make the technology competitive, including gate oxide scaling, high voltage technology, technology transfer and process control monitoring (PCM). Some tests should be conducted to find out the failure rate dependence on oxide thickness. In addition, experiments need to be done to improve the traditional HV device, like LDMOS (laterally diffused MOS), in order to ensure reliability. Some specific design rules and PCM strategies should be optimized to ensure manufacturability and reliability at future technology nodes. In spite of the challenges it faces, eNVM in standard CMOS processes is still one of the best solutions for RFID applications due to its better data retention, lower cost and higher yield.

4.3 Baseband of RFID Tag A conventional RFID tag contains three essential components: (1) the front end; (2) the baseband; and (3) the non-volatile memory. The front end obtains energy from radio waves, receives and transmits signals from and to interrogators. The non-volatile memory stores information, and the baseband executes operations required by protocols. In this section, the attention is focused on the low-power design of RFID tag baseband. Initially, an introduction to baseband including its function and blocks is given, and later low power design techniques, attempting to reduce both clock rate and supply voltage, are discussed in detail.

4.3.1 Introduction The baseband of an RFID tag should manage to execute the following operations: 1. Waveform coding and decoding: EPC C1G2 protocol defines the interrogator-to-tag link uses PIE (Pulse Interval Encoding) code, and the tag-to-interrogator link uses FM0 baseband signal or Miller (M = 2, 4, or 8) modulation of a subcarrier signal to encode the backscattered data. The baseband should have the ability to decode the PIE coded command from the interrogator and to encode data to FM0 or Miller waveform.

116

RFID Systems

2. Command decoding and tag state transmission: The baseband should recognize commands from interrogator, reply (or not) to interrogator and change current state according to EPC C1G2 protocol. The baseband should keep its current state until the next command arrives while powered up. 3. Link timing: The EPC C1G2 protocol specifies the link timing requirements such as T1 (time from interrogator transmission to tag response) and T2 (interrogator response time required if a tag is to demodulate the interrogator signal, measured from the end of the last bit of the tag response to the first falling edge of the interrogator transmission). The baseband should provide the accuracy to meet timing requirements. 4. Non-volatile memory interface: The interrogator and the tag have the ability to exchange data. The baseband should incorporate an interface with the non-volatile memory, through which reading from and writing to the memory are performed. 5. Pseudo-random number generator (PRNG): EPC C1G2 requires the use of a random or pseudo-random number. To achieve the balance between power consumption and randomness, a PRNG is usually implemented in the baseband. The seed of PRNG may be generated by a True-Random Number Generator (TRNG), which is an analog circuit and measures a physical noise such as clock jitter. Figure 4.15 shows the block diagram of a baseband in a RFID tag chip. Since the signal of the interrogator-to-tag link is strong and with high SNR, non-coherent demodulation is widely used in the tag when it comes to PIE demodulating. According to the EPC C1G2 protocol, the length of RTcal is equal to the length of symbol data-0 plus that of symbol data-1, and the length of symbol data-1 should be 1.5 to two times of that of symbol data-0. The demodulator of a tag measures the length of RTcal and divides it by two, the result of which is called pivot, and is considered as the decision threshold. Any subsequent PIE symbol shorter than a pivot will be decided as data-0, otherwise data-1. The performance of non-coherent demodulation is affected mainly by the frequency of the clock, and how to determine the clock frequency is discussed in the subsequent section.

4.3.2 Low Power Baseband Design

PIE Code Demodulator

CRC

Backscattered signal to RF/Analog Front End

FM0/Miller Modulator

Clock & Timing Management

Decoder State Machine

PIE coded signal from RF/Analog Front End

Backward Frame Shape

Memory Interface

PRNG

Figure 4.15 Block diagram of an RFID tag baseband.

Non-Volatile Memory

Power and cost are the two permanently hot topics for RFID tag chips. Power consumption in CMOS circuits consists of two parts: static power and dynamic power. Reduction of

TRNG

RFID Tag Chip Design

117

the former in a digital circuit depends mainly on the advancement of manufacturing aspects. We will focus here on reducing dynamic power consumption. Dynamic power is comprised of short-circuit power and switching power. Conventionally dynamic power can be approximated by: 2 Pdyn ∝ α0→1 Vdd CL fclk

(4.4)

Here Vdd is the supply voltage, CL is the equivalent capacitance load, fclk is clock rate, and α0→1 is the activity factor. From Equation (4.4) it makes sense that reducing clock rate and supply voltage are common techniques to achieve low power in RFID tag chips. First, we will discuss the minimum clock rate required by EPC C1G2 protocol. The clock rate is mainly restricted by the requirement of PIE decoding accuracy and backscatter link frequency (BLF) tolerance. Second, methods to balance clock rate and power performance are introduced. The method called clock management ensures that function blocks operate at their lowest clock rate. A clock gating method can reduce short-circuit power consumption. In addition, adopting dual edge triggered (DET) flip-flops at the circuits that require the highest clock can help cut the highest clock rate to half, which can reduce the power of the oscillator significantly. The sub-threshold technique and the adiabatic technique, which are methods aimed at low-power low-speed application, such as RFID tag chip, are also introduced in this section.

4.3.3 Clock Rate The maximum clock rate of the baseband is restricted by two factors: power requirement, and time resolution during non-coherent demodulation of PIE codes. The effect of the former factor is somewhat clear because lower frequency leads to both a lower dynamic power of digital circuit and a lower current consumption of oscillator. The influences of time resolution lies in ambiguity between symbols data-0 and data-1, and BLF offset. As the relationship between power consumption and clock rate can easily be found in any books and papers which talk about low power design, here more concern is focused on analyzing how the EPC C1G2 protocol restricts the clock rate to produce lower. In general, the clock rate of the tag is an integer multiple of the maximum BLF, which is 640 KHz defined in EPC C1G2 protocol. Impinj Inc. [41] first performed research on this problem and concluded that 1.92 MHz is more reliable than 1.28 MHz when demodulating clock rate is taken into consideration. As mentioned above, the tag prefers non-coherent demodulation of PIE encoded waveform from the interrogator. The tag measures the length of RTcal (which should be equal to the sum of length of data-0 and data-1) and each data symbol, and compares the length of data symbol to half of that of RTcal (the so-called Pilot). The difference value between the pilot and the worst data symbol (the minimum length of data-1 or the maximum length of data-0) is the margin for decoding, the so-called decode margin. However, quantization errors introduced while measuring the length can inevitably deteriorate the decode margin. When the decode margin is zero, the demodulator cannot distinguish data-0 and data-1 from each other. Figures 4.16 and 4.17 show the decode margin deteriorated by quantization error when the clock rate is 1.28 MHz and 1.92 MHz. They show that it is safe to use 1.92 MHz as

118

RFID Systems

decode margin for data-0 14

12

12

10

10 decode margin

decode margin

decode margin for data-1 14

8 6

8 6

4

4

2

2

0 5

Figure 4.16 circle).

10

15 Tari / us

20

0 5

25

decode margin for data-1

20

25

decode margin for data-0 25

20

20 decode margin

decode margin

15 Tari / us

Decode margin for 1.28 MHz with 5% clock uncertainty (violations marked with the

25

15

10

15

10

5

5

0

10

5

10

15 Tari / us

20

25

0

5

10

15 Tari / us

20

25

Figure 4.17 Decode margin for 1.92 MHz with 5% clock uncertainty.

the demodulating clock rate, because the decode margin may be zero when the clock rate is 1.28 MHz (marked with the circle). According to the EPC C1G2 protocol, tag backscatters its data at a rate specified by TRcal and a parameter called DR (Divide Ratio), both of which are sent by interrogator in a Query command. The tag computes its BLF by the following equation: BLF =

DR TRcal

(4.5)

RFID Tag Chip Design

119

BLF error when DR = 64/3

0.2

0.2

0.1

0.1 BLF error

BLF error

BLF error when DR = 8

0

0

−0.1

−0.1

−0.2

−0.2 50 100 150 value of TRcal / us

200

50 100 150 200 value of TRcal / us

Figure 4.18 BLF error when the clock rate is 1.28 MHz (violations marked with the circle).

Here DR = 8 or DR = 64/3 as specified by the interrogator and TRcal denotes the length of TRcal . The BLF frequency offset occurs since the length of TRcal is quantized to the integer times of the period of sampling clock while the tag measures TRcal by sampling the  can be PIE coded signal at a sampling rate of fclk . Then the measured TRcal (TRcal) represented by:  = nTclk = TRcal + eTR TRcal

(4.6)

TRcal where Tclk is the reciprocal of fclk , n = TRcal Tclk or n = Tclk + 1, where x denotes the largest integer not greater than x and eTR is the quantization error of TRcal . Figures 4.18 and 4.19 show the maximum possible BLF error due to quantization error when clock rate is 1.28 MHz and 1.92 MHz respectively. The bolder line is the error limit defined in the EPC C1G2 protocol. All violations are marked with the circle. They show that when clock rate is 1.28 MHz, BLF error may outrun the bound specified by EPC C1G2 protocol.

4.3.4 Clock-Related Low-Power Techniques Reduction of the clock rate is a widely used low power technique. According to the above analysis, the baseband contains three clock domains: the sampling clock domain, the backscattering clock domain and the controlling clock domain. To obtain adequate time resolution, the demodulator and timing control module should operate at the clock rate computed in Section 4.3.3. This can easily be taken from the waveform of FM0 and Miller that the modulator module should operate at a frequency which is twice of BLF. Other modules, such as the finite state machine and the non-volatile memory interface

120

RFID Systems

BLF error when DR = 64/3

0.2

0.2

0.1

0.1 BLF error

BLF error

BLF error when DR = 8

0

0

−0.1

−0.1

−0.2

−0.2 50 100 150 value of TRcal / us

Figure 4.19

200 50 100 150 value of TRcal / us

200

BLF error when the clock rate is 1.92 MHz.

can operate at the lowest frequency. Through grouping clocks, power and performance can be well balanced. Not all registers work at every clock cycle. The technique of disabling the clock of registers when the values are not changed, known as clock gating, is effective in lowering power consumption of digital circuits. Figure 4.20 illustrates clock gating. When the signal EN is inactive (low), no clock pulse can be propagated to the clock pin of registers. Compared to its equivalent implementation using the multiplexer (left), clock gating outperforms it at power and area cost. The latch (sometimes the register) placed before the AND gate is used to eliminate glitches. Clock gating can be applied at every level of hierarchy, and it can be inserted both manually by RTL coding and by EDA tools such as Design Compiler of Synopsys during synthesis [41]. Another low power technique is based on DET FF (dual edge triggered flip-flop). Using DET FF in the demodulator and timing module can release the highest clock rate to a half. Though this will introduce more dynamic power, it can significantly reduce the power of oscillator in the analog front end and contribute to the overall decrease of power consumption.

D Q

D

EN

EN CLK

CLK DFF with MUX

DFF with clock gating

Figure 4.20

Clock gating.

Q

RFID Tag Chip Design

121

4.3.5 Sub-Threshold Digital Circuit Standard CMOS logic operates less than a voltage supply greater than the sum of threshold hold voltages of PMOS and NMOS FET. The sub-threshold digital circuit is a technique that lets the MOS FET operate at a voltage slightly higher (100 mV greater) or no higher than the threshold voltage. In this case, the current in the sub-threshold region is:     V T ,g W KT 2 VGSkT−V − kTDS /q /q IDsub = 1−e (4.7) µn Cox n e L q where VT ,g = VT + nkT /q, VT is the threshold voltage, n is empirical parameters, with n ≥ 1. In sub-threshold technology, circuits are derived by the IDsub . The propagation delay of an inverter is: Td =

KC g Vdd I0,g

VGS −VT ,g e nVth

(4.8)

where, Cg is the effective capacitance load. For a digital circuit, whose critical path delay is ξ times of inverter propagation delay, the maximum operating frequency is VGS −VT ,g

fmax

I0,g e nVth 1 = = ξ · Td ξ KC g Vdd

(4.9)

Sub-threshold technology is more than just decreasing the power supply voltage. Circuits operating in sub-threshold region should be carefully designed. Wang [42] gives an example of how to balance the leakage current by modifying the logic architecture. In contrast to sub-threshold analog circuit design, the difficulty of sub-threshold digital circuit design is integrating it into existing EDA flows. The key point of being compatible with existing EDA flow is to design sub-threshold libraries for synthesis, automatic placing, routing and static timing analysis.

4.3.6 Adiabatic Circuit The adiabatic circuit is also called the energy-recovery circuit. It achieves low power consumption by restricting the currents to flow across devices with low voltage drop and by recycling the energy stored in their node capacitors using an AC-type power supply rather than a DC type [43]. The adiabatic circuit must follow the following key rules: 1. A device can be turned on only when the source-drain voltage is zero. 2. The source-drain voltage can be changed only when the device is off. 3. Any voltage change must be done gradually. An example of adiabatic CMOS logic circuits is shown in Figure 4.21. The equivalent diagram of charging and discharging process is shown in Figure 4.22. Instead of holding a constant supply voltage, the power source of adiabatic circuit rises linearly during the charging process. The voltage drop on the effective resistance of the driven device is

122

RFID Systems

Input

Pull-up Network

Input

Pull-down Network

Input

Pull-up Network

Input

Pull-down Network

Vdd Vdd

Power Clock

CL

Standard CMOS Logic (b)

Adiabatic CMOS Logic (a)

Figure 4.21

Example of adiabatic CMOS circuit (a) versus standard CMOS Logic (b).

V V

Vdd

V0 R Vdd

CL T

2T

0 (a)

(b)

Figure 4.22 Energy dissipation in the adiabatic circuit.

small; energy dissipation on it, therefore, is small too. The overall energy dissipated in the transition is reduced to: E≈2

RC CV 2 T

(4.10)

When T is sufficiently larger than RC, energy dissipation during the transition can become almost negligible. The EDA for adiabatic digital circuits design encounters a number of challenges. Because the power clock differs from conventional concept of clocks, there is still a long way to make it compatible with current timing-driven synthesis and placing and route EDA tools.

4.4 RFID Tag Performance Optimization In terms of the applications of RFID, high performance and low cost are two equally important aspects that impose trade-offs in the design of an RFID tag. New technologies

RFID Tag Chip Design

123

need to be explored to improve the RFID tag performance. This section mainly summarizes the corresponding optimization technologies, and mentions the possible future technologies in the RFID tag design. Finally, several products are introduced and comparative results of their features are presented, showing the level of development of tag design.

4.4.1 Low Power Tag performance directly influences the success of the whole system, including read range, singulation speed, read rate, etc. [44]. Low power design is always an important issue in backscattering RFID system. It can improve the tag’s performance, including communication reliability and read/write range. However, accounting for decreasing power consumption and battery-assisted/semi-passive system, the read range is not only determined by the tag’s power consumption, but also by the reader’s sensitivity. Cooperative design between the reader and the tag in RFID system is very important. At the same time, the equivalent impedance of antenna is influenced by the environmental temperature, the surrounding objects, and so on. The impedance matching between tag antenna and chip will be altered so that the read range is shortened. An automatic impedance matching system (AIMS) may be considered a good solution to this problem. AIMS includes an impedance mismatch detector, a tunable matching network, a control unit, etc., which are always used for transceiver antenna impedance matching. Recently, the technology has been developed in RFID NFC system [45]. An on-chip tuner should be implemented to adjust for variations in antenna impedance. The tag system measures the output voltage of the rectifier to estimate the impedance mismatching, then dynamically adjusts the capacitor array according to the negative feedback network, and finally it provides the optimal matching state and increases the read range. The singulating speed is determined by different anti-collision algorithms, the communication speed, the decoding arithmetic, etc. A detailed analysis has been given in [44]. Now the singulating speed is calculated based on tag-read operation. But for some extra applications, a high tag write speed is very important. For example, in the WULIANGYE application, which is a famous wine in China, RFID tags on wine bottles need to be encoded quickly when they move along a bottling line, so that the singulating write speed becomes a key factor. The read rate is defined as a percentage of successful read operation. In real applications, RFID readers are not always able to access tags on a 100% basis. Currently, research is being conducted to resolve this problem, and analysis and related improvement methods have been given in [44]. These methods include the Swiss cheese effect, power management and state recovery technology, and automatic impedance matching.

4.4.2 Low Cost Reducing the RFID tag cost is the key element to widen the applications of RFID technology. Now, the average price of passive UHF RFID tag is several cents, and printed RFID tags have a lower cost. In general, the tag chip accounts for 55% of the total tag cost, and antenna manufacture and package costs occupy 45%.

124

RFID Systems

Tag Antenna

Feed Loop Tag Antenna

Tag Chip

Feed Loop

(a)

Figure 4.23

Tag Antenna

Tag Chip

Feed Loop Tag Chip

(b)

(c)

The different packages of on-chip antenna and tag chip.

Besides reducing the chip area, low cost can be achieved by using: • standard CMOS process including eNVM memory and Schottky diode always used in rectifier circuit design; • an on chip antenna (OCA). RFID tags with on chip antenna (OCA) have several advantages including: reducing the assembly cost of RFID tags, providing more universal applications, especially item-level tagging (ILT) which puts a strict demand on tag size and cost [46]. Figure 4.23 presents the different packages of on-chip antenna and tag chip for the different applications. The detailed design technology is related to antenna theory, and is beyond the scope of this chapter. Next, we compare several UHF tag (chip) features in Table 4.5. All five tag chips are the most up-to-date products from Impinj, Texas Instruments, Philips, Alien, and Quanray (a rising RFID design house in China). Different characteristics are compared including read/write sensitivity, programmable non-volatile memory capacity and the chip area. From the view of power consumption, Higgs-3 is the lowest one, which is only Table 4.5 Comparison of several UHF tag chip products. Frequency range

Read/write sensitivity

Programmable NVM

Area (µm2 )

[47] [48] [40]

860–960 MHz 860–960 MHz 860–960 MHz

256-bit 192-bit 512-bit

720 × 720 – 450 × 450

[2] [50]

860–960 MHz 860–960 MHz

−15 dBm/−12 dBm −13 dBm/−9 dBm 7 m, US 6.6 m, Europe −18 dBm −14 dBm

800-bit 2560-bit

650 × 650 720 × 720

Manufacturer

RFID Tag Chip Design

125

−18 dBm. If we suppose that the conversion efficiency of the rectifier is about 30%, and the power supply is 1.5 V, then the total read current is 3.17 µA.

4.5 Conclusion This chapter has described RFID tag chip design. Design aspects of the tag architecture and of every module are given in detail, and some design considerations are also explained. At the same time, the design techniques introduced in this chapter are also applicable to semi-passive and active RFID tag chip design. Low-power and low cost design techniques are always the most important research directions in order to widen the applications of RFID systems. Since some design techniques are immature, such as eNVM compatible with the standard CMOS process and adiabatic circuit technology, the tag chip using these techniques is still under development. However, the authors believe that, with the advancements in research, these techniques will be used in tag chip products, and more new design technologies will be developed for the application in RFID systems.

Problems 1. How to set parameters TRcal and DR of Query command to let tag replay at a BLF equal to: 640 KHz 320 KHz 160 KHz 80 KHz 2. Calculate CRC16 word for the following data (in Hex); 0000 0800_1111 1000_1111_2222 1800_1111_2222_3333 3. Explain why there should be a latch in clock-gated architecture. 4. Try to construct a DET using a positive edge DFF and a negative edge DFF. 5. List the advantages and disadvantages of the Miller Code. 6. Please list three main factors which affect the power of tag baseband, and the corresponding low-power techniques. 7. List the methods to round a number to an integer. What is the effect of employing different rounding methods in tag baseband? 8. The power threshold for tag Ptag,th is −14 dBm, and power consumption of the circuits Pload as a whole is 12 µW. What is the PCE η of the rectifier? 9. Please check the system architecture of Figure 4.1. A multi-level supply voltage generation method is to be applied in the design, and supply voltage requirements for these building blocks are 0.65 V for Demod, 0.8 V for RNG, 0.8 V for POR generator,

126

RFID Systems

0.85 V for bias and regulator, 0.6 V for clock generator, 0.6 V for baseband, and 1.5 V for EEPROM. How to group the building blocks in order to get the systematic low voltage performance? 10. What is the operation region of biasing transistor Mb in the cloud chart of Figure 4.6, and why? 11. If a 1.28 MHz system clock is applied to the circuit in Figure 4.8 as fast clock, why should the sampling circuit be a T flip-flop instead of D flip-flop? NB Solutions to the problems are provided on the book’s website.

References [1] Zhu, Z. (2004) RFID analog front end design tutorial. Available at: http://autoidlab.eleceng.adelaide.edu .au/Tutorial.html. [2] Alien (n.d.) Datasheet: Higgs-3 EPC Class 1 Gen 2 RFID Tag IC. [3] Barnett, R., Balachandran, G., Lazar, S., Kramer, B., Konnail, G., Rajasekhar, S., and Drobny, V. (2007) A passive UHF RFID transponder for EPC Gen 2 with −14 dBm sensitivity in 0.13 µm CMOS, IEEE International Solid-State Circuits Conference, Digest of Technical Papers, pp. 582– 583. [4] Giustolisi, G., Palumbo, G., Criscione, M., and Cutr`ı, F. (2003) A low-voltage low-power voltage reference based on subthreshold MOSFETs, IEEE Journal of Solid-State Circuits, 38(1): 151–154. [5] Serra-Graells, F. and Huertas, J. (2003) Sub-1-V CMOS proportional-to-absolute temperature reference, IEEE Journal of Solid-State Circuits, 38(1): 84–88. [6] Che, W., Yan, N., Yang, Y., and Min, H. (2008) A low voltage low power RF/analog front-end circuit for passive UHF RFID tag, Chinese Journal of Semiconductors, 29(3): 434– 437. [7] Hu, J. and Min, H. (2005) A low power and high performance analog front end for passive RFID, in IEEE Workshop on Automatic Identification Advanced Technologies, pp. 199–204. [8] Facen, A. and Boni, A. (2006) A CMOS analog frontend for a passive UHF RFID tag, in International Symposium on Low Power Electronics and Design, pp. 280–285. [9] Mandal, S. and Sarpeshkar, R. (2007) Low-power CMOS recti?er design for RFID applications, IEEE Transaction on Circuits and Systems – I: Regular Papers, 54(6): 1177– 1188. [10] Dickson, J.F. (1976) On-chip high-voltage generation MNOS integrated circuits using an improved voltage multiplier technique, IEEE Journal of Solid State Circuits, 11: 374–378. [11] Karthaus, U. and Fischer, M. (2003) Fully integrated passive UHF RFID transponder IC with 16.7-µW minimum RF input power, IEEE Journal of Solid State Circuits, 38(10): 1602– 1608. [12] Kotani, K. and Ito, T. (2007) High efficiency CMOS rectifier circuit with self-vth-cancellation and power regulation functions for UHF RFIDs, IEEE Asian Solid-State Circuits Conference, pp. 119– 122. [13] Wang, X., Jiang, B., Che, W., Yan, N., and Min, H. (2007) A high efficiency AC-DC charge pump using feedback compensation technique, in IEEE Asian Solid-State Circuits Conference, pp. 253– 255. [14] Sasaki, A. and Ito, T. (2008) Differential-drive CMOS rectifier for UHF RFIDs with 66% PCE at −12 dBm input, in IEEE Asian Solid-State Circuits Conference, pp. 105– 108. [15] Bucci, M., Germani, L., Luzzi, R., Trifiletti, A., and Varanonuovo, M. (2003) A high-speed oscillatorbased truly random number source for cryptographic applications on a smart card IC, IEEE Transactions on Computers, 52(4): 403– 409. [16] Nakamoto, H., Yamazaki, D., Yamamoto, T., Kurata, H., Yamad, S., Mukaida, K., Ninomiya, T., Ohkawa, T., Masui, S., and Gotoh, K. (2006) A passive UHF RFID tag LSI with 36.6% efficiency CMOSonly rectifier and current-mode demodulator in 0.35 µm FeRAM technology, in IEEE International Solid State Circuits Conference, pp. 1201– 1210. [17] Balachandran, G. and Barnett, R. (2008) A 440-nA true random number generator for passive RFID tags, IEEE Transactions on Circuits and on Circuits and Systems – I: Regular Papers, 55(11): 3723– 3732. [18] Chen, W., Che, W., Bi, Z., Wang, J., Yan, N., Tan, X., Wang, J., and Min, H. (2009) A 1.04 µW truly random number generator for Gen2 RFID tag, in IEEE Asian Solid State Circuits Conference, accepted.

RFID Tag Chip Design

127

[19] Luo, Q., Guo, L., Li, Q., Zhang, G., and Wang, J. (2009) A low-power dual-clock strategy for digital circuits of EPC Gen2 RFID Tag, in IEEE International Conference on RFID, pp. 7–14. [20] Holleman, J., Otis, B., Bridges, S., Mitros, A., and Diorio, C. (2006) A 2.92 µW hardware random number generator, in Proceedings of the 32nd European Solid State Circuits Conference, pp. 134–137. [21] Ruhkin, A., Soto, J., Nechvatal, J., Smid, M., and Barker, E. (2001) A statistical test suite for random and pseudorandom number generators for cryptographic applications, NIST, Tech. Rep. 800–22. [22] Nakamoto, H., Yamazaki, D., Yamamoto, T., Kurata, H., Yamada, S. Mukaida, K., Ninomiya, T., Ohkawa, T., Masui, S., and Gotoh, K. (2007) A passive UHF RF identification CMOS tag IC using ferroelectric RAM in 0.35-um technology, IEEE Journal of Solid State Circuits, 42: 101–110. [23] Derbenwick, G.F. (2008) Invited paper FR008 embedded ferroelectric memory for RFID tag applications, 17th IEEE International Symposium on the Applications of Ferroelectrics, vol. 2. [24] Lee, M.C., Barsatan, R., and Chan, M. (2007) OTP memory for low cost passive RFID tags, in IEEE Conference on Electron Devices and Solid State Circuits. pp. 633– 636. [25] Barsatan, R., Man, T.Y., and Chan, M. (2006) A zero-mask one-time programmable memory array for RFID applications, in IEEE International Symposium on Circuits and Systems, pp. 975–978. [26] Science News (2007) Next-generation RAM: Remembering the future, ScienceDaily. Available at: http://www.sciencedaily.com/releases/2007/12/071221174912.htm. [27] Dixian, Z., Na, Y., Wen, X., Liwu, Y., Junyu, W., and Hao, M. (2008) A low-power single-poly nonvolatile memory for passive RFID tags, Journal of Semiconductors, 29(1): 99–104. [28] Pesavento, A. and Hyde, J.D. (2007) PFET nonvolatile memory, United States Patent, Patent No.: US, 7,221,596B2. [29] Ohsaki, K., Asamoto, N., and Takagaki, S. (1994) A single poly EEPROM cell structure for use in standard CMOS processes, IEEE Journal of Solid State Circuits, 29: 311– 316. [30] Raszka, J., Advani, M., Tiwari, V., Varisco, L., Hacobian, N.D., Mittal, A., Han, M., Shirdel, A., and Shubat, A. (2004) Embedded flash memory for security applications in a 0.13 µm CMOS logic process, in IEEE International Solid-State Circuits Conference, Digest of Technical Papers, vol.1. [31] Kee-Yeol, N. and Kim, Y.-S. (2006) High-performance single polysilicon EEPROM with stacked MIM capacitor, IEEE Electron Device Letters, 27: 294–296. [32] Schuster, S.E. and Matick, R.E. (2009) Fast low power Edram hierarchical differential sense amplifier, IEEE Journal of Solid State Circuits, 44: 631– 641. [33] Makosiej, A., Nasalski, P., Giraud, B., Vladimirescu, A., and Amara, A. (2008) Double-gate sub-32nm CMOS SRAM current and voltage sense amplifiers, insensitive to process variations and transistor mismatch, IEEE International SOI Conference, pp. 63–64. [34] Li, N., Huang, Z., Jiang, M., and Inoue, Y. (2008) High efficiency four-phase all PMOS charge pump without body effects, in International Conference on Communications, Circuits and Systems, pp. 1083– 1087. [35] Hsu, C.-P. and Lin, H. (2007) Analysis of power efficiency for four-phase negative charge pumps with body potential control, in IEEE International Conference on Integrated Circuit Design and Technology, pp. 1–4. [36] Yan, N., and Min, H. (2006) High efficiency all-PMOS charge pump for low-voltage operations, IET Electronics Letters, 42(5). [37] Bin, W. (2006) Graded-junction high-voltage MOSFET in standard logic CMOS. United States Patent, Patent No.: US, 7145203. [38] Bianchi, R.A., Monsieur, F., Blanchet, F., Raynaud, C., and Noblanc, O. (2008) High voltage devices integration into advanced CMOS technologies, in IEEE International Electron Devices Meeting, pp. 1–4. [39] Sun, W.F. and Shi, L.X. (2003) High reliability HV-CMOS transistors in standard CMOS technology, in Proceedings of the 10th International Symposium on the Physical and Failure Analysis of Integrated Circuits, pp. 25–28. [40] Impinj, Inc. (2008) PRODUCT BRIEF AEON MTP EEPROM Architecture. Gen 2 Tag clock rate – what you need to know, Impinj White Paper. Available at: www.impinj.com/WorkArea/downloadasset.aspx?id = 2541. [41] Impinj, Inc. (2008) Clock Gating Methodology for Power and CTS QoR, Synopsis White Paper. [42] Wang, A. and Chandrakasan, A. (2005) A 180-mV Subthreshold FFT processor using a minimum energy design methodology, IEEE Journal of Solid-State Circuits, 40(1): 310– 319. [43] He, Y. and Min, H. (2007) Adiabatic circuit applied for LF tag, Auto-ID Labs White Paper WPHARDWARE-041.

128

RFID Systems

[44] Miles, S.B., Sarma, S.E., and Williams, J.R. (2008) RFID Technology and Applications. Cambridge: Cambridge University Press. [45] Roland, M., Witschnig, H., Merlin, E., and Saminger, C. (2008) Automatic impedance matching for 13.56 MHz NFC antennas, Communication Systems, Networks and Digital Signal Process, CNSDSP 2008. 6th International Symposium on, pp. 288– 291. [46] Jingtian, X., Na, Y., Wenyi, C., Xiao, W., Hongyan, J., and Hao, M. (2009) On-chip antenna design for UHF RFID, Electronics Letters, 45(1). [47] Impinj (n.d.) Datasheet: EPCglobal Generation 2 RFID Monza. [48] Texas Instruments (n.d.) Datasheet: UHF Gen2 STARP RI-UHF-STRAP-08. [49] Philips (n.d.) Datasheet: SL3 ICS 10, UCODE EPC G2. [50] Quanray (n.d.) Datasheet: Qstar EPC C1G2 RFID Tag IC.

5 Design of Passive Tag RFID Readers Scott Chiu Intel Corporation

5.1 Overview With the recent advances in global standardization, RFID has become an important tool for supply chain management [1] while expanding into many possible consumer applications through initiatives like Mobile RFID [2]. Similar to the evolution of other innovative technologies, many systems exist in the field which can be characterized by: 1. The operating frequency band: In general, RFID relies on unlicensed bands for communications. For example, ISM1 bands in the United States, HF (13.5 MHz), VHF (40 MHz), or UHF (860–960 MHz, or 2.4 GHz). Currently HF is more prevalent in applications while UHF has been gaining popularity with lower cost tags, and the flexibility of a longer reach. 2. How the tag and the communication link are powered: Passive tags are powered by harvesting the RF energy transmitted from the reader, and the tag-to-reader communication is maintained by modulating (backscattering) the reader-transmitted RF energy; active tags are battery-powered, and are capable of sending signals back to the reader; semi-active tags fall in between – these tags are battery-powered but rely on backscattering for uplink communications. The trade-off here is cost versus reach. With on-tag power source, the active or semi-active tags have a longer range, but with a higher cost and a shorter life-span (dependent on how the battery power is consumed). 1

Industrial, Scientific, and Medical bands.

RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c Chapter 5,  Intel Corporation, 2009.

Edited by Miodrag Boli´c, David Simplot-Ryl,

130

RFID Systems

3. The coupling mechanism: Near-field with inductive or electrical field coupling, or far-field with electro-magnetic wave coupling. This chapter will focus on the emerging UHF (860–960 MHz) ISO 18000-6C compliant passive tag RFID reader design. Readers with an interest in the RFID evolutions should refer to [1] for detailed discussions of legacy systems. In the following, Section 5.2 offers an introduction to passive RFID operations using the ISO 18000-6C [3] air interface. The emphasis is to make a link from the issues encountered in the field to the mitigation specified in air interface. An example of an inventory round is then used to demonstrate the tag access. Section 5.3 describes how the application impacts the design decisions, and the approach to derive the design specifications. A generic implementation is used to demonstrate the challenges of reader design, and methods to mitigate the non-idealities. Section 5.4 discusses more advanced design topics of the reader designs, and Section 5.5 closes with a conclusion.

5.2 Basics of Passive RFID Operation The RFID system is an ad hoc network formed between a reader and the tags covered by its RF energy field for a duration controlled by the reader. The underlying reader (interrogator) selects the target tag to establish a link and the start/end time of the communication – it serves as the master in the system and co-exists with other reader systems in the same physical location. A passive tag is powered by energy harvested from the reader transmitted carrier power [4–6]. As shown in Figure 5.1, the reader initiates an inventory round (the period CW RF Carrier RFID Tag Assembly

UHF RFID Reader

RFID Tag IC

transceiver

Interrogator Modulation

Protocol and application stack processing

Power harvesting logic memory

The RFID Tag backscatters the Reader’s TX Carrier Local Area network Application server

internet

Database

Figure 5.1 A RFID reader interacting with a tag. Based on a figure taken from S. Chiu et al., “A 900 MHz UHF RFID reader transceiver IC,” IEEE Journal of Solid State Circuits, vol. 42, no. 12, pp. 2822–2833, Dec. 2007.  2007 IEEE.

Design of Passive Tag RFID Readers

131

needed to read all tags within its field of view) by sending an unmodulated carrier signal to “energize” the tags. The unmodulated carrier (CW) is sent throughout the inventory round, except during the period the reader talks, to keep the tags alive. Once the tags are powered up, the reader enters “talk” mode to initiate the communication by modulating its carrier to issue a command. After sending a command, the reader stops modulation to enter “listen” mode. During this period, the reader sends a CW while turning on the receiver to read tag responses. The tag of interest responds by backscattering the reader’s CW signal. The backscattering is achieved by reflecting the incoming carrier power by changing the state of the tag antenna impedance. Note that the use of “talk” and “listen” is to emphasize the asymmetric nature of this half-duplex link.

5.2.1 An Introduction to ISO 18000-6C Air Interface All RFID readers must abide by two types of rules for interacting with the tags: (1) local regulations established by a governing organization, or a government entity; (2) an air interface standard defining the protocols for the reader and tags to interact. The aim of the local regulation is to ensure that the RFID reader and tag operation does not interfere with existing public or commercial services. In the United States, UHF RFID readers operate at the 902–928 MHz ISM (Industrial, Scientific, and Medical) band as defined by FCC (Federal Communications Commission) title 47 part 15 [7]. In Europe, the European Telecommunication Standard Institute (ETSI) set the guideline with ETSI 302 208 [8] in the 865–868 MHz band. These publications define the spectrum mask and maximum out-of-band emissions – for example, an ETSI compliant device can emit up to 33 dBm of power within 865.6 to 867.6 MHz with spectrum mask as shown in Figure 5.2. In terms of reader implementation, the in-band spectrum mask performance is usually dominated by the baseband filtering and transmitter linearity; the out-of-band emission is mostly determined by the inter-modulation of the carrier (which is at high power), and by spurious tones coupling into the frequency generation and transmit paths. The latter effect is often neglected in the early design process, and results in great pain late in a product development. Additionally, the adjacent and alternate channel spectrum masks are defined in reference to an absolute power level. This means the linearity requirement can be substantially relaxed for readers operating at a lower transmit power level. The air interface consists of a set of rules governing the communication between the reader and the tags. There are two active international standard bodies for air interface definitions: the International Standard Organization (ISO) which defines RFID specifications including ISO 18000-6A, B, and C types of tags [9], and EPC global which is behind the EPCglobal Class 0 [10], Class 1 Generation 1 (C1G1) [11], and Class 1 Generation 2 (C1G2) [12]. The standards of ISO 18000-6C and EPC C1G2 are harmonized to reach a global standard, and are referred to as ISO 18000-6C in this chapter. ISO 18000-6C consists of the physical layer interface and the command sets for implementing the protocol. The physical interface defines the modulation (waveform and spectrum mask), encoding scheme, data rate, link timing, and message constructs (preamble type, length, and message body). Table 5.1 summarizes the encoding and modulation types – the difference between reader to tag and tag to reader encoding schemes shows the asymmetric nature of the RFID system.

132

RFID Systems

ISO 18000-6C PR-ASK Tari: 12.50 x: 0.5 0 −10 −20

Magnitude [dBc]

−30 −40 −50 −60 −70 −80 −500

−400

−300

−200

−100 0 100 Offset Frequency [kHz]

200

300

400

500

Figure 5.2 Transmit spectrum mask example: PR-ASK modulated reader with an ETSI 302-208 mask (Tari = 12.5 µs, with data ‘1’ 0.5 Tari longer than data ‘0’). Table 5.1 Summary of ISO 18000 6-C encoding and modulation specifications. Reader to tag

Tag to reader

Encoding

Modulation

Encoding

Modulation

PIE (Pulse interval encoding)

DSB-ASK PR-ASK SSB ASK

FM0, Miller subcarrier

DSB-ASK PSK

For the downlink, all three reader-to-tag modulation schemes are forms of Amplitude Shift Keying (ASK) for the ease of tag demodulation. Double Side Band ASK (DSB-ASK) is a real-value modulation which offers the easiest implementation and the possibility of supply modulation [13]. Single-Side Band ASK (SSB-ASK) and Phase-Reversal ASK (PR-ASK) offer better spectrum efficiency with an increase in modulation complexity (more components in the transmit path). Based on the air interface standard, the bandwidth required is not absolute since the reader can lower the signaling rate to reduce the spectrum occupied (note that the signaling rate range is specified to be between 6.25 and

Design of Passive Tag RFID Readers

133

25 µs, and is controlled by reader through reader to tag preamble). This allows DSB-ASK to operate in the allocated channel except at a reduced data rate. Pulse Interval Encoding (PIE) is used for transmit coding since (1) the reader maintains the carrier wave power except for 1/2 Tari time slot at the end of a symbol. Tari is the basic reader to tag signaling time unit which is equal to length of a symbol “0.” This allows the tag to harvest the maximum amount of power during an inventory cycle; (2) it allows the tags to decode by using its internal clock to differentiate an encoded “one” from a “zero” symbol through counting. Since the tag demodulates the signal by tracking amplitude variations, excessive ripples can trigger false transitions while slow edges can blur the difference between a “1” to a “0.” Thus the air interface defines a waveform mask in additional to the spectrum mask. Figure 5.3 shows the coding and modulation using a PR-ASK modulated reader to tag packet as an example. As depicted in Figure 5.3 a packet is composed of a preamble and the succeeding data symbols. The preamble of the packet consists of a delimiter, a “0” symbol, and two special PIE symbols RTcal and a TRcal. RTcal is equivalent in length to a “0” plus a “1” symbol for setting the tag symbol slicing threshold. TRcal defines the tag-to-reader link signaling rate with a length equal to the inverse of the link frequency. For the uplink, both Amplitude Shift Keying (ASK) and Phase Shift Keying (PSK) are allowed in tag implementations. When the tag backscatters through changing the real portion of the antenna load, the modulation appears as ASK. When the scattering is done with changing the imaginary portion of the antenna load, the modulation appears as PSK. There are two tag encoding schemes specified in ISO 18000-6C: FM0 or Miller modulated subcarrier (an example is shown in Figure 5.4 with various signaling rates). Use of the tag encoding scheme is chosen by the reader initiating the session. Both encoding schemes avoid signal contents at DC. FM0 is a baseband encoding which spreads the signal power around the carrier frequency. Miller uses frequency shifts to offset the signal

Data-0

RTcal

TRcal

Delimiter 1 0.8 0.6 0.4 0.2 0 −0.2 −0.4 −0.6 −0.8 −1 100

150

200

250 Time [µs]

300

350

Figure 5.3 An example of a RFID reader-to-tag communication packet.

134

RFID Systems

tag to Reader spectrum plot: FM0 −50

tag to Reader spectrum plot: Miller (subcarrier: 320KHz) 40K 80K 320K

−55

−50 −55 dBW/Hz

dBW/Hz

−60 −65 −70

−60 −65

−75 −80 −85

M = 8,40Kbps M = 4,80Kbps M = 2,160Kbps

−70 −5

−4

−3

−2

−1

0 Hz

1

2

3

4 x 105

(a)

1

2

3 Hz

4

5

6 x 105

(b)

Figure 5.4 ISO 18000-6C tag modulation spectrum plot in reference to the carrier frequency: FM0 modulated (a); Miller subcarrier modulated (b), the subcarrier is shifted to 0 Hz.

power into a sub-carrier frequency. For example, for an 80-Kbps data with M = 4 Miller subcarrier modulation, the center of the signal spectrum is 320 KHz offset from the actual carrier. Miller modulated subcarrier encoding is especially useful in a dense reader environment due to the following: (1) offset tag responses to the subcarrier frequency to avoid the noise coupled from the reader transmission. To keep the tag powered throughout the inventory round, the reader is transmitting CW (carrier) at full power while receiving tag responses. Since the received tag power can be 90 dB2 smaller than the transmitted power, the close-in noise of the transmitter dominates the receiver noise floor even considering a very good transmit-to-receive isolation. Moving the tag response away from the carrier mitigates this effect on tag sensitivity, especially in a dense-reader environment where other readers may be transmitting carrier at the same time; and (2) the subcarrier encoding provides more transitions in the same symbol period, which helps the receiver track the tag data rate variation at link frequency of 300 KHz for up to 15% initial tolerance and 2.5% drift within one data packet.

5.2.2 Tag Singulation and Access After understanding the basic means of establishing a link between one reader and one tag, this section illustrates the case of communicating with many tags using the interface protocol. Before the reader powers up the passive tags in its RF range, it does not have any prior knowledge of the tag population. The air interface protocol thus defines three sequential steps to establish the links with the targeted tags: select, inventory, and access, each with associated commands for managing a tag population. 2 Assuming the transmitted carrier is at 30 dBm, with a tag backscattered power at −75 dBm, and 15 dB of transmit to receive isolation (in the case of using a bi-static configuration; to be discussed later in the chapter). The transmitter power is 90 dB higher than the tag reflection. However, depending on the targeted performance, this value can be high or lower in a specific system.

Design of Passive Tag RFID Readers

135

1. Select: During this phase, the reader commands the tag population to identify the group included into an inventory round. The Select command has comparison criteria specified by an array of four entries [MemBank, Pointer, Length, Mask]. MemBank determines the memory block under comparison – it can be UII (Unique Item Identifier) memory which contains a CRC-16 generated at power up, the protocol control word, and EPC (Electronic Product Code), TID (Tag Identification) memory for tag class information, or the user memory. CRC-16 is a 16-bit cyclic redundant check code computed to ensure the UII data integrity. Pointer and Length specify the target location and size; Mask is the bit string to compare against. 2. Inventory: The reader identifies individual tags from the selected group using a slotted ALOHA algorithm with (reader) query and (tag) acknowledgement. The Query command starts a new inventory round with a 16-bit slot count parameter Q. Each tag receiving the Query command generates a slot count between 0 to 2Q -1. The tags with slot count 0 reply with its RN16 (a 16-bit random number from the tag pseudorandom number generator). To continue on the same inventory round, the reader issues QueryRep to decrement tags’ slot counter by 1, and listens to the reply from tags with slot count 0. Selection of Q determines the statistical performance of this operation. Choosing a large Q avoids tag collisions (many tags reply in the same slot) at the expense of many unused slots; choosing too small the Q with respect to the tag population increases collisions and retries, which exhibits a degraded performance. Unfortunately, the tag population is not known before an inventory round. The air interface provides a command QueryAdjust to allow the reader to increment or decrement Q value in the current inventory round. This allows adaptive changes of Q during this phase based on heuristics (for example, from the collision statistics). 3. Access: Once a tag is uniquely indentified, the reader can choose to perform different operations ranging from read/write to the specified memory bank using a Read or a Write command, to lock the tag for secure access (using Lock command to prevent or allow access to the RFID password or a specific memory band), or even permanently disable the tag (using Kill command). A possible sequence of reader and tag interactions is shown in Figure 5.5. In this example, the reader starts by picking a selected group of tags in a tag population, and begins an inventory round. The first query attempt results in collisions with multiple tags responding at the same time. In practice, the reader will follow an adaptive procedure to determine whether to collect more statistics or to change the Q – this example is a simplified case – the reader then sends a QueryAdjust command with a larger Q, and gets a tag reply RN16 (a 16-bit random number generated by the tag) successfully. The reader sends an ACK command with the same random number attached for tag to validate the data integrity. After verifying the RN16 is a match, the tag sends its unique item identifier with CRC-16. At this point, the reader can decide to continue on the inventory process, or access the identified tag. In this example, the reader begins the access by issuing a req_RN command to “open” a tag. The tag replies a different RN16 (RN16_2 as seen in Figure 5.5) to be used as a key for the following access operations. The reader asks for the contents by issuing a read command with a specific memory bank in tag memory. After the tag replies with its stored data, the reader “closes” the selected tag, and resumes the inventory round with a QueryRep.

136

RFID Systems

select

reader sends select command to choose a subset of the tags

reader sends Query command with a slot count parameter Q

Tag with slot count = 0 acknowledge with a 16-bit random number RN16_1 Many tag responds at the same time

reader sends QueryAdjust command with a larger Q

Tag with slot count = 0 acknowledge with a 16-bit random number RN16_1

reader sends ACK command containing the RN16_1

Tag replies with EPC (Electronic Product Code)

inventory

reader sends Req_RN command

Tag replies with a new RN16_2 (tag is in OPEN state)

access reader sends read command with RN16_2 to access the tag memory

inventory

Figure 5.5

Tag reply with its memory contents

reader sends QueryRep

Example of a protocol followed between a reader and a tag.

5.3 Passive RFID Reader Designs Depending on applications, commercially available UHF RFID readers have form factors ranging from the size of a notebook to palm top as illustrated in Figure 5.6. In this section, we will discuss the difference in design from a technical perspective by discussing the system specifications. A generic RFID reader implementation is then provided as an example to provide a link between the requirement and the circuit to support the functionality.

Design of Passive Tag RFID Readers

137

Figure 5.6 A few example implementations of UHF ISO 18000-6C compliant RFID readers. Reproduced by permission of  Impinj Inc.

5.3.1 RFID Reader Read Range and Transmitted Power In principle, creating an UHF RFID reader specification is a practice no different from that of other wireless systems. Guided by a requirement document derived from standards, local regulations, and specific customer applications, the specification should encompass the parameters on dynamic range, linearity, and phase noise, etc. [14]. In the following, we will focus on deriving the relationship between the required reader sensitivity, read range, and transmitter power, which is specific to passive tags. The performance of an RFID reader is characterized by its read range – unlike the conventional wireless communication devices, the read range of a passive RFID system is affected by three factors: (1) the power delivered and rectified to wake up the tag; (2) the power reflected from the tag; and (3) the reader sensitivity. Pr =

P t · Gt · Gr   d 2 4π λ

(5.1)

Equation 5.1 shows a general range equation describing the relationship of the received power at the receiver (Pr) to the transmitted power (Pt) referenced to an isotropic radiator, the transmitter antenna gain3 (Gt), the receiver antenna gain (Gr), and the path loss (which is the term in the denominator based on the distance between them: d, and the radio wavelength: λ). This equation defines the power delivered to the tag. 3

Antenna gain is to be discussed further in Section 5.3.2.1.

138

RFID Systems

Referring back to Figure 5.1, the path from a reader to a tag is traversed twice between transmit and receive: once with the carrier sent from the reader, and again with the received power reflected from the tag; thus the pass loss doubles in dB comparing to other wireless systems. In addition, a loss term Ltag is added to account for the limitation in its antenna construction, polarization loss, and matching: Pr =

P t ∗ Gt ∗ Gr ∗ Ltag   d 4 4π λ

(5.2)

For example, for an RFID system operates at 900 MHz UHF band, Ltag loss of 15 dB (3 dB of polarization loss and 12 dB of reflection loss for Raleigh range ∼ λ−4 [1]), 3 dB gain on both transmit and receive antenna, 30 dBm transmitter, and the tags consuming 3.5 µW on the average with a 20% power conversion (harvest) efficiency, a read range of 9 m can be achieved when receiver sensitivity is better than −80 dBm. In the above, two criteria are met to budget for an RFID system: (1) deliver enough power to keep the tag alive; and (2) designate a reader sensitive enough to accept the signal. Figure 5.7 depicts the read range versus receiver sensitivity and available power to the tag with the above assumptions. Note that the tag power consumption is a major part of the equation – if the tag consumes more power to stay alive (implementation needs more power or the conversion efficiency is lower), the link becomes downlink dominated – further improvement to receiver sensitivity will not yield any range increase. Though the range is limited by the tag on downlink currently, the RFID reader design should accommodate the improvement in tag performance over time. Also implied in the equations is that the transmit power can be raised to improve the reader range. Higher transmit power requires much linear transmitter – a 1 dB increase in power means 3 dB increase in third order linearity requirement. This approach results in a less efficient power amplifier plus a larger case and board to ease thermal considerations.

RFID reader read distance vs. receiver sensitivity and available tag power −35.00

45.0 1

2

3

4

5

6

8

9

10 40.0

receiver sensitivity

−45.00 receiver sensitivity (dBm)

7

available rectified power to tag

−50.00

35.0 30.0

−55.00

25.0

−60.00 20.0

−65.00

15.0

−70.00 −75.00

10.0

−80.00

5.0

−85.00

available power for tag (uW)

−40.00

0.0 distance (m)

Figure 5.7

Sensitivity and available tag power versus reader range (from the example).

Design of Passive Tag RFID Readers

139

The other way to improve the system performance is to increase the reader or tag antenna gain. The reader antenna can be optimized based on size, cost, and applications. The fixed reader on the dock door usually has a larger size comparing to the lower performing hand-helds. In addition, tags are not created equal – at the same operating band, larger tags with more sophisticated antenna design tend to have better antenna efficiency. Therefore, it is important to select the tags for a given application.

5.3.2 RFID Reader Implementation An UHF RFID reader is composed of four basic building blocks as depicted in Figure 5.8: 1. Antenna interface is where the RFID reader converts the electrical signal into electromagnetic waves and vice versa. Depending on the application, several antenna elements can be connected to a single reader through an antenna multiplexer. As an example, Figure 5.8 shows two antennas covering different locations accessing items on a conveyer belt. 2. Transceiver is the physical front-end translating bits into radio frequency transmission, or recovering backscattered tag responses into bits. To keep track of the operation status, a radio controller is used to: (1) monitor the reflected power level due to antenna impedance mismatch (in the case of someone disconnecting the transmit antenna); (2) transmit the power level for closed loop power control; (3) initialize the radio to operate at a target channel; and (4) ensure link timing conformance to the standard. 3. Protocol and application processor issues commands to the radio transceiver for tag inventory and read/write accesses described in Section 5.2.2. By embedding smartness to achieve performance on top of what is suggested in the air interface standards, the manufacturers differentiate their products in protocol processing, for example, using an improved Q algorithm to achieve faster inventory rounds [15]. In addition, one or more layers of the RFID system software stacks reside on the RFID reader with the rest located remotely based on applications. An example is the RFID system software architecture proposed by Microsoft [16] (we will leave a detailed description of the system software stack implementation to other chapters of this volume).

transceiver

Protocol processing Power

Power Regulation

Ethernet

controller Network Processor

LO Section

Indicators GP I/O

Flash

demodulator

Transmit Section Receive Section

Antenna MUX

Serial

Antenna interface

Radio

SDRAM

RFID reader

Figure 5.8 Block diagram of a generic RFID reader implementation. Reproduced by permission of  Intel Corporation.

140

RFID Systems

4. Network IO interface is the link between the reader and the server(s) processing a higher level of the software stacks. Depending on the applications, one or more of the standard interfaces are offered – Ethernet is the predominant choice, especially with the possible use of power over Ethernet for supply generation for fixed-location readers. USB or PCMCIA are the choices for attaching an RFID reader to a portable device. In the following, we will introduce the RFID specific hardware implementation of these components. 5.3.2.1 Antenna Interface The antenna is the physical interface converting between electrical signals and electromagnetic waves. Depending on the application environment, the RFID reader designer need to prioritize different considerations based on cost, material, package, allowed space or geometric shapes. This multi-dimensional optimization makes antenna selection often determined by “engineering judgments.” In this section, we will instead provide an introduction to the critical parameters with the aim of understanding the antenna used in RFID test/validations, followed by a discussion of antenna diversity techniques commonly used in current RFID readers. The important parameters for an RFID reader antenna are: • Radiation pattern and directivity: Isotropic radiation (the radio power emits evenly like a globe) exists only in theory as a reference. In real life, power emission varies depending on the direction looking into the radiator. The variation of the power as a function of the direction away from the antenna observed from the far-field is defined as the radiation pattern. To visualize the three-dimensional pattern in a two-dimensional plot, it is specified or plotted with respect to elevation and azimuth angles. Figure 5.9 depicts two antenna pattern plots – the narrow one is from a 10-element Yagi antenna, and the wider one is from a 3-element directional antenna. Directivity is a measure of how “directional” an antenna’s radiation pattern is in reference to an isotropic radiator. 0 dB directivity means an isotropic antenna, and 3 dB equals a peak radiation at twice the power level as compared to an isotropic radiator. The “peak directivity” is usually specified for antenna. • Beam width: The angle between two points (in the radiation pattern plot) where the magnitude of the radiation decreases by 50% (3 dB) from the peak of the main lobe. Using the same example from Figure 5.9, the 3-element antenna has a beam width of 70 degrees, while the Yagi’s beam width is close to 40 degrees. In ETSI 302 208 [8], the antenna needs to have a beam width ≤ 70◦ in the horizontal direction for transmission between 500 mW to 2 W in a channel. • Efficiency: The efficiency is the ratio between the power emitted from and the power delivered to the antenna. The loss is due to the resistive loss of the antenna or the impedance mismatches. For high power, high performance RFID systems, the efficiency of the antenna has a large impact on the overall power consumption. • Gain: This is the ratio between the power transmitted in the radiation direction and an isotropic radiator (as a reference). The “peak” antenna gain is usually specified.

Design of Passive Tag RFID Readers

Copyright © 2007 G R Freeth www.g4hfq.co.uk

141

0 dB

7/26/2009 5:58:35 PM

−2 −4 −6 −8 −10

−20 −30

−30 −20

−10 −8 −6 Collected by:

−4 −2

Grow−>

Figure 5.9 Illustration of antenna radiation pattern. Reproduced from PolarPlot by permission of  Bob Freeth.

• Polarization loss factor: An antenna is linearly polarized if the antenna radiation pattern follows a specific plane of emission. When two orthogonal radiators are combined for the transmission, the polarization rotates either counter-clockwise or clockwise (“right hand” or “left hand” circularly polarized ). Since no excitation can happen between two linearly polarized antenna placed orthogonally to each other, a linearly polarized reader antenna may run the risk of losing tags with linearly polarized tag antenna. The polarization factor is used to account for the potential loss of signal power due to the different orientation of the tags from the reader. Given a reader with a circular polarized antenna and a tag with linearly polarized antenna, the polarization loss factor is 3 dB in the worst case. The key point here is that each antenna design has a specific pattern of coverage – by understanding these parameters, we can specify the right type based on cost, coverage area, and size. A common practice in RFID system design is to place the complexity in the reader antenna design to trade off the cost of the tag. In addition to the individual

142

RFID Systems

+X –circulator isolation dBm

Tx/Rx

Band select Filter

Reader RX Reader TX

+X dBm (a) Rx +X –TX/RX antenna isolation dBm

Band select Filter

Reader RX

Tx

+X dBm

Reader TX (b)

Figure 5.10 Mono-static and bi-static configuration. Based on a figure taken from S. Chiu et al., “A 900 MHz UHF RFID reader transceiver IC,” IEEE Journal of Solid State Circuits, vol. 42, no. 12, pp. 2822–2833, Dec. 2007. 2007 IEEE.

antenna performance, the use of more than one antenna has been explored to advance the system operation. Currently two types of diversity techniques are in use: 1. Spatial diversity: By placing multiple directional antennas (usually limited to less than four per reader) aiming at different angles from several locations, for example, of a dock door. This approach enhances the coverage of the reader, especially when tags are located in different spots in a pallet, or being obstructed by a metal shield in some direction. 2. Transmit and receive diversity: This is shown in Figure 5.10(b) as a bi-static configuration with a transmit antenna and a receive antenna, comparing to a mono-static configuration (Figure 5.10(a)) which uses a single antenna. In the case of a mono-static configuration, a circulator (an expensive cost adder) or an attenuator (a reduction in sensitivity) is added in the receive path to avoid saturation from transmit to receive coupling. The bi-static configuration provides much better isolation to enable higher performance by reducing the large signal compression of the receiver frontend and the transmit noise coupled into the receiver. If the transmit/receive antenna produces 25 dB of isolation, this example shows at least 10 dB difference compared to using a circulator of 15 dB isolation. In practice, the difference can be larger due to the component margin and board impedance mismatches. However, transmit/receive diversity doubles the antenna cost and size. Currently bi-static configuration is limited to high performance systems, and mono-static configuration dominates less demanding applications. 5.3.2.2 RFID Transceiver Serving as the physical layer interface, the RFID transceiver consists of three main functional blocks: the transmit path, the receive path, and the frequency generation section, as

Design of Passive Tag RFID Readers

143

ADC DC removal

RX

Q

RF detector

Demodulator

Receive

I

LNA

A D C

ADC

Tag bits (tag to reader)

Frequency generation Xtal Osc

0/90

÷2

PLL

÷2

TX

MUX

+

I Q

DAC

DAC

modulator

Transmit LO

Transmitted Bits (reader to tag)

Figure 5.11 A RFID transceiver implementation. Based on a figure taken from S. Chiu et al., “A 900 MHz UHF RFID reader transceiver IC,” IEEE Journal of Solid State Circuits, vol. 42, no. 12, pp. 2822–2833, Dec. 2007.  2007 IEEE.

depicted in Figure 5.11 using a direct conversion implementation. The transmit path carries out the task of baseband modulation and passband signaling by frequency up-conversion; the receive path performs frequency down-conversion and demodulation; the frequency generation section generates and distributes the carrier. These functions have been discussed extensively in wireless transceiver literatures [17, 18], and will not be repeated here. In the following, we will focus on the implementation differences originated from the unique attributes pertaining to a RFID system. Receive Path The biggest differentiation is the requirement to handle a large transmitter leakage during tag reception. Depending on the transmit to receive isolation, and the transmitter power level, the receiver typically needs to tolerate transmitter leakage signals (self-jammer) greater than 5 dBm (assuming a 30 dBm transmitter with 25 dB transmit to receive isolation). This requires a receiver front-end with high compression point, and limits the amount of front-end gain to less than 0 dB. Since there is no effective gain before the received signal reaches baseband, any noise added before the baseband amplifier impacts the receiver noise figure. Thus the receiver front-end implementation requires high compression point and low noise. Due to the large self-jamming signal residing at the carrier frequency, direct conversion architecture is commonly employed to remove the self-jammer by mixing down it to DC. There are four types of DC offset cancellation methods used in practice [19] as illustrated in Figure 5.12: (a) using a simple RC-high pass filter; (b) using a high pass filter with sample and hold to memorize the low frequency signal contents; (c) using active cancellation with DC estimated by a low-pass filter; and (d) cancel the DC after converting the signal to digital. The first method creates a conflict between passing the

144

RFID Systems

in

out

in

out

High pass filter High pass filter with sample and hold

(a)

in

+

out −

Low pass filter

(c)

in

+

(b)



Signal processing

(d)

Figure 5.12 DC offset correction. Based on a figure taken from S. Zhou and M.C.F. Chang, “A CMOS passive mixer with low flicker noise for low-power direct-conversion receiver,” IEEE Journal of Solid State Circuits, vol. 40, no. 5, pp. 1084–1093. May 2005.  2005 IEEE.

signal contents and responding to DC changes from transmit power fluctuations. Since the tag response can be very close to DC due to the modulation schemes used, the high pass filtering corner needs to be set low, and incurs a long transient whenever the self-jammer level changes. For example, when the carrier goes from modulated to unmodulated, and vice versa. The second method eliminates this issue with an additional sample and hold control. The DC level during receive is sampled and held during transmit (when the modulated transmit changes the DC level). This approach allows a quick turnaround time at the expense of a large sampling capacitor to minimize the charge leakage impact. The third method removes DC (down converted self-jammer) at the mixer output, and reduces the mixer output compression requirement. However, the active circuitry adds noise into the system before the received signal can be amplified. This impacts the receiver noise floor, and lowers the receiver sensitivity. The last approach is not feasible for RFID receiver due to the large dynamic range required with a high power self-jammer. Besides the risk of driving the receiver front-end into compression, the noise of the self-jammer can substantially degrade the receiver sensitivity. For example, the equivalent receiver noise figure will be 60 dB under a 6 dBm self-jammer with −120 dBc/Hz phase noise (6 + (−120) − (−174) = 60 dB at room temperature). To mitigate this effect, the same LO (local oscillator) source is used for both transmit up-conversion and receive down-conversion as shown in Figure 5.11. This provides the basis of correlating the phase noise down to DC4 at the receiver mixer output. Furthermore, to capture the phase 4

Multiplying a sinusoid with itself produces a signal component at DC and one at twice the frequency. It is assumed that the high frequency content is removed by low pass filtering in the signal conditioning path.

Design of Passive Tag RFID Readers

145

noise added by the components after the transmit up-mixer, an external LO (as shown in Figure 5.11) can be used to couple the transmitter power amplifier output for use as the receiver LO. Tags are designed for low cost with a loose margin to account for process and device variations. For example, ISO 18000-6C specifies a Miller modulated tag to respond within ±15% of nominal data rate and ±2.5% drift over a tag response with 250 KHz link frequency. The receiver demodulator needs to handle these variations by: (1) keeping the baseband filter bandwidth wide to cover all the tag data rate variations while narrow enough to reject neighboring channel interferences; this requires a balance between filter complexity and receiver selectivity to cover the tag reply power spectrum; and (2) tracking the data rate changes using a continuous timing recovery circuitry in the demodulator. Frequency Generation The carrier generation specification is determined by analyzing air interface standards and applicable local regulations. For ETSI applications [8], the most stringent phase noise requirement is −116 dBc/Hz at 200 kHz offset5 originated from adjacent channel interferer cross-mixing, and −144 dBc/Hz at 3.6 MHz from out-of-band spurious emission regulation. This is tough to meet especially for highly integrated RFID transceivers. The good news is: settling time is not explicitly defined by any local standards, though it is desirable to keep it to within a few hundred µsec to enable fast monitoring of available spectra. Based on these requirements, an integer-N PLL can be used for the best spectral performance and implementation simplicity; though a low noise Fractional-N PLL is applicable in less demanding applications. In addition, the VCO frequency should be selected to be two or four times the carrier frequency to avoid high power transmit output coupling back to the VCO. If this is not possible, extra attention should be paid to isolate the VCO from noise coupling at the carrier frequency. Transmit Path Direct conversion is commonly used to lower implementation costs. The data bits are streamed in from the protocol processor, encoded in pulse interval encoding, and shaped to limit the spectrum of the transmission. The baseband signal is then up-converted to passband through IQ mixing and amplified through a power amplifier to the antenna interface. There are several issues to be noted for a RFID transmitter: 1. Receive to transmit link timing: Specified as the time interval between the reader responds and the reply from tags ends, this includes the group delay through the receiver path as well as that of the transmit path. Referred to as “T2” time in ISO 18000-6C link timing, this is defined with respect to the tag modulation symbol period. As an asymmetrical communication system, the tag response data rate can be high while the reader transmit data rate is low (for example, with Tari = 25 µsec) to limit spectrum usage. Since the transmitter group delay scales with the transmit symbol Given −116 dBc/Hz phase noise, the noise power within 200 KHz (53 dB) with a −35 -dBm adjacent channel interferer is −98 dBm. Assuming 12 dB is required to demodulate FM0 , and 1 dB implementation loss, the receiver can maintain a target sensitivity at −85 dBm.

5

146

RFID Systems

time basis (Tari), “T2” is increasingly hard to hit with higher tag data rate. One way to get round this problem is pre-empt transmission which attempts to send the transmit preamble before fully decoding the tag reply. 2. Transmit linearity: Incurred by a modulated signal passing through a nonlinear circuit, spectral re-growth increases the original baseband signal bandwidth by mixing among its different frequency components. For RFID readers, −40 dBc IM3 (third order inter-modulation) of the transmitter front-end [13] is required to meet various spectrum mask requirements at 30 dBm output level. To meet the stringent linearity requirement, a low efficiency Class A or Class AB amplifier (∼at 10% power added efficiency or below) is used due to the large back-off needed. Another option is to explore the drain modulation with a non-linear switch mode power amplifier [20]. There are two common design challenges: (1) the signal bandwidth of drain modulation is roughly 10 times larger comparing to that of I/Q modulation. This poses as a limitation on wide bandwidth protocols, but presents a lesser challenge for the narrow band RFID modulations; and (2) non-linearity in the modulation path – using the power amplifier linearization techniques developed can mitigate the problem. Detailed treatment of this topic is beyond the scope of this chapter (the interested reader can refer to [21]). 3. Transmitter AM noise: Even when the receiver front-end is not compressed, the receiver sensitivity can still be impacted by the phase and amplitude noise of the selfjammer (transmit to receive coupling). As an example, assuming a 6 dBm self-jammer carries −120 dBc/Hz phase noise, and −140 dBc/Hz amplitude noise, the equivalent receiver noise figure caused by the phase noise will be 60 dB, and by the amplitude noise 40 dB, respectively. In the case of phase noise, using a correlated frequency source for receive LO can attenuate the phase noise substantially. However, since the amplitude information is not maintained through the receive mixer LO, the coupled AM noise does not get attenuated, and becomes the next obstacle to achieve good receiver sensitivity. Aside from reducing transmit AM noise though minimizing transmit mixer noise (for example, increasing the bias current) using self-jammer cancellation offers another option which will be discussed in the next section.

5.4 Advanced Topics on RFID Reader Design This section discusses the current challenges of RFID reader design, and the approaches undertaken to address these issues. It should be recognized that the UHF RFID development is relatively new so far, and new problems in addition to the ones reviewed may occur as the deployments grow.

5.4.1 Integrated Transceiver Due to the stringent performance requirements, the reader market used to be dominated by readers constructed with discrete components. With the introduction of fully integrated RFID reader transceivers [13, 22, 23], the reader has benefited from the reduction in size, power, and cost. It is expected that the trend of integration will continue and move towards CMOS implementations.

Design of Passive Tag RFID Readers

147

To obtain a high performance CMOS implementation, two RFID specific issues should be resolved: 1. Low frequency signal contents of the tag reply overlaps with the CMOS 1/f noise range: Since the transceiver can provide very limited RF gain, the device 1/f noise of the mixer weighs heavily on the receiver noise figure within a few hundred KHz for Miller encoded and a few tens KHz for FM encoded tags. There are many developed architecture and circuit techniques to deal with 1/f noise – of these options, the passive mixer [19, 24] seems the most promising for RFID transceiver implementation. 2. Low breakdown voltage of the CMOS device at advanced CMOS process nodes: For RFID applications with transmit power up to 30 dBm, the transistor drain stress limits the conventional power amplifier design to high-voltage tolerant RF processes. To increase the possibility of integration, the use of Distributed Active Transformer (DAT) [25] by power combining is a possible option for further study. On the receiver end, the stress on the transistor bounds the mixer input and output compression. This is handled by reducing receiver RF front-end gain and/or attenuating the signal before it reaches the receiver. However, a better alternative is to actively cancel the self-jammer which will be discussed in the following section.

5.4.2 Cancellation of Transmitted Carrier Leakage For a mono-static configuration, a direct coupling path exists from transmit to receive by sharing the same antenna. The receiver is kept from saturation by adding either a circulator or an attenuator in the signal path. Using an attenuator is less expensive, but at the cost of reducing the received signal strength by the same amount. With bi-static configuration, the leakage is minimized by antenna isolation (for example, pointing the antenna in different directions). Depending on the application scenario, however, the antenna isolation may be substantially less than what can be achieved. Going forward, this issue is expected to become more serious with shrinking reader sizes and converting towards mono-static design for cost reduction. Since the coupling source is known, an alternative is to cancel the transmitter leakage with a replica from the source, the transmitted carrier. This is a simplified case of the general carrier cancellation being explored for amplifier linearization and noise reduction [26]. The formulation of this problem into an optimization loop is depicted in Figure 5.13 with four stages: (1) picking the replica source; (2) devising the method to vary the phase and amplitude of the replica; (3) deriving the error detection scheme; and (4) finding the search mechanism for an optimal solution. Without going into details of every possible implementation, the following discussion focuses on important points specific to RFID reader designs: • The replica’s phase can be changed by a phase shifter or by varying the relative I/Q amplitude in the following equation: cos(wt − w0) = cos(wt) cos(w0 )+ sin(wt) sin(w0 ). The angle shifted is represented by the ratio between sin(w0 ) and cos(w0 ). The latter method is more suitable for integrated circuit implementation with a digital-to-analog converter.

148

RFID Systems

+

From receive antenna



Replica jammer generation

replica Jammer source

To reader transceiver Front-end

Phase Amplitude change

Error Detection

Loop Filter/solution search

Figure 5.13 Self-jammer cancellation problem formulation.

• Error detection can be performed by observing the DC residue after the frequency downconversion (assuming a direct conversion receiver). However, co-channel interferers, located at the same frequency as the self-jammer, can confuse the calibration circuit during initialization. This can limit the cancellation performance especially in a dense reader environment. Instead of relying on detecting the residual carrier by measuring the DC amplitude, more sophisticated modulation can be applied to the transmitted carrier during calibration phase to offset the residual power away from DC to avoid detection interferences. Two implementation examples are shown in Figure 5.14 to illustrate the cancellation procedure. In Figure 5.14(a), the replica jammer is tapped from the transmitter output using a coupler. Two orthogonal vector signals are generated by an IQ phase splitter. The phase/amplitude change is implemented with two programmable variable attenuators commanded by the radio controller. The amplitude change is made by varying the amplifier gain on the I/Q paths in tandem, while the phase change is done by changing the gain of I relative to that of Q path. A power combiner is added before the received signal hits the received input to cancel out the self-jammer. The algorithm for detecting the error and close the control loop is implemented in software running on the radio controller by checking the DC amplitude level from the receive I and Q mixer outputs. A similar implementation for cancelling jammer with polar (phase and amplitude) control is illustrated in Figure 5.14(b). Rather than transforming the replica into orthogonal signals, the amplitude and phase are directly operated on using an amplifier and a phase shifter. There is no fundamental difference between these two implementations, except that care must be taken to ensure the replica signal generation path must have much lower noise than the receiver path. Otherwise, the cancellation mechanism can potentially add more noise to the system.

5.4.3 Dense Reader Operations It is easy to conceive the coexistence of many readers in one physical location as RFID starts to be integrated into our daily life. When many readers establish concurrent links

Design of Passive Tag RFID Readers

149

Receive from antenna

+ −

I/Q phase splitter

+

RX

Reader transceiver

+

Protocol processor

TX

transmit

Amp l Amp Q Phase/amplitude Change (Cartesian)

(a) Receive from antenna

+ −

Phase shifter

Replica jammer source

RX

Reader transceiver

Protocol processor

TX

transmit

phase

Noise filter

Noise filter

amp

Phase/amplitude Change (polar)

(b)

Figure 5.14

Transmitter carrier leakage cancellation implementation.

with the same tag population, we have to build the reader with smart intelligence to handle the following scenarios: • Reader transmission interferes with tag responses: Reader transmission is usually much higher than tag responses – the response can be easily overwhelmed even when the reader is not talking (readers still transmit carrier waves). During the dense reader operating mode, ISO 18000-6C air interface separates the tag response from the reader carrier spectrally. In the US, the tag response uses Miller subcarrier coding M = 4, 62.5 Kbps to focus the backscattering energy right in between the center of two channels. In Europe, the standard recommends using only channels 1, 4, 7, and 10 with Miller subcarrier coding M = 4, 75 Kbps to avoid the reader and tag operating in the same channel. • Reader-to-reader interference: When there are many readers working at the same time, two readers can choose the same channel, called co-channel interferers, or right at the next channel, called adjacent channel interferers. When two co-channel readers are talking at the same time, the tag population cannot differentiate where the commands are from unless they are physically separated by the channel reuse range of a cell. Since there is no base station in the system, the interference is mitigated through some combination of antenna isolation [27], synchronization [28], and interference avoidance. Synchronization describes the agreement (1) in timing for either reader talking or tag responding so that reader and tag messaging do not overlap in time; or (2) in coordinated channel usage to avoid the same channel being reused by closely located readers. This

150

RFID Systems

Reader B Tag reply collision (to different readers using the same channel)

(a) Reade A Tag replying to Reader B Tag replying to Reader A

Reader B

Reader A

(b)

Tag replying to Reader B

Tag reply collision (to different readers) Tag replying to Reader A

Figure 5.15 Tag reply collisions.

is done either with a central controller if all readers on the same network, or through side-band communications. For interference avoidance, for example, in the US, the local regulation requires the reader to randomly hop within the allocated channels from 902–928 MHz. Before hopping into the next channel and starting to power up tags, the reader can listen to the target channel to decide whether there is a possibility for collision or adjacent channel interference. • Tag to tag interference: Depending on the spectral location, we can determine two different types of interferences: (1) the tag responses come from responding to two readers occupying in the same channel (depicted in Figure 5.15(a)). From the reader perspective, it is similar to tag collision as described before. (2) The tag responses come from two readers occupying in adjacent channels (as shown in Figure 5.15(b)). As can be seen from Figure 5.15, the collision is one-sided for the tag responses. A reader can thus deduce the tag signals by throwing away the collided portion with filtering.

5.5 Conclusion This chapter provides an overview of UHF passive reader hardware design with a focus on ISO 18000-6C. Similar to the spread of the bar code technology, the proliferation of passive RFID technology would depend on continuing cost reduction while building up the infrastructure. Two of the most prominent issues encountered today are: (1) the active power consumption of the reader is still too high to enable general portable and embedded usage models. This chapter has introduced a few problems and approaches, but further progress in technology is necessary to enable the applications; (2) the integration

Design of Passive Tag RFID Readers

151

Table 5.2 Spurious emission requirement for ETSI 302 208 during reader operation.

Emission limit

47–74 MHz, 87.5–118 MHz, 174–230 MHz, 470–862 MHz 4nW

Other frequencies below 1 GHz

250 nW

Frequencies above 1 GHz 1 µW

of RFID into the application environment – progress is being made both on software and hardware, but the application experience is still limited for this new technology. Despite the challenges, with the key technical/application problems being addressed and resolved over the last few years, it is expected that we will see reader implementation reach maturity to enable an explosive uptake of adoption in the near future.

Problems 1. Spurious emissions are transceiver spectral emissions at frequencies other than the intended carrier frequency. For ETSI 302 208, the spurious emission requirement is partially specified in Table 5.2. (a) What is the spurious emission represented in dBc when the reader is transmitting 30 dBm? (b) There are four proposed RFID reader channels (865.7 mHz, 866.3 MHz, 866.9 MHz, and 867.5 MHz). What is the minimum phase noise requirement for the RFID reader at 3.7 MHz from the carrier? (c) Does the transmit noise requirement change if the reader is operating at 20 dBm? 2. Receiver sensitivity is defined as the minimum detectable signal power level; given Eb/No = 15 dB is required to demodulate the incoming tag signal at a satisfactory packet error rate (for example, 10−3 ), the data rate is 75K bits per second, with a receiver noise figure of 15 dB, what will be the receiver sensitivity under a white Gaussian noise channel? 3. An RFID reader operates at 2.4 GHz with transmit power of 20 dBm, and reader transmit antenna gain of 3 dB. (a) Using the range equation (Equation 5.1) what is the reader-to-tag path loss when the distance between tag and reader is 1 m and 2 m? (b) What is the available power (assuming tag antenna gain = 3 dB, and a 3 dB polarization loss) for the tag to harvest when the distance between tag and reader is 1 m and 2 m? (c) With the range equation (Equation 5.2), and a total of 19 dB of tag loss (Ltag ), what is the minimum receiver sensitivity of an RFID reader operating at 2.4 GHz at 1m away? At 2m away? (d) What can you do to improve the operating range given that the reader transmit power is limited to 20 dBm? Please assume two different scenarios (1) the tag does not have enough power to activate; (2) the backscattered power form the tag is too small to demodulate.

References [1] Finkenzeller, K. (2003) RFID Handbook: Fundamentals and Applications in Contactless Smart Cards and Identification, 2nd edition, Chichester: John Wiley & Sons, Ltd.

152

RFID Systems

[2] Park, J.S. (2007) Trends for mobile RFID reader SoCs, developed by Korean ASIC companies, RFID Workshop, IEEE RFIC Symposium. [3] Information technology-Radio frequency identification for item management, Part 6: Parameters for air interface communications at 860 MHz to 960 Mhz, ISO-IEC CD 18000-6C, 2007. [4] Balachandran, G.K. and Barnett, R.E. (2006) A 110 nA voltage regulator system with dynamic bandwidth boosting for RFID systems, IEEE Journal of Solid State Circuits, 41(9): 2019– 2028. [5] Karthaus, U. and Fischer, M. (2003) Fully integrated passive UHF RFID transponder IC with 16.7-µW minimum RF input power, IEEE Journal of Solid State Circuits, 38(10): 1602– 1608. [6] Pillai, V., Heinrich, H., Dieska, D., Nikitin, P.V., Martinez, R., and Rao, K. V. S. (2007) An ultra-lowpower long range battery/passive RFID tag for UHF and microwave bands with a current consumption of 700 nA at 1.5 V, IEEE Transactions on Circuits and Systems I: Regular Papers, 54(7): 1500– 1512, July. [7] Operation within the bands 902– 928 Mhz, 2435– 2465 MHz, 5785– 5815 MHz, 10500– 10550 MHz, and 24075– 24175 MHz, FCC title 47 part 15. [8] Electromagnetic compatibility and radio spectrum matters (ERM): Radio frequency identification equipment operating in the band 865 MHz to 868 MHz with power levels up to 2 W. Part 1: Technical requirements and methods of measurement, ETSI EN 302– 208-1, 2007. [9] Information technology-Radio frequency identification for item management, Part 6: Parameters for air interface communications at 860 MHz to 960 MHz, Amendment 1: Extension with Type C and Update of Types A and B, ISO-IEC CD 18000-6, 2004. [10] Protocol specification for a 900 MHz class 0 radio frequency identification tag. MIT Auto-ID Center, Feb. 2003. [11] 860 MHz-930 MHz class I radio frequency identification tag radio frequency & logical communication interface specification candidate recommendation. ver. 1.0.1, MIT Auto-ID Center, 2003. [12] EPC UHF radio frequency identification protocols: Class 1 generation 2 UHF RFID. Ver. 1.2.0, EPCglobal, 2007. [13] Chiu, S. et al. (2007) A 900 MHz UHF RFID Reader Transceiver IC, IEEE Journal of Solid State Circuits, 42(12): 2822– 2833. [14] Sklar, B. (2001) Digital Communications: Fundamentals and Applications, 2nd edn. Englewood Cliffs, NJ: Prentice-Hall. [15] Maguire, Y. and Pappu, R. (2009) An optimal Q-algorithm for the ISO 18000-6C RFID protocol, IEEE Transactions on Automation Science and Engineering, 6(1): 16–24. [16] MSDN Architecture Center. Available at: http://msdn.microsoft.com/en-us/library/aa479362.aspx#rfidtech over topic5. [17] Lee, T. (2003) The Design of CMOS Radio-Frequency Integrated Circuits, 2nd edn. Cambridge: Cambridge University Press. [18] Razavi, B. (1997) RF Microelectronics. Englewood Cliffs, NJ: Prentice Hall. [19] Zhou, S. and Chang, M.-C.F. (2005) A CMOS passive mixer with low flicker noise for low-power direct-conversion receiver, IEEE Journal of Solid State Circuits, 40(5): 1084– 1093. [20] Berglund, B., Johansson, J., and Lejon, T. (2006) High efficiency power amplifiers, Ericsson Review , 3: 92–96. [21] Raab, F. et al. (2003) RF and microwave power amplifier and transmitter technologies – Part 4, High Frequency Electronics, Nov. pp. 38–49. [22] Khannur, P.B. et al. (2008) A universal UHF RFID reader IC in 0.18-µm CMOS technology, IEEE Journal of Solid State Circuits, 43(5): 1146– 1155. [23] Austria Microsystems. Available at: http://www.austriamicrosystems.com/eng/content/view/full/7540. [24] Valla, M., Montagna, G., Castello, R. Tonietto, R., and Bietti, I. (2005) A 72-mW CMOS 802.11a direct conversion front-end with 3.5-dBNF and 200-kHz 1/f noise corner, IEEE Journal of Solid State Circuits, 40(4): 970– 977. [25] Aoki, I., Kee, S.D., Rutledge, D.B., and Hajimiri, A. (2002) Fully integrated CMOS power amplifier design using the distributed active-transformer architecture, IEEE Journal of Solid State Circuits, 37(3): 371–383. [26] McNeilage, C., Ivanov, E.N., Stockwell, P.R., and Searls, J.H. (1998) Review of feedback and feedforward noise reduction techniques, in Proceedings of the 1998 IEEE International Frequency Control Symposium, May, pp. 146–155.

Design of Passive Tag RFID Readers

153

[27] Leong, K.S., Ng, M.L., and Cole, P.H. (2006) Positioning analysis of multiple antennas in a dense RFID reader environment, in International Symposium on Applications and the Internet Workshops, Jan., pp. 56–59. [28] Leong, K.S., Ng, M.L., Grasso, A.R., and Cole, P.H. (2006) Synchronization of RFID readers for dense RFID reader environments, in International Symposium on Applications and the Internet Workshops, Jan, pp. 48–51.

6 RFID Middleware: Concepts and Architecture Nathalie Mitton, Lo¨ıc Schmidt, and David Simplot-Ryl INRIA Lille-Nord Europe, France

6.1 Introduction The RFID middleware is a central point in the integration process of any RFID solution [1–3]. There are several kinds of RFID tags and consequently several kinds of readers. If the end-user application is directly connected to readers, the management of all of them can be a hard task. Furthermore, this application would have to deal with a larger amount of data than necessary. Indeed, the RFID readers have to regularly scan their environment to know which tags are present. A middleware would enable data selection, that is, the application would only receive the information it requires. The RFID middleware is a set of components which aims to manage RFID readers, deals with RFID-events and data, and is connected to end-user applications. This chapter describes the general architecture of such a middleware. We will start by presenting the management of readers. This means reader protocol/interface, monitoring, and settings. The reader protocol deals with interactions between readers and controllers. It offers a way to insulate the host from the technical aspects of how a reader communicates with tags. There are two main ideas in such a protocol. Firstly, it has to provide an abstract syntax for messages between the controller and the reader. Secondly, it must implement different transport layers (Bluetooth, wifi, serial, TCP/IP, . . . ). An issue here is how to define a protocol which can interface with multiple kinds of readers, not only RFID, but also sensors, bar-codes, etc, in a transparent way for upper layers [4]. RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

156

RFID Systems

The next point is the data management. Indeed, readers can be configured to be aware of everything happening in a warehouse. This results in a large amount of data, which must be treated and stored intelligently in a database. The important point here is the data “filtering and collection.” It means transforming a tag reading flow in a comprehensive “what, when and where” record. The way to retrieve data has a direct impact on the ease of developing and connecting the system with end-user applications. This is the role of the ALE standard ratified by EPCglobal Inc. [5]. The application level events (ALE) is an interface between business applications and the middleware data. Finally, we will present how data are stored in the system and shared between multiple sites, multiple partners. The idea is to provide databases (EPCIS) [6] and a way to retrieve and query the right one. The main component here is the object name server (ONS) [7]. This component provides a mechanism to retrieve information about an object from its identification number. When an application asks the middleware more details about an object, the system first checks its local database. The EPCIS acts like a database, storing EPC-related events. If the object requested is not registered in it, the system then forwards the request to the ONS owning the database which contains this information. The ONS acts like the well-known Internet domain name server (DNS) [8], with an ONS root. Having a unique ONS root is both a technical and political problem. The challenge here is to define a multi-root structure for ONS providing load-balancing and scalability for this service, but also a shared governance for the political concern. Another way of retrieving information is the discovery service. It has yet no standards defining this service, but it aims to offer the ability to find all servers sharing RFID events about a product.

6.2 Overview of an RFID Middleware Architecture 6.2.1 The Need for a Middleware A middleware is an intermediate software layer between the applications and the network, allowing the dialog between heterogeneous applications. It aims to accomplish technical tasks for business applications connection and data exchange. In RFID systems, such a middleware has to manage readers, to deal with RFID events and to be connected to end-user applications. In some cases, there is no need for a ‘middleware’, as in small and unique applications like “count number of tags read”, or “list the tags in the scope of a reader.” But when the need to share information coming from readers between several applications is rising, then messages from readers need to be application specific. For example, a conveyor application does not just need the list of tags read, it also needs to know which of these read tags is from the pallet and which are tags from products. In most cases, physical details are not useful for business logic. Therefore, a middleware becomes mandatory when the RFID applications become more complex. It is a unique access point for collected data. It associates read items with an activity and a location. It homogenizes the access of the data via a standardized interface which allows: • Reader management (decoupling physical and logical readers). • Data management (formatting, aggregation, filtering). • Call to application functions or results sending.

RFID Middleware: Concepts and Architecture

157

Business Application

Business Application

Application Interface Middleware Reader Protocol Readers Hardware/Software component Interface Middleware

Figure 6.1

Outside a RFID middleware.

• Rules and exceptions management. • Data collection and treatment in real time or in asynchronous mode. RFID middleware is considered one essential intelligence-added component of any RFID system and could be linked with other company information systems such as external databases, business partner information systems or warehouse management systems, etc. Figure 6.1 shows the position of a middleware in an RFID system as well as two key concepts from an external point of view: the reader protocol and the application interface. Indeed, the middleware is the link between the readers and the business applications. Let’s now have a look at this middleware black box to see other key concepts of a RFID middleware and how they are used to perform the tasks of the middleware. The middleware specification and challenges are nowadays partially addressed in the scope of the Architectural Framework of EPCglobal [9]. EPCglobal1 is a subscriberdriven organization comprised of industry leaders and organizations focused on creating global standards for the EPCglobal Network. EPCglobal has developed a collection of interrelated standards for hardware, software, and data interfaces, together with core services that are operated by its delegates. In this chapter, we will focus on RFID middleware functionalities and services and will mostly refer to the standards developed by EPCglobal.

6.2.2 Architecture Connecting readers directly to applications raises several main problems: • Applications are reader-dependent. This means that if the need to change the readers appears and if the new readers do not use the same communication protocols as the former ones, part of the application also needs to be recoded. 1

http://www.epcglobalinc.org

158

RFID Systems

Business Application

Business Application

Application-Level Events (ALE)

Reader Management

Filtering and Collection

Reader Management

Reader Protocol (RP, LLRP)

Readers Hardware/Software component Interface(EPCglobal Standard) Middleware

Figure 6.2 Inside a RFID middleware.

• Every application connected to the readers receives the same information about the tags read. Moreover, this information is raw (it is a list of tag identifiers). Nevertheless, it may need only a part of these information, which may differ from one application to the next. Applications are thus flooded with too much information. • Finally, if there is a great amount of readers or/and applications, the system is likely to fail since it is saturated and overloaded. Thus, connecting readers directly to applications is not scalable and wastes resources uselessly. Using a middleware between readers and applications solves these problems. To do so, the middleware is divided into several blocks, as shown in Figure 6.2. This figure shows a part of the EPCglobal Architecture Framework [9]. The reader protocol interface allows the first problem regarding the heterogeneity of the readers to be solved. This interface provides a way to monitor different readers, can configure them, check their status (reading, idle, etc.), possibly correct any operational problem, etc. Reader protocols will be detailed in Section 6.3. The two other problems are managed through both the Application-Level Events (ALE) interface and the Filtering and Collection module. This latter module can also be seen as the ALE engine, that is, it allows operations on data, based on the events defined by applications through the ALE interface. These modules (ALE interface and Filtering and Collection) will be described in Section 6.4. Indeed, thanks to the middleware, information about tags is sent to applications only if the latter have emitted a data request through the ALE Interface. Then, instead of forwarding raw information, this is filtered, grouped, aggregated by the ALE engine before being sent to applications. In this way, applications directly receive exploitable data.

RFID Middleware: Concepts and Architecture

159

Then, information retrieved from tags through the middleware may need to be persistent, that is, one should be able to access the data afterwards. This is, for instance, the case for traceability business applications. Indeed, we need to record the data for later use, to translate ALE events into business events, that is, to give them a business signification. For instance, a tag’s reading states that Object A has been packed with Object B, which can be translated as “Order number 10 is ready for expedition from date 2009-03-12 at 12:28.” This event needs to be stored. This illustrates the role of the EPCIS (EPC Information Services). The goal of EPCIS is to enable disparate applications to leverage Electronic Product Code (EPC) data via EPC-related data sharing, both within and across companies. In order to perform these operations, the EPCIS provides interfaces to capture (store) and to query (retrieve) information. Finally, a business application may need to retrieve a product name or manufacturer from its EPC identifier that is not stored in the company EPCIS. To do this, it needs to query an Object Name Service (ONS), which, in a DNS fashion, will answer the request.

ONS

EPCGlobal Network Services

ONS Interface

EPCIS Accessing Application

Local ONS

ONS Interface Partners

“Pull” or “Push” mode “Pull” or “Push” mode

EPCIS Query Interface EPCIS Repository

Optional bypass for realtime 'Push'

EPCIS Capture Interface EPCIS Capturing Application ALE Interface Filtering and Collection

Reader Management

Reader Protocol Interface

Reader Management Interface

RFID Reader Hardware/Software component Interface (EPCglobal Standard)

Figure 6.3

EPCglobal Network Architecture.

160

RFID Systems

All these modules such as EPCIS or ONS have to be connected, directly or not, to the middleware as shown in Figure 6.3. We detail these modules and their interaction with the middleware in Section 6.5.

6.3 Readers Management The aim of the reader protocol and of the reader management are respectively to provide: • a uniform communication interface between different readers and middleware so that the application can function independently of the hardware; • a way to manage and monitor readers. In this section, we firstly present the role of the reader protocol and the two standards ratified by EPCglobal, the Reader Protocol and the Low-Level Reader Protocol. The second part describes the management of readers: it exposes the need of a management tool for readers and presents EPCglobal Standards which deal with the reader management (the Reader Management Standard and the Discovery, Configuration and Initialization Standard). Reader management should also include the management of interference between readers. Techniques like reader anti-collision protocols or activity scheduling can be applied [10, 11]. These techniques are described later in this book.

6.3.1 Reader Protocol/Interface The reader protocol defines interactions between readers and application softwares (or hosts). Figure 6.4 shows the links between the host, a reader and tags. There are a lot of parameters to give to readers in order to work efficiently, such as the number of antennas connected to it, the transmitter power, etc. The role of the protocol is to provide an interface, or API, to control and command the reader to launch reading actions (tag ID or tag additional data), but also writing, killing or locking ones, etc. The most common case is that reader’s vendors provide their own protocol, their own API to communicate specifically with their own readers. The problem here is the difference between these hardware specific control interfaces. Different architectures,

Air Interface HOST

Reader Protocol Ethernet Wireless Serial etc.

Tags

RF Signal

Figure 6.4 A host-to-reader connection through reader protocol.

RFID Middleware: Concepts and Architecture

161

different protocol options, etc. increase the maintenance task cost when new drivers or new hardwares need an application modification. This cost is also significantly increased when large number of readers have to be operated by various application softwares. The role of the middleware here is thus to provide a Hardware Abstraction Layer (HAL) between readers and host. The HAL goal is to provide an API, offering a way to send commands to readers independently of the reader specific API commands. An underlayer will then translate these commands into reader-readable commands and send them through a specific transport protocol to the reader. EPCglobal defines two reader protocols in order to insulate the host from details of the reader and tags communication. The reader protocol deals with the RFID reader’s commands such as inventory tags, read, write, kill or lock tags. The low-level reader protocol is low-level because of its awareness of the air protocol between readers and tags. This section describes these protocols in more details. 6.3.1.1 Reader Protocol The EPCglobal reader protocol [12] standard defines three layers (see Figure 6.5): • Reader Layer defines the abstract syntax used in messages between the readers and host. • Messaging Layer specifies how the message is encompassed in the network transport dependent message. • Transport Layer corresponds to the networking facilities provided by the operating system or equivalent. Defined from a conceptual object view, the RP specifies commands in the Reader Layer that readers may perform. In the specification, the object ReaderDevice implements the command getEPC which returns the EPC of the reader. No matter how it is implemented by manufacturer, the protocol knows that this command performs this operation. A messaging/transport binding (MTB) is composed by one messaging layer and one transport layer, providing different ways of transport (e.g. TCP/IP versus Bluetooth versus serial). The role of a MTB is to transport the RP command between host and readers. It defines transformations of messages from a Reader Layer before sending it to the Transport Layer.

Reader Layer

Messaging Layer MTB Transport Layer

Figure 6.5

Layers of the EPCglobal Reader Protocol.

162

RFID Systems

It also specifies which Transport Layer to use and how to set up connections. The reader protocol provides different MTBs but others can be defined. 6.3.1.2 Low-Level Reader Protocol In contrast, the Low-Level Reader Protocol (LLRP, also defined by EPCglobal [13] is sensitive to the details of the air-interface protocol (between readers and tags) and provides specific parameters and controls in order to reveal the command and timing parameters of the RFID air protocol to the controller. The control of readers is based on specifications (called ROSpecs for Reader Operation Specifications). These specifications are sent to and stored in the reader, so readers can perform operations without overloading the network with command and status data. In general, a reader can store more than one specification. This implies a significant cost to the reader because of the amount of computational and memory resources needed by the ROSpecs. The ROSpec describes a reader configuration, including the reader’s ID, the antennas to be used and the air protocol at each antenna, start and stop triggers for inventory action, transmit power and receiver sensitivity, and a priority. When a trigger condition is met, the reader performs operations of the corresponding ROSpecs. Thanks to the priority level, some of these specifications can be pre-empted by a higher priority ROSpecs. The LLRP standard defines parameters and controls for the UHF Gen 2 air protocol, but it also provides mechanisms for additional commands or additional parameters. Readers can ignore such kind of extensions, but they have to return errors if these extensions are not supported. To conclude this part on reader protocols, we can say that events generated by readers through these protocols are of the form “Reader A saw EPC X at time T” each time a tag is read. This may seem meaningless in a business context, and may overload the network with useless messages to end-user applications. Therefore, filtering and data aggregation are needed in order to provide application-specific events. This part of the middleware is described in Section 6.4.

6.3.2 Manage and Monitor In systems with a lot of readers, having a way to manage them and to monitor them can be very useful. Indeed, there is a lot of parameters to monitor in a reader. Therefore, we need a reader management tool that offers a way to monitor readers in a RFID middleware. It permits retrieval of information from readers such as their identity, their number of antennas, the air protocol configured, the number of tags read, the status of the communication channels, etc. The reader can also send some alerts for an operational problem. In large RFID deployments, interference between readers can occur when readers placed in the same area attempt to send messages to RFID tags. Regarding this issue, the communication channel’s configuration and monitoring of readers is important. This manages reader transmission simultaneously in order to prevent interferences for example. Due to the number of different vendors of different RFID readers, the managing and monitoring task is very complex. Indeed, readers do not have same hardware capabilities or configuration, control and monitor parameters. A hardware abstraction layer is also required more here.

RFID Middleware: Concepts and Architecture

163

Reader Management Command Set XML Serial

Figure 6.6

SNMP TCP

UDP

Protocol layers mapping.

The EPCglobal Reader Management standard [14] (RM) defines management protocol specifications. Based on the same structure as the reader protocol (Figure 6.5), this standard defines MTBs for the management process of reader health. Figure 6.6 shows two MTB examples. The command set described in the RM specification allows the host to set or get a description from the reader, the location description, the operational status, etc. It also defines a lot of commands for the management and monitoring of readers, antennas, notifications, alarms, etc. All these operations can be categorized into three types of operations according to the task they perform: • DO-type The reader performs an action. • SET-type The reader changes the internal variable state. • GET-type The reader sends the internal variable state to the host. Some of these operations require special communication patterns, such as request/ response, or asynchronous messages. The specifications of the RM Protocol define three kinds of communication channels with their own behavior. Used for communication between the Reader Layer and the Message Layer, channels are independent. To be compliant with these specifications, readers may support one or more of each of the following channels: • Command Following the request/response pattern, this channel is used for requests from the host to the reader, and for responses from readers. • Alarm This channel allows asynchronous messages from the reader to the host only. • Notification Notification channels are used for delivering tag data. The RM standard aims to provide mechanisms to monitor reader’s operational status and to notify the host of potential operational problems. Another standard used for reader management is the Discovery, Configuration and Initialization [15] (DCI). Using the Control and Provisioning of Wireless Access Points (CAPWAP) protocol [16], the DCI specifies the device called the Access Controller. This device can perform some initial operations such as reader’s identification, configuration, and network connectivity management. The aim is to provide a way for readers to discover hosts and hosts to discover readers. The Access Controller can also configure and initialize readers.

164

RFID Systems

Other Network Services Step 1 Reader Step 2 Access Controller

Figure 6.7

Step 4 Step 3 Host

DCI overview.

Configuration includes of course configuration of the reader, but also updates the software and the firmware of readers. The initialization step provides parameters to the reader in order to begin the operation. Figure 6.7 describes the steps of DCI. The first step is the reader network addresses assignment and the access controller addresses determination. Step 2 represents the discovery of host and reader initialization. Thirdly, the access controller gives indication to the host, and finally, the communication starts (Step 4). Now we can manage readers, plug them and retrieve data from tags through a reader protocol and reader management protocol. The next important component in a RFID system is a component that filters and aggregates all tag data collected from readers.

6.4 Data Management and Application-Level Events In a middleware RFID, we find the ALE Interface and the ALE Engine, also called Filtering and Collection. The ALE specification [5] is a software specification indicating required functionality and behavior, as well as a common API expressed through XML Schema Definition (XSD) and Web Services Description Language (WSDL). The ALE is thus an interface and provides all the functionalities this implies. Through this interface, clients may interact with filtered, consolidated EPC data and related data from a variety of sources. Therefore, the ALE interface cannot be decoupled from the data management held by the ALE engine. The goal of the ALE (interface and engine) is to reduce the volume of data that comes directly from EPC data sources such as RFID readers into coarser “events” of interest to applications. In this layer, common optimizations are efficient RFID data management and load-balancing methods [17, 18], indexing continuous queries [19], enhancement and contextualization of information [20, 21] . In this section, we focus on the way the data can be handled and the Application Level Event or ALE, a standard created by EPCglobal. We first highlight the role of the ALE as well as the functionalities it provides. Then, we detail the specification standard of the ALE. Note that the standard gives only the specification of the ALE and the functionalities it has to offer. It does not define the way these functionalities should be carried out.

RFID Middleware: Concepts and Architecture

165

6.4.1 Data Management and ALE Functionalities The role of the ALE is to provide independence between the infrastructure components that acquire the raw data (like RFID readers, for instance), the architectural component(s) that filter and count that data, and the applications that use the data. This allows changes in one without requiring changes in the other, offering significant benefits to both the technology provider and the end-user. ALE standard specifies an interface (not an implementation) which involves: • • • •

receiving data from one or more data sources (readers); accumulating data over intervals of time; filtering to eliminate duplicated data or data with no interest; counting and grouping.

ALE allows aggregation and filtering of tag data over a period of time. An ALE server specifies when to start collecting data, when to stop collecting data, how to organize and sort the data and when to send the data to interested parties. An ALE client allows communication with any compatible ALE server to define data requirements and receive reports. To illustrate the benefit of the ALE, let’s assume a smart shelf as depicted in Figure 6.8(a). It is equipped with three RFID readers which all have two antennas (each antenna reads a unique board of the shelf). We assume that there are objects lying on the shelf, all equipped with a RFID tag. If there is no middleware or ALE, every reader of the shelf communicates continuously the list of tags/objects it has on it. This information is clearly neither useful not exploitable. Indeed, if an application is not written in a manner to handle this large throughput of data, it can suffer from severe scalability problems, perform poorly, or at worst, crash. This is where the ALE specification comes in. It defines a configurable set of data gathering techniques that higher order business

SCREEN

SCREEN

A

A

B

B A

Reader 1

SCREEN

B

Reader 2

(a) Physical view.

Reader 3

Reader 1

Reader 2

(b) Logical View 1.

Figure 6.8

Smart shelf.

Reader 1

Reader 2

(c) Logical View 2.

Reader 3

166

RFID Systems

applications can then use to receive more specialized set of tags to process. ALE provides a standard way to describe and define these data gathering techniques to help decouple a business application from the source of its RFID data. First, ALE configures logical readers, which are discorrelated from the physical readers and antennas. This can be traduced by the fact that according to the application, we can configure logical readers in several ways. For instance, we may need to read and manage independently the right and left sizes of the shelf as one can see on Figure 6.8(b). In such a case, we configure two logical readers on the shelf: one reading the left side, composed by Reader 1 and Antenna A of physical Reader 2 and one monitoring the right side of the shelf, composed by physical Reader 3 and Antenna B of physical Reader 2. Another application having access to the same shelf may need to manage the object of the shelf according to height of the board on which they are. In this latter case, we would configure three logical readers as shown by Figure 6.8(c). The first logical reader is thus composed of Antenna A of both physical readers 1 and 3, the second is composed by Antenna B of both physical readers 1 and 3, the third logical reader is physical Reader 2 itself. This then allows a clearer view of the shelf by the application and allows it to query only one logical reader at once. Second, thanks to the ALE, we will be able to configure, aggregate and filter the events we need. For instance, we may need one report every minute. In this case, the ALE aggregates data over time and over all readers and antennas and filters duplicated data. ALE sends a report every minute instead of continuously. We may also need to know only which objects which have been removed or/and added in a period of time. In such a case, ALE filters every parasite data and reports only objects that have (dis)appeared during the last period of time. Finally, ALE allows grouping and counting objects. For instance, a request would be to know how many objects of each kind are available on the shelf. In this case, ALE groups the objects kind by kind and counts them. The shelf then only reports a line by kind of objects (for instance, at time T , shelf had 3 books and 10 pens). All this reduces the volume of data as soon as possible (reader or middleware) and makes data coming from RFID systems understandable by the information services.

6.4.2 Specs and Reports In order to provide every functionality mentioned, several aspects have to be specified. Indeed, to allow the middleware to manage the data (filtering, aggregating, etc.), we first need to define a cycle during which data are acquired and managed. For it, we first define a reader cycle. A reader cycle is the smallest unit of interaction between the application and the (logical) readers. Nevertheless, this unit is highly readerdependent. For example, for an ALE implementation directly embedded in an RFID reader device, the reader is here represented by the communication pathway between the ALE subsystem and the RF protocol subsystem. In such a case, a reader cycle might represent one iteration of the RF protocol used to communicate with RFID tags. But if the ALE implementation is provided by a middleware which communicates with an outboard RFID reader device, a reader cycle is a unit of interaction defined by the protocol used to interact the middleware and the reader. This latter may correspond to one or several RF protocol iterations.

RFID Middleware: Concepts and Architecture

167

Then, we define the event cycle (for reading API) or command cycle (for writing API). An Event Cycle (resp. Command Cycle) consists of a boundary which indicates when to start collecting data (resp. writing data) and when to stop collecting data (resp. writing data on tags). This start and stop conditions can be time, stability or trigger-based. A timebased condition is simply a number of milliseconds to wait before starting or stopping the boundary. A stability-based condition is only valid for stopping a cycle and is used to indicate stability in the set of tags detected (no new tags added or removed). A triggerbased condition allows an external controlling process to start or stop the cycle. Note that triggers are an expected extension point in the ALE specification as the ALE specification does not define any standard trigger mechanisms. If we take the example of the smart shelf again, a time-based condition event cycle example would be “Read from Reader A for five seconds and Report every tag whose EPC belong to the A class.” The event cycle in such a case is 5 seconds. At the conclusion of an event cycle (resp. a Command Cycle), data that has been collected (resp. tags that have been operating as well as the operation result) during the cycle is filtered and grouped. A set of reports are generated and sent to the ALE client. The ALE standard defines specs and reports linked to these cycles. “Specs” define a language, a format to send the client request, to make it understandable by the system. For instance, we will use spec definition to send “Read from Reader A for five seconds and Report every tag whose EPC belong to the A class.” “Reports” define the way the result of the request should be sent. These Specs and Reports are translated in a XML language Specification Definition [22–24]. There are four specs and reports: • • • •

Event Cycle Spec (ECSpec): ALE Client request in Reading API. Event Cycle Report (ECReport): ALE response to an ECSpec. Command Cycle Spec (CCSpec): ALE Client request in Writing API. Command Cycle Report (CCSReport): ALE response to a CCSpec.

6.4.2.1 Event Cycle Specs and Reports An ALE Client’s Declarative Specification (ECSpec) includes three fields: • Location (where): Specifies the logical reader(s) required. • Boundaries (when): Specifies the event cycle type (Continuous, Interval or Trigger) and boundary (Duration, Field status or Trigger). This defines when to start and stop collecting data. • Report contents (what): Specifies what kind of events should be reported. The report content is driven by the following options: • Output Set: Specifies what tags should be reported from the last reading: only current tags, new tags (addition) or missing tags (deletions). • Filtering: Applies a set of patterns to the tags being reported to filter them in or out. Filtering is optional and more than one or both types can be used. • Grouping: Applies a set of patterns to the tags being reported to group logical units together. For instance, you could group all products by a certain company or all products of a certain type together.

168

RFID Systems

• Data Format: Indicates how the data should be reported: only a count of data (very useful for grouping), or the actual tag ID values in one of their supported formats (as defined by the EPC Tag Data Standard). • If empty: Specifies what to report if there is no EPC to send. A report (ECreport or CCReport) is then composed of two fields: the header and the output of the report. This latter is the answer content to a given ECSpec or CCSpec. The former includes several fields of which the most important are the ECSpec or CCSpec it answers, the start and stop condition and time of the event cycle, the readers handled and if any, the readers that ALE failed to question. To illustrate this, let’s go back to our smart shelf. Let’s assume that we have the following request: “Report once per minute about things added and removed from Shelf Reader number 1.” The ECSpec to be built for such a request should thus contains two different reports, one for new objects and one for objects taken away. Note that spec report names are arbitrary and are only used to identify different reports. The final ECSpec will thus be: Logical Readers Time Boundaries

Reader 1 Start Continuous Stop Duration = 60 seconds Report Spec 1 ‘‘New’’

Output Report Spec 2 ‘‘Taken away’’

Output set Content Report if empty Output set Content Report if empty

This ECSpec will be translated in a XML language into: Reader1 60000

Additions Field epc (epc-uri) False Deletions Field epc (epc-uri) False

RFID Middleware: Concepts and Architecture

169

To which the shelf will answer with a ECReport: Header

Reader Start time Duration Report 1 Tag1

Output Tag2

1 Timestamp = 2009-03-11T16:37:00Z 60000 ms ID Designation ID Designation

10210 Cookies 31210 Chocolate

An ALE Event Cycle Spec is presented to the API in one of the three following modes: • Subscribe Mode: Defines the ECSpec ahead of time like Poll. It does not require the ALE client to continually request data from the server. Instead the ALE client adds a subscription to the ECSpec that the ALE server communicates. One example of an ALE subscription would be an ALE client that opens a TCP port that the server can connect to and send reports through it. • Immediate Mode: Allows the client to send an ECSpec and have it be processed and fulfilled immediately, returning the reports when the ECSpecs first Event Cycle Boundary completes. This option requires the client to continually send the ECSpec definition to the server with every request. • Poll Mode: The Poll mode of interaction with an ALE server is similar to Immediate, except the ALE client defines the ECSpec on the server ahead of time. It gives the ECSpec a name which allows the ALE client to request data from it using the name instead of sending the ECSpec every time. 6.4.2.2 Command Cycle Specs and Reports Command Cycle Specs and Reports concern writing API on tags. With a CCSpec, an ALE client describes which operations to perform on tags during a command cycle. Each active CCSpec gets exclusive access to the tags concerned. As for the ECSpecs, a CCSpec defines the location (which reader(s)?) and the start and stop conditions of the command cycle. It also defines some specifications, such as the filters to apply (if any), the kind of report if no tag to handle and a list of operations. Each operation to be performed specifies the operation type, the field to operate upon and the data input format. These writing operations are the following: • • • • • • • • •

Initialize: Prepare a bank for using variable fields. Read : Read contents of a field. Write: Write contents of a field. Add : Add a new field. Delete: Remove field. Lock : Change access permission. Kill : Kill tag. Password : Provide password to enable subsequent ops. Check : Check memory state consistency.

170

RFID Systems

Then, every operation is performed in the same way as the event cycles. For instance, we need to write in a tagged product which enters the warehouse. The ALE CCSpec sent will be: Logical Readers Time Boundaries

Reader 1 Stop Tag count = 1 Filter

Commands

Command Spec 1

None INIT PRODUCT Commands ADD FIELD ADD FIELD

afi = 0xC1 Warehouse = #1 Date = 2009 − 03− 12T 12 : 57 : 00Z

This CCSpec will also be translated into a XSD standard. It is then followed by a CCReport as following:

Header

Reader Start time Duration Report 1

Output Tag1

1 Timestamp = 2009-03-12T12:57:00Z 135 ms Op1 SUCCESS Op2 SUCCESS value = 1 Op2 SUCCESS value = 2009-03-12T12:57:00Z

6.4.3 Research Challenges The ALE may be connected to several readers and these readers may read several tags. In such a case, the ALE may have too many tags to filter and aggregate. In order to provide a scalable middleware, the task of filtering and aggregation should be distributed over different entities. This would allow the ALE engine to manage more readers. Nevertheless, this is a challenging task as it has to remain transparent to the application, that is, the latter should not have to change its behavior when using this distributed ALE engine. A simple solution would be to duplicate the ALE engine to manage readers, but it is not transparent for the application since it has to send one ECSpec by ALE. In addition, this solution would not be scalable. Another drawback is that the application has to handle the fact that a tag may be read by two readers belonging to different ALEs. The tag ID will appear on both ECReports and the application has to work on the reports to delete duplicated tags. In [25], the authors propose a Global-ALE which upon receiving an ECSpec from the application splits it and distributes it to the sub-ALEs. The Global-ALE receives ECReports from sub-ALEs and builds one global ECReport for the application. This solution provides transparency regarding the application. It also limits the overload of readers events, but another bottleneck may appear between sub-ALEs and the Global-ALE. An efficient distributed ALE will contain both solutions. On one hand, it must be transparent for the application, that is, the application sends only one ECSpecs and receives only one ECReport. On the other hand it should be distributed in a peer-to-peer manner so that it minimizes the bottleneck problem. This is one of the hot issues on which researchers are currently working.

RFID Middleware: Concepts and Architecture

171

6.5 Store and Share Data In order to store and share data between different sites of the same company or between companies, EPCglobal defines three standards (two of them are ratified, one is in development). The first one, called EPC Information Services [6], provides interfaces between applications that capture data and those that need it and can store the data captured. The second standard defines an Object Naming Service [7] that performs a simple operation: return an EPC related ressource (the EPCIS address which contains data of a given EPC, html link, web services, . . . ). The last standard (in development) describes the EPC Discovery Services [26], which search and return locations storing EPC-related data.

6.5.1 EPC Information Services The EPCIS standard is organized into three layers: • Abstract data model layer: specifies the generic structure of EPCIS data with general requirements for creating data definitions. • Data definition layer: specifies which data is exchanged via EPCIS. • Service layer: specifies service interfaces: capture and query. The EPC Information Services (EPCIS) standard defines three interfaces, the EPCIS Query Callback Interface, the EPCIS Query Control Interface (both part of the EPCIS Query Interface) and the EPCIS Capture Interface, which interact with two modules: the EPCIS repository and the EPCIS Capturing Application. Figure 6.9 positions these interfaces between components. We explain these five new notions in this section (the ALE Interface was described in Section 6.4). The EPCIS Capturing Application aims to process ALE events with business information. This is the component that can check for business context problems and react

EPCIS Accessing Application Partners

“Pull” or “Push” mode EPCIS Query Interface EPCIS Repository

Optional bypass for realtime “Push”

EPCIS Capture Interface EPCIS Capturing Application ALE Interface

Figure 6.9 EPCIS interfaces.

172

RFID Systems

to them. For example, it can divert a bad case in the network area in a conveyor system. It captures ALE events ECReports and CCReports (see Section 6.4) and rearranges them, that is, it translates ALE events into EPCIS events. It adds some business context and creates one or more business events. It allows the multiple ALE events to continue. Moreover, it allows the aggregation of other data sources that the ones coming from the filtering interface like user inputs or business information systems. It seems to have the same role as the ALE Engine except that the EPCIS Capturing Applications knows about the business steps context. The EPCIS Capture Interface is the interface that defines how EPCIS events are delivered to the higher levels of the system (to the EPCIS Repository, to EPCIS Accessing Applications, or to a partner). It consists of one method capture with a list of EPCISEvent for single argument. This method returns nothing or void. The standard from EPCglobal defines a binding via an HTTP URL to send events from the capturing application to the higher levels through the capture interface. The EPCIS Repository stores EPCIS events in order to offer persistence to EPC-related data. This data can be accessed through an EPCIS Query Interface by EPCIS Accessing Application or partners. It acts as a database and can be implemented in different ways using known database systems such as MySQL, Oracle, . . . An EPCIS event is a registering of something that occurs in the real world. Most of the time, but not necessarily, it is triggered by the reading of a RFID tag. An event has four dimensions: • • • •

What: what object is concerned? When: when did it happen? (timestamp) Where: where the event happened? (location identifier) Why: from what business process is it issued?

An event is either: • Object Events: Observation of a list of data tag during a business step in a given place and time These objects were read at the distribution center #9 at 10:01AM, in receiving mode. • Aggregation Events: Physical association of a set of tags with a parent one at a given place and time. These objects were assembled on Pallet #12 at Pallettizer #4 at 12:32PM . • Quantity Events: Declaration of a certain quantity of a type of products (object class) at a given time and place. 200 bottles of “Chateau La Pompe” were identified in the stock of shop #234 today at 3:20PM or • Transaction Events: Association of registered objects with a given transaction. Order #123 was set up with objects x, y and z . The EPCIS Query Interface is divided into two parts: the EPCIS Query Control Interface and the EPCIS Query Callback Interface. The EPCIS Accessing Applications retrieves EPC-related data through the control interface in either synchronous or asynchronous mode. In asynchronous mode, the answer of the query is “pushed” to the corresponding Accessing Application(s) via the callback interface. This latter interface is also used to send data immediately from Capturing Application(s) to Accessing Application(s) (the “optional bypass” arrow on Figure 6.9).

RFID Middleware: Concepts and Architecture

173

The EPCIS Accessing Application is the business application that uses EPC-related data. It is the module which generates and sends ECSpec and CCSpec (see Section 6.4). For instance, it can be the bill module of a company which checks whether an order has been expedited before editing the bill. An EPCIS Accessing Application can be either a part of the enterprise system or a partner application. For instance, this may be the system of a delivery company which checks whether an order is ready before coming to retrieve it. A partner can access the EPC-related data of another enterprise through the EPCIS Query Interface. But partners have to know the EPCIS Repository address to query it. There are two EPCglobal standards (one ratified and one in development) in order to retrieve partners’ EPCIS Repository address. These two standards are presented below.

6.5.2 Object Naming Service The Object Naming Service (ONS) is a central look-up service of the EPCglobal Network [7]. Its main function is the address retrieval of manufacturer information services for a given Electronic Product Code (EPC) identifier. It locates the EPCIS of the issuing authority for the EPC (usually a manufacturer). Today, the ONS is built on top of the Domain Name System (DNS) [8], that is, it is a hierarchical organization as shown in Figure 6.10. To use it, the EPC is first translated into an URI format, equivalent to a URL address. If the local root does not know this address, it forwards it to the ONS Root. The ONS Root contains the address of the local ONS for each EPC manager number. For instance, it knows the IP address of the local ONS hosting every EPC starting with 061414. Thanks to the DSN mechanisms, the requested is forwarded to the local ONS containing the EPC requested. In its turn, based on the same mechanism, the local ONS routes the request towards the proper EPCIS. A local ONS contains the address (URL) of EPCIS of every EPC (e.g. SGTIN, SSCC, . . . 2 ). Therefore, the local ONS is able to find the right EPCIS thanks to the DNS and to redirect the request to it. Finally, the EPCIS contains data about a specific EPC.

Figure 6.10 2

ONS look-up service.

Two standards explain the different Electronic Product Codes [29, 30].

174

RFID Systems

Nowadays, the number of requests to an ONS remains low due to the relatively low number of RFID systems currently deployed. Therefore, till April 2008, there existed only six duplicated ONS roots in the world operated by Verisign. This solution is not practicable or scalable as the overhead for preserving data consistency between roots is high together with memory overhead. Furthermore, this solution is not compliant with every country’s systems. Since the number of RFID systems will increase in the coming years, the EPCglobal network needs several other ONS roots in other countries, but the standard actually states the existence of a unique central root directory. Therefore, at the GS1 France initiative, EPCglobal will focus on designing and evaluating a multi-roots ONS system that will take into account security, stability, performance as well as interaction with DS (Discovery Services). The aim is to show that several ONS roots can work together and safely share the management and the governance of the network.

6.5.3 Discovery Services Discovery Services simplify the data exchange process by offering a service that links information about RFID-enabled products as they move through the supply chain. The addition of Discovery Services to the EPCglobal Network offers trading partners the ability to find all parties who have possession of a given product and to share RFID events about that product. The main difference with the ONS is that each EPCIS having information about a specific EPC register themselves on the DS. The company can query the DS which will return the list of all previously registered EPCIS containing information on this EPC. Then the company can retrieve all the knowledge of the entire network for this EPC by querying these EPCISs one after the other. Nevertheless, nowadays, there exists no standards or norms. This is in the scope of EPCglobal who defines Discovery Services as a standard in development.3 Today, several protocols are proposed to feed the standard by companies such as IBM [27] or Afilias [28] but the standard is still not in production.

6.6 Example We can illustrate the EPCGlobal Network architecture through the two following examples. The first example displayed by Figure 6.11 takes place in a closed architecture of a manufacturer network. A manufacturer builds some products, each product labeled with an EPC. To facilitate the delivery of his products, the manufacturer gathers them in a pallet also equipped with an EPC. In order to keep an history of his products, the manufacturer records the link between the “containing” entity (pallet) and the “contained” objects (manufactured products) through an EPCIS event in his EPCIS repository (called AggregationEvent). So the manufacturer, via his business application sends the ECSpecs to configure the ALE (Figure 6.11 1 ). The ALE configures the readers (Figure 6.11 2 ). When the ALE receives the readers events with the relevant EPC (Figure 6.11 3 ), it sends the ECReport to the EPCIS through the Capture Interface and Capture Application (Figure 6.11 4 ). 3

http://www.epcglobalinc.org/standards/discovery

RFID Middleware: Concepts and Architecture

175

APP

2

1

3

ALE

Figure 6.11

EPCIS

4

Manufacturer network.

EPCGlobal Network Services

Root 6

ONS

5

Manufacturer Network

Retailer Network

ONS

ONS

4

APP

7

2 3 1

EPCIS

EPCIS

ALE

Figure 6.12 Across networks.

The second example plotted by Figure 6.12 shows how and when data can be exchanged and shared, across enterprises. The same manufacturer as above sends his pallet to the retailer. When the retailer receives the pallet, its EPC is read with or without the contained EPC-labeled objects (Figure 6.12 1 ) and sent to the application (Figure 6.12 2 ). The retailer needs to check the content of the pallet. His application will first query the local retailer EPCIS “Have you recorded an event based on this EPC?” (Figure 6.12 3 ) but as we saw in the first example, the correlated AggregationEvent is indeed stored in the manufacturer EPCIS. Then, two possible options: (i) either the retailer knows the entry for the Query Interface of the manufacturer’s EPCIS, so the application can directly query this EPCIS (Figure 6.12 7 ); or (ii) the retailer’s application can ask the ONS “Give me the address of the local ONS requested by this EPC” (Figure 6.12 4 and 5 ), then the request is directed to the manufacturer’s local ONS (Figure 6.12 6 ). Finally, the retailer can retrieve the EPCIS address for the event record. Now he can compare the list of

176

RFID Systems

objects contained by the pallet with the manufacturer record (Figure 6.12 7 ) and check the consistency of his order.

6.7 Conclusion This chapter has drawn up a description of the functionalities offered by a middleware and highlighted the necessity for such a middleware when the amount of data and applications dealt with increase. For all components of such a middleware, EPCGlobal gives some standards but the way to implement it is still challenging. In the actual economic context, RFID applications have been finding their place and the coming years will witness their development. Researchers are currently proposing new models to implement every brick of the middleware in a still easier, faster and simpler way that will appear very soon. With the support of more and more companies and researchers, EPCGlobal is developing the standards, to give access to the largest number of companies.

Problems 1. Who is the entity which produces RFID standards? 2. What is a RFID middleware and what is its purpose? 3. What are the three possible modes for querying a source? 4. What is the entity in charge of reader management? 5. What principles do the current ONS rely on?

References [1] Liu, S., Wang, F. and Liu, P. (2006) Integrated RFID data modeling: An approach for querying physical objects in pervasive computing, in Proc. 15th ACM Int. Conf. Information and Knowledge Management (CIKM’06), Arlington, Virginia, USA, poster. pp. 822–823. [2] Wang, F., Liu, S., Liu, P. and Bai, Y. (2006) Bridging physical and virtual worlds: Complex event processing for RFID data streams, in Proc. 10th Int. Conf. Extending Database Technology (EDBT’06), pp. 588–607. [3] Weinstein, R. (2005) RFID: A technical overview and its application to the enterprise, IT Pro pp. 27–33. [4] Schmidt, L., Mitton, N. and Simplot-Ryl, D. (2009) Towards unified tag data translation in the internet of things, in Proc. 1st Wireless Communication Society, Vehicular Technology, Information Theory and Aerospace & Electronics Systems Technology (VITAE’09), Aalborg, Denmark. [5] EPCglobal (2008a) The Application Level Events (ALE) specification. [6] EPCglobal (2007a) The EPC Information Services (EPCIS) standard, version 1.0.1. [7] EPCglobal (2008c) Object Naming Service (ONS) Standard, version 1.0.1. [8] Mockapetris, P. (1987) Domain names – concepts and facilities. IETF RFC 1034. [9] EPCglobal (2007b) Epcglobal Architectural Framework, version 1.2. [10] Eom, J. B. and Lee, T. J. (2008) RFID reader anti-collision algorithm using a server and mobile readers based on conflict-free multiple access. In Proc. IEEE Int. Performance, Computing and Communications Conference (IPCCC 2008), pp. 395–399, Austin, Texas, USA. [11] Joshi, G., Mamum, K. A. and Kim, S. (2009) A reader anti-collision MAC protocol for dense reader RFID system, in Proc. WRI Int. Conf. Communications and Mobile Computing (CMC’09), Kunming, Yunnan, China. pp. 313–316. [12] EPCglobal (2006) The Reader Protocol (RP) standard.

RFID Middleware: Concepts and Architecture [13] [14] [15] [16] [17]

[18] [19]

[20] [21] [22] [23] [24] [25] [26] [27]

[28] [29] [30]

177

EPCglobal (2007c) The Low-Level Reader Protocol (LLRP) standard. EPCglobal (2007d) The Reader Management (RM) standard. EPCglobal (2009a) The Discovery Configuration and Initialization (DCI) standard for reader operations. IETF (2009) rfc5415 Control and Provisioning of Wireless Access Points (CAPWAP). Park, S. M., Song, J. H., Kim, C. S. and Kim, J. J. (2007b) Load balancing method using connection pool in RFID middleware, in Proc. 5th ACIS Int. Conf. Software Engineering Research, Management & Applications (SERA 2007), pp. 132– 137. Wu, J., Wang, D. and Sheng, H. (2007) ECA rule-based RFID data management, in Proc. 1st Annual RFID Eurasia, pp. 1–5. Park, J., Hong, B. and Ban, C. (2007a) Efficient transformation scheme for indexing continuous queries on RFID streaming data, in Proc. 2nd Int. Conf. Systems and Networks Communications (ICSNC 2007), Cap Esterel, France. pp. 41–46. Moon, M., Kim, Y. and Yeom, K. (2006) Contextual events framework in RFID system, in Proc. 3rd Int. Conf. Information Technology: New Generations (ITNG’06), Las Vegas, Nevada, USA, pp. 586– 587. Wang, F. and Liu, P. (2005) Temporal management of RFID data, in Proc. 31st Int. Conf. Very Large Data Bases (VLDB 2005), Trondheim, Norway, pp. 1128– 1139. Biron, P. and Malhotra, A. (2001) XML shema part 2. W3C recommendation. Bray, T., Paoli, J., Sperberg-McQueen, C. M., Maler, E. and Yergeau, F. (2004) EXtensible Markup Language (XML) 1.0 (third edition), W3C recommendation. Thompson, H., Beech, D., Maloney, M. and Mendelsohn, N. (2001) XML Shema part 1. W3C recommendation. Liu, F., Jie, Y. and Hu, W. (2008) Distributed ALE in RFID middleware, in Wireless Communications, Networking and Mobile Computing, 2008. WiCOM ’08. 4th International Conference, pp. 1–5. EPCglobal n.d. The Reader Management (RM) standard (to appear). Rantzau, R., Kailing, K., Beier, S. and Grandison, T. (2006) Discovery services enabling RFID traceability in EPCglobal networks, in Proc. of the 13th International Conference on Management of Data (COMAD), Delhi, India. Afilias (n.d.) Afilias provides free discovery services for RFID pilots. EPCglobal (2008b) Epcglobal Tag Data Standard (TDS). EPCglobal (2009b) Epcglobal Tag Data Translation (TDT) standard.

Part Two Tag Identification Protocols

7 Aloha-Based Protocols Kwan-Wu Chin and Dheeraj Klair University of Wollongong, Australia

RFID readers aim to identify tags quickly. The key constraint, however, is that a tag can only be read one at a time. Hence, in scenarios with multiple tags, for example, tagged items on a pallet, reading becomes problematic when multiple tags reply simultaneously because their respective signals collide and corrupt one another. RFID readers, therefore, run an anti-collision protocol to arbitrate tag replies so that they experience minimal or no collision, and hence achieve fast identification. Unfortunately, anti-collision protocols face a key challenge. That is, they are unaware of the identity (ID) or the number of tags in a reader’s interrogation zone. If this information is available, then an anti-collision protocol can easily instruct each tag to transmit at a time that does not overlap with other tags’ transmission. For example, if there are two tags with ID X and Y, an anti-collision protocol can inform the tags to transmit at time t1 and t2 respectively, where t2 occurs immediately after the reader has acknowledged receiving tag X’s transmission. However, in practice, both these bits of information are unknown. As a result, anti-collision protocols have difficulty ensuring tag replies are collisionfree. Moreover, an anti-collision protocol needs to ensure tags’ replies are consecutive in order to maximize reading rate. This also means the reader does not wait idly for a tag reply. Otherwise, it will experience prolonged identification delays, and also energy and bandwidth wastage. To date, researchers have proposed two categories of anti-collision protocols: Aloha and tree. Each protocol category comes with its own advantages and disadvantages. In general, tree protocols promise deterministic identification but are complex, incur significant memory overheads, and require complex hardware [1, 2]. In contrast, Aloha protocols have simpler reader designs, lower protocol complexity and bandwidth requirements, smaller number of reader to tag commands, and are able to adapt dynamically to varying tag population. RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

182

RFID Systems

Table 7.1 RFID standards and products that use Aloha-based tag reading protocols. Standards

Protocol

ISO 18000-3 “MODE 1” ISO 18000-3 “MODE 2”

Pure Aloha or Framed Slotted Aloha Tags select from eight channels and use Slotted Aloha during transmission Dynamic Framed Slotted Aloha Frame Slotted Aloha with muting and early-end Basic Framed Slotted Aloha with early-end Dynamic Framed Slotted Aloha

ISO 14443-3 Type B ISO 18000-6A EPCGlobal Class 1 Philips I Code

Table 7.1 presents the standards that use Aloha-based protocols. These standards are managed by two bodies: EPCglobal and the International Standards Organization or ISO. The former develops standards targeted at supply chain networks whereas the latter specifies the air interface for tracking cattle, payment systems and smart cards. Apart from these organizations, there are also propriety standards by Philips. In general, standards over low bandwidth air interface rely on Aloha variants. Otherwise, systems have the flexibility to either choose tree or Aloha variants. Henceforth, this chapter presents a comprehensive review of Aloha-based anti-collision protocols. In Section 7.1, we will present protocols based on Pure Aloha (PA). Then, in Section 7.2, we introduce Slotted Aloha (SA) variants and show how they effectively address the partial collision problem suffered by PA variants. This is then followed by an analysis comparing all Pure and Slotted Aloha variants in Section 7.2.1. This is followed by an introduction to Framed Slotted Aloha (FSA) protocols in Section 7.3, where tags are read using frames with either a fixed or varying number of slots. Moreover, we show how each frame is sized using a tag estimation function. Besides that, in Section 7.3.3, we review protocols that combine the advantages of both Aloha and tree algorithms. Finally, Section 7.4 presents future works and concludes the chapter.

7.1 Pure Aloha Figure 7.1 shows three tags in a reader’s interrogation zone. Upon being energized, each tag sets itself to respond randomly after receiving a read request from the reader. If the reader receives a tag response successfully, it transmits an acknowledgment (ACK). On the other hand, if there is a collision, the reader transmits a negative ACK (NACK), which causes a tag to retransmit its ID after a random delay. RFID systems based on Pure Aloha, however, suffer from the well-known partial collision problem that limits their throughput to 18% [3]. To improve the performance of Pure Aloha, researchers have proposed various optimizations to reduce collision and increase its tag identification rate [4]: • Pure Aloha with Muting. In this variant, the reader sends a mute command after identifying a tag. As a consequence of receiving the mute command, the identified tag stops responding to future read requests. This means every successful tag identification

Aloha-Based Protocols

183

Interrogation Zone

T3 Reader

T2 T1

Figure 7.1 Reader and tags interactions.

T1 T2

T3

T1

T2 Time

Collision

Mute

Figure 7.2

Mute

Mute

Pure Aloha with muting.

reduces the number of tags and the reader’s offered load. Figure 7.2 shows the behavior of Pure Aloha with muting. Initially, the transmission from tags 1 and 2 partially overlaps, which results in a collision. These tags then retransmit after a random delay. After receiving a tag ID successfully, for example, tag 3, the reader sends a “mute” command to the tag. • Pure Aloha with Slow Down. In this variant, once a tag is identified, it is instructed to increase its back-off time using a slowdown command. In Figure 7.3, if slowdown is not used, we see tag 3’s reply interfering with the transmission of tag 1 and 2. However, with slowdown, the reader instructs tag 3 to back off for a longer period of time. Consequently, the reader has a higher probability of identifying tags 1 and 2 successfully. • Pure Aloha with Fast Mode. The reader transmits a “silence” command after detecting the start of a tag transmission. This command has the effect of muting other tags. Tags are allowed to transmit again after the reader has sent an ACK command or until their waiting timer expires. Figure 7.4 shows Pure Aloha with fast mode. Once the reader detects a transmission from tag 2, tag 1 and tag 3 are silenced until the tag finishes its transmission. Otherwise, these tags would have collided with the reply from tag 2. • Hybrids. Lastly, it is possible to combine the above features to create other Pure Aloha variants. Namely, Pure Aloha with fast mode and muting, and Pure Aloha with fast mode and slowdown. In the former variant, tags are temporarily silenced whenever a tag has started its transmission. A tag is then muted after identification. On the other hand, instead of muting tags, the latter variant slows the reply of identified tags so that their replies are less likely to collide with those from unidentified tags.

184

RFID Systems

T3 T3

T3 No Slowdown

T1

T2

Time Collision

T3

T2

T1

T3 Slowdown Time

Slow Down

Slow Down

Figure 7.3

Slow Down

Pure Aloha with slowdown.

T3 Silenced T1

T2

T1

T3

Silence

Silence

Silence

Time

Figure 7.4 Pure Aloha with fast mode.

7.2 Slotted Aloha The main problem that limits the reading rate of Pure Aloha systems is partial collisions. To overcome this fundamental problem, RFID systems can employ Slotted Aloha, which has a maximum throughput of around 36% [3]; viz. doubles that of Pure Aloha. Instead of replying on a continuous timeline, tags are now required to respond at pre-defined slots. This means the reader and tags are tightly synchronized and tags are only allowed to transmit at the beginning of a slot. If there is a collision, tags wait for a random number of slots before retransmitting again. The key performance gain obtained by Slotted Aloha is due to the fact that collisions only occur at the start of each slot as opposed to any time in Pure Aloha. Similar to Pure Aloha, there are numerous variants: 1. Slotted Aloha with Muting or Slowdown. This variant has the same operating principle as Pure Aloha with muting or slowdown, but instead operates in a slotted manner. 2. Slotted Aloha with Early End . A reader closes a slot if it does not detect any transmissions at the beginning of a slot. Two commands are used: start-of-frame (SOF) and end-of-frame (EOF). The former is used to start a reading cycle, and the latter is used by the reader to close an idle slot early. Figure 7.5 depicts how early end is used to terminate two idle slots. As a result, tags can transmit sooner, leading to a higher

Aloha-Based Protocols

185

Closed Idle Slots

T2

T1

T3 Time

Figure 7.5

Slotted Aloha with early end.

reading rate. Moreover, a reader is able to conserve energy as it can start receiving the next reply sooner [5]. 3. Slotted Aloha with Early End and Muting. After identifying a tag, the reader sends a mute command to the tag, thereby removing the tag from contending in subsequent slots. In addition, as required by the early end feature, idle slots are closed early using the EOF command. 4. Slotted Aloha with Slow Down and Early End : This variant combines slowdown with the early end feature. In other words, as well as closing idle slots, identified tags are instructed to slow their replies.

7.2.1 Pure versus Slotted Aloha Variants Klair et al. [6] and Rivera et al. [7] have conducted extensive analytical and simulation studies of Pure and Slotted Aloha variants. The key simulation parameters used are as follows. The tag’s data rate is 26 kbps as per ISO 15693. They modeled the communication between a tag and the reader as a Poisson process. Specifically, the mean arrival time between tag responses is λ1 , where λ is the average duty cycle of tags. Each tag ID is 112 bits in size. They bound the retransmission delay to K random slots; Schwartz [3] found K = 5 to be optimal as higher values prolonged identification delays. From Figure 7.6, we see that the throughput for Pure Aloha with fast mode is as high as Slotted Aloha variants – throughput is defined as the number of tags identified per second. In addition, Pure Aloha protocols using the fast mode feature approach the maximum throughput when there is a large number of tags. Therefore, when protocols are operating at their maximum system efficiency, fast mode variants of Pure Aloha operate below their maximum system throughput. This means, given the same number of tags, Pure Aloha variants using fast mode experience more free channel time as compared to other protocols, hence tags’ responses are less likely to collide. Note, the performance of Pure and Slotted Aloha protocols are limited by collisions. To clarify, Pure Aloha protocols suffer from partial collisions, which limit their performance to 18%. On the other hand, Slotted Aloha protocols have a maximum throughput of 36% because collisions can only occur at the start of a slot [3]. Figure 7.7 plots the average number of collisions when reading n tags. Among Pure Aloha variants, Pure Aloha has the highest collision count because of its vulnerability

186

RFID Systems

0.4

0.35

System Throughput

0.3 PA SA PA−Mute

0.25

SA−Mute PA−Slow SA−Slow PA−Fast PA−Fast+Mute PA−Fast+Slow

0.2

0.15

SA−Early SA−Mute+Early SA−Slow+Early

0.1

0.05

0

0

5

10

15

20

25

30

35

40

45

50

Number of Tags (n)

Figure 7.6 λ = 20, K = 5, where λ is the offered load and K is the maximum retransmission delay [6]. Computer Communications. Reproduced by permission of  2008 Elsevier B.V. All rights reserved.

period of 2T, where T is the transmission time of a tag ID. On the other hand, Pure Aloha with fast mode and muting has the lowest number of collisions. This is because both fast mode and muting reduce collisions. Among slotted Aloha variants, conventional slotted Aloha has the highest number of collisions. On the other hand, slotted Aloha with muting achieves the lowest collision count. Note that, early end has no effect on the number of collisions. Overall, Pure Aloha fast mode variants have the best performance, that is, they outperform all slotted Aloha variants. The trade-off, however, is the increased system complexity because a separate channel is used to transmit the fast mode command. Figure 7.8 plots the number of idle slots that occur when reading n tags. We can see that for Pure Aloha, Pure Aloha with muting, and Pure Aloha with slowdown, with increasing tag population, the number of idle slots reduces proportionally. This is because the probability that at least one tag choosing a given slot increases with tag numbers. For Pure Aloha, in particular fast mode variants, the number of idle slots increases initially, and then begin to reduce when the tag population exceeds 35. This is because with fast mode, collision reduces, and therefore, there are fewer tag retransmissions, which explains the initial small number of idle slots. However, as expected, when we increase the number of tags, the offered load to the reader becomes higher, which reduces the probability of idle slots.

Aloha-Based Protocols

PA SA PA−Mute SA−Mute PA−Slow SA−Slow PA−Fast PA−Fast+Mute PA−Fast+Slow SA−Early SA−Mute+Early SA−Slow+Early

105

Average number of collisions when reading n tag

187

104

103

102

101

100

10−1

10−2 0

10

20

30

40

50

60

70

80

90

100

Number of Tags (n)

Figure 7.7

Number of collision to read n tags.

Similar to Pure Aloha variants, slotted Aloha variants initially observe a lower number of idle slots before peaking, and gradually dropping with increasing number of tags. In summary, the key results from [6] and [7] are: • Fast mode variants experience the smallest number of collisions. This means, a reader using these variants will be able to identify the highest number of tags quickly. • Early end variants reduce energy consumption from idle listening significantly. However, its use does not contribute to any reduction in collisions. Apart from that, as the number of tags increases, idle listening reduces, and hence it does not provide any benefits in terms of energy savings. • Pure Aloha with fast mode and muting can read the highest number of tags. Overall, this variant experiences the smallest number of collisions, and hence the highest read rate.

7.3 Framed Slotted Aloha A major problem with Pure and Slotted Aloha variants is that tags reply at least once in a reading cycle. To address this problem, framed slotted Aloha (FSA) variants restrict

188

RFID Systems

Average idle slots when reading n tag

100

10−1 PA SA PA−Mute SA−Mute PA−Slow SA−Slow PA−Fast PA−Fast+Mute PA−Fast+Slow SA−Early SA−Mute+Early SA−Slow+Early

10−2

10−3

0

10

20

30

40

50

60

70

80

90

100

Number of Tags (n)

Figure 7.8 Number of idle slots encountered when reading n tags.

tag reply to only once in each frame. As a result, the offered load to the reader reduces, and consequently so do collisions. Moreover, it allows a reader to estimate the number of tags, and consequently, the number of slots or frame size required to minimize collisions and idle slots. To date, there are three categories of FSA-based anti-collision protocols, as determined by a reader’s ability to adjust the frame size used to read tags. The following sections review readers that use (i) a fixed/basic frame, where the same frame size is used in each read cycle; (ii) a dynamic frame, where the size of the frame is determined by a tag estimation function; and (iii) an enhanced/hybrid, where a reader adjusts its frame size dynamically as well as segregates tags into small groups.

7.3.1 Basic Basic FSA (BFSA) anti-collision protocols can be classified according to their use of the following features: muting and early end. In other words, (i) BFSA-no-muting,

Aloha-Based Protocols

189

(ii) BFSA-with-muting, (iii) BFSA-no-muting-with-early-end, and (iv) BFSA-mutingearly-end. These variants work similarly to those in Section 7.2 but operate on a frame-by-frame basis. A fundamental problem with BFSA-no-muting is that a reader’s identification delay increases exponentially when the number of tags is significantly bigger than the frame size used to read them. To address this problem, Hwang et al. [8] propose segregating tags into small groups so that the number of tags replying is smaller than the frame size. To achieve this, the reader transmits a “comparison” bitstring whenever the ratio of slots with collisions to frame size exceeds 0.5. Tags then compare this bitstring against a part of their ID; for example, the 20th to 25th bit of their ID. Tags with a smaller result reply in the current frame. The disadvantage of Hwang et al.’s work is that it does not limit the number of tags in a group. Given that a reader does not know the tag IDs in its interrogation zone, the “comparison” string used may yield a group with no tags, or in the worst case, a group that encompasses all tags.

7.3.2 Dynamic The main problem with BFSA variants is the use of a fixed frame size. A reader will experience many idle slots when the number of tags is small. Conversely, many collisions will occur when the tag population increases beyond the frame size. Therefore, there is a need to adjust the frame size dynamically to match the number of tags. The Q algorithm [4] uses a simple frame adjustment technique. A reader initially broadcasts a slot counter Q and sets its frame size to 2Q , where Q is an integer between zero and eight. Tags then choose a slot randomly from 0 to 2Q − 1. The reader then monitors each slot. An idle slot causes the reader to decrease Q by a constant c, whereas slots with collisions cause the reader to increase Q by the same constant. Here, 0.1 ≤ c ≤ 0.5. Before the start of the next read cycle, Q is rounded and sent to tags. This technique is computationally inexpensive and quickly converges to the optimal frame size. Lee et al. [9], however, showed that this technique can be improved further as follows. First, to reduce the algorithm’s signaling overheads, they propose to send the value of Q to tags only when it changes, that is, when rounding yields a new Q value. Secondly, a reader can use a different constant and scale to update the value of Q. Specifically, they propose to decrease Q by Ci for each idle slot, and to increase Q by Cc for slots with collisions. These constants are computed as follows. If the number of tags, m, is known, Cc = −0.0491 × ln(m) + 0.534

(7.1)

Ci = (e − 2) · Cc

(7.2)

Otherwise, the constants are calculated as: Cc = any value between 0.1 and 0.5

(7.3)

Ci = (e − 2) · Cc

(7.4)

where e is the mathematical constant 2.71828. . . . The frame size of most Dynamic FSA (DFSA) variants, however, is controlled by a tag estimation function. As shown by Schoute [10], the maximum throughput of FSA occurs

190

RFID Systems

when the frame size matches the number of tags. Hence, this function plays a critical role in the performance of FSA protocols. A tag estimation function uses the status of slots in the last frame to compute a tag estimate. Specifically, the number of slots filled with zero (c0 ), one (c1 ), and multiple (ck ) tag responses. A tag estimation function then computes the optimal frame size for the next read round using c0 , c1 , and ck as its parameters. Note, a read round is considered complete at the end of each frame. To date, researchers have proposed a number of tag estimation functions. Vogt [11] presents two tag estimation functions, which we denote as Vogt-I and Vogt-II. The former sets the tag estimate to be c1 + 2ck as it assumes at least two tags are involved with every collision. On the other hand, the latter is based on Chebyshev’s inequality and aims to minimize the distance between the actual and theoretically computed result. In other words, justify  N,t     a c0   0  N,t  c  (7.5) εvd (N, c0 , c1 , ck ) = min a1  −  1  t    a N,t ck  k

where εvd corresponds to the error between an actual and theoretical read result as represented by vector < c0 , c1 , ck > and < a0N,t , a1N,t , akN,t > respectively. In other words, Vogt-II aims to determine a tag number estimate t that minimizes εvd . The elements of the vector < a0N,t , a1N,t , akN,t > correspond to the expected number of empty slots, slots filled with one tag, and slots with collisions, respectively. To calculate akN,t , given a frame size of N , and t, the expected number of slots filled with k responding tags is given by,     k  1 t−k t 1 1− (7.6) akN,t = N × N N k Vogt also proposes frame sizes that yield low identification delays for a given tag range; see Table 7.2. For example, a frame size of 16 is considered optimal when the number of tags is in the range of one to nine. Zhen et al. [12] propose a function that computes the expected number of tags replying in a collided slot. According to Zhen et al., on average, 2.39 tags are involved in every collision. Thus, the number of estimated tags is c1 + 2.39ck . In addition, Zhen et al. Table 7.2 Optimal frame sizes for a given tag range. Frame Size (N)

Low (n)

16 32 64 128 256

1 10 17 51 112

High (n) 9 27 56 129 ∞

[11]. Reproduced from  2002 IEEE.

Aloha-Based Protocols

191

propose to over-estimate the tag population because doing so lowers identification delays. Based on their experimentations, they propose 1.4 × (c1 + 2.39ck ) as a tag estimate. On the other hand, for RFID systems that support muting, instead of using 1.4 as a scaling factor, they use 0.65. Cha and Kim [13] present two tag estimation functions for muting based RFID environments. Cha-I estimates the number of tags n by solving the following equation,     n 1 n Cratio = 1 − 1 − 1+ (7.7) N N −1 where Cratio is computed after a read round as Cratio = cNk . The tag estimate in Cha-II is simply 2.39ck . The DFSA variant proposed by Khandelwal et al. [14] estimates the number of tags using, c  0 log  N  n= (7.8) 1 log 1 − N As before, N is the current frame size. Note, Equation 7.8 cannot be applied when c0 = 0. When this happens, the tag estimate is n = c1 + 2ck . Lastly, Khandelwal et al. propose setting the frame size to 1.943 × n times the estimated number of tags. Floerkemeier and Wille [15] and Floerkemeier [16] present two estimation functions: Floerkemeier-I and Floerkemeier-II. These functions estimate tags based on the Bayesian transmission strategy proposed by Rivest [17]. In Floerkemeier-I, a reader not only considers read results in the current read round, but also records those in the last frame to determine the frame size to be used in the next read round. Specifically, Floerkemeier et al. select the frame length L to be used in the next round by maximizing the following equation, n max 

U (η = i, N )P (η = i)

(7.9)

i=0

The function U (η, N ) determines the frame size N that yields the maximum throughput for a given number of tags η, and it is derived by [10] as,   η 1 η−1 U (η, N ) = (7.10) 1− N N The function P (η = i) returns the probability that there are i tags in the reader’s interrogation zone. To derive P (η = i), Floerkemeier et al. make use of information from the last t frames, and also the last j slots of the current frame. In addition, they adjust P (η = i) to account for newly arriving and departing tags. The difference between the function Floerkemeier-I and Floerkemeier-II is that the latter calculates P (η = i) on a slot-by-slot basis, as opposed to on a frame-by-frame basis. Moreover, Floerkemeier-II terminates and restarts a frame if it is found to be non-optimal. As mentioned by Floerkemeier [16], this scheme is computationally expensive given that the reader must iterate

192

RFID Systems

nmax times every slot or frame to compute the optimal frame size. Hence, both functions are not suitable for resource constrained RFID readers. Kodialam and Nandagopal [18] propose an estimation function that computes the expected number of idle and single response slots by inserting k = 0 and k = 1 in Equ. 7.6. The resulting equations are then used to derive two estimators, called zero estimator (ZE) and collision estimator (CE), c0 ZE = e−(n0 /N ) = (7.11) N  ck nk  −(nk /N ) e (7.12) = CE = 1 − 1 + N N In Equations 7.11 and 7.12, n0 is the tag estimate obtained from ZE, and nk is the tag estimate computed from CE, respectively. The value of c0 and c1 is obtained by observing the number of idle slots, and slots with a successful transmission in a given frame. They are then used to solve ZE for n0 and CE for nk . If n0 < nk , then the tag estimate is n0 , otherwise it is nk . The authors assume that tag estimation occurs in a phase that precedes the identification phase. In addition, slots in the estimation phase are only 10-bits long. Chen and Lin [19] introduce two estimation functions: Chen-I and Chen-II. In the former, the authors compute the probability of having exactly k tags in m slots as [20], p (k, m) =

(−1)m N !n! m!N n 

min(N,( n/k))

×

j =m

(−1)j

(N − j )n−j k (j − m)! (N − j )! (n − j k)! (k!)j

(7.13)

Using Equation 7.13, the authors calculate the probability of exactly m slots with zero tag responses, that is, k = 0. The actual value of m is c0 , which is obtained from the reader’s feedback. The probability equation is then solved for the value of n, which is the tag estimate. The Chen-II function, however, computes the expected number of slots filled with zero and a single tag using Equation 7.6. The results, denoted as E and S, are then inserted into the following equation, S (7.14) E where N is the frame size. Equation 7.14 is then solved for the tag estimate n. In addition to the above works, Wang et al. [21] propose preceding each reading or jump frame with a short collision detection (CD) frame; see Figure 7.9(a). Tags pick a random slot in the CD frame and transmit a 4-bit Manchester encoded packet. A reader then determines whether there are any collisions in a slot; see Figure 7.9(b). Tags that respond successfully are instructed to reply in the jump frame. Note, the number of slots in the jump frame is equal to the number of tags that successfully transmitted in the CD frame. The key advantage of this scheme is that collisions are short, and a reader does not waste much energy receiving bits corrupted by collisions. In addition, given that the CD frame is effectively a reservation frame, tags are able to transmit collision-free in the jump frame. Having said that, it remains important to dimension the CD frame appropriately to minimize collisions, as the number of slots in the jump frame is dependent only on the number of successful transmissions in the CD frame. n = (N − E − 1)

Aloha-Based Protocols

193

0

1

0

1

1

0

0

1

?

?

Tag 1

Tag 2 Collision Detection Frame

Jump Frame Receiver

(a)

(b)

Figure 7.9 (a) A CD frame precedes each jump frame. The number of slots in each jump frame corresponds to the number of successful transmissions in the CD frame; in this example, only two tags successfully transmitted their ID. Tags that transmitted successfully in the CD frame then send their full ID in the jump frame. (b) Collision detection using Manchester encoding, where no transition in the middle of a bit indicates a collision.

7.3.3 Enhanced/Hybrid A fundamental problem in all FSA protocols is that their frame size grows exponentially with increasing tag numbers. Hence, existing DFSA variants have a maximum frame size; for example, 256 [11] or 512 [22]. Unfortunately, persistent collisions occur when the tag population exceeds a variant’s maximum frame size. Lee et al. [23] address the aforementioned limitation by proposing enhanced-DFSA or EDFSA; an anti-collision protocol that grows the number of slots used to read tags linearly. The key idea is to determine whether the estimated number of tags would yield the maximum system efficiency given the current frame size. If not, tags are divided into M groups. Table 7.3 shows the value of M for a given tag range. Also shown are frame Table 7.3 EDFSA frame sizes. n denotes the number of tags, N is the frame size, and M is the number of tag groups. Number of tags (n) 1–11 12–19 20–40 41–81 82–176 177–354 355–707 708–1416 1417–2831

Frame Size (N)

M

8 16 32 64 128 256 256 256 256

1 1 1 1 1 1 2 4 8

[23]. Reproduced from  2005 IEEE.

194

RFID Systems

sizes for varying tag ranges that achieve maximum system efficiency. The reader then sends a number to the tags, which the tags then use with their ID in a modulo operation. Only tags that yield a remainder of zero are allowed to reply. Peng et al. [24] noticed two limitations with EDFSA. First, after one round of reading, there will be fewer unread tags in each group. Secondly, EDFSA will not adjust its frame size until the number of unread tags fall below a given threshold. They, therefore, conjecture that these observations lead to sub-optimal system efficiency. Henceforth, in order to restore system efficiency, they propose a protocol that re-adjusts the number of groups after the first round of reading to better match the current frame size. To date, researchers have also proposed hybrid protocols that combine the advantages of both FSA and tree protocols. For example, Bonuccelli et al. [25] propose Tree Slotted Aloha (TSA), an enhanced FSA protocol that forms a tree-like structure during its reading cycles. TSA works by quickly segregating and identifying tags that have experienced collision in the following manner. A reader starts with an initial frame size l0 . Tags then select and record a random slot number, and transmit in the chosen slot. At the end of each frame, the reader broadcasts the slot numbers with collisions. Tags that transmitted in these slots are informed to reply in the upcoming frame. This process repeats at the end of each frame until a frame is collision-free. Bonuccelli et al. [25] compute the size of an upcoming 1 frame as  n−c ck , where n is a tag estimate obtained using Vogt’s [26] function; recall that ck and c1 correspond to the number of slots with and without collision respectively. In a different work, Klair and Chin [27] propose ResMon, an energy-efficient DFSA protocol that supports both tag identification and monitoring. ResMon is conceptually similar to Wang et al.’s [21] protocol, but adds an additional collision-free frame that is used solely for monitoring tags; see Figure 7.10. To clarify, ResMon uses three frames. The first frame, Rframe , is used by tags to reserve a collision free slot in the Bframe . Tags send a short, randomly generated bitstring to the reader. Collision is then detected as per Figure 7.9. If a tag’s transmission is collision-free, the reader allocates the tag a slot in the upcoming Bframe , which the tag uses to transmit its full ID. As part of the reservation process, the reader also allocates tags a slot in the Mframe , which they then use to send the same 3-bit Manchester encoded bitstring to indicate they are still in the reader’s interrogation zone. A reader transmits an acknowledgment if it receives this bitstring successfully. Otherwise, if there is an idle slot, the reader sends a negative acknowledgment (ACK) to compress the frame in order to remove the idle slot. This is

Collision Free RFrame

BFrame

Sized by Vogt-I

Full ID Transmission

Figure 7.10

MFrame

ResMon Frames.

Aloha-Based Protocols

195

carried out as follows. Assume there are four tags with transmission slot one, two, three, and four respectively. If the number two slot is idle, indicating a missing tag, the reader sends a negative ACK to inform all tags to reduce their slot number by one. As a result, in the next Mframe , these tags will transmit one slot sooner. Note, tags that have transmitted before the idle slot are not affected by the negative ACK as they are muted after transmission. Hence, only tags following an idle slot will re-adjust their transmission time. The remaining two hybrid works make use of the Query Tree (QT) protocol [28]. Briefly, the QT protocol works as follows. Assume there two tags with ID 01 and 10 – each tag has a prefix matching circuitry. The reader first transmits a null string, which matches the prefix of both tags. As a result, both tags reply to the reader simultaneously and cause a collision. The reader then pushes the prefix “0” and “1” onto its stack. The next read cycle begins with the reader popping and transmitting the prefix “0”. This causes tags with ID matching the prefix “0” to reply, that is, tag 01. After that, the reader transmits prefix “1”. As only tags with prefix “1” can reply, the reader only receives a reply from tag 10. The main advantage of QT is its ability to systematically and deterministically identify tags. However, QT may require multiple iterations, that is, large tree depth, in order to identify tags. Specifically, the tree depth required to identify all tags may equal the length of tag ID. For example, as shown in Figure 7.11, the tree depth is four. Notice that if tags A and C or B and D are segregated into different groups, the resulting tree is shorter due to their unique prefix. To this end, researchers have devised various methods to reduce the probability of tags with the same prefix replying simultaneously. Shin et al. [29] propose two algorithms that combine FSA protocols with the QT protocol. In the first protocol, called framed query tree, the reader transmits a set of frames, which tags then randomly choose one to reply in. As a result, there are fewer tags replying in each frame. Moreover, each of them may have a different prefix. The reader then identifies tags in each frame using the QT protocol. In the second protocol,

′′′′ 1

0 01

00 ′′′′

011

010

A

C

001

(i) No Segregation

000

1

0

0001

0000

01

00

D

B

A

B (ii) Segregation

Figure 7.11 An example showing QT being used to identify the following tags: A (0111), B (0000), C (0101) and D (0001). In scenario (ii), only tags A and B contend with each other. Tag C and D will contend at a different time.

196

RFID Systems

that is, query tree Aloha, the reader transmits a prefix, for example, “0”, and a frame size. Tags with an ID that match the prefix randomly select a slot, and reply in the chosen slot. Upon detecting a collision, a new prefix is sent, for example, “01”, to segregate the collided tags, and a new frame is then used to read tags. On the other hand, Namboodiri and Gao [30] introduce three anti-collision protocols that combine QT with DFSA. The key idea in all three protocols is to associate a frame or multiple slots with each QT query. In other words, with each query or a node on the tree, the reader sends a frame with multiple slots for tags to choose from. As a result, the reader is able to read tags with the same prefix. Thus, more tags can be read with the same number of queries as compared to conventional QT. 7.3.3.1 Performance Klair et al. [31] have studied the performance of FSA variants that use the muting and early-end feature. Briefly, the simulation parameters used in their Matlab implementation are as follows. The tag to reader data rate is 26 kbps (ISO 15693). Hence, with a tag ID 96 size of 96 bits, each slot is of duration 26000 . For non-muting experiments, the RFID reader is assumed to continue reading until all tags are read with 99% confidence level. On the other hand, when muting is enabled, it continues reading until there are no collisions in a given read cycle. To evaluate the performance of each tag reading protocol, they conducted experiments with increasing number of passive, static tags, and repeatedly read these tags for 100 simulation runs. They then recorded the mean delay, and the average number of idle and collision slots experienced by the reader when reading a given tag set. Lastly, they assume negligible propagation delay and channel error. Figure 7.12 shows the delay incurred by each variant to read a given number of tags. We see that among non-muting variants, DFSA with early end has the lowest reading delay. This is because the early-end feature ensures the delay incurred by idle slots are minimized. Similarly, for muting variants, those that combine the early end yield the quickest reading time. The advantage of having a dynamic frame size is particularly critical for high tag numbers. For example, when there are more than 60 tags, the reading delay of BFSA variants grow much quicker than DFSA and EDFSA variants. The superiority of EDFSA variants is also evident. In particular, the reading delay of EDFSA with muting and early end is a few orders of magnitude lower than its DFSA counterpart. Figure 7.13 shows the average number of collisions encountered when using framed Aloha variants. DFSA with muting has the lowest number of collisions as compared to other variants. On the other hand, BFSA non-muting variant has the highest number of collisions. Note, early end has no effect on collisions as it is used only to shorten idle slots. Figure 7.14 shows the average number of idle slots that occur when reading n tags using framed Aloha variants. DFSA non-muting variants have the highest number of idle slots for a given number of tags. This is because of their varying frame size, and the use of muting – both of which reduce and increase the probability of collisions and idle slots respectively. On the other hand, BFSA muting variants have the lowest number of idle slots. This is because they use a fixed frame, and hence this causes a dramatic reduction in the number of idle slots. A key issue with DFSA and EDFSA variants is the accuracy of their tag estimation function. In particular, an inaccurate estimate may lead to persistent collisions or a high

Aloha-Based Protocols

197

Total delay to read n tags (seconds)

101

100

10

BFSA–Non Muting BFSA–Muting BFSA–Non Muting–Early End BFSA–Muting–Early End DFSA–Non Muting DFSA–Muting DFSA–Non Muting–Early End DFSA–Muting–Early End EDFSA–Non Muting EDFSA–Non Muting–Early End EDFSA–Muting EDFSA–Muting–Early End

−1

10−2

10

20

30

40

50

60

70

80

90

100

Number of tags

Figure 7.12 Reading delay of BFSA, DFSA, and EDFSA variants.

number of idle slots. To this end, Klair et al. [32] have compared the accuracy of the following functions: Vogt-I, Vogt-II, Cha-I, Cha-II, and Zhen et al. [12]. Their results show Vogt-II achieves the best accuracy for a wide range of tags. On the other hand, Cha-I is more accurate when the number of tags increases beyond the current frame size. In general, functions that derive tag estimates using probabilistic or statistical methods are more accurate with increasing tag numbers. These include Chen-I, Chen-II, Cha-II, Vogt-II, Floerkemeier-I, Floerkemeier-II, and the functions proposed by Kodialam and Nandagopal [18], and Khandelwal et al. [14]. Their better accuracy stems from not relying on a fixed multiple of ck . Moreover, they are more robust in a wide range of tag population, whereas static estimation functions are suitable only for low tag ranges [11, 19, 26]. Apart from that, a critical observation is that frame sizes can only be a power of two. For example, when there are around 60 tags, a tag estimation function may set the frame size to 128. This reduces collisions, but at the expense of idle slots or system efficiency. Therefore, the early-end feature must be used to offset any frame size inaccuracies. Apart from that, tag estimation functions have varying computational requirements. Some involve only additions and multiplications; for example, Vogt-I, Cha-II, Q-algorithm, and those proposed by Zhen et al. [12]. On the other hand, Chen-II, Cha and Kim [13], Floerkemeier-I and Floerkemeier-II require the reader to compute

198

RFID Systems

104

Average collision when reading n tags

103

102

BFSA–Non Muting BFSA–Muting BFSA–Non Muting–Early End BFSA–Muting–Early End DFSA–Non Muting DFSA–Muting DFSA–Non Muting–Early End DFSA–Muting–Early End EDFSA–Non Muting EDFSA–Non Muting–Early End EDFSA–Muting EDFSA–Muting–Early End

101

100

10−1

10−2

0

10

20

30

40

50

60

70

80

90

100

Number of tags

Figure 7.13 Average number of collisions encountered when reading n tags.

factorials and fractions, and hence they have a need for more computational power. Vogt-II, Cha-I and Chen-I have the highest requirement because all of them involve recursion. To date, only two works have investigated the performance of anti-collision protocols in mobile tags scenarios, for example, tagged items on a conveyor belt. The key concern is whether an anti-collision protocol can quickly read all tags before they exit a reader’s interrogation zone. In Lee et al. [33], the authors found DFSA protocols that use Zhen et al. and Cha-I with muting are able to quickly identify tags on a conveyor belt moving at varying speeds; their experiments involve, on average, 200 tags taking 0 to 30 milliseconds to transit a reader’s interrogation zone. The key advantage of these algorithms is their ability to mute tags after identification. This prevents read tags from contributing to the offered load and colliding with unread tags. An important consideration when reading mobile tags is their varying signal over time, which causes some tags to be unreadable [34]. In this respect, Floerkemeier-I and Floerkemeier-II are designed to adapt to varying tag population as they re-adjust their frame size after each slot or frame. Lastly, Namboodiri and Gao [30] conclude that using FSA with QT lowers the energy expenditure of a RFID reader. Similarly, Shin et al. [24] demonstrate that combining QT with FSA leads to low identification delays. Bonuccelli et al. [25] show that TSA

Average idle slots when reading n tags (seconds)

Aloha-Based Protocols

199

102 BFSA–Non Muting BFSA–Muting BFSA–Non Muting–Early End BFSA–Muting–Early End DFSA–Non Muting DFSA–Muting DFSA–Non Muting–Early End DFSA–Muting–Early End EDFSA–Non Muting EDFSA–Non Muting–Early End EDFSA–Muting EDFSA–Muting–Early End

101 0

10

20

30

40

50

60

70

80

90

100

Number of tags

Figure 7.14

Average number of idle slots encountered when reading n tags.

achieves a higher system efficiency as compared to DFSA, EDFSA, QT, and QT with an aggressive enhancement when there are more than 60 tags. On the other hand, when the number of tags is below 50, QT with the aggressive enhancement has the highest system efficiency. The above results validate the advantages of hybrid protocols. Moreover, given the emergence of novel tree and Aloha variants as well as tag estimation functions, we expect researchers to propose better hybrid protocols in the future.

7.4 Conclusion The performance of Aloha-based protocols increases as we move from Pure Aloha to DFSA variants. This, however, is at the expense of increased system cost due to the use of tag estimation functions that vary in complexity. Tag requirements also differ significantly. Pure Aloha protocols only require tags to have a timer, whereas Slotted Aloha and DFSA variants require tags to have a random number generator, synchronization circuits, and a timer. Having said that, DFSA variants are the most promising of all Aloha-based protocols because of their ability to read varying number of tags. This is particularly critical in practice because the vagaries of the wireless medium cause some tags to remain silent. In other words, the tag population will inevitably change with each read as the

200

RFID Systems

signal strength from the reader varies with each read. This is in addition to tag population changes caused by tags being muted and newly arrived tags. Other than that, the vagaries of the wireless channel also cause the capture or near-far effect, where tags with stronger signal strength are more likely to be read or detected by a reader. This is an interesting issue that has thus far received little attention. A key observation is that varying signal strength reduces tag population or contention level at a given reader. Moreover, if a reader supports muting, it will first identify tags with a strong signal followed by those with weaker signal strength. Hybrid protocols that combine Aloha and tree protocols are becoming popular in the research community. These protocols take advantage of the probabilistic and deterministic nature of Aloha and tree protocols respectively to quickly identify tags. In particular, the key strength of tree protocols is their ability to systematically segregate collided tags, thereby preventing the occurrences of persistent collisions – a key problem in Aloha-based protocols. Consequently, hybrid protocols are very promising for future RFID systems. Given that there are only a handful of hybrid protocols to date, and the vast number of Aloha and tree protocols, a challenging research direction is therefore to determine the best combination that yields a high performing tag reading protocol. The commercial availability of RFID readers such as SkyeTek’s RFID reader [35] that mates with Crossbow’s MICA2Dot sensor motes [36] has made it possible to create pervasive networks that can be used to track RFID tagged objects. A key research challenge in these networks is the limited resources on embedded devices. For example, according to Klair et al.’s [6] analysis of a sensor mote with an RFID reader, the energy consumed by a reader scanning 96-bits of tag ID is higher than a sensor node receiving and transmitting the same number of bits. Moreover, as a reader’s scanning/reading duration increases, so does its energy consumption. Therefore, the commercial success of such RFID networks is dependent on having energy-aware, computationally inexpensive, low signaling anti-collision protocols that are capable of both identification and monitoring. In this respect, besides [27], few works have considered anti-collision protocols that can both identify and monitor tags efficiently. Accordingly, we believe the development of anti-collision protocols that operate efficiently in resource-constrained environments to be the next frontier in tag reading protocol research.

Problems 1. Show that pure Aloha’s throughput is approximately 18%. 2. Show that slotted Aloha’s throughput is approximately 36%. 3. Consider a reader using the Q algorithm. Assume an initial Q and c value of 4 and 0.5 respectively. Determine the frame size to be used in the next round if the reader experiences 4 collisions and 2 idle slots in the current round. 4. A reader, using a frame length of 8, estimates the vector < c0 , c1 , ck > after a read round to be < 4, 2, 2 >. Determine the tag estimate if the reader uses (i) Vogt-I, (ii) Cha-II, and the function by (iii) Zhen et al. [12], and (iv) Khandelwal et al. [14] with N = 8.

Aloha-Based Protocols

201

5. Use Vogt-II to estimate the number of tags in a reader’s interrogation zone for the following vector: < 4, 2, 2 >. Assume the frame length is 8. 6. Repeat question five, but using Cha-I. 7. Draw the resulting query tree for tags with the following ID: 1111, 01010, 1010, 0001. 8. Consider a reader using ResMon that has allocated tag A, B, C and D slot 1, 2, 3, and 4 in the Rframe , respectively. Show how ResMon adjusts tag D’s slot number after tag B and C leave. 9. Consider 16 tags with ID ranging from 0000 to 1111 in a reader’s interrogation zone. Determine the number of tags competing in each frame if a reader using EDFSA transmits a modulo of two (0010). NB Solutions will be provided on the book’s website.

References [1] Finkenzeller, K. (2003) RFID Handbook: Fundamentals and Applications in Contactless Smart Cards and Identification. Chichester: John Wiley and Sons Ltd. [2] Hush, D.R. and Wood, C. (1998) Analysis of tree algorithms for RFID arbitration, in The IEEE International Symposium on Information Theory, pp. 107– 114. [3] Schwartz, M. (1988) Telecommunication Networks Protocols, Modeling and Analysis. New York: AddisonWesley. [4] EPCglobal (2005) Class 1 gen 2 RFID specifications. White paper. http://www.alientechnology.com/docs/ AT wp EPCGlobal WEB.pdf. [5] Klair, D.K., Chin, K.-W., and Raad, R. (2007b) An investigation into the energy efficiency of pure and slotted Aloha based RFID anti-collision protocols, in IEEE WoWMoM , Helsinki, Finland. [6] Klair, D., Chin, K.-W., and Raad, R. (2009) On the energy consumption of pure and slotted Aloha based RFID anti-collision protocols. Computer Communications, 32(5): 961–973. [7] Rivera, A., Klair, D., and Chin, K.-W. (2009) A simulation study on the energy efficiency of pure AJD slotted Aloha based RFID tag reading protocols, in The 6th IEEE Consumer Communications and Networking Conference, Las Vegas, USA. [8] Hwang, T.-W., Lee, B.-G., Kim, Y.S., Suh, D.Y., and Kim, J.S. (2006) Improved anti-collision scheme for high speed identification in RFID system, in Proceedings of the First International Conference on Innovative Computing, Information and Control , pp. 449–452. [9] Lee, D., Kim, K., and Lee, W. (2007) Q+ -algorithm: An enhanced RFID tag collision arbitration algorithm, in The 4th International Conference on Ubiquitous Intelligence and Computing, Hong Kong, China. [10] Schoute, F.C. (1983) Dynamic frame length Aloha. IEEE Transactions on Communications, 31(4), 565–568. [11] Vogt, H. (2002b) Multiple object identification with passive RFID tags, in The IEEE Intl. Conf. on Man and Cybernetics, pp. 6–13. [12] Zhen, B., Kobayashi, M., and Shimizu, M. (2005) Framed Aloha for multiple RFID objects identification, IEICE Transactions on Communications, E88-B, 991– 999. [13] Cha, J.-R. and Kim, J.-H. (2005) Novel anti-collision algorithms for fast object identification in RFID system, in The 11th Intl. Conference on Parallel and Distributed Systems, pp. 63–67. [14] Khandelwal, G., Yener, A., Lee, K., and Serbetli, S. (2006) ASAP: a MAC protocol for dense and time constrained RFID systems, in IEEE International Conference on Communications (ICC’06). [15] Floerkemeier, C. and Wille, M. (2006) Comparison of transmission schemes for framed ALOHA based RFID protocols, in Proceedings of the International Symposium on Applications on Internet Workshops. [16] Floerkemeier, C. (2007) Bayesian transmission strategy for framed ALOHA based RFID protocols, in IEEE International Conference on RFID, Grapevine, Texas, USA.

202

RFID Systems

[17] Rivest, R. (1987) Network control by Bayesian broadcast, IEEE Transactions on Information Theory, IT-33(3), 323–328. [18] Kodialam, M. and Nandagopal, T. (2006) Fast and reliable estimation schemes in RFID systems, in SIGMOBILE: ACM Special Interest Group on Mobility of Systems, Users, Data and Computing, pp. 322– 333. [19] Chen, W.-T. and Lin, G.-H. (2006) An efficient anti-collision method for RFID system, IEICE Trans. Communications, E89(B), 3386– 3392. [20] Feller, W. (1970) An Introduction to Probability Theory and its Applications. New York: Addison-Wesley. [21] Wang, J., Zhao, Y., and Wang, D. (2007) A novel fast anti-collision algorithm for RFID systems, in International Conference on Wireless Communications, Networking and Mobile Computing (WiCom 2007), pp. 2044– 2047. [22] Auto-ID Center (2003) 13.56MHz ISM band Class 1 radio frequency identification tag interface specification, version 1.0. HF RFID standard. http://www.epcglobalinc.org/standards/specs/. [23] Lee, S.-R., Joo, S.-D., and Lee, C.-W. (2005) An enhanced dynamic framed slotted Aloha algorithm for RFID tag identification, in The 2nd Intl. Annual Conference on Mobile and Ubiquitous Systems: Networking and Services, pp. 166– 172. [24] Peng, Q., Zhang, M., and Wu, W. (2007) Variant enhanced dynamic frame slotted Aloha algorithm for fast object identification in RFID system, in The IEEE International Workshop on Anti-Counterfeiting, Security and Identification, China. [25] Bonuccelli, M.A., Lonetti, F., and Martelli, F. (2006) Tree slotted Aloha: a new protocol for tag identification in RFID networks, in International Symposium on World of Wireless, Mobile and Multimedia Networks, pp. 603– 608. [26] Vogt, H. (2002a) Efficient object identification with passive RFID tags, in The International Conference on Pervasive Computing (PerCom), Forth Worth, USA. [27] Klair, D.K. and Chin, K.-W. (2008) A novel anti-collision protocol for energy efficient identification and monitoring in RFID-Enhanced WSNs, in IEEE ICCCN , St Thomas, US Virgin Islands, USA. [28] Law, C., Lee, K., and Siu, K.-Y. (2000) Efficient memoryless protocol for tag identification (extended abstract), in Proceedings of the 4th International Workshop on Discrete Algorithms and Methods for Mobile Computing and Communications, pp. 75–84. [29] Shin, J.-D., Yeo, S.-S., Kim, T.-H., and Kim, S.K. (2007) Hybrid Tag Anti-Collision Algorithms in RFID Systems. Berlin/Heidelberg: Springer. [30] Namboodiri, V. and Gao, L. (2007) Energy-aware tag anti-collision protocols for RFID systems, in The 5th Annual IEEE International Conference on Pervasive Computing and Communications (PerCom), pages 23–46, NY, USA. [31] Klair, D.K., Chin, K.-W., and Raad, R. (2007c) On the suitability of framed Aloha based RFID anticollision protocols for RFID-Enhanced WSNs, in IEEE ICCCN . Honolulu, Hawaii, USA. [32] Klair, D., Chin, K.-W., and Raad, R. (2007a) On the accuracy of tag estimation functions, in Sixth IEEE International Symposium on Communications and Information Technologies, Sydney, Australia. [33] Lee, W., Choi, J., and Lee, D. (2008) RFID Handbook: Applications, Technology, Security and Privacy, Chapter 9: Comparative Performance Analysis of Anti-Collision Protocols in RFID Networks, pp. 161–179. Boca Raton, FL: CRC Press. [34] Floerkemeier, C. (2006) Transmission control scheme for fast RFID object identification, in The 4th Annual Intl. Conference on Pervasive Computing and Communications Workshops. [35] SkyeTek (2009) Skyemodule M1-Mini. Datasheet. http://www.skyetek.com/Portals/0/SkyeModule M1 Mini 060426.pdf. [36] Crossbow (2009) The mica2 mote. Datasheet.

8 Tree-Based Anti-Collision Protocols for RFID Tags Petar Popovski Department of Electronic Systems, Aalborg University

8.1 Introduction Radio Frequency Identification (RFID) systems are increasingly and ubiquitously digitalizing the physical environment by attaching tiny tags to objects and people [1]. They are expected to support a large volume/diversity of applications, ranging from logistics, transport, retail services, access control as well as many yet-to-be-imagined applications belonging to the “Internet of Things” paradigm [2]. The two key components of an RFID system are tags and readers. A tag is a small microchip equipped with antenna which is attached to the object or a person. A tag can be passive (not using a battery), semi-active (using a battery for sensing/processing, but not for communication), or active (battery-powered tag with very low power consumption). The readers, also known as interrogators, are transceivers that can communicate with the tags, by reading information from tags or writing information to them. Although a variety of RFID systems has been used for a considerable time [3], in recent years RFID technology has exhibited immense growth [4]. There are several reasons for such growth. Tags and readers are becoming less expensive, therefore their number and ubiquity increase. Clearly, inexpensive tags will be present in much larger numbers relative to readers. In particular, RFID systems with passive tags that operate in the UHF (Ultra-High Frequency) band are gaining importance due to the increased read range. The potential of such systems has increased dramatically in recent years due to the RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

204

RFID Systems

increased knowledge and applications of wireless networking. However, passive, batteryless tags pose unique challenges in the area of wireless networking due to the modest processing/storing capabilities of tags. Passive tags use the energy beamed from the reader to power their circuitry and also to transmit back to the reader using backscattering [3]. A tag can be attached to a sensor and act as a communication interface, thus creating a RFID sensor. One can easily conceive future systems comprising RFID tags and sensors that feature a combination of all types of tags; with passive tags far outnumbering active and semi-passive tags. The communication mode in RFID systems can be described as follows: the reader sends a probe to a set of tags within its radio range. Based on the content of the probe, a tag decides whether to send its response by backscattering the power supplied by the reader. If more than one tag should respond to the probe, then multiple replies will simultaneously arrive at the reader, thus giving rise to the tag collision problem. If the replies of two or more tags collide, then there is very high probability that the signal will be corrupted and unreadable by the reader. Note that the tag collision problem occurs at the reader. If a tag is in the range of more than one reader, then a reader collision can occur at the tag [5]. This occurs when two or more readers transmit simultaneously, so that a tag that is in range of both readers cannot receive the probe sent by either readers (Figure 8.1). Tag collision is resolved by running an anti-collision protocol (also called arbitration protocol or collision resolution protocol). The objective of an arbitration protocol is to make a transmission schedule for the tags, so that eventually each tag manages to send the reader a successful reply. Alternatively, instead of resolving the collision completely, the reader may need only partial resolution. For example, the reader may run a sequential decision process and gather data from the tags in order to carry out a statistical test of certain hypothesis and stop the arbitration process as soon as it has a sufficient set of responses. In general, the requirement of the arbitration protocol is determined by the higher-level task that is imposed on the reader network. Tag collision can be a particularly acute problem in UHF RFID systems, due to the larger reading range and thus the possibility of having larger number of tags in the reader’s range. Reader collision becomes a significant problem in environments with high reader density. Hence, the problem is to allocate communication resources (time, frequency) to the readers in order to minimize interference among them. For example, in Colorwave [6], the readers communicate with each other with the air interface that is also used to communicate with the tags and independently coordinate to minimize reader collisions using a t4

t1

t2

RA

(a)

t3

t1

RA

t3

RB

t2

(b)

Figure 8.1 Illustration of (a) tag collision and (b) reader collision. Tag collision occurs at the reader RA . The reader collision occurs at tags τ2 , τ3 , but not at τ1 , τ4 .

Tree-Based Anti-Collision Protocols for RFID Tags

205

distributed time-division anti-collision algorithm. A comprehensive survey of the collision problems and collision resolution procedures in RFID systems is given in [7]. This chapter introduces the main ideas and perspectives of a class of anti-collision protocols termed tree algorithms. Tree protocols represent an important paradigm in the design of arbitration protocols for random access. These protocols have been introduced for a single channel with multiple access, where a single reader gathers replies from a population of tags that are in the range of that reader [8–10]. In essence, upon encountering collision, tree-based protocols recursively resolve the collision until all the tags initially involved in the collision have successfully sent their replies to the reader. Tree-based arbitration protocols represent one of the main approaches to resolve the tag collision problem in RFID systems [7, 11]. In recent years there have been a lot of studies that propose new variants of the tree algorithms [12–19]. The objective of this chapter is not to provide a detailed account of the different tree-based proposals and/or their quantitative comparison, but rather provide insight into the mechanisms used in tree-based approaches and possibilities to further optimize these algorithms in various scenarios. This chapter is organized as follows. Section 8.2, where the principles of tree protocols are discussed in a generic setting, not strictly limited to RFID systems. This section first introduces several basic variants of the tree protocol, followed by techniques for improving the basic variants and finally an alternative arbitration framework is introduced which permits estimation of the tag population. Section 8.3 describes tree protocols that are used in specifications for UHF RFID systems. Section 8.4 presents practical issues that need to be considered, such as transmission errors and treatment of moving/late arriving tags. Section 8.5 discusses how tree protocols can be generalized to scenarios with multiple cooperative readers. The last section concludes the chapter.

8.2 Principles of Tree-Based Anti-Collision Protocols This section introduces the principles and basic ideas used in tree protocols in a setting that is not necessarily limited to RFID systems. For example, we do not account for the computational/processing limitations of tags, error-prone communication, etc. The system model used in this section is idealized, but sufficient to introduce the mechanisms that are pertinent for understanding tree protocols. Relation to real-life systems and operation under non-idealized assumptions is given in Sections 8.3 and 8.4.

8.2.1 System Model In this section we describe the context to present the basic ideas of tree-based anti-collision protocols. At the radio level, we assume that if a tag is within a distance D from the reader, then it can always receive probes from that reader without errors. Vice versa, if a single tag in the range of the reader transmits, then its packet is received successfully by the reader. We use the collision model for multiple access channel, which means that if two or more tags that are in the reader range transmit simultaneously, then the reader does not receive any of the transmitted packets. Note that these are rather strong assumptions for passive RFID systems, as many factors can cause transmission errors, such as fading, tag orientation, obstacles between the reader and the tag, etc. Nevertheless, this idealized MAC-layer model is sufficient to convey the concepts that are pertinent to the tree-based

206

RFID Systems

probe slot 1 probe slot 2 probe 1 collision 2 collision 3 (t1 t2 t3)

(t1 t2)

slot 3 idle ()

probe 4

slot 4 single (t1)

time

Figure 8.2 Time diagram that represents the channel over which the tag-reader arbitration and communication are done. Probes are sent by the reader, while in the other slots the reader receives. For example, in slot 1 tags τ1 , τ2 , τ3 collide upon sending replies simultaneously.

protocols. In Section 8.4 we will relax the assumptions and discuss how various error types affect anti-collision protocols. A tag sends a reply only if it is asked to do so by a probe sent by the reader. This is illustrated in Figure 8.2, where tags send replies in response to the probe. For example, upon receiving probe 1, tags τ1 , τ2 , τ3 send their replies. By enabling replies from certain subsets of tags, the reader arbitrates the collisions on the channel. Note that the duration of the packet sent by each tag is constant and equal to a slot, so that if two tags transmit simultaneously their packets are completely overlapped. The reader then piggybacks feedback to tags in the next probe. For example, in Figure 8.2, probe 2 sent by the reader carries feedback that informs the tags of the outcome (collision) in slot 1. When k tags transmit in the same slot, then the interrogator perceives the channel in that slot as: • Idle (I ) if k = 0 that is no tag replies. • Successful reception (S) or tag resolution if k = 1. • Collision (C) if k ≥ 2. In order to introduce tree-based arbitration protocols, we will assume that each tag is capable of producing random bits when asked to do so by the reader. This assumption will be revised in Section 8.4, where we will also discuss how the bits from the binary vector that describes tag’s ID can be used for arbitration. In the absence of errors, the efficiency of the arbitration protocols is measured both in terms of time and number of messages. If there are n tags, then we are interested in the average time T¯n that the protocol consumes to identify all the tags. Let Tn be a random variable that stands for the time consumed when a particular instance of the arbitration protocol is executed. From Figure 8.2 it can be seen that Tn has two components: (1) the time used to send the probes; and (2) the time slots used for transmission from tags. We will assume that the duration of a probe sent by the reader is zero, and only focus on the time slots consumed for backscattering. With zero-length probe, the time efficiency of an arbitration protocol is defined as ηn = T¯n , see [20]. Note that another measure that n is traditionally used to assess the performance of tree protocols is the average number of messages M¯ n sent during the arbitration process. This would be relevant if active tags were considered, as in that case each transmitted message consumes energy from the tag. On the other hand, when dealing with passive tags, M¯ n is rather irrelevant, as the tag energy is supplied by the reader and the average energy spent by the reader is linearly proportional to the average duration of the arbitration process.

Tree-Based Anti-Collision Protocols for RFID Tags

207

8.2.2 Basic Tree Protocols Tree-based protocols (also known as splitting-tree protocols or tree-walking protocols) have emerged as a solution to the multiple access problem over a shared medium. Several research groups have invented practically the same approach rather simultaneously [8–10]. The motivation for proposing these algorithms can be explained through a simple example with a single reader and two tags. The reader sends a probe requesting replies from the tags. Given that the reader does not know the tags’ ID, this probe solicits an answer from unspecified tag and hence both tags are entitled to send a reply. After observing collision, the reader knows that there are at least two tags that have sent a reply. However, the number and the ID of these tags are unknown to the reader, giving rise to the symmetry problem, that is, all tags involved in the collision appear the same to the reader. In order to break this symmetry, the reader uses randomization. This means that after the collision, both tags toss a fair coin to generate a 0 or 1 value and in the next probe the reader asks only the tag that obtained 0 to send a reply. This procedure is repeated until the two tags have obtained different values for the random bits. Figure 8.3 shows, through an example, how this basic idea can be generalized to more than two tags. In the example, the total number of tags τi is equal to n = 8. We define collision multiplicity (or conflict multiplicity) to be the number of tags that transmit when a collision is observed. The node labeled sj refers to the outcome in the j -th slot. For example, the collision in the first slot s1 has a multiplicity of eight, because all the tags sent their replies when responding to the initial probe. The level of a tree node is the path length from that node to the root of the tree. Each tree node is uniquely associated with a string called address. The address of a tree node is determined by the tossing outcomes for the tags belonging to that node. Using the tree representation, we can say that an

s1 0

1

C

s2 0 C 1 s3

t1

s7 C

0

1

s8 0C 1

s4 0 C 1

s11 0 C 1

s5

s6

s9

s10

s12

t2

t3

t4

t5

I

s13 0 C 1 s14

t6

s15 0 C 1 s16

s17

t7

t8

Figure 8.3 An instance of the binary tree algorithm for collision multiplicity of N = 8. Each vertex represents a slot that can be in one of the three states: Idle (I), Single (S) or Collision (C). For channel state “S,” τi denotes the resolved or singulated tag.

208

RFID Systems

address (a node) is enabled in slot si if the tags that belong to that node are allowed to transmit in si . At slot s1 the tree root (level zero) is enabled, all eight tags transmit and the reader observes a collision. The address of the root node is empty, denoted by . Each tag tosses a fair coin and, in this instance of tree protocol, τ1 , τ2 and τ3 obtained value 0. At slot s2 , the left descendant of the root node is enabled and τ1 , τ2 , τ3 are enabled to transmit. Note that the node enabled at slot s2 is at level one of the tree and has address 0. The tags τ4 − τ8 that obtain coin value 1 after slot s1 belong to the other node at level one, with the address 1. Having collision in slot s2 , the tags τ1 , τ2 and τ3 again toss fair coins. Only τ1 obtained 0, while τ2 and τ3 obtained 1. Thus, when the node with address 00 (at level two of the tree) is enabled, only τ1 sends a reply that is correctly received by the reader. Having transmission from a single tag, at slot s4 another node at the same level is enabled, with address 01. Based on this explanation, the reader can go through the other steps of the example in Figure 8.3. Figure 8.5 shows the timing diagram of the tree algorithm from the example on Figure 8.3. It can be seen that the reader recursively resolves collisions until it obtains single (successful) answer from each tag. Figure 8.4 shows a possible implementation of the basic variant of binary tree protocol by using a pseudocode. The code for the algorithm run at the reader represents one complete tree traversal. By setting a = empty, the reader can restart the traversal from the root of the tree. The pseudocode at the tag assumes that each probe carries the complete address. Note that the original tree algorithms proposed by [10] run in a completely distributed manner, without centralized probes, by using stack-based implementation.

a = empty; end=no; while(end == no) transmit probe with address a; L = length(a); // if a is not empty, then a = (a1a2··· aL) receive tag-reply; if tag-reply = IDLE transmit feedback; if tag-reply == COLLISION if L == 0 a = (0); else a = (a1a2··· aL0) else if a == empty end = yes; else while aL == 1 delete aL; L = L-1; if L == 0 end = yes; else aL = 1;

a = empty ;end = no; while (end == no) receive pr; %this is the address in the probe if a == pr; transmit; get feedback at the end of the slot; % current address is a = (a1a2...aL) if(collision) set b = randombit; set a = (a1...aLb); else if(single) set end = yes;

(a) The algorithm run at the reader

(b) The algorithm run at a tag

Figure 8.4 Pseudocode of the algorithms run at the reader and at a tag for the basic variant of binary tree protocol.

Tree-Based Anti-Collision Protocols for RFID Tags

209

init. slot 1 probe slot 2 probe slot 3 probe slot 4 probe slot 5 probe slot 6 probe slot 1 probe C 0 C 00 S 011 S 1 C S 01 C 010 (t1-t8)

(t1-t3)

(t1)

(t2-t3)

(t2)

(t3)

(t4-t8) time

Figure 8.5 Timing diagram for part of the protocol execution described in Figure 8.3. The notation (τi − τj ) below a slot denotes the set of tags τi , τi+1 , τi+2 . . . τj that transmits in that slot.

Tree algorithms belong to the class of collision resolution protocols, rather than collision avoidance protocols, where a representative of the latter are the protocols that use random backoff, such as the MAC protocol for IEEE 802.11 [21]. When there are no other errors except collisions, the tree algorithm is guaranteed to collect a reply from each tag that is in the range of the reader. An interesting dividend is that the algorithm creates ad hoc addresses for tags. Namely, after the collision resolution, the reader might need to communicate with a particular tag. Instead of using the full address of that tag, the reader can assign a short address to the tag, and this address is represented by the address of the tree node in which this tag has been successfully resolved or singulated. For example, in Figure 8.3 the short ad hoc address of the tag τ3 is 011.

8.2.3 Improvements to the Basic Tree Protocol In this section we describe several ideas that improve and generalize the basic tree protocol, described in the previous section. At first, note that the described tree algorithm constructs a tree by binary branching as dictated by bit flipping. The branching can be generalized to M rather than two outcomes in the randomization process. Larger M decreases the probability of collision, but increases the probability of obtaining an idle slot. If the idle slot has the same duration as a single/colided slot, then it is clear that increasing M does not increase efficiency. However, an idle slot can have a shorter duration than a slot in which one or more tags are transmitting. This is because the reader, in principle, does not need to wait for the whole duration of the slot to detect that no tag has transmitted. Hence, the duration TI of an idle slot can be lower than the duration of the slot with tag transmission, denoted by TS . Such a model for collision resolution is termed a Carrier Sense Multiple Access (CSMA) model. If TI  TS , then an idle slot is very “cheap” in terms of duration and the tree protocol should be engineered in a way to produce idle slots with higher probability as compared to the probability of occurrence of collided slots. Nevertheless, in this text we will not specifically treat the CSMA model. Possible optimizations of the tree-based algorithms in CSMA setting are discussed in [22] and [18]. Massey in [23] and Tsybakov and Mikhailov in [10] proposed a simple way to improve the basic variant of the binary tree algorithm. They noticed that there are some tree nodes that certainly contain more than one tag and will thus certainly result in collision if the node is enabled. These nodes should be skipped during the traversal of the tree. For example, in Figure 8.3, after the collision in slot s11 and an idle slot s12 , it is clear that the enabling of the address 111 results in certain collision. Hence, after the idle slot s12 , the terminals belonging to node 111 toss a coin immediately and in slot s13 the enabled node is 1110. This algorithm will be referred to as Modified Binary Tree (MBT) algorithm.

210

RFID Systems

However, as discussed in Section 8.4, this small optimization has a very high price if transmission errors occur with non-zero probability. Early on, researchers observed that knowing the multiplicity (number) of tags involved in a collision can speed up the resolution of that collision. Capetanakis in [9] observed that binary tree algorithms are most efficient for conflicts of small multiplicity and applied this observation to devise a dynamic tree algorithm for scenarios with Poisson arrivals. The idea is to divide the initial set of n tags into smaller groups and then apply a tree protocol within each group. Multiplicity estimation to speed up tree protocols has been used in [24] and [20]. In both works, the proposal is a hybrid algorithm that consists of two phases. The first phase is devoted to the estimation of multiplicity. After obtaining an estimate n, ˆ the second phase starts. The unresolved terminals are randomly split into approximately nˆ groups and each group is resolved using the basic tree algorithm. Note that both [24] and [20] have an explicit phase for estimating nˆ and the accuracy of such an estimate is using algorithm-specific parameters that are chosen in advance. On the contrary, in [18] we have introduced a new framework for tree-based arbitration protocols where a running estimate nˆ of the initial population is obtained, which becomes increasingly accurate as the collision resolution progresses. Before proceeding to the details of the framework from [18], we introduce the Clipped Binary Tree (CBT) algorithm, which is the main ingredient of tree algorithms with a running estimate. CBT algorithm has been independently introduced by several authors in the context of random access for an infinite population of terminals that generates transmission requests with Poisson arrivals [25]. It is identical to the MBT algorithm except that it is stopped, that is, the tree is clipped whenever two consecutive successful transmissions follow a conflict. CBT resolves the batch conflict partially, since not necessarily all nodes of an initial batch are resolved during the execution of CBT. For example, referring to the tree in Figure 8.3, the first instance of the CBT algorithm starts in slot s1 and terminates in slot s6 , resulting in three resolved tags. Now, the standard binary tree algorithm would continue to enable the node labeled with s7 , that is, all the tags that have flipped 1 after the initial collision. The key observation from [18] can be applied to this example as follows. After the termination of the first CBT algorithm, it is noted that 3 tags have flipped 0, so that the expected value of the number of tags that have initially flipped 1 is 3. This implies that the enabling of the node s7 , which belongs to the level 1 of the tree, will highly likely result in collision and therefore the next probe should directly enable a node that has a higher level in the tree. For example, if the node with address 10, from level 2, is enabled after the termination of the initial CBT algorithm, then one slot is saved by skipping the node s7 at level 1. In such case, the next instance of the CBT algorithm would start by enabling the node with address 10 and terminate when tag τ5 successfully sends its reply, that is, after the address 101 has been enabled. In the next section, we describe the framework that systematically estimates which tag subsets should be enabled for transmission after a single instance of CBT is finished.

8.2.4 General Arbitration Framework for Tree-Based Protocols The key part of the framework proposed in [18] is the interpretation of the randomization procedures used in a tree protocol. Let us assume that, upon the transmission of the initial probe, each tag generates a random real number, uniformly distributed in the interval

Tree-Based Anti-Collision Protocols for RFID Tags

211

[0, 1). This random number is referred to as a token and let ri denote the token generated by the tag τi . Let xi = (xi1 xi2 xi3 . . .) be the binary representation of fractional part of ri . Then each ri is an infinite string of 0s and 1s and: ri = r(xi ) =

∞  xij j =1

(8.1)

2j

The token ri can be understood as an infinite reservoir for generating fair coin tosses, since each xij gets a value 0 or 1 with probability 0.5. For example, in Figure 8.3, after the initial collision, the node with address “0” is enabled, so that all the tags that have generated their tokens within the interval [0, 0.5) are entitled to send a reply. This principle is generalized as follows. The act of enabling a tree node with a certain address from Figure 8.3 corresponds to the act of enabling transmission by tags that have generated their token in a certain sub-interval of [0, 1). For example, when the tree node with address 10 is enabled in slot s8 , it corresponds to enabling transmission by tags that have their tokens generated in [0.5, 0.75). More generally, if the node with address: a = a1 a2 . . . aK

(8.2)

is enabled, then this corresponds to enabling the interval [b, c) where: b=

K  ai i=1

2i

c=b+

1 2K

(8.3)

Hence, instead of using a binary tree, the arbitration process can now be represented using a sequence of enabled intervals. Figure 8.6 represents the example from Figure 8.3 using a sequence of enabled intervals. Such a representation brings a qualitatively new observation: after the termination of one instance of the CBT algorithm, we can obtain an estimate of nˆ of the tag population size n and this estimate can be used to find out how many tags are still there to be resolved. For each slot on the abscissa of Figure 8.6, there is a rectangle that stands for the enabled interval. A shaded interval denotes that there is a collision in the slot, for example, when the enabled interval is [0, 0.5) in the second slot. On the other hand, in slot 3 the rectangle is marked by τ1 and it denotes that the enabled interval [0, 0.25) contains only the tag τ1 . The enabled interval in slot 12 is empty. We proceed to explain how the multiplicity estimation can be used in this framework. After the termination of the first CBT instance in slot 6, the reader observes 3 tags in the interval [0, 0.5). If n is the total number of tags in [0, 1), then the expected number k of tags in [0, 0.5) is k = 0.5n. Since k = 3, we can estimate nˆ = 0.5 = 6, so that the expected number of tags in the unresolved interval [0.5, 1) is 3. The second instance of the CBT algorithm is finished when tag τ5 is resolved, after which the reader observes that 5 tokens are located in the interval [0, 0.75). Hence, a new estimate can be made of 5 the initial tag multiplicity: nˆ = 0.75 = 6.67, while the number of tags left to be resolved 5 in the interval [0.75, 1) is estimated to be 0.75 0.25 = 1.67. In general, assume that the arbitration is at the stage in which there are k resolved tags with tokens in [0, p). Then there are n − k unresolved tags (with n unknown) with their tokens in the interval [p, 1). As stated before, knowing the multiplicity n − k can speed

RFID Systems

Tokens in the interval (0,1)

1 r8x r7x r6x

enabled interval

212

t7

t6

t8

t5

r5x

t4

r4x

t3

r3x

t2

r2x

t1 r1x 0

1

2

3

4

5

6

7

8

9 10 11 12 13 14 15 16 17

C

C

S

C

S

S

C

C

S

S

C

I

C

S

C

S

slot S state

Figure 8.6 Binary tree from Figure 8.3 represented in the framework that uses tokens and sequence of enabled intervals. A rectangle denotes the interval enabled in a given slot. A shaded rectangle denotes collision; a rectangle labeled with τi denotes that only tag τi has its token in that particular enabled interval.

up the arbitration protocol for the remaining tokens. Considering the token generation process, the reader can make the following observation: From n tokens that are uniformly picked from  the interval [0, 1), where n is unknown, k tokens have been picked from the interval 0, p). The probability of that event, conditioned on n, is   n k P (k|n) = p (1 − p)n−k (8.4) k For a given k, p, we use the maximum likelihood (ML) estimator: nˆ =

k p

(8.5)

which is not necessarily an integer. The estimator is unbiased, since E[n|n] ˆ = n, while its variance is: n Var[n|n] ˆ = −n (8.6) p As the arbitration protocol progresses, the value of p increases, which decreases the variance Var[n|n] ˆ and therefore the obtained estimate becomes more accurate. Using a framework based on token generation, the authors of [18] propose several arbitration protocols. Each of those protocols achieves faster collision resolution compared to basic tree protocols. In particular, the protocol termed Interval Estimation Conflict Resolution (IECR) has been reported to be the fastest known collision resolution protocol.

Tree-Based Anti-Collision Protocols for RFID Tags

213

The algorithms in [18] operate by iteratively applying CBT. Let us assume that upon a termination of a CBT instance, the reader observes k resolved tokens in [0, p). Based on the estimation of the conflict multiplicity, the next enabled interval is [p, p + p). If a collision is detected, a new CBT instance starts and the process is repeated. If in [p, p + p) there is a single token or no tokens, then the next slot is treated as a termination slot for that CBT instance and the operation of estimation/enabling new interval is reapplied. The speed of IECR is due to the optimized selection of the length of the enabled interval upon termination of a CBT instance. The authors in [18] show that the speed of IECR in resolving batch conflicts is asymptotically equivalent to the speed by which the collision resolution for Poisson arrivals is performed using the First Come First Serve (FCFS) tree algorithm [25]. In [18] it is shown that, when n is large, the length of the next enabled interval should be: 1.26p k

(8.7)

where k is the number of tokens found in the interval [0, p) after the termination of an CBT instance. In the same work the authors proved that such a selection of the enabled interval results in asymptotic time efficiency of the IECR algorithm that is equal to: n = 0.487 lim τn = lim n→∞ n→∞ Tn which is identical to the efficiency of the FCFS algorithm [25] for Poisson arrivals. The implementation of the optimized IECR algorithm in RFID setting is not straightforward, since the limits of an enabled interval [a, b) are in general real numbers and each of them has a binary representation with infinite length. We say that a binary representation of a real number x ∈ [0, 1), written as x = 0.x1 x2 x3 . . ., has infinite length if there is no finite integer i so that for all j > i, the decimals are xj = 0. The selection of an optimized length p may result in a sequence of enabled intervals that cannot be efficiently contained in the probe sent by the reader, as the interval is described by infinite-length numbers. The observation in the previous paragraph makes another variant of the estimating tree algorithms, termed Estimating Binary Tree (EBT) [18], more suited for usage with RFID tags. In EBT, the length of the enabled interval is always equal to 2−l , where l is an integer. The advantage of such a choice is that each enabled interval corresponds to a single node in the evolution of the binary tree, so that the enabled interval can be represented in a compact way, using the correspondence between the address in the tree and the actual interval, as described by Equations (8.2) and (8.3). The only thing that needs to be decided is how to select l, after a CBT instance is terminated. For example, a straightforward way is to select l in the way that it maximizes the probability of obtaining a single response:  p   l ∗ = argmin 2−l −  (8.8) k Finally, a priori information about the size of the tag population can additionally improve the efficiency of tree protocols. For example, if it is known that the number of tags is at least nmin > 1, then the initial probe does not need to enable the whole interval

214

RFID Systems

Table 8.1 Time efficiency ηn =

n T¯n

of various conflict resolution algorithms.

Number of tags n

MBT

EBT

EBT with nmin = 64

IECR

IECR with nmin = 64

Optimized IECR with known n

50 100 1000

0.379 0.376 0.375

0.452 0.460 0.463

0.472 0.471 0.465

0.463 0.467 0.483

0.467 0.479 0.486

0.499 0.494 0.488

  [0, 1), but it can start, for example, with the interval 0, n1.26 . Such an approach can min avoid several of the collisions that occur in the initial stage, until the first CBT instance is terminated and the first estimation nˆ is made.

8.2.5 Numerical Illustration In this section we provide some performance figures to highlight the speed by which the described tree algorithms resolve the collisions in the cases without transmission errors. Table 8.1 illustrates the time efficiency, defined as ηn = T¯n , where n is the number of n tags and Tn is the average number of slots required to resolve all tags. The performance numbers are given for several algorithms. As a reference, we have also provided the time efficiency for the reference IECR version in which the number of tags n is known a priori . This reference algorithm operates analogously to the described IECR algorithm, except that it uses n instead of estimating n, ˆ and after the termination of each CBT instance, it knows exactly the number of tags that are left in the unresolved part of the interval [p, 1). It can be seen that estimating tree algorithms are superior to the Modified Binary Tree, which is in turn superior to the basic variant of binary tree protocol. If no a priori information is known about the tag population size, then the efficiency of EBT and IECR rises as the tag population increases. The required the number of resolved tags k that can produce a reliable estimate nˆ grows slower than n. Therefore, tree protocols with running multiplicity estimation bring benefits already when a relatively low fraction of the tags is resolved. It is also seen that EBT pays the price in terms of time efficiency relative to IECR. For example, when n = 50, the average number of slots required to resolve the tags is Tn = 110.62, while it is Tn = 106 with IECR and around Tn = 132 with MBT. Having some or complete a priori knowledge about n improves performance. The results in Table 8.1 show that when n is known a priori , efficiency decreases as n increases. This is because for small n, the a priori information about n is more valuable compared to the estimation of the number of remaining tags in the unresolved interval. Such an advantage of the a priori knowledge diminishes as n increases.

8.3 Tree Protocols in the Existing RFID Specifications Having introduced the main ideas used in tree protocols, in this section we turn to actual realizations of tree protocols in RFID systems. Although many of the principles are retained, there are changes that are necessitated by the limitations of passive UHF tags.

Tree-Based Anti-Collision Protocols for RFID Tags

215

We describe two tree-based protocols used in EPCglobal Generation 1 tags: Class 0 and Class 1.1 In many applications, Generation 1 tags and the associated protocols have been displaced by Class 1 generation 2 tags [26]. Besides these two protocols, another standardized tree protocol for UHF tags is ISO 18000-6A [27].

8.3.1 Tree Protocol for EPCglobal Class 0 Thus far we have assumed that the tags can generate random numbers, which can in turn be used as tokens in anti-collision protocol. EPCglobal Class 0 tags (“Class 0 tags”) do not use random bits, but arbitration relies on the unique ID of each tag. If the tag ID consists of L bits, then each ID x1 x2 . . . xL is uniquely represented by a token in the interval [0, 1). The token is a real number that has a binary representation 0.x1 x2 . . . xL , albeit not necessarily uniformly distributed in [0, 1). We can use again Figure 8.3 to exemplify arbitration based on tag ID. Tags τ1 , τ2 , τ3 are allowed to transmit in slot s2 since each of them has the first bit of the ID x1 = 0, while the other tags τ4 − τ8 have x1 = 1. Thus, from Figure 8.3 it can be inferred that the ID of τ1 is 00 · · ·, the ID of τ6 is 1110 · · ·, etc. Tree protocol for Class 0 tags uses bit-by-bit query response rather than having probe packet from the reader followed by packets from tags. The reader starts arbitration by transmitting a special NULL sequence. Tags that receive this sequence backscatter the first bit x1 of their ID. If x1 is either 0 or 1 for all tags that backscatter, then the reader receives 0 or 1, respectively, after which it echoes that bit. On the other hand, if some tags transmit x1 = 0 and other tags transmit x1 = 1, then, due to the modulation format used for backscattering, the reader detects that both 0 and 1 have been sent. Note that this is different from the idealized collision model, used in Section 8.2.1. If the reader receives both 0 and 1, then it decides randomly to echo either 0 or 1. In the next step, tags that have x1 equal to the bit echoed by the reader transmit the value of the bit x2 . On the other hand, tags that have x1 different from the bit echoed by the reader are muted until the next transmission of a NULL sequence. We can reuse the example from Figure 8.3 to illustrate the operation of the tree protocol used for Class 0 Generation 1 tags. The bit used to label each of the branches is the bit echoed by the reader. Assume that the tags have 5-bit addresses. The addresses of the tags are, for example, X(τ1 ) = 00100, X(τ2 = 01010), X(τ3 ) = 01100, etc., where X(τi ) stands for the address of τi . Each slot has duration equal to one bit. Assume that after initiating the tree at slot s1 , the reader chooses to echo 0, and after s2 it chooses to echo 0. Here comes a change compared to the execution shown in Figure 8.3: τ1 sends in sequence x3 = 1, echoed by 1; it proceeds analogously for bits x4 = 0 and x5 = 0. Clearly, since τ1 is the only tag that sends reply to the reader in these last three queries, each bit is received and echoed correctly by the reader, since we are neglecting noise-induced errors. After its fifth bit is being read, τ1 goes into a dormant state. The reader sends again the NULL signal, receives 0 and 1, echoes 0, receives 1, echoes 1, receives 0 and 1, echoes 0, end proceeds to get the last two bits of τ2 . The resolution of the remaining tags proceeds in a 1

At the time of writing, the specification for Class 1 Generation 1 tags was no longer available. Nevertheless, a “stable” reference for both protocols is [26].

216

RFID Systems

similar way. One peculiar thing to be noted about this version of the tree protocol is that there are no slots with idle reply, such as slot s12 in Figure 8.3. The reason is that there is no collision, but instead a combination of 0 and 1, the reader always echoes bit value of an existing tag, so that there is always at least one tag that sends a reply. Although Class 0 tree protocol is easy to implement and efficient in terms of transmitted symbols, it has several significant drawbacks. The drawbacks that stem from the physicallayer phenomena are treated in [26], here we outline some of the problems solely related to MAC-layer operation. The operation described above is rather a simplified version of the full protocol. In reality, each tag is in a default Dormant state and is wakened up by a Reset signal from the reader. After the Reset signal, the reader transmits a sequence of symbols to calibrate the tags, and only after such a calibration can a tag receive commands from the reader. Note that, besides the symbols 0, 1, and NULL, the specification defines other commands that can be sent by a reader. Note that the tags that arrive after the Reset/calibration commands have been sent, such as in a scenario with conveyor belt, cannot participate in the execution of tree protocol and should wait until another set of reset/calibration commands is issued. Hence, in order to be able to read late arriving tags, the reader should frequently issue calibration commands, which hinders the speed of the anti-collision protocol. Another remark regarding the tradeoff in time efficiency of the algorithms is in order. The protocol saves execution time by using short transmission with a symbol-level duration. On the other hand, short commands (0, 1, or NULL) used by the reader necessitate that the tree traversal is restarted from the root of the tree each time a tag is resolved. In general, short reader commands decrease the flexibility of the protocol to move within the tree and such flexibility is necessary to implement some of the time-efficient protocols, such as EBT.

8.3.2 Tree Protocol for EPCglobal Class 1 A serious drawback of Class 0 tags is that they are factory-written, such that the protocol does not make provisions for writing new data in the tags. This has been one of the main reasons for introducing EPCglobal Class 1 Generation 1 tags. Different from Class 0, in Class 1 the reader sends commands in packets and the tags reply to commands by whole packets. Each reader command has a synchronization sequence, such that, in principle, late arriving tags can join the arbitration protocol. Having reader commands with more content offers flexibility in traversing the tree, but the price is that the duration of a single protocol slot (containing reader command and tag reply) is much larger than the symbol duration. The Class 1 specification provides commands to support a tree-based protocol, but does not specify a particular way to traverse the tree [26]. The reader can use the PingID command, which is in fact a probe sent to a particular subset of tags. This command consists of: a pointer [PTR], which points to a location (or bit index) in the tag identifier, and [VALUE], which is a bit mask of length [LEN]. Tags that have a matching bit patters to VALUE in their ID at a location indicated by [PTR], are entitled to send reply to the probe. For example, if PTR = 1 and the bit mask is “11,” then the tags that have ID x11, where x can be either 0 or 1, can send a reply to the probe. Interestingly, not all tags with a matching bit mask send a reply immediately. The reader delineates eight bins after the probe termination and the bins are labeled 000, 001, 010, . . . 111. A tag that can match the required bit mask, uses the next three bits of its ID to determine in which

Tree-Based Anti-Collision Protocols for RFID Tags

217

bin to send its reply. For example, the tag with address x11010 · · · will send a reply in the third bin after a probe with mask x11. Described in terms of arbitration framework with tokens, a probe simultaneously enables eight intervals. In addition, the set of bit masks that are used can be adapted to incorporate the ideas of multiplicity estimation, such as in EBT, and speed up tag resolution. Both Class 0 and Class 1 tags from EPCglobal Generation 1, along with the associated arbitration protocols, have been gradually displaced by Class 1 Generation 2 tags [26], which use a variant of the ALOHA protocol. One of the main reasons for abandoning these tree protocols was the fact that they have difficulty in dealing with late arriving tags. Therefore, in Section 8.4.3 we will discuss which provisions can be made to create tree protocols capable of dealing with moving tags and late arriving tags.

8.4 Practical Issues and Transmission Errors 8.4.1 Token Generation The two practical tree protocols for UHF RFID, described in Section 8.3 do not rely on random number generation at tags. Generation of random bits at a passive tag is not a problem and has been used for the ALOHA-based protocol for Class 1 Generation 2 tags. Hence, a future version of tree-based RFID protocol can rely on random bit generation rather than use the ID bits. This can be important from a security point of view, since the masks/addresses used in the probes sent by the reader are not correlated with the tags ID. If ID bits are used for arbitration, then the tokens that correspond to the tags ID may not be uniformly distributed in [0, 1), which poses problem for the estimation methods used in, for example, EBT. Note that EBT provides a better average performance than BT when the tokens are uniformly distributed in [0, 1). If the ID of tags in a given set are not uniformly distributed on the interval [0, 1), then the statistical estimation utilized is not valid. Thus, in such a case EBT may be even slower than the basic binary tree protocol. We illustrate this by an example. The set has five tags and their ID is in the interval [0.25, 0.5). If the basic binary-tree algorithm is applied, then when all the tags are resolved, the next enabled interval is [0.5, 1), which will result in an idle response and thus the algorithm will terminate. On the contrary, if EBT is applied, then after the tokens are resolved, the algorithm continues to estimate the cardinality of the tag set and enables intervals smaller than [0.5, 1), so that it takes more time to terminate the arbitration protocol. A way to randomize the identity of the tags is that the reader uses a predefined randomized permutation of the bits in the ID. This permutation can be used to scramble the addresses prior to the start of the EBT algorithm and thus randomize the prefixes used in collision resolution.

8.4.2 Transmission Errors The described tree protocols operate under the assumption that there are no channel errors. The channel errors affect these protocols in different ways. In order to develop a systematic view on the error impact on tree protocols, we first divide the errors into errors that occur at a tag and reader.

218

RFID Systems

8.4.2.1 Errors at the Tag An error at a tag causes the tag not to send a reply back to the reader. We differentiate two error types at the MAC protocol layer: static and dynamic errors. Static error models the case in which, during a whole collision resolution session, the tag is in a “blind spot”, so that none of the probes in the whole session is received by the tag. The effect of the blind spot is as if the tag is outside the range of the reader. Let ps denote the probability that a tag experiences a static error during a given collision resolution session. On the other hand, a dynamic error occurs so that each probe sent by the reader is independently received with probability (1 − pd ) and received incorrectly with probability pd . Looking from a physical-layer perspective, static errors result from deep signal fade, while dynamic errors are caused by noise. Probability of static (dynamic) errors at tag τi is denoted by psi (pdi ). In general each psi = psj and pdi = pdj when i = j . The impact of static errors on the protocol is simple – the tag does not participate in the reader session. One might think that a static error occurs when a tag is at a “blind spot” – a spot at a close distance to the reader that has a poor radio link to the reader. If a tag is at a blind spot in one session does not necessarily mean that the tag will be in blind spot in another reader session. For example, before the next session with the same reader, tags may be physically displaced, such that the radio link to the reader at the new location is uncorrelated with the link at the old location. Furthermore, if the next session is performed by another reader, its radio link to a tag can be also decorrelated, so that the tag is not at a blind spot for the second reader. This latter case can be represented by the same model used to represent readers with different reading ranges, as shown in Figure 8.1, the tag τ4 is not in the range of reader RA but is it in the range of reader RB . The model in Figure 8.1 can be conceptually extended to account for tags that are in the blind spot of both readers, and hence they will be missed after both readers terminate their sessions. Reference [28] introduces several methods that rely on the described error model in order to approximately determine the probability of having missed tags after both readers terminate the sessions. The impact of dynamic errors on the tree protocols is more involved. Assume that the reader is running the basic tree protocol, without any estimation or the optimization introduced by the MBT. Consider the example in Figure 8.6 and assume that tag τ3 has not received the probe correctly in slot s6 , but it receives the probe in slot s7 correctly. Accordingly, the reader observes an idle slot s6 . One can easily verify that until the end of the collision resolution session, this tag will not receive a probe that entitles it to send a reply. In terms of enabled intervals, starting from the slot s7 , the reader treats the interval [0, 0.5) as resolved, while the token of the unresolved tag τ3 lies in that interval. Note that, in the errorless case it is impossible for an unresolved tag to see a probe which informs the tag that the interval in which it belongs is resolved. There are two approaches to mitigate the effects of dynamic errors. In the first approach, even after the interval [0, p) is assumed to be resolved, the reader redundantly enables this interval or some of its subintervals in order to probe whether there is any missing tag left. In the second approach, if a tag that has a token in [0, p) has not been singulated, while the reader sends out a signal to indicate that it considers the interval [0, p) resolved, then the tag generates its token to be uniformly distributed within [p, 1). The second approach

Tree-Based Anti-Collision Protocols for RFID Tags

219

implies that the estimates used in the framework with enabled intervals, such as for IECR or EBT, need to be changed, since the unresolved interval is statistically different from the resolved interval. Furthermore, if the resolution is purely identity-based, then the token re-generation may lead to unresolvable conflicts. For example, let τ1 have identity 1x2 x3 . . . and let τ2 have 0x2 x3 . . ., that is, they differ only in the first bit. If tag τ1 is missed after the interval [0,0.5), the tag can regenerate its token in [0.5, 1) by ignoring the first bit–however, doing so means it cannot be differentiated from τ2 in the collision resolution process. Hence, it is more feasible to leverage on the approaches in which the reader re-enables some of the previously resolved intervals. The basic binary tree algorithm is robust with respect to the errors at the tag, in a sense that these errors are not fatal to the algorithm operation. On the other hand, modified binary tree can experience a fatal error. Consider the situation with two tags τ1 , τ2 , with their respective tokens r1 = 0.1 and r2 = 0.2. Assume that the first probe, which enables [0, 1) has been received correctly by both τ1 and τ2 , so that the reader observes collision. Next, assume that the second probe, which enables [0, 0.5) has not been received by either of the two tags. One can check that the MBT algorithm will enter an endless session, where the reader enables smaller and smaller subintervals from [0.25, 0.5), getting only idle replies. On the other hand, if an estimating tree algorithm is applied, for example’ EBT, without using the modification of the MBT, then the algorithm stays robust to errors, but the estimators will be less precise. With static errors, the probability that a reader misses a tag is ps . On the other hand, when there are dynamic errors that occur with probability pd , the probability of missing a tag is not in a straightforward relation with pd . We show through a simple example that the probability that a tag is missed depends on the other tags. Let there be two tags only, and let the token of tag τ1 be in [0, 0.5), while the token of τ2 be in [0.5, 1). Let µi denotes the event that τi is not read at the end of the reader session. Let us assume that the probability of dynamic error pd is identical for both tags. Thus, Pr[µ1 µ¯ 2 ] denotes the probability that τ1 is missing and τ2 has been successfully read in the session. The probability that both tags will be missed P r[µ1 µ2 ] is given as: Pr[µ1 µ2 ] = pd2 + (1 − pd )2 pd2

(8.9)

where the first pd2 is the probability that both tags do not receive the probe that enables [0, 1). The second term comes from the probability that both tags receive the first probe, which is (1 − pd )2 and then each tag misses the probe that enables its subinterval – τ1 misses the probe for [0, 0.5) and τ2 misses the probe that enables [0.5, 1). In a similar manner, the following probabilities are determined: Pr[µ1 µ¯ 2 ] = pd (1 − pd ) + (1 − pd )2 pd (1 − pd )

(8.10)

Pr[µ¯ 1 µ2 ] = pd (1 − pd ) + (1 − pd ) pd (1 − pd )

(8.11)

Pr[µ¯ 1 µ¯ 2 ] = (1 − pd )4

(8.12)

2

The marginal probability that τ1 is missing is given by: Pr[µ1 ] = Pr[µ1 µ2 ] + Pr[µ1 µ¯ 2 ] = pd [1 + (1 − pd )2 ] = Pr[µ2 ]

(8.13)

220

RFID Systems

where the equality with Pr[µ2 ] is due to symmetry. We can now try to determine the conditional probability that τ1 is missing given that τ2 is missing: Pr[µ1 |µ2 ] =

Pr[µ1 µ¯ 2 ] = pd Pr[µ¯ 2 ]

(8.14)

which is clearly different from Pr[µ1 ]. From this example, it is confirmed that the tag has less chance of being missed if its resolution requires the reader to go deeper in the tree, that is, towards smaller enabled intervals. This is because, for a deeper tree level, more probes are reaching a tag and thus the chance to respond is higher. In that sense, one can expect that tree protocols will more reliably cope with reading errors at the tag, as these protocols are intensively using probes, as opposed to the ALOHA protocol family, where relatively fewer probes are sent. 8.4.2.2 Errors at the Reader The effect of non-ideal receptions at the reader cause erroneous interpretation of the channel state at the reader. For example, if only one tag transmits and the packet is not received correctly by the reader, it might be interpreted either as idle (not sufficient received power) or collision (transmissions are present, but are not decodable). Figure 8.7 shows the general error model for errors at the reader [23]. The value puv stands for the probability that the channel state u is perceived as channel state v. In the ideal model, assumed so far, we have puu = 1 and puv = 0 if u = v. Note that the model assumes that the probability of producing a valid single output when there is no transmission is pis = 0. The actual values of the probabilities puv depend on the underlying physical phenomena, such as noise and fading. It can be verified that the basic variant of the binary tree algorithm stays robust also with respect to the errors at the reader. Reference [23] describes an example in which MBT can again exhibit fatal behavior if psc > 0. Similarly to the case when the errors occur at the tag, the errors at the reader affect the correctness of the estimates made by the estimating tree algorithms. If the basic binary tree is used, with or without running multiplicity estimation, then when psc > 0, pic > 0, while all the other puv = 0, for u = v, the protocol will not produce missing tags. A tag can be missed if either of psi , pci , pcs is positive. In practice, errors will occur both at tags and at the reader. In order to mitigate the problem of missed tags, a combination of the two approaches described above should pii

idle

pic single

collision

psi pss

psc pci

idle

single

pcs pcc

collision

Figure 8.7 General model that represents the errors at the reader. puv stands for the probability that the channel state u is perceived as channel state v.

Tree-Based Anti-Collision Protocols for RFID Tags

221

be used. The protocol designer might prefer to utilize approaches in which the reader re-enables some of the intervals that have already been enabled. This is because those techniques require minimal increase of the intelligence in the tag.

8.4.3 Dealing with Moving Tags

belt position

A particularly important scenario for RFID tags is one in which tags are physically moving in and/or out of the coverage region for a given reader. Such a case occurs when tagged items are put on a conveyor belt. This also makes conditions for protocol operation nonideal, but in a way that is different from the impact that the static/dynamic errors have on the protocol. If we ignore other error types, then we can consider that the channel is ideal for the tags in range and no communication is possible with tags that are outside the reader’s range. Figure 8.8(a) depicts the situation in which the reader covers a total length L of the conveyor belt on which the tags are moving. It is again useful to represent the collision resolution process by using a two-dimensional arbitration space, depicted in Figure 8.8(b). When a given interval in the token space is enabled, a stripe of length L is enabled in the two-dimensional arbitration space, as depicted in Figure 8.8(b). If the tag density is uniform across the conveyor belt, then one stripe always contains the same expected number of tags, irrespective of the dynamics of the belt. The basic binary tree algorithm is, again, robust with respect to the tags coming in and out of range, since these effects are equivalent to the reading errors at the tag. Consider the example in Figure 8.9. At time t1 the reader observes a collision due to transmission by τ1 and τ2 and starts to resolve this collision. The token of τ1 is in [0, 0.5), while the tokens of τ2 and τ3 are in [0.5, 1). At the time t2 it resolves τ1 . Unfortunately, at time t3 , when the reader enables the interval which would have resolved τ2 , the tag τ2 is already out of range. Moreover, a new tag t3 has arrived.

reader

L (a)

0 0.25 0.5

1

token

(b)

Figure 8.8 (a) RFID tags on a conveyor belt. The reader covers length L on the belt. (b) Equivalent representation in the two-dimensional arbitration space in one time snapshot. The abscissa represents the random token in [0, 1). The light shaded stripe is the coverage area of the reader in that snapshot, while the dark shaded stripe is the area enabled when the tree protocol enables the interval [0.25, 0.5).

222

RFID Systems

t3

t1

t2

time t1

t4

t3

t1

t2

time t2

t4

t3

t1

t2

time t3

Figure 8.9 Movement of the tags on a conveyor belt, represented through three snapshots in time. The shaded area represents the reader coverage.

An important parameter for the performance of the collision resolution algorithms is the tag density in terms of number of tags per time unit that enter in the coverage area of the reader. Due to balance, this density is equal to the number of tags per time unit that exit the coverage area. When the tag density is small, the collisions are resolved quickly and there is a low probability that a tag will be missed. As the density increases, the average percentage of unresolved tags increases. The setting is reminiscent of the original setting in which the tree algorithms have been proposed–the MAC protocol for an infinite population of terminals with Poisson packet arrivals [22]. In those scenarios, the maximal stable throughput of a given algorithm is represented by the average packet arrival rate which keeps the number of buffered packets finite. In the case of the conveyor belt, we again observe an infinite population of transmitting terminals, but if the packet is not delivered until a given deadline (i.e. until the tag leaves the coverage area of the reader), it is dropped. The packet deadline is determined as  = Lv , where v is the speed of the conveyor belt. It is of interest to find the packet dropping rate as a function of the tag arrival density and the deadline . Regarding estimating the tree algorithms, it is interesting to find how the estimate, obtained after a CBT instance is terminated, is related to the tag density. But perhaps the most important parameter to determine by such an analysis is the density of outgoing tags that are not resolved. Namely, the problem of missing tags should be tackled by deploying multiple readers across the conveyor belt. The density of outgoing unresolved tags for reader RA represents the input tag density for the next reader RB . Thus, for a given number of tags per unit time on the conveyor belt, one can determine the required number of readers in order to keep the fraction of missing tags below a certain value.

8.5 Cooperative Readers and Generalized Arbitration Spaces The described framework for tree protocols is suitable to define the arbitration space for a given random access protocol. The arbitration space is a set of points (real, integers, etc.), so that the collision resolution process can be represented via a sequence of enabled subsets from that arbitration space. In the approach described in the previous section, the arbitration space is the interval [0, 1) and is one-dimensional. In the following sections we will show that it is useful to generalize the ideas of arbitration space to higher dimensions.

Tree-Based Anti-Collision Protocols for RFID Tags

223

8.5.1 Two-Dimensional Arbitration Space The two-dimensional arbitration space can be conceptually introduced by considering two readers, as shown in Figure 8.1(b). The readers RA and RB can share information over a dedicated link, which uses a wire or radio interface that is different from the one used to communicate with tags. Alternatively, one can also think of the two readers as one common controller with distributed antennas. The readers cooperate with each other in the following ways: (a) they coordinate their transmissions in order to avoid reader collision and (b) they exchange information related to the tag reading process. Let Tu denote the set of tags in the range of reader u = A, B. In general, the sets TA and TB are different. We assume that the density of the tags is uniform within the area covered by the two readers. This assumption is rather strong for RFID systems, where the tag density depends significantly on the usage scenario and the physical setup, such as a warehouse or shop. Nevertheless, such an assumption is very useful to introduce the principles of multi-dimensional arbitration spaces. Discussion on how the assumption can be relaxed is given in Section 8.5.2. We consider the following arbitration algorithm. First RA sends an initial probe, labeled A, and the tags from TA reply. Next RB sends another probe labeled B and the tags from TB reply. We define the following sets: S1 = TA \ TAB , S2 = TAB , and S3 = TB \ TAB . These two initial probes are used to make each tag aware about the set Sj it belongs to. For example, a tag that received the probe from RA , but not from RB , belongs to S1 . After the two initial probes, the arbitration should continue as three independent collision resolution processes, one for each set Si , respectively. The arbitration for the set S2 can be carried out by either of the readers. Let a session for the set Si be the uninterrupted time during which the arbitration protocol is run only for the tags from the set Si . There can be more than one session per set, such that τij denotes the j -th session for the i−th set. For example, the readers can coordinate to run the following sequence of sessions: τ2,1 , τ3,1 , τ2,2 , τ1,1 , . . .. Let the cardinality of set Si be denoted by ni and let kij be the number of resolved tokens during the session τi,j . The spatial division, introduced by the different coverage of each reader, is helpful for collision resolution as it inherently segments the tags into smaller groups. Such a segmentation suggests the following straightforward solution: run three sessions τ1,1 , τ2,1 , τ3,1 and in each session all tags from a given set are resolved, that is, ki1 = ni . Such a solution features very limited cooperation between the readers and the estimation of the multiplicities n1 , n2 , n3 is done independently. Nevertheless, if the spatial statistics of the tags is known, the readers can enhance their cooperation in estimating the population size. The key idea of the two-dimensional arbitration space is, in addition to the dimension where random tokens are generated, another (spatial) dimension is utilized to resolve the collisions among tags. As an analogous notation to Si , let S1 be the area covered only by RA , S2 be the area covered by RA and RB , and S3 be the area covered only by RB . Since the tags are uniformly distributed in the total area covered by the two readers, the probability that a tag belongs to Si is qi = SSi , where S = S1 + S2 + S3 . The considered readers are not mobile and they have already run a certain initialization process during which they have estimated the overlapping of their range and have thereby estimated qi . Figure 8.10 depicts a representation of the two-dimensional arbitration space. The

RFID Systems

spatial position

224

1

RA

tokens in range of RB

RB q2

tokens in range of RA and RB

q1 tokens in range of RA

area S1

area S2

0

area S3

0

p1 p2 p3

1 token

Figure 8.10 Representation of the compound random process: the tags are randomly distributed in a two-dimensional space and their tokens are randomly distributed in interval [0, 1). For a given tag, the random token represents its x-coordinate and the random placement its y-coordinate.

abscissa represents the dimension of random tokens, while the ordinate represents the spacial position of a tag. If two tags are in the same spatial set Si , then the only way they can be differentiated in the arbitration process is through usage of random tokens. On the other hand, if tag τ1 ∈ Si , τ2 ∈ Sj and i = j , then these two tags do not need to use random tokens to be differentiated. To facilitate the presentation, let us assume that initially, the system of readers runs independent instances of for example the IECR protocol for each set separately. For example, the collision resolution starts by having the reader RA run the IECR protocol over the tags from the set S1 . The reader has stopped when an instance of the CBT algorithm is terminated after k1 tags have been resolved in the token interval [0, p1 ). Subsequent IECR instances are run for each of the Si , i = 2, 3, and an instance is stopped if after the termination of a CBT there are ki tags within the interval [0, pi ). Instead of making separate estimates nˆ i = pkii , the following approach can be used. Referring to Figure 8.10 and considering that the tags are uniformly distributed within the total coverage area of the two readers, the readers can share the information they have at observation time and one can obtain the following estimate for the total population of tags n = n1 + n2 + n3 : nˆ =

k1 + k2 + k3 p1 q1 + p2 q2 + p3 q3

(8.15)

Using this estimation of the total population, an estimate can be obtained for the cardinality of each tag set Si as follows: ˆ i nˆ ci = nq

(8.16)

This estimate (using the superscript c ) will be referred to as a cooperative estimate of ni . In order to assess its correctness, its variance can be estimated as follows:2 In the reference [19] the expression used for the variance Var[nˆ ci |n] is incorrect, this error is corrected by the expression in the present chapter.

2

Tree-Based Anti-Collision Protocols for RFID Tags

 Var[nˆ ci |n]

= nqi

225

 qi (1 − 2pi ) +1 p1 q1 + p2 q2 + p3 q3

(8.17)

An alternative would be to use non-cooperative estimation, in which the cardinality of each set Si is calculated separately: noi =

ki pi

(8.18)

and the variance of this estimate is determined as follows:   1 o Var[nˆ i |n] = nqi − qi pi

(8.19)

On can check that Var[nˆ ci |n] ≤ Var[nˆ oi |n], so that, on average, cooperative estimation provides more accurate estimates. It is important to point out that this is not the case, for example, if we fix n1 , n2 , n3 (instead of fixing only n) and if p2 = 1: then we clearly have nˆ o1 = n1 . Numerical illustration of the estimate variances is shown in Figure 8.11. The values of the resolved intervals at the moment of observation are p1 = 0.1, p2 = 0.25, p3 = 0.3. The distance d between the readers controls the amount of overlap between coverage areas, and thereby the values of the probabilities qi . Due to symmetry, q1 = q3 and q2 = 1 − 2q1 . Each variance is linearly proportional to n, the total number of tags, and this figure plots the variances normalized with n. When the distance between the centers is d = 0, then q2 = 1 and q1 = q3 = 0, hence the variance of nˆ 2 is maximal, while the variance of nˆ 1 is 0. When d = 2R, where R is the coverage radius, then q1 = q3 = 12 and q2 = 0. Figure 8.11 clearly shows that cooperative estimation significantly decreases estimate

Normalized variance of the estimate

5 4.5 4 3.5 Cooperative for n1

3

Non-Cooperative for n1

2.5

Cooperative for n2 Non-Cooperative for n2

2 1.5 1 0.5 0 0

0.5

1

1.5

2

Normalized distance d between the centers

Figure 8.11

Different estimates of normalized distance by the total number of tags n.

226

RFID Systems

variance of each set. As discussed before, such a correctness directly impacts the speed by which the tags within a certain subset are resolved. To summarize, in the described method of arbitration, a CBT instance is run for one of the sets Si at the time, while the estimation of the cardinality after each CBT instance is cooperative. From Figure 8.10 it can be seen that instead of enabled interval, during the arbitration for the set Si , there is an enabled area, rectangle of length p and height qi .

8.5.2 Further Remarks and Multi-Dimensional Arbitration In the previous example, the shape of a single enabled area has been restricted to a rectangle. In addition, we have assumed that the readers are unable to have an active role in determining the spatial division. If such a spatial control is possible, for example, by using multiple antennas and beamforming, then the enabled areas may have a more general shape. From the example shown in Figure 8.12, the multi-dimensional arbitration space can be utilized even with a single reader. In this example, the reader can control its transmission and create a beam that covers a sub-area of its total coverage area. Hence, the reader can spatially “address” smaller circles within its coverage area. The arbitration protocols for each subarea can be run in parallel (i.e. by interleaving the sessions in time) and the estimation can be made by aggregating the information from the different arbitration instances, as described in Section 8.5.1. In this particular case, it is suitable to use a three-dimensional arbitration space: two dimensions (x and y) are represented by the coverage area of the reader and the possibility to enable different spatial patterns within the coverage area, while the third dimension z is represented by the random tokens. Thus, if the possible spatial patterns are circles, then the enabled objects in the arbitration space are cylinders. Additional acceleration of arbitration protocols can be achieved if the reader sessions take advantage of the spatial reuse and simultaneously inquire some of the tag subsets. For the example in Figure 8.10, the probing of set S1 and S3 can occur simultaneously. This can be generalized to multiple (>2) readers, where the sessions in the non-overlapping

RA

overall coverage possible coverage region by beamforming

Figure 8.12 Possible spatial arbitration patterns for a system with a single reader and beamforming. The arbitration space is three-dimensional, one dimension being the random token and the other two dimensions are represented by the circle.

Tree-Based Anti-Collision Protocols for RFID Tags

227

sets are scheduled to occur simultaneously. In the figure, the scheduling of the sessions is done in a round-robin manner, which in general is not optimal. In principle, a group of simultaneously scheduled sets that covers a larger area should be scheduled more frequently. Optimization of the arbitration speed achieved by the tree-based anti-collision protocols is difficult even in the case of one-dimensional arbitration space [18]. These difficulties are aggravated in the case of multi-dimensional arbitration. For example, in the case of two-dimensional arbitration, this optimization has two components: (1) selection of the length of each enabled interval in Figure 8.10; and (2) decision on the schedule used to query the three sets Si . A way to approach this optimization is to use Markov Decision Processes [29]: after a CBT instance is terminated, decide the next area to be enabled by the arbitration protocol, so as to minimize the expected duration of the protocol, while taking into account the conditions for spatial reuse and simultaneous transmissions. It is interesting to note that, if the tag density in the coverage area of each reader is identical (or at least known), then the readers can cooperate even if their regions do not overlap. For the example in Figure 8.10, this is equivalent to setting q2 = 0. This observation allows us to define a different problem: if tags are distributed uniformly in a certain area, we need to find which deployment of the readers produces the most efficient coverage of the area in terms of minimized expected duration of the collision resolution time. Besides the described example of three-dimensional arbitration space for Figure 8.12, there are other ways to increase the dimension of the arbitration space. For example, in addition to the spatial dimension and the random token dimension, an additional feature of the RFID tags is taken into account when discriminating among the different tags. For example, tags may be attached to temperature sensors, so that the third dimension of the arbitration space is represented by the temperature that is read at a tag. The temperature range is mapped on the interval [0, 1) and the reading of each tag is a token distributed in this interval. This interval can be used as the third dimension for arbitration, in addition to the random token and the spatial position of tags. Now in the single step of the arbitration process, a 3-D region is enabled. Note that token distribution in the temperature dimension need not to be uniform, which should be reflected in determining the estimator of the tag population. Nevertheless, the task given to the reader network might require a partial resolution (e.g. estimate the average temperature or the temperature distribution with a sufficient accuracy) which should result in suitable selection of the sequence and shape of the enabled 3-D regions. In general, when there is an arbitration dimension across which the tags are not uniformly distributed, the width of the enabled intervals for that dimension should be adapted to the actual distribution of tags.

8.6 Conclusion The problem of tag collision is fundamental in RFID systems and there are a number of protocols that deal with this problem. In this chapter we have described an important class of protocols for arbitration among collided tags, termed tree protocols. These protocols are based on a recursive solution and a query structure that forms a binary tree. We have introduced the basic principles of tree protocols, as well as several variants of treebased algorithms. Furthermore, we have described an alternative framework for arbitration and representation of the tree-based algorithms from [18]. We have also described two variants of the tree protocol that have been subject to standardization by EPCglobal.

228

RFID Systems

These practical RFID protocols show that there are many choices and approaches to implement the principles of tree-based protocols in real-life systems. Considering that the RFID systems are prone to transmission errors, we have discussed the impact that the transmission errors have on the tree-based arbitration protocols. Finally, we have discussed how tree protocols can be re-engineered in order to be adapted to a scenario with multiple cooperative readers. For that purpose, we introduced a generalized class of tree protocols via the notion of arbitration space. Such a generalization of the arbitration space offers a novel and versatile framework for developing future anti-collision protocols for RFID systems. By adjusting the arbitration space, these algorithms can be tailored to different applications, for example, in RFID sensor networks.

Problems 1. Let there be n = 3 tags in the range of the RFID reader. Assume that there are no transmission errors and the ideal collision model can be applied. Using the framework with enabled intervals, find the average duration of an anti-collision protocol if: (a) A basic binary tree algorithm is used. (b) A modified binary tree (MBT) algorithm is used. (c) Consider the situation in which the reader knows in advance that there are n = 3 tags. How can we use this information to optimize tree protocol? Note that since the reader knows that there are n > 1 tags, it will not enable the interval [0, 1), as it will certainly result in a collision. Furthermore, assume that the algorithm is optimized in the following way: if the reader knows that in certain interval there are n tags, it enables 1/n-th of that interval. 2. The pseudocode in Figure 8.4 assumes that the way in which the tree is traversed is decided by the reader, since it explicitly sends the address of the next enabled node. Let us assume that the “intelligence” of the tree protocol is transferred to the tags and the reader only provides: (a) the signal to initiate tree traversal; and (b) feedback to the tags. Write a pseudocode that describes the operation at the tag for this case. 3. In some models for collision resolution it is assumed that the multiple-access receiver, that is, the RFID reader, cannot distinguish between an idle slot and a collision slot. How should the binary tree algorithm be modified in that case? 4. The framework with enabled intervals has been applied to describe/design tree algorithms. How would an ALOHA-based protocol be represented in the same framework? 5. Describe how the ideas of an Estimating Binary Tree (EBT) algorithm can be applied to a tree a protocol for Class 1 Generation 1 tags. 6. Using the error model in Figure 8.7, explain why having only errors, where single or idle are interpreted as collisions, does not produce missing tags.

References [1] Want, R. (2006) An introduction to RFID technology, IEEE Pervasive Computing, Jan. p. 25. [2] Buckley, J. (2006) From RFID to the Internet of Things: Pervasive networked systems, DG Information Society and Media Conference, Brussels.

Tree-Based Anti-Collision Protocols for RFID Tags

229

[3] Finkenzeller, K. (2003) RFID Handbook: Fundamentals and Applications in Contactless Smart Cards and Identification. Chichester: John Wiley & Sons. Ltd. [4] Juels, A. (2006) RFID security and privacy: A research survey, Journal of Selected Areas in Communication, vol. 24, no. 2, pp. 381– 394. [5] Engels, D. W. and Sarma, S. E. (2002) The reader collision problem, in Proc. IEEE International Conference on Systems, Man and Cybernetics. [6] Waldrop, J. Engels, D. W. and Sarma, S. E. (2003) Colorwave: A MAC for RFID reader networks, in Proc. IEEE Wireless Communications and Networking Conference (WCNC03). [7] Dong-Her, S. Po-Ling, S. David, Y. and Shi-Ming, H. (2006) Taxonomy and survey of RFID anti-collision protocols, Computer Communications, 29(11): 2150– 2166. [8] Hayes, J. F. (1978) An adaptive technique for local distribution, IEEE Trans. Commun., 26: 1178– 1186. [9] Capetanakis, J. I. (1979) Tree algorithms for packet broadcast channels, IEEE Trans. Inform. Theory, IT-25: 505–515. [10] Tsybakov, B. S. and Mikhailov, V. A. (1978) Free synchronous packet access in broadcast channel with feedback, Probl. Peredach. Inform., 14 (4): 32–59. [11] Hush D. R. and Wood, C. (1998) Analysis of tree algorithms for RFID arbitration, in Proc. IEEE International Symposium on Information Theory, Boston, USA. [12] Micic, A. Nayac, A. Simplot-Ryl, D. and Stojmenovi´c, I. (2005) A hybrid randomized protocol for RFID tag identification, in Proc. 1st IEEE Int. Workshop on Next Generation Wireless Networks (WoNGeN 05). [13] Bonuccelli, M. A. Lonetti, F. and Martelli, F. (2006) Tree slotted Aloha: a new protocol for tag identification in RFID networks, in Proceedings of the 2006 International Symposium on World of Wireless, Mobile and Multimedia Networks (WOWMOM). [14] Chiang, K. W. Hua, C. and Yum, T.-S. P. (2006) Prefix-randomized query-tree protocol for RFID systems, in Proc. IEEE Int. Conf. on Communications (ICC). [15] Myung, J. Lee, W. and Srivastava, J. (2006) Adaptive binary splitting for efficient RFID tag anti-collision, IEEE Commun. Lett., 10 (3): 144– 146 [16] Choi, J. H. Lee, D. and Lee, H. (2006) Bi-slotted tree based anti-collision protocols for fast tag identification in RFID systems, IEEE Commun. Lett., 10 (12): 861– 863 [17] Ryu, J. Lee, H. Seok, Y. Kwon, T. and Choi, Y. (2007) A hybrid query tree protocol for tag collision arbitration in RFID systems, in Proc. IEEE Int. Conf. on Communications (ICC). [18] Popovski, P. Fitzek, F. and Prasad, R. (2007) A class of algorithms for collision resolution with multiplicity estimation, Algorithmica, 49 (4): 286– 317. [19] Popovski, P. (2008) Tree protocols for RFID tags with generalized arbitration spaces, in Proc. 10th Symp. on Spread Spectrum Techniques and Applications (ISSSTA). [20] Cidon I. and Sidi, M. (1988) Conflict multiplicity estimation and batch resolution algorithms, IEEE Trans. Inform. Theory, IT-34: 101–110. [21] Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specification, IEEE Std. 802.11, Aug. (1999). [22] Berstekas D. and Gallager, R. (1992) Data Networks, 2nd edn. New Jersey: Prentice-Hall. [23] Massey, J. L. (1981) Collision-Resolution Algorithms and Random-Access Communications, ser. CISM Courses and Lectures. Berlin: Springer-Verlag, no. 265, pp. 73–137. [24] Greenberg P. F. A. G. and Ladner, R. E. (1987) Estimating the multiplicities of conflict to speed their resolution in multiple access channels, J. ACM , 34 (2): 289– 325. [25] Gallager, R. G. Conflict resolution in random access broadcast networks, in Proc. AFOSR Workshop Commun. Theory Appl ., Provincetown, MA, Sep. 1978 pp. 74– 76. [26] Dobkin, D. M. (2007) The RF in RFID. Oxford: Newnes. [27] ISO/IEC 18000: Information Technology Automatic Identification and Data Capture Techniques - Radio Frequency Identification for Item Management Air Interface, International Organization for Standardization. Std., 2003. [28] Jacobsen, R. Fyhn, K. Popovski, P. and Larsen, T. (2009) Reliable identification of RFID tags using multiple independent reader sessions, in Proc. IEEE RFID 2009 . [29] Puterman, M. L. (1994) Markov Decision Processes: Discrete Stochastic Dynamic Programming. New York: John Wiley & Sons, Ltd.

9 A Comparison of TTF and RTF UHF RFID Protocols Alwyn Hoffman, Johann Holm, and Henri-Jean Marais Northwest University, Potchefstroom, South Africa

9.1 Introduction The design of a communications protocol can be viewed as the challenge to find an optimal compromise between the required functionality of intended applications, current state-of-the-art capabilities of electronic technology, and constraints implied by regulatory issues, moderated by the creativity of the designer. Long-read range passive RFID represents a very interesting variation on this theme, as the capabilities of technology and the availability of bandwidth must be stretched to the ultimate in order to allow the candidate protocols to provide acceptable solutions to the envisaged applications, with RFID tag technology cost as an important constraint. Several texts have appeared in recent years describing the design of UHF RFID systems in general [5, 7]. Much has also been reported in the literature regarding proposed schemes to address the anti-collision problem that is commonly experienced with long-read range passive RFID systems ([8–11] and many others not cited in this text). This chapter will focus only on aspects related to the choice of a suitable protocol. While a brief overview will be provided of RFID protocols in general, the main focus is on the challenges to find a suitable protocol for long-read range passive UHF RFID. The motivation for focusing on UHF RFID is that both short-range magnetic induction-based passive RFID and active UHF RFID have long been established and are thus currently not facing the same challenges as passive UHF. As the working of existing ISO standardized protocols are covered elsewhere in this volume, the focus will furthermore be on candidate protocols that are not yet part of the ISO18000-6 standard for passive UHF RFID, and how they

RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

232

RFID Systems

compare with the most popular standardized protocol (ISO18000-6C or EPC Class 1 Generation 2). The chapter starts with an overview of the requirements for RFID protocols, to provide a framework within which candidate protocols can be evaluated and compared against the existing standards forming part of ISO18000-6. The different approaches that can be taken in the implementation of an RFID protocols are briefly covered, followed by an overview of the requirements for the effective operation of RFID, with the focus mainly on passive UHF. A detailed description is then provided of the operation of more than one stochastic protocol, including protocols that do not require modulated interrogation signals from the reader, and the potential benefits of such protocols are highlighted. The theoretical and practical performances of such protocols are then compared against the performance achieved by the ISO18000-6 protocols, more specifically the ISO18000-6C or EPC Class 1 Generation 2 protocol. Finally, conclusions are drawn based on the comparison between the different candidate protocols.

9.2 Requirements for RFID Protocols The requirements to be satisfied by RFID protocols are largely dependent upon the type of functionality and performance intended to be offered by a specific type of RFID. It therefore does not make sense to state one set of requirements for all types of RFID – a more productive approach would be to divide RFID into the major categories and state the requirements for the protocol based on the target applications for each category.

9.2.1 Categories of RFID Technology For the purpose of this discussion it is sufficient to divide RFID into three primary categories: • Short-range magnetic induction-based passive RFID, operating in the LF and HF bands. • Medium to long-range passive RFID operating in the UHF band. • Long-range active RFID operating in different UHF bands. Again, the main focus of this chapter is on passive UHF protocols, primarily because this is the technology that has attracted most attention over the past decade. The protocols used by the other two categories of RFID have been deployed in practice for a substantial period of time and seem to be largely stable technologies, with limited recent effort from industry to launch new technologies justifying new standards within those categories. 9.2.1.1 Passive LF and HF Initial deployments of passive RFID operating in the LF band mostly involved short-range applications that require only one tag to be read at one point in time. Based on this point of departure, a number of standards have been adopted by ISO over the years for the LF band, including ISO 11784, ISO 11785 and ISO 14223-1. These protocols were, in general, not designed to provide any significant anti-collision capability, as the intended

Comparison of TTF and RTF UHF RFID Protocols

233

applications at the time generally assumed that only one tag would be visible to the reader at one point in time. The low data rates that can be supported by 125–134 kHz band carriers make LF fundamentally unsuitable for implementing fast anti-collision protocols involving substantial collections of tags. This limiting factor, combined with the lack of read range beyond 0.5–0.6 m, largely excludes LF technology from the debate about more effective anti-collision protocols. The situation is somewhat different for HF (typically around 13.56 MHz). While the read range that can be achieved with HF, using readers that are moderate in terms of both complexity and cost, is similar to that of LF, the much higher carrier frequency (13.56 MHz) does in principle allow the implementation of very fast and efficient anticollision protocols. This has led to the adoption of three major ISO standards for this band, namely ISO14443, ISO15693 and ISO/IEC 18000 Part 3 Mode 2. The latter is primarily implemented through the product range of [19], and claims very fast anti-collision for short-range applications such as the counting and identification of gambling chips. The combination of this powerful anti-collision protocol with technological implementations that allow tags to be read when stacked directly on top of each other has resulted in some very successful commercial deployments, the most prominent case in point being the gambling industry. The issue to be considered regarding the relevance of anti-collision protocols developed for HF beyond short-range applications, is whether this technology does have any significant scope for applications that require medium to long read ranges (typically supply chain applications such as bulk stock control). HF has been selected by prominent global business entities, such as Benetton, for pilot implementation in the garments industry, as well as by a number of global pharmaceutical companies for pilots that focus on product authentication. Neither of these initiatives has yet resulted in major commercial deployments, partly due to privacy issues (the Benetton garments tagging project was cancelled several years ago) and partly due to cost pressures in the case of pharmaceuticals [12]. Vendors such as Magellan have made claims of achieving read ranges that allow the implementation of portals 8 feet wide, but it would appear that such implementations are very costly (as effective screening has to be constructed to screen off potential sources of noise). The limiting factor in this case has nothing to do with the protocol as such, but results from the limited strength of the energizing magnetic field that can be set up by HF readers and that comply with existing spectrum regulations, combined with the fact that the HF band suffers from a very wide variety of unintentional radiators. These electro-magnetic culprits cause high noise levels that compete with relatively weak tag signals at the reader’s receiver. These unintentional radiators include, for example, most kinds of induction motors, PC power suppliers, noise radiated by power distribution lines, in addition to the noise generated by the energizing signals from the reader itself. In this respect, HF is in a much worse position compared to UHF, and therefore cannot be realistically expected to be a serious future contender for RFID applications requiring read ranges beyond 1 meter. An interesting technology to take note of in the LF/HF band is the so-called dual frequency (DF) technology developed by IPICO Inc., with EM-Microelectronics acting as manufacturer of transponder chips to support this technology [13]. Dual frequency technology is based on the known technique of using different frequencies for the energizing of tags and for tag-to-reader communications (e.g. [21]), In the case of this specific dual frequency implementation (that will be referred to as DF for the purpose of this

234

RFID Systems

discussion), the energizing signal is generated at 125 kHz (or a frequency close to 125 kHz), while the tags respond at a frequency in the HF band (in this case, 6.8 MHz). The carrier frequency for tag communications, which is generated in the transponder chip as a harmonic of the energizing signal, is comparable to the carrier frequency used by 13.56 MHz systems, and thus has the ability to achieve similar anti-collision capabilities and signal-to-noise performance. The fact that the energizing signal is located at 125 kHz, however, allows the powering up of tags at distances roughly double the ranges that can be achieved using standard HF systems. As DF readers do not directly contribute to the noise levels visible to their own receivers, DF systems will in principle suffer somewhat less from the effects of noise, allowing them to practically achieve the longer read ranges allowed by their longer energized ranges (compared to HF). Given the fact that the operation of DF systems is based in total on magnetic induction, it is for all practical purposes not affected by the presence of liquids or humidity (unlike UHF systems), and thus creates opportunities for this technology to be deployed in supply chain applications that require the covering of standard warehouse doors. HF systems that are required to read through liquids, or highly humid media, do not perform as well as DF systems that will be able to energize and read at longer ranges, while UHF systems are affected even more adversely by such conditions. A prominent example is the fresh produce industry, where product is typically transported in plastic bins that are stacked in multiple layers. This application represents a very challenging scenario for both HF and UHF, while it falls relatively easily within the capabilities of DF. The air-interface protocol currently used by the above DF implementation is not part of existing ISO standards. It is based on the same IP-X tag-talk-first approach as used by IPICO Inc. in its range of UHF products [13]. This protocol will be discussed in detail in the sections that focus on alternative UHF protocols, and will therefore not be covered in any depth in the section. It should be mentioned that the transponder bit rate for DF tags is 128 kbaud, compared to a bit rate of 256 kbaud as implemented in the latest version of UHF IP-X products. The reason for this is the fact that the carrier frequency used for DF products is 6.8 MHz, which limits the maximum baud rate that can be supported. The results displayed in the section on UHF IP-X should therefore be translated by a factor of two when the anti-collision capabilities of the DF implementation of the IP-X protocol are evaluated, for example, when the number of tags read per second or the time required to detect a specific proportion of all tags is determined. 9.2.1.2 Passive UHF Passive RFID is viewed as the primary candidate technology to fulfill the aim of ubiquitous tracking of items, thereby realizing the dream of “the Internet of things” [14]. Compared to magnetic induction-based versions of passive RFID, UHF comes much closer to fulfilling the functional requirements that will make the widespread use of RFID a reality. UHF achieves much longer read ranges compared to LF or HF and the physical construction of UHF transponders also allows the production of low cost tags. Recent efforts show that the ideal of the so-called “5 cent tag” is already very close to reality [15, 16, 17]. This does not, however, mean that the effort to produce UHF RFID technology (that fulfills these requirements in practice) is without challenges. A great deal of resources has been spent over the past 10 years to develop chip, transponder and reader technologies

Comparison of TTF and RTF UHF RFID Protocols

235

to the level where both cost and functionality expectations can be met, hence enabling the widespread deployment of a variety of RFID-based tracking and identification systems. Among these efforts is counted the process to drive RFID standards to a point where at least one widely supported standard is now in place, namely ISO18000-6C, also referred to as EPC Gen 2. While it has been demonstrated that this standard comes close to satisfying most requirements for UHF RFID, there is still by no means complete consensus regarding the suitability of this standard to successfully support the deployment of important applications in all regulatory jurisdictions. The primary debate in defining a global standard for passive UHF RFID has revolved around issues of the most suitable approach to find a compromise between RFID system performance and regulatory restrictions. These regulatory restrictions are defined in terms of the allowed output power and amount of frequency spectrum allocated to passive UHF RFID in some very prominent jurisdictions, including the European Union and China. Earlier versions of the EPC protocol were defined based mainly on the regulatory situation in the USA, where UHF RFID can be used in 52 bands, each of 500 kHz bandwidth, between 902 and 928 MHz. The situation in the EU is very different, allowing only 10 channels of 200 kHz each. The successful deployment of systems within this regulatory environment hence presents a far greater challenge, specifically in terms of restricting reader modulation spectra from causing interference in adjacent channels, as well as allocating channels to readers in such a way that large numbers of readers operating in close proximity do not interfere with each other. These issues will be discussed in more detail in Section 9.3. 9.2.1.3 Active RFID The technical challenges posed by the air-interface protocols, used by active RFID systems, are very different from those applicable to passive RFID. First, active transponders that have their own power sources, are characterized by much more complex functionality compared to passive transponders, allowing the use of protocols that also require more processing power (e.g. CDMA) than what is viable for passive RFID. Second, active readers do not need to transmit as powerful signals as required for passive readers, since the reader signals are only used for communication, while passive reader signals must in the first place provide power to transponders. Given the much longer read range of active RFID systems, transponders will normally have more time to respond to reader signals compared to passive transponders, making the time required to successfully identify transponders less critical. More emphasis is placed on techniques to conserve transponder power in order to maximize battery lifetimes, which is not an issue with passive RFID. Given the relatively low power levels emitted by both active readers and transponders, the technical challenges to provide active RFID systems with sufficient bandwidth to support the deployment of large systems is quite different from those applicable to passive UHF RFID. While the long read ranges of active RFID systems do lead to the problem of crossreads between different readers (i.e. two readers picking up the same transponder), the average amplitude of reader and transponder signals are of the same order of magnitude. The presence of another active RFID reader in the same frequency band and transmitting at the same time will therefore not prevent an active reader from successfully detecting the desired transponder signals. This is in contrast to passive UHF RFID, where time

236

RFID Systems

and frequency overlap of reader and transponder signals may cause severe interference problems, due to the much smaller amplitudes of passive transponder signals. Active RFID protocols are mainly covered by ISO standard ISO 18000-7, while several application-specific standards also exist (e.g. ISO 18185 for electronic shipping container seals based on active RFID transponders). For the reasons discussed above, the detailed working of these protocols will not be covered in this chapter.

9.2.2 Requirements for Passive UHF RFID As recent efforts to arrive at a global standard for RFID have mostly focused on passive UHF, the remainder of this chapter will be devoted to this topic. Before the different approaches to arrive at an efficient UHF protocol are investigated, it will be useful to study the requirements to be satisfied by passive UHF RFID, as derived from some of the most promising application domains. In this respect two extreme scenarios will be considered: • Supply chain management, where the major challenges to RFID involve the identification of large numbers of tagged items, moving at low to moderate speeds, and including potentially large numbers of readers operating in close proximity. • Transportation, where the challenges revolve around the reading of tags moving at high speeds relative to readers, in most cases involving only a small number of tags visible to a reader at one time, and with small to moderately sized collections of readers operating in one location. 9.2.2.1 Supply Chain Applications For supply chain applications, the primary performance criterium that is directly impacted by the air-interface protocol can be viewed as the overall throughput achieved by the total RFID system operating in one location. The definition of “one location” can be stated as an area of such geographic extent that readers can potentially impact on the performance of other readers operating within that same area. In the process of calculating effective throughput, it will be necessary to estimate the total number of tags that can be identified successfully by a number of independently operating sets of readers, all within a specified time period. A typical example will be readers deployed in different warehouses that could be located around a large port, assuming that it will not be possible to time-synchronize the operation of sets of readers operating on different premises. A typical reader will be monitoring a warehouse door and could be observing relatively large collections of tags (up to 1,000 tags, but more typically 50 to 100 tags per batch) that are moving past readers within a few seconds and at random time intervals. The challenge presented to the air-interface protocol in this case can be summarized as follows: • The anti-collision capabilities must be sufficiently powerful to allow the maximum number of tags that may be encountered in one group to be identified in the available period of time (worst case up to 1,000 tags to be read within 1–2 seconds). • It must be possible for large numbers of readers to operate independently of each other, in order to allow the warehouse operations to continue without disruptions while tags

Comparison of TTF and RTF UHF RFID Protocols

237

are being identified. Up to 200 readers could be found per warehouse, and up to 50 warehouses located in an area small enough for different readers located at the extremes of this area to potentially interfere with each other if no precaution is taken, that is, up to approximately 1,000 potentially interfering readers. It should be noted that the above criteria are not independent of each other: increasing the rate at which readers can identify tags implies an increase in the required data rates for reader–tag communications, which implies more bandwidth used by each reader, thus leaving fewer communication channels for the remaining readers, given that the overall regulated bandwidth will be fixed. The challenge is therefore to find a compromise between an acceptable rate of identifying tags in large tag populations, and maximizing the number of readers that can operate independently at one site. Other factors, such as the directionality of reader antennas and the degree to which reader signals can be isolated from adjacent channels, as well as the average duty cycle of a reader, will all impact upon the effective throughput of the aggregated system. 9.2.2.2 Transport Applications In the case of transportation systems, the primary challenge is the ability to read tags moving at high speeds, implying a short period of time that any tag will be visible to a particular reader. Road safety regulations typically require readers to be mounted between 5 and 6 meters above the road surface. The radiation patterns of patch antennas – that are normally used with long-range passive UHF readers – result in reader fields with an effective width of around 4 m at the typical height at which tags will move through the beam. Currently, RFID will mostly be used for law enforcement within this application domain, and it is essential for the RFID system to be able to identify tags that are moving at speeds that are well in excess of legal speed limits. A typical requirement will be to detect tags at speeds of up to at least 220 km/h (more or less 140 mph). In addition to reading a tag’s unique ID, it may be required to also extract additional information from such tags at speed. This is done in order to allow a law enforcement officer to verify the legal status of a vehicle without first performing an online verification on a central database. A typical need could therefore be to read at least 256 bits of information from any tag at maximum speed. Given the physical nature of dense traffic environments, it can be assumed that less than 10 tags will be visible to a reader at any point in time. However, when tagged cargo is verified at high speed, the number of tags in a reader’s field may increase significantly – this is currently not a viable application of UHF RFID when regulatory limitations are taken into consideration. The nature of typical transportation read stations will imply the presence of only a moderate number of readers in one location, as different independently operating read stations will typically be located several hundred meters if not several kilometers apart. At the same time it will, however, be possible for a highway read station to be located within close proximity to several warehouse management systems, for example, in the case of a highway running past a large port. It will therefore be necessary to determine the minimum required spatial separation between different read stations to prevent readerto-reader interference. Once again, the need to read tags at high speeds will imply the use of high communication data rates, which will increase the bandwidth requirements

238

RFID Systems

per reader, thus leaving fewer bands for the operation of other read stations. The optimal solution will once again have to be a compromise between these conflicting requirements. The question may be asked why the above challenges to the deployment of passive RFID in transport applications are not avoided altogether by rather reverting to the use of active or even semi-passive technologies for transport applications. The answer to this lies mainly in the cost issue: many transport applications can only deliver maximum benefits if all vehicles in a specific jurisdictions are obliged to carry the same type of RFID transponder. A good example in this respect is the use of RFID as part of vehicle licensing. As vehicle licenses are frequently renewed, there is much cost pressure on the choice of technology, and passive RFID obviously has a significant cost benefit over both active and semi-active RFID. It is therefore logically expected, in spite of technical challenges, that passive RFID will become the technology of choice for mass deployments of automated identification in the transport market.

9.3 Different Approaches Used in UHF Protocols The different air-interface protocols that have been developed over the years for UHF RFID, are mostly derivatives of protocols that were developed for use in general computer networking applications. The adaptations that are required to arrive at an efficient RFID protocol are necessitated by those factors that characterize passive RFID reader–tag communications: • Tags must be assumed to be simple devices incapable of performing the type of data processing required by most computer networking protocols. • As readers have to power tags in addition to receiving and transmitting communications messages, reader signals are in general orders of magnitude stronger than tag signals. The protocol therefore has to ensure that reader and tag communications are separated in either the time domain or the frequency domain, or both, to allow readers to successfully detect tag signals against the background of potentially interfering signals from other readers operating in the same geographic location. • Tags will communicate only with readers, not with other tags, so instead of peer-to-peer type of communications, the nature of RFID protocols will tend to be master–slave, with one reader (the master) communicating with many tags (the slaves). • The nature of communication messages are in general not ad hoc communications, as in the case of computers on a network exchanging information with one another as and when required, but rather short bursts of communications from tags to reader to inform the reader about their presence. This normally happens in a short time period as tagged items move past a reader, with more complex two-way exchange of information mostly only happening when tags are initially programmed by a reader, in which case the communication will mostly be one-on-one at close range. From the above discussion, it is clear that the air-interface protocol must primarily satisfy two important criteria: 1. When a large number of tags moves past a reader in a short period of time, the anticollision capabilities of the protocol must allow all tags to successfully communicate

Comparison of TTF and RTF UHF RFID Protocols

239

their presence to the reader at least once. Alternatively the read scenario may involve a small number of tags but moving at high speeds, thus leaving the tags very little time to communicate with the reader. 2. When a number of readers are deployed at the same location (i.e. sufficiently close to each other so that signals from other readers can potentially interfere with the tag signals that a particular reader needs to detect), the protocol must allow all readers to successfully communicate with the tags that they are supposed to detect, even in the presence of interfering signals from other readers communicating with their own sets of tags. In general (at least in the case of reader-talk-first protocols), these two primary criteria tend to pose conflicting requirements on the system as a whole: the successful detection of a large number of tags in a short period of time will benefit from a reader that sends out very fast communication signals to tags; such high speed reader communications, however, directly result in a wider reader transmission spectrum, increasing the possibility of interference between different readers, which is in conflict with the second criterion of limiting potential reader interference. The paragraphs below discuss different approaches that have been taken to arrive at a satisfactory solution to the above problem.

9.3.1 Deterministic versus Stochastic Most communication protocols can be categorized as being either deterministic or stochastic. Deterministic protocols follow an exact and repeatable algorithm, which can be guaranteed to result in the successful transmission of a message within a predetermined time interval, as long as the physical communication link is working. Stochastic protocols are statistical in nature, and there is a non-zero chance that any particular effort to transmit a message may not be successful or may be delayed until a later stage, with the exact time delay also being stochastic in nature. While the deterministic approach may appear to be the preferred one, in practice, superior performance is often obtained using the stochastic approach when all practical limitations and restrictions are taken into consideration. This can be seen from the definition of protocols that have successfully passed a rigid standards process. For this reason, commonly used protocols in the computer networking domain are mostly stochastic in nature (e.g. the Ethernet protocol), as the stochastic nature provides it with the flexibility to optimally utilize available resources (in this case, available bandwidth) without prior information about the number of network participants or the nature of the messages that will be transmitted. The drawback of deterministic protocols, in general, is the fact that a rigid procedure must be followed that will also cater for worst-case scenarios, even though such scenarios will seldom appear. As a result, deterministic protocols have not survived the same rigid standards processes that stochastic protocols have passed. Good performance (in the case of RFID, the fast identification of tags by readers) can thus only be guaranteed through brute force, that is, high speeds of communications. Some early stage passive UHF protocols that followed the brute force approach, like the EPC Class 0 protocol as implemented in the products of Matrics during the early years of UHF RFID, required high reader data rates. These high rates resulted in readers that transmitted high power modulated signals outside of the regulated band of the carrier, and hence in systems

240

RFID Systems

that were not compliant with regulations that restrict reader communications to specific regulated bands. More recent protocols therefore did not persist with this approach. Stochastic protocols, on the other hand, follow a more flexible approach that makes realistic assumptions in respect of the likely scenarios to be encountered in practice. These protocols are designed to effectively adapt to each practical scenario, either by changing the rate of tag communications (in the case of tag-talk-first protocols) or by adapting the queries sent out by the reader (in the case of reader-talk-first protocols). In this way, close to optimal performance can be achieved for a wide range of tag identification scenarios without having to rely on overly fast reader communications, as is the case with, for example, the dense reader mode of the EPC Class 1 Generation 2 or ISO18000-6C protocol. As an example of the benefits offered by stochastic protocols, the stochastic approach is more bandwidth efficient, as the only modulation signals generated by this protocol come from the RFID tags and not the reader (the reader emits a fixed carrier frequency used to power up tags). The signals backscattered from tags would normally fall below the regulated power levels. This approach offers the potential of providing good performance with limited available spectrum (as is the case in several important regulatory jurisdictions). More will be said about the I-PX protocol in the next section.

9.3.2 RTF versus TTF The distinction between reader-talks-first (or RTF) and tag-talks-first (or TTF) protocols is even more fundamental than the distinction between deterministic and stochastic protocols, as discussed in the previous section. The arguments on which RTF protocols are based, can be summarized as follows: • The objective to detect large populations of tags within acceptable time periods (e.g. when reading items leaving a warehouse) is only achievable if the reader is allowed to interrogate the tags, using in most cases a combination of deterministic and stochastic principles, rather than leave it to the tags to decide when to communicate with the reader (which implies the use of some stochastic algorithm to prevent tag collisions). In a sense, a “communication session” is opened between tags and readers until all tag information has been read. • It is, at least in principle, possible to deploy large RFID systems that involve large numbers of readers that operate in relatively close proximity, and to restrict the interference between such readers within all important regulatory jurisdictions. This can be done when each reader interrogates its own set of tags independently from other sets of readers and tags. • TTF protocols suffer from too many limitations in terms of lack of flexibility as readers do not have the option to control the way in which tag detections occur but have to rely on the behavior of tags with very limited built-in intelligence. This results in systems that are not sufficiently intelligent to support fast universal adoption. TTF protocols, on the other hand, depart from the following set of arguments: • The bandwidth available to UHF RFID readers in some regulatory jurisdictions is limited compared to the number of readers that must operate independently and within

Comparison of TTF and RTF UHF RFID Protocols

241

“hearing range” of each other. This effectively makes it impossible to deploy large numbers of readers in close proximity without either running out of bandwidth or severely restricting the duty cycle of individual readers (e.g. by using some kind of listen-before-talk measure). The solution to this dilemma is to not require readers to interrogate tags in the normal course of operation and to leave it to the tags to make their presence known to readers through some stochastic algorithm that is programmed into the tags. • The time that a tag will be visible to a reader may in some cases be very short (e.g. in the case of tagging vehicle on a highway), and the field through which a tag moves is often very non-uniform [18] due to the inherent nature of UHF multipath propagation. Faster effective identification of tags will be possible through a stochastic TTF protocol compared to an RTF protocol that operates within regulatory limitations, as RTF protocols tend to be less efficient in respect of the time constraints (more specifically when also faced with limited available bandwidth). • The implementation of TTF protocols, from a protocol point of view, is less complicated compared to RTF protocols, which results in less complex and potentially cheaper transponder chips, and therefore more affordable systems overall. Both of these two sets of competing arguments contain at least some element of truth, and it is easy to demonstrate that both approaches can offer superior performance for specific applications. It is also interesting to note that the weight of some of the arguments supporting the two opposing points of view is closely related to the current state-of-theart of integrated micro-electronic and RF technology. As technology has progressed over the past decade, the disadvantages posed by the additional complexity of RTF systems have become less of a challenge, and in some cases even a moot point. The verdict is, however, still out regarding the all-important question: “Which of these two approaches will eventually prove to be good enough to support the deployment of successful largescale RFID systems (by satisfying the two primary criteria of the marketplace: low cost and reliable operation)?”

9.4 Description of Stochastic TTF Protocols TTF anti-collision protocols must rely on the stochastic behavior of transponders to avoid collisions in transponder communications, as the reader does not directly control when transponders will transmit their IDs. Such stochastic protocols theoretically suffer from the disadvantage that there is no upper limit to the period of time that it may take for a tag to be detected, as each stochastic effort by the tag to communicate may end in a collision (although this is unlikely), without either reader or tag being aware of this situation. Figure 9.1 describes the behavior of a typical free-running TTF protocol using stochastic tag delay times. Several alternative variations on the standard TTF free-running protocol have been patented to relieve the above problem, and some have been practically implemented over the past 10 years. Two of these will be described in more detail, namely Supertag [2], which represents a collection of alternative approaches to TTF operation, and IP-X [3], which is actually a special case of the Supertag protocol.

242

RFID Systems

Tag 1 Tag 2 Tag 3 Reader Collision between transmissions

No collision between transmissions

Figure 9.1 Tag transmissions operating under a free-running TTF protocol [3].

9.4.1 Supertag The earliest implementation of a TTF-type UHF RFID protocol was the so-called Supertag protocol as patented by Atkins, Marais and Smit in [2], which is an example of an unslotted Aloha protocol. The Supertag patent provides a number of alternative and complementary approaches to achieve optimal performance by using the underlying approach that tags are not initially prompted by an interrogator, but start transmitting their IDs in pseudo-random fashion as soon as they are energized. Supertag allows for both free-running versions of the protocol (i.e. tags behave totally independently from reader instructions) or versions where the reader may instruct tags to change their behavior once tag communications have been detected. The common denominator between the different approaches that are proposed in the Supertag patent is the fact that the tags are prompted to start communicating by the energizing signal of a reader, which can be a purely continuous wave (excluding any modulation), and that the tags then transmit their IDs in pseudo-random fashion. Each tag is programmed with an initial maximum waiting interval between consecutive transmissions of its own ID. Upon wake-up a tag will generate a pseudo-random number and use this number to calculate a pseudo-random waiting interval, which will be a fraction of the above maximum waiting interval. After waiting for this pseudo-random period, the tag ID is transmitted and the process of generating another pseudo-random interval is then repeated successively. This system hence relies on the fact that pseudo-random time delays are generated on each tag, that allows all tags to eventually be identified by the reader based on the time diversity achieved through the stochastic nature of tag transmissions. It is accepted that, depending on the number of tags present and the duty cycle of each tag transmission, a specific proportion of tag communications will end up in collisions. The options allowed for in the Supertag protocol are differentiated by (1) the way in which the tags respond once they have started transmitting their IDs, as well as by (2) the way in which the reader is allowed to control the behavior of tags once at least some of them have been successfully identified. The first possibility involves the reader retransmitting the tag IDs that have been successfully detected. As soon as a tag receives its own ID from the reader, it realizes that it has been detected and stops further transmissions of its own ID until it shuts down (due to loss of energy) and wakes up again (typically upon leaving and then re-entering a reader beam or sometimes moving through a null in the electro-magnetic field). The fast version of Supertag involves the reader sending out a mute signal to all transponders once it has

Comparison of TTF and RTF UHF RFID Protocols

243

started to detect a tag transmission. The transmitting tag will continue its transmission while all other tags will suspend their waiting period countdowns for 128 clock cycles to allow the current tag transmission to be detected successfully before any other tag will start transmitting. Another option, patented as part of the Supertag protocol, is the fact that each tag may have to alter its maximum waiting time until the next ID transmission takes place to allow the adaptation of the protocol to specific circumstances in order to optimize performance. This includes the possibility of dynamically altering the waiting time upon reception of an instruction from a reader. If a reader detects that there are initially too many tag collisions (which will be based on the number of failed tag detections due to tag signal overlap in time), it may send out a signal to tags requesting them to increase the maximum waiting period in order to reduce congestion, as depicted in Figure 9.2 [2]. In this example, initial signals from tags 1, 2 and 3 clash and therefore cannot be identified. The reader, sensing that the protocol is over-congested, sends out a “slow down” instruction, resulting in tags doubling the maximum waiting period which increases the chance of successful detection. The switch-off variation of the Supertag protocol allows tags to be switched off by the reader once the respective tag has been successfully detected. Fast switch-off involves a combination of the fast protocol described above and the switch-off approach. While both of these contribute to a significant reduction in collisions and average faster detection times, it also increases the error rate. The primary problem is the issue of when to allow a switched-off tag to start transmitting again [1]. If the switched off state is terminated at power-on reset, the tag may start transmitting again after moving through a null in the reader field, in the process defeating the objective of reduction in unnecessary tag transmissions. An alternative approach is for tags to be woken up by a reader signal, but due to the unreliability of reader-to-tag communications (following from to the normally non-uniform nature of the reader field), there is no guarantee that a switched-off tag will

Overcrowded tags

Less congestion

Tag 1 Tag 2 Tag 3 Tag 4 Tag 5 Reader

Slow down instruction Collision between transmissions

Figure 9.2

Acknowledgements

No collision between transmissions

Description of dynamic changing of the waiting period under the Supertag protocol [1].

244

RFID Systems

Tag 1 Tag 2 Tag 3 Tag 4 Tag 5 Reader

Switch off instructions Collision between transmissions

Figure 9.3

No collision between transmissions

Switch-off Supertag protocol sequence [1].

in fact wake up when instructed by a reader. This will result in an unacceptable proportion of tags not being detected at some read station subsequent to being switched off at an earlier read station. Should frequent wake-up calls by the reader be employed to avoid this situation, the TTF protocol will effectively start behaving like an RTF protocol, losing the benefits of operating in TTF mode while not being as effective in other respects as true RTF protocols. Figure 9.3 describes the switch-off mode for a TTF protocol. Yet another variation provided for by the Supertag patent is for tags to automatically change the waiting time, for example, by doubling the initial maximum allowed waiting time. This may involve the tags incrementing or decrementing the waiting time successively in a number of stages. One option would be to initially transmit with a shorter maximum waiting time, and to subsequently increase the maximum waiting time by repeatedly doubling the period, based on the assumption that at least one of the initial ID transmissions has been successfully detected by the reader. This will lead to a very simple implementation of the TTF protocol, as the reader is not required to instruct transponder to change their behavior, thus retaining all of the inherent benefits of a TTF protocol. The benefits associated with this option are discussed in more detail in the next section.

9.4.2 IP-X The IP-X protocol, as implemented in a series of transponder chips produced by EMMicroelectronics, can be viewed as a special case of the Supertag protocol as it also implements a free-running unslotted Aloha protocol. IP-X transponders wake up in socalled free-running TTF mode (actually TTO or tags-talk-only, as no communication from the reader is required to read tags), and will start to transmit their IDs at a pseudo-random rate without waiting for any signal from a reader. The pseudo-random delays between subsequent tag ID transmissions are achieved as follows: The transponder ICs are programmed with a maximum delay period. Upon wake-up, the transponder will generate a pseudo-random number between 0 and 1 that is used to determine a pseudo-random fraction of the maximum delay period, and will then transmit its ID after waiting for this

Comparison of TTF and RTF UHF RFID Protocols

245

pseudo-random time period. In order to initially speed up the rate at which readers will detect such tags, a so-called acceleration version of the Supertag protocol is used. This involves the transponder initially transmitting with a maximum delay period that is an 8th of the eventual maximum delay time for which the chip was programmed. After every second transmission the maximum delay period is doubled until the programmed maximum delay period is reached, after which transponders keep on transmitting at this maximum delay interval. The benefit of this approach is that transponders get the opportunity of being detected very soon after waking up (when the tag is transmitting in accelerated mode), while tags that have already transmitted their IDs slow down in order to reduce the chance of collisions with tags that woke up more recently. The non-uniform nature of the reader field, combined with the fact that spatial diversity will ensure that all tags do not wake up at the same time, assists the IP-X protocol in avoiding collisions during the start-up phase when tags are transmitting in accelerated mode. This very simplistic anti-collision protocol (which is a special case of one of the simpler options provided for in the Supertag protocol, avoiding instructions from the reader to change the rate of tags transmissions) provides surprisingly good performance when applied either to fast-moving tags or to tag populations of up to 200 tags at a time. In the series of chips developed by IPICO Inc. in cooperation with EM-Microelectronics that support this protocol, provision has been made for four different maximum delay intervals (measured in terms of the number of bits that can fit into the maximum interval): 1 kbits, 4 kbits, 16 kbits and 64 kbits (Table 9.1). This allows transponders to be implemented with different transmission characteristics suited to different applications. The fast variation (1 kbit maximum delay), combined with the acceleration protocol, allows the detection of tags traveling at speeds of theoretically up to 600 km/h. The other extreme (64 kbits maximum delay) can accommodate large numbers of tags given the small tag duty cycle – theoretically up to 800 tags can be allowed before the protocol will saturate, with tags travelling at 5 m/s [22, 23]. Another benefit offered by the TTF approach is the fact that user programmable memory fields, in addition to the tag ID, can be read without any reader communications. Assuming that the tag memory has already been programmed (this would typically happen at the point where tags are issued and associated with specific tagged objects), the tag can be configured to transmit not only its ID in TTO mode, but also additional user-defined memory fields. While this will increase the tag duty cycle – requiring either a reduction in the number of tags or an increase in the maximum delay time – it still offers the benefit

Table 9.1 Protocol version V1 V2 V3 V4 V5

Different implementations of the IP-X protocol. Baud rate

Maximum interval

64 k 64 k 256 k 256 k 256 k

4k 16 k 4k 16 k 64 k

 EM Microelectronic-Marin SA.

246

RFID Systems

Power-on reset

Boot

Wait random time

Transmit ID

No

Reader command?

Yes No

Wait command time

Correct ID?

Yes

Execute command

Figure 9.4 iP-X R/W protocol flow diagram. Reproduced from Van Eeden, H.L., “Passive UHF RFID systems,” Ph.D. thesis, Northwest University, Potchefstroom, South Africa, 2004.

of operating readers in a mode where only energizing signals have to be transmitted. The fundamental benefit of using very little spectrum for reader operation is therefore retained. Alternatively, the reader has the option, after successfully detecting a specific tag, to interrupt the anti-collision sequence of all tags through a mute command, and then to communicate with the specific singulated tag by using the last 48 bits of the tag ID as identifier [1], as displayed in Figure 9.4. While this approach provides the user with more flexibility in terms of for example, the specific area of tag memory to address, it also results in the loss of the benefits of communicating in TTO mode as the reader will have to transmit instructions and, as a result, will occupy similar frequency spectrum as is the case for RTF protocols.

9.4.3 TOTAL The TOTAL protocol (“tag only talks after listening”) was proposed to ISO as an addition to the existing set of ISO standards for UHF RFID (i.e. ISO18000-6A, B and C). TOTAL involves the option of adding a TTF protocol, based on the IP-X protocol, to ISO18000-6

Comparison of TTF and RTF UHF RFID Protocols

247

compliant readers in addition to one or more of the other ISO18000-6 protocols. The specific motivation for adding the TOTAL option is to make provision for readers that do not generate any reader modulation, and that therefore utilize very little spectrum as such readers only transmit a continuous wave energizing signal when listening for tags. As TOTAL has to be compliant with the operation of other protocols (such as EPC Gen 2) that, at this stage, are already part of the ISO18000-6 standard, it is not possible to simply operate in the same way as the standard IP-X protocol. The reason for this is the fact that an IP-X tag that is present within a population of RTF tags, and that is energized by an RTF reader, will transmit its ID in stochastic fashion while the RTF reader tries to interrogate the RTF tags. This will result in significant disruption of the operation of the RTF protocol. Practical tests have proven the degree of such disruption to be unacceptable. It was therefore necessary to add an additional aspect to the behavior of standard TTF tags: when energized, the TTF tags must first listen for interrogations from an RTF reader. If such reader communication is detected, the TTF tag must either respond in the required RTF fashion (if it was designed to be a dual-protocol tag) or else it must keep quiet for a prescribed period of time before it can again listen for reader communications. If no reader communication is detected, it can commence to transmit its ID in the same fashion as it would have for a standard TTF tag. The critical aspect of the TOTAL protocol is obviously the required waiting time for the tag while it is verifying whether an RTF reader is present. Should this time period be too long, the TOTAL protocol will lose one of the primary benefits of a TTF protocol, that is, the ability for a tag to be detected within a very short time period after the tag has been energized. This is a requirement to allow the tag to be read reliably when traveling at very high speeds. If, however, this waiting period is too short, the risk is that an RTF interrogation process may in fact have been ongoing while the TOTAL tag was listening, resulting in disruption of the RTF process as described above. The main features of the TOTAL proposal that are being considered by the ISO Committee at the present time, can be summarized as follows: • It is based on the IP-X anti-collision protocol with a LBT (listen-before-talk) feature: tags listen for RTF modulation and remain mute if such modulation is detected in order to minimize interference with RTF applications. • The minimum listening time is 5 ms after power-up, minimum 25 ms mute time if reader modulation is detected, and a maximum tag ID duty cycle of 2%. • There is no forward (reader-to-tag) link specification: any of 18000-6 A, B, C or proprietary forward links may be used. • The return (tag-to-reader) link must be either PPE (pulse position encoding) or Miller encoding at 256 kbit/s. • Structured or unstructured data is allowed. If the IP-X options described in the previous section are studied against the background of the above constraints for TOTAL, the following observations can be made regarding the expected performance of a TOTAL tag compared to an IP-X tag: • The fast version of IP-X (V3 with 4k maximum delay time and 256 kb/s bit rate) has a maximum standard delay time of between 15 and 16 ms, but an initial maximum delay time (when implementing the acceleration part of the protocol) of just below 2 ms.

248

RFID Systems

The prescribed listening time of 5 ms of the TOTAL protocol will therefore make this protocol significantly slower compared to standard IP-X in terms of the ability to detect fast-moving tags. • The duty cycle of standard IP-X tags programmed according to the fast V3 mode is just more than 2% when responding in standard (non-accelerated) fashion, which is already marginally in excess of the duty cycles that are allowed by TOTAL. The acceleration portion of IP-X will increase this duty cycle by a factor of 8 for the initial two tag transmissions. It is therefore likely that the acceleration portion of IP-X will be deemed to be illegal under TOTAL for tags programmed to run in the 4k mode (as is the case for the V3 version). The conclusion at this early stage must therefore be that the proposed TOTAL protocol will not be able to achieve quite the same performance for fast-moving tags as the IP-X protocol. At the same time it must be mentioned that the V3 version of IP-X can theoretically detect tags – up to 6 tag reads maximum – at speeds of around 600 km/h (375 mph). A legal implementation of TOTAL, with a total transmission length of 96 bits (including preamble and CRC), baud rate of 256 kb/s, that has an initial listening time of 5 ms and then a maximum delay time of 16 ms to satisfy the 2% duty cycle limit (compared to the initial 2 ms maximum delay of IP-X V3 in accelerated mode) will still be able to read tags at speeds in excess of 200 km/h (125 mph for a maximum of 3 tag reads). If a theoretical limit of maximum 6 tag reads is deemed to be the criterion for reliable detection in potentially high density traffic, the maximum speed for a TOTAL tag will be just above 100 km/h (around 63 mph). This will further deteriorate with the addition of further data pages, over and above the tag ID.

9.4.4 Comparison between Different TTF Protocols Figure 9.5 displays the typical average reading time distribution of a free-running TTF protocol (i.e. no interruption from the reader to modify the behavior of tags). The long “tail” of the distribution illustrates that the longest possible time to detect a tag is potentially much longer than the average time, which is the primary drawback of this version of TTF protocols. In order to improve the maximum reading time, different approaches are possible. The Supertag switch-off and fast switch-off protocols will instruct tags that have been detected to stop transmitting, hence reducing the length of the “tail” of the distribution by eliminating unnecessary tag transmissions (see Figures 9.6 and 9.7). The effect of saturation of TTF protocols with an increasing number of tags is displayed in Figure 9.8. It is clear that for the 64 kb/s version of a free-running protocol, saturation takes place after about 50 tags, while the switch-off protocols do not show significant saturation. In the case of a 256 kb/s implementation of the same protocol, saturation will take place at around 200 tags as the faster bit rate will reduce the tag duty cycle by a factor of 4. The drawback of the switch-off protocols is, however, the increase in tag error rate that is encountered, resulting from tags being switched off without successful detection (Figure 9.9). As the optimal switch-off period to ensure reliable detection is difficult to implement, this variation of TTF protocols is not suitable for applications requiring read

Comparison of TTF and RTF UHF RFID Protocols

249

Probability mass function of tag read times 0.14

Probability mass

0.12 0.1 0.08 0.06 0.04 0.02

502

481

440 460

419

399

378

358

337

316

296

275

255

234

193

214

172

152

131

111

90

70

49

0

28

0

Time (ms)

Figure 9.5 Average reading time distribution for the Supertag Free-running protocol (10 tags, 64 k baud rate, 64 k interval) generated using a protocol simulator, with 1,000 samples, resulting in average read time of 172.7 ms, longest read time of 480.9 ms and standard deviation of 70.2 ms. Courtesy of [1]. Reproduced from Van Eeden, H. L., “Passive UHF RFID systems,” Ph.D. thesis, Northwest University, Potchefstroom, South Africa, 2004.

Probability mass function of average tag reads 0.18 0.16 Probability mass

0.14 0.12 0.1 0.08 0.06 0.04 0.02 1030

981

931

882

832

783

733

683

634

584

535

485

435

386

336

287

237

187

138

88

39

0

0

Time (ms)

Figure 9.6 Average reading time distribution for the Supertag Switch-off protocol (10 tags, 64 k baud rate, 64 k interval) generated using a protocol simulator, with 1,000 samples, resulting in average read time of 107.0 ms, longest read time of 1030.4 ms and standard deviation of 52.6 ms. Courtesy of [1]. Reproduced from Van Eeden, H. L., “Passive UHF RFID systems,” Ph.D. thesis, Northwest University, Potchefstroom, South Africa, 2004.

250

RFID Systems

Probability mass function of average tag reads 0.6

Probability mass

0.5 0.4 0.3 0.2 0.1 0 0

46

56

66

77

87 97 Time (ms)

108

118

129

139

Figure 9.7 Average reading time distribution for the Supertag Fast Switch-off protocol (10 tags, 64 k baud rate, 64 k interval) generated using a protocol simulator, with 1,000 samples, resulting in average read time of 72.6 ms, longest read time of 138.9 ms and standard deviation of 8.1 ms. Courtesy of [1]. Reproduced from Van Eeden, H. L., “Passive UHF RFID systems,” Ph.D. thesis, Northwest University, Potchefstroom, South Africa, 2004. Protocol saturation (64 kbaud, 4k interval) 4.5 4

Read time (s)

3.5 3 P4022 freerunning

2.5

P4022 Switch-off

2

P4022 Fast Switch-off

1.5

iP-X V1

1

iP-X V3

0.5 0 0

Figure 9.8

10

30 20 Number of tags

40

50

Protocol saturation in TTF protocols [22].  EM Microelectronic-Marin SA.

reliabilities of close to 100%. A better alternative would be to use faster implementations of the IP-X protocol. Figure 9.10 displays average reading times for IP-X implementations differing with respect to baud rate and maximum delay interval as previously described in Table 9.1. It can be seen that the V3 version, designed for very fast moving tag applications, has a

Comparison of TTF and RTF UHF RFID Protocols

251

Error rate (64 kbaud, 4k interval) 2.5

Error rate (%)

2 P4022 freerunning

1.5

P4022 Switch-off 1

P4022 Fast Switch-off

0.5

0 0

30 20 Number of tags

10

40

50

Figure 9.9 Error rates for TTF protocols [22].  EM Microelectronic-Marin SA. Average reading times (IP-X) 160 140

Number of tags

120 100 V1 80

V2

60

V3 V4

40

V5 20 0 0

0.2

0.4 0.6 Read time (s)

0.8

1

Figure 9.10 Average reading times for different implementations of the iP-X protocol [23].  EM Microelectronic-Marin SA.

very low reading time for low tag populations but quickly saturates as the number of tags increase beyond about 20. The V4 and V5 versions, designed for large tag populations, only start displaying saturation for much higher numbers of tags. Using these versions of IP-X tags, more than 100 tags can on average be detected in less than one second. Figure 9.11 displays the maximum read rate using different implementations of the IP-X protocol. For small tag populations, more than 300 tags can be read per second

252

RFID Systems

Maximum reading rates (IP-X) 800 700

Number of tags

600 500 V1 400

V2

300

V3 V4

200

V5

100 0 0

50

150 100 Read time (s)

200

250

Figure 9.11 Maximum reading rates for the iP-X protocols [23].  EM Microelectronic-Marin SA.

Maximum tag speeds (IP-X) 160 140

Speed (m/s)

130 100 V1 80

V2 V3

60

V4

40

V5 20 0 0

20

40 60 Number of tags

80

100

Figure 9.12 Maximum tag speeds for the iP-X protocol [23].  EM Microelectronic-Marin SA.

using the V3 and V4 versions. Similarly, Figure 9.12 displays the maximum tag speed for different IP-X implementations for different tag populations. If this is converted to km/h, it can be seen that tag speeds in excess of 200 km/h can be handled if only a small number of tags are visible to the reader.

Comparison of TTF and RTF UHF RFID Protocols

253

9.4.5 TTF Performance with Additional Data Pages A feature of the TTF approach is the fact that, in addition to the tag ID, any number of additional memory pages can be transmitted by tags in TTO mode (that is, without any interrogation from the reader). This can be achieved by pre-configuring tags to always transmit a predetermined number of memory pages in addition to the tag ID. This provides TTF protocols with an additional benefit over RTF protocols, as the latter requires additional tag interrogations, over and above the normal operation of the protocol, to extract additional data from tags. An obvious question is: “How much additional user-defined data can be extracted from tags in TTO mode, and to what extent will the reliability of tag communications be impacted by the addition of memory pages to be transmitted?” Before the experiment involving different numbers of memory pages is described, the tag memory fields provided for by the different protocols must be explained. In the case of the EPC Gen 2 protocol, provision is made for a compulsory user programmable EPC code, for a compulsory tag identifier or TID [20] as well as for optional additional general purpose user-defined memory. While the EPC code will identify the specific type of product to which the tag will be attached, the TID identifies the tag manufacturer, and will typically be hard-coded at the factory to ensure the integrity of tag identities. In the case of the IP-X protocol, only a unique factory programmed tag identifier is compulsory, while additional user programmable memory can be used, for example, to store an EPC code. A practical experiment was performed using IP-X protocol tags, configured in a mode that allows the transmission of a selectable number of data pages in TTO mode, in addition to the tag ID [4]. In this case, no specific provision was made in the hardware implementation of the protocol on the transponder chips to inform the reader regarding the number of additional data pages that could be expected. This increases the possibility that a reader that has successfully received the tag ID plus some data pages from a particular tag will confuse further data pages with data pages received from a different tag that has also started transmitting. For each experiment, the tags were configured to each transmit its tag ID plus a specific number of additional pages, with repetition of the experiment for different numbers of additional pages, this number ranging from 0 to 5. Three tags at a time, configured in this way, were exposed to a reader for 60 seconds, and the number of times that each page was successfully received was recorded. Figure 9.13 displays the results that were achieved (repeat rate indicates the number of times the respective data page was received). As could be expected, there is a more or less proportional decline in the number of times that each page is received with an increase in the number of pages transmitted. What was, however, noticed was that, with an increase in number of pages, in addition to the tag ID, a marked reduction in the reliability of detection of pages 4 and 5 was observed. This is evident from Figure 9.14, where all pages were detected with more than 98% reliability compared with the tag ID for up to 3 additional pages, whereas this figure dropped below 70% for the 5th of 5 additional pages. This is likely to be a result of the fact that the additional data pages contain no information linking them to a particular tag ID – it is up to the reader to make this association based on the time differences between consecutive pages that are detected after the initial detection of a tag ID. As more than one tag is present, the probability of another tag starting with a transmission while the first tag is still transmitting some of its data will obviously increase with the number of additional pages transmitted. The longer the time lag between the tag ID and

254

RFID Systems

Repeat rate with increasing page reads (IP-X) 900 800 700 600 500 400 300 200 100 0

TID only

TID and 1 page TID and 2 pages TID and 3 pages TID and 4 pages TID and 5 pages

Figure 9.13 Average repeat rate of tag backscatter transmissions with an increase in the number of user pages. Courtesy of Naud´e and Marais [4].  2008 Northwest University. Degradation with increasing page reads (IP-X) 100 90 80 % degradation

70 60 50 40 30 20 10 0 TID only

TID and 1 page TID and 2 pages TID and 3 pages TID and 4 pages TID and 5 pages

Figure 9.14 Degradation of read performance with an increase in the number of user data pages. Courtesy of Naud´e and Marais [4].  2008 Northwest University.

an associated data page, the higher the likelihood that the reader will not associate the additional page with the correct tag ID due to data from other tags being received in the meantime. It would therefore seem that, for collections of 3 tags at a time, 3 additional pages is the maximum number that should be used in practical applications if a read reliability of more than 98% is required. With an increase in size of tag population, this

Comparison of TTF and RTF UHF RFID Protocols

255

number will decrease even further unless a more elegant and robust approach is used to correctly associate each data page with the correct tag ID. The obvious question is: “How many additional data pages will normally be required for practical applications?” While the tag ID will suffice for many applications, there are cases that require a high level of security where it will be beneficial to extract encrypted codes from tags at high speed (e.g. in the detection of tags that form part of vehicle license discs, where reads must take place at speed to identify illegal vehicles). Given the length of encryption keys of adequate strength, the number of bits to be extracted per tag may be as high as 256, which in this case would require 4 pages if no CRCs are included in the user data, and 6 pages if CRCs are provided for. Such applications may therefore suffer from marginal read performance as the read rates reflected in Figures 9.13 and 9.14 are expected to significantly deteriorate in dense traffic environments. Because of the progressive deterioration observed in the reliability of tag transmissions with the increase in the number of additional data pages, the proposed TOTAL standard provides for two mechanisms aimed at improving the reliability of reading additional data pages, namely, (1) CRC bits to be included in each data page (to allow the reader to verify if a page has been successfully received); and (2) for so-called link bits that inform the reader about the number of data pages that should be expected in addition to the tag ID. The availability of CRC and link bits will enable the reader to at least be aware if it should be searching for additional data pages in subsequent free-running transmissions from the tag, should it not have successfully received all data pages, possibly due to collisions with transmissions from other tags.

9.5 Comparison between ISO18000-6C and TTF Protocols The ISO18000-6C protocol (better known as EPC Class 1 Generation 2) is currently the UHF RFID protocol that has achieved the widest adoption in the global RFID community. The purpose for which this protocol was developed, was to provide a universal standard that would be suitable for all “open loop” applications of UHF RFID, that is, applications where systems used by different end-users and supplied by different vendors must be fully interoperable. The two most prominent application domains for UHF RFID are supply chain management and transportation or traffic management. Both of these tend in general to be “open loop” in nature, with potentially large numbers of independent stakeholders that could benefit from accessing the same RFID data, hence justifying a global standard.

9.5.1 Areas of Comparison ISO18000-6C, like all other RTF protocols, potentially suffers from a number of deficiencies, both in terms of performance as well as in terms of the ability to deploy large systems while complying with frequency spectrum regulations. Three prominent areas of performance can be mentioned where RTF protocols can potentially suffer from deficiencies when compared with TTF protocols: 1. The reliability of reading tags in large populations, when moving through very non-uniform RF fields: The potentially inferior performance of RTF protocols under

256

RFID Systems

such conditions is caused by the much more complex process flow required for successful execution of an RTF protocol compared to the much simpler TTF protocols. This can be appreciated by comparing the state diagram for a typical free-running TTF protocol (Figure 9.15) with the state diagram for the ISO18000-6C protocol in Figure 9.16. The much more complicated RTF state diagram implies that the transponder chip must be able to remember in which of a number of states it last was when it has to respond to the next reader command. While this will not be an issue in relatively uniform RF fields, it does become an issue in non-uniform fields where the tag not only can temporarily lose its communication link with the reader when moving through a weak spot in the field, but can even switch off and wake up again without knowing what its state last was in terms of the execution of the protocol. This can result in tags responding in the wrong way or not responding at all, with the implication that tags may pass undetected even though they were visible to the reader for a significant period of time. In comparison, a TTF tag only has to be seen once for a very short period of time, during which period it can transmit its ID in one brief communication packet, without having to wait for the completion of a fairly complex interrogation process where potentially hundreds of tags must respond in coordinated fashion for the detection process to be completed successfully. While earlier versions of RTF protocols definitely suffered from this phenomenon, the question is to which extent the more sophisticated EPC Gen 2 protocol managed to address this weakness of earlier versions. The other question is to what extent improvements in technology have made UHF RFID protocols more immune against the fluctuation nature of RF

Power on reset

Boot

Send ID

Wait random time

Figure 9.15 State diagram for the IP-X free-running TTF protocol [1]. Reproduced from Van Eeden, H.L., “Free running RF identification system with increasing average inter-transmission intervals,” United States Patent, Patent Number 6,154,136, Date of patent: November 28, 2000.

Comparison of TTF and RTF UHF RFID Protocols

Power down Power up and killed

257

Power down

Power up

Valid Kill with handle, password Backscatter handle Select QueryRep Valid ACK Backscatter {PC+EPC+CRC} Valid REq_RN, Read, Write, Lock

Query [with matching flags]

QueryRep Decrement slot counter

Ready

Secured

Arbitrate Select

Valid Access with handle, password Backscatter handle

Select QueryRep QueryAdj

Select QueryRep QueryAdj

Select

No valid ACK

Query [with matching flags] Backscatter RN 16 slot = 0

Reply

Open

QueryAdjust [slot = 0] Backscatter new RN 16

Valid ACK Backscatter {PC+EPC+CRC} Valid Req_RN, Read, Write, Lock Valid Req_RN access password ≠ 0 Reply with handle

Acknowledged

Valid ACK Backscatter {PC+EPC+CRC}

Valid ACK Backscatter {PC+EPC+CRC} Invalid Req_RN (no reply)

Figure 9.16 State diagram for EPC as adapted from the ISO18000-6C standard [20]. Reproduced from EPC Radio Frequency Identity Protocols, Class-1 Generation-2 UHF RFID, Protocol for Communications at 860–960 MHz, Version 1.0.9, January 2005.

fields – more specifically the significant reduction in RF field strength that more recent tag versions need to wake up and stay awake compared to earlier versions. 2. The ability to reliably detect tags moving at high speeds: The second area of potential weakness for RTF protocols compared to TTF is the ability to reliably detect tags moving at high speeds. For similar reasons as discussed above, RTF protocols tend to take a longer period of time to complete an entire interrogation cycle compared to TTF protocols, where tags respond in their own time in quick bursts with low duty cycle. The comparison that must be done in this case is to measure the time required for an RTF reader to successfully search through a tag population of specific size, against the time required for a similar number of TTF tags to be successfully detected against the background of possible tag collisions. While both supply chain applications (involving large numbers of tags moving at relatively low speeds) and traffic applications (involving small numbers of tags moving at high speeds) can place severe constraints on RFID protocols, the scenario for which a comparison is easiest is the case of a small number of tags moving at high speeds, which allows the reader

258

RFID Systems

limited time in which to successfully detect tags. In the case of traffic law enforcement, the emphasis would be on detecting vehicles moving at speeds high above the legal speed limit and to do so in dense traffic environments. 3. The spectrum requirements to support the respective protocols: The third area for comparison, and possibly the most fundamental limitation on the ability to successfully deploy large RFID applications, is the spectrum requirements to support the respective protocols. As explained in Section 9.4, an inherent benefit of free-running TTF protocols is their ability to detect tags without requiring interrogations from the reader, resulting in very limited spectrum requirements. This comes at a cost, though, since the RF sections of the readers have to be more robust and sufficiently accurate in terms of frequency in order to co-exist. This attribute will allow the operation of potentially large numbers of readers in the same channel and at the same location. Doing the same with RTF readers can potentially result in overlap between the spectral content of reader communications and the spectral content of unrelated tag communications (i.e. tags communicating to other readers in the immediate vicinity), which can be relieved by physically synchronizing RTF readers, which may not always be practical. If the regulatory limitations on RFID channels do not allow the complete spectral separation of tag and reader communications, the implication is that only one reader will be able to operate in a channel and within an area of such size that other reader signals will be of the same order of magnitude or larger than tag signals. The physical size of such an area can be determined by calculating the reduction in secondary reader power density at the primary reader with increase in distance between them (using the 1/r2 rule), and comparing this power level with the typical power received from a tag at the primary reader. With no specific measures to limit the size of unwanted reader signals in the tag communication band (i.e. if the reader communication spectral band fully overlaps with the tag communication band), the required separation to prevent interference can be as large as 7 km [1]. Such a situation will make the deployment of large RFID systems totally impractical, against the background that some regulatory jurisdictions, for example, ETSI, only provide for as few as 10 channels. For this reason, the ISO18000-6C standard provides for different types of spectral masking applicable to reader signals, ensuring that unwanted reader radiation will fall below the level of tag signals for much smaller distances between different readers.

9.5.2 The Impact of Progress on Technology During the early years of UHF RFID the factors discussed in the previous section provided TTF protocols with an undeniable advantage over RTF protocols, even taking into account potential issues around the undetermined maximum time period to detect a TTF tag. An early implementation of the IP-X protocol was the first to be applied in large applications in the European regulatory jurisdiction, based on the fact that the limited spectrum availability did not impact on the operation of this protocol. During this same period (2002–2005), the early version EPC protocols could not effectively operate outside of FCC regulatory areas, as the nature of their reader communications made these protocols illegal to operate in Europe and in other jurisdictions with similar spectrum regulations. A number of important technological advances have, however, taken place in the past 10 years, resulting in a drastic change in the UHF RFID landscape. A number of these

Comparison of TTF and RTF UHF RFID Protocols

259

developments and their implications for the comparative situation of RTF versus TTF will be discussed below. 9.5.2.1 Improvements in Silicon Technology The more simplistic logic of TTF protocols compared to RTF protocols results in TTF transponder chips requiring a factor of 3 to 4 times fewer transistors to implement the protocol (typically 3,000–4,000 transistors for a TTF chip compared to approximately 12,000 chips used in an EPC Gen 2 chip). This results in smaller size, lower energy consumption and lower cost. This was also the primary reason for early commercial successes for TTF protocols in Europe. TTF tags implemented in 0.5 micron processes were approximately 0.25 mm2 in size compared to RTF chips with areas of around 1 mm2 . The 0.5 mm × 0.5 mm TTF chip sizes were ideal for flip-chip manufacturing operations, as this was about the smallest size that inlay production lines could handle. They required less power to operate, resulting in more reliable operation in the presence of weak and fluctuating RF power, as they switched on in weaker fields (requiring around −1 dBm in power to operate, about 3 dB less than their RTF counterparts). Over the past five years three significant technological advances have changed this scenario: • Silicon processes moved from 0.5 micron to 0.35 micron and then on to 0.18 micron process technology. The immediate implication for RFID chips was a reduction in size almost by a factor of 10, resulting in even more complex new RTF chips (e.g. EPC Gen 2 chips) shrinking in size to the inlay manufacturing limit of around 0.5 mm × 0.5 mm. TTF chips did not benefit to the same degree as they were already approaching optimal size in the older 0.5 micron technology. The cost benefit of TTF was therefore reduced. • Voltage levels for transponder chips to switch on dropped from around 2 V to less than 1 V, resulting in tag power-up levels to drop to −12 dBm and in some cases even less. This development largely eliminated the unsatisfactory behavior of RTF protocols in the presence of fluctuating RF fields, as the tendency of tags to switch off was greatly reduced in the type of RF fields practically experienced by tags. • Transponder antenna technology became more sophisticated, with the old model linear dipole type antenna being replaced by much more sophisticated designs that further improved the performance of tags in non-ideal environments. This includes more robustness with respect to environmental change, wider bandwidths than what dipoles antennas can achieve, and a reduction in size as opposed to dipoles. This improved performance resulted in more reliable operation of the complex RTF protocols. Once again, the much simpler TTF protocols did not benefit to the same degree from these developments, as they already performed fairly reliably (within their own set of limitations) with older technology. 9.5.2.2 Improvements in DSP Technology As is evident from the above discussions, RTF protocols will suffer severely from reader interference if a scheme is not devised to effectively separate reader communication signals from tag signals. As discussed before, such separation can take place either in the time or in the frequency domain. Without such separation, the problem can be as bad

260

RFID Systems

as requiring physical separation of unpractical proportions between co-channel readers, making the deployment of large systems impractical. As early versions of the EPC Gen 2 protocol did not provide for proper spectral separation between reader and tag signals for co-channel readers, efforts in Europe were aimed at deploying systems based on time separation using LBT (listen-before- talk) measures. These, however, met with limited success as it implied that readers could only communicate with very low duty cycles, less than what is required by either supply chain or traffic management systems. These low duty cycles result from the fact that most of the time an LBT reader, that would like to interrogate tags, will have to keep quiet due to activity detected from other readers in all available LBT channels. It is therefore clear that large deployments of RTF-based protocols are only possible if sufficient spectral separation can be achieved between reader and tag signals – a form of frequency multiplexing. The effective separation of reader and tag communications requires high performance filters that can only be implemented using digital signal processing techniques. These filters must typically satisfy strict spectral requirements, which implies not only a sharp cut-off but also very low ripple in stop bands by smoothing the shape of pulses as required by the respective modulation technique. The real-time implementation of DSP techniques for such applications and for sampling rates of typically 5 Msamples/sec is beyond the capabilities of low-end DSP processors, and requires implementation of the DSP algorithms on FPGA chips of sufficient capacity. While such FPGA components would have added significantly to the cost of a reader 5 to 10 years ago, is does not currently provide any significant obstacles in terms of implementing high performance readers at acceptable cost. It was therefore practically possible for the ISO18000-6C standard to incorporate spectrum masks that allow separation of 60 dB between reader and tag communications, even for 200 kHz channels as required by ETSI regulations. Readers that are compliant with the so-called “EPC dense interrogator mode” can therefore operate within ETSI regulations and still achieve 60 dB suppression of unwanted reader signals in the tag communication band. This drastically reduces the required physical separation between readers to prevent reader interference by as much as a factor of 1000, making practical reader co-existence a real possibility. 9.5.2.3 More Intelligent Utilization of the Available Spectrum The first attempts at creating standards for UHF RFID did not meet much success, partly because those efforts tried to achieve too much with a single uniform standard. The drastic differences between UHF RFID spectrum allocation in different regions alone were reason enough why this was not an achievable objective. The more thorough approach followed in the compilation of the ISO18000-6C protocol addressed this issue by allowing for several different options in the same standard, with the optimal choice for a specific jurisdiction depending on the spectrum that is available as well as on the type of RFID deployments that are expected to occur: • Single interrogator mode is based on the assumption that there are a sufficient number of independent channels available so that a reader is unlikely to encounter other readers in adjacent channels. It therefore does not impose any requirements on a reader, in addition to those already imposed by the regulatory authority.

Comparison of TTF and RTF UHF RFID Protocols

261

• Multiple-interrogator mode is designed for those cases where the number of simultaneous collocated readers is modest compared to the number of available channels, and imposes some constraints on the transmitted spectrum, sufficient to minimize interference in adjacent or second-adjacent channels [5]. • Dense-interrogator mode is designed to allow successful tag reading even when every channel is occupied by a reader. Another innovation in ISO18000-6C to limit reader interference is the use of Millermodulated subcarrier (MMS) encoding. Miller encoding has the effect of displacing the tag spectrum on a subcarrier away from the carrier. The benefit of this approach is to increase the spectral separation between reader and tag signals, in the process reducing the possibility of interference between tag signals and unwanted reader signals. In the ETSI regulatory environment a scheme has already been proposed whereby readers may only occupy every third reader channel in order to limit adjacent channel interference. The implication is that the associated tag communications will effectively take place in channels where no reader communications are allowed (as the MMS technique will effectively displace the tag spectrum from the reader carrier into the adjacent channels). Any number of readers can therefore operate in the designated reader channels, as their signals will have no spectral overlap with tag spectra occurring in the adjacent channels. The problems of reader interference and of too few available channels are effectively eliminated in the process. Shown in Figure 9.17 is a 25 us TARI spectrum and ETSI spectral mask at 30 dBm output. The spectral mask was scaled for 30 dBm reference power at the output connector of the reader vs 33 dBm ERP that is normally used. All ETSI power measurements, both spurious and intentional, are specified as ERP and the power at the connector of the reader would be 3 dB lower than the normally radiated power. Thus, 3 dB must be subtracted from the spectral limits (e.g. −36 dBm becomes −39 dBm). The continuous wave is shown as a reference.

9.5.3 A Comparison between RTF and TTF for Fast Moving Tags In the domain of ISO standards, RTF-based protocols have until now been preferred over TTF-based protocols within the global RFID community. The primary reason for this is most likely the fact that RTF-based protocols provide more flexibility in handling a variety of scenarios for reading tags. The fact that the reader can instruct tags how to respond allows the RFID system to moderate tag behavior in a way which suits the requirements of any specific read scenario. In contrast, TTF-based protocols require the system to be designed around a more rigid set of rules, as tags must be configured prior to entering the system to respond in a specific way, while readers can only observe the presence of tags, without being allowed to respond to potentially dynamic system requirements. As described in earlier sections, the TTF approach does offer specific benefits to compensate for a somewhat less flexible system response. The primary benefits will be summarized again for the sake of completeness: 1. More robust behavior in the presence of fluctuating RF field strengths. 2. Faster tag response times, allowing tags to be read at higher speeds. 3. Less spectrum required per reader, allowing a larger number of readers to be deployed within the same set of regulatory constraints.

262

RFID Systems

Marker

R S

Ref Lv 1 30.8 dBm

1 [T2] − 62.77 dBm 865.20200401 MHz

RBW VBW SWT

3 kHz 300 Hz 2.8 s

30.8 1 [T2]

10.8 dB Offset 20

RF Att Unit

40 dB dBm

−62.77 dBm 885.20200401 MHz

A

10 0 −10

1 MA 2 MA

1VIEW 2VIEW

−20 −30 −40 −50

ETS12

−60 −70 −80 −89.2 Center 865.7 MHz

100 kHz /

Span 1 MHz

Figure 9.17 Spectrum of a Miller-encoded EPC reader with a spectral mask. Courtesy of Impinj, Inc.

In order to compare the performance of RTF and TTF protocols against at least some of the above claimed benefits for TTF protocols, a simulation was conducted using the CISC RFID protocol simulator. The simulation was aimed at determining the maximum speed at which a tag can be reliably detected using either an RTF or a TTF protocol. As EPC Gen 2 or ISO18000-6C is currently the RTF standard enjoying most support, this was the choice of RTF protocol included in the simulations. To represent current contenders for TTF protocols, IP-X V3 (non-standards based) and the proposed TOTAL protocol (potentially part of the ISO18000-6 standard) were included. 9.5.3.1 Set-Up and Procedure for Simulations A conveyor belt set-up was used to simulate tags moving at specific speeds. The simulations were done so that only a single tag is present within the interrogation zone at a specific time. Multipath effects were not taken into account, thus the assumption was made that the tag entering the interrogation zone will always be energized.

Comparison of TTF and RTF UHF RFID Protocols

263

The physical read scenario applicable to both the IP-X and the EPC Gen 2 protocols was set up using default settings with the following changes: • • • •

Interrogation zone length = 4 m. Distance between tags = 4 m. Speed of conveyor belt varies between 60 km/h to 250 km/h. Tag population comprised of 4 tags.

For the IP-X protocol, the V3 2kbaud version was used for the evaluation, that is, the version with the fastest repeat rate that is most suitable for high speed applications. For the EPC Gen 2 protocol, some additional settings were required as this protocol provides for several parameters to be adjusted based on the expected read scenario. A short explanation of the relevant parameters is provided before the selected values are given: • Q: a parameter that the reader uses to regulate the probability of a tag response. Q is an integer in the range 0–15; the probability of a tag responding in the next slot equals 2−Q [20]. • C: a value that is added to the floating point version of Q in order to adjust the probability of a tag response [20]. • Data link rates: these are the bit rates of communication for the downward link (readerto-tag) and the upward link (tag-to-reader) in kbits/s. Parameter values used for this evaluation were: • Data link rates evaluated: 128/640, 40/160, 30/40, 32/64 and 32/53. • Starting Q values evaluated: 0, 3, 6, 9, 12 and 15. • C values evaluated: 0.1 to 0.9 in intervals of 0.1 (where applicable). The EPC parameters described in the set-up section above were evaluated for each combination of speed of the conveyor belt, algorithm, data link rate, starting Q, and C. From these simulations, the EPC combination that performed the best at each movement speed was chosen for comparison with IP-X. Figure 9.18 displays the number of singulations for EPC Gen 2 tags at different tag speeds and for different link rates. It is clear that the 128/640 link rate (128 kb/s for reader downlink, 640 kb/s for tag uplink) provides more than adequate performance for speeds of up to 250 km/h. The 40/160 link rate also provides acceptable performance for speeds up to 160 km/h, but starts failing as from 200 km/h. The slower link rates provide acceptable performance up to about 120 km/h, but are down to only one tag singulation at a speed of 250 km/h, which will result in unreliable performance in practical scenarios. Next an experiment was performed using a population of 4 tags and using different link rates. The result in Figure 9.19 shows that only the 128/640 link rate was able to detect all 4 tags for speeds in excess of 200 km/h. Figure 9.20 compares the most successful of the EPC Gen 2 options (128/640 link rate) with a pure IP-X implementation of the TOTAL concept (to be contrasted with a TOTAL implementation that is fully in line with the current ISO proposal). It is clear that, while both protocols successfully detected all tags for speeds of up to 250 km/h, the IP-X

264

RFID Systems

Repeat rate for different tag speeds (EPC) 40 35 Link rate

Repeat rate

30 25

30/40 20

40/160 128/640

15

32/53 10

32/64

5 0

60 / 37.5

120 / 75

160 / 100

200 / 125

250 / 156

Tag speed (kph / mph)

Figure 9.18 Total number of EPC tag singulations for different link rates.

Hit rate for different tag speeds (EPC) 4.5 4 Link rate

3.5

Hit rate

3 30/40

2.5

40/160

2

128/640 1.5

32/53

1

32/64

0.5 0

60 / 37.5

120 / 75 160 / 100 200 / 125 Tag speed (kph / mph)

250 / 156

Figure 9.19 Number of tags detected from population of 4 using EPC Gen 2 with different link rates.

Comparison of TTF and RTF UHF RFID Protocols

265

EPC vs TOTAL at different tag speeds 450 400

Singulations

350 300 250 200

EPC

150

TOTAL

100 50 0 60 / 37.5

Figure 9.20

120 / 75 160 / 100 200 / 125 Tag speed (kph /mph)

250 / 156

Comparison of EPC and IP-X in terms of total number of tag singulations.

protocol outperforms all EPC Gen 2 options, achieving about 10 times the number of tag singulations compared to EPC Gen 2. This will result in more robust performance in dense traffic environments where the practical number of tag singulations tends to decrease significantly compared to the theoretical figure. A fully legal TOTAL implementation is, however, expected to perform similarly to EPC Gen 2 using a 128/640 link rate, as it will have to exclude the IP-X acceleration protocol.

9.6 Conclusion This chapter provided an overview of TTF and RTF protocols and performed a comparison of these protocols both in philosophical and empirical terms. The above comparison shows that protocols other than those currently included in the ISO18000-6 standard for UHF RFID do have some interesting properties that justify more than passing attention. For specific applications, most significantly in the case of small populations of fast moving tags, TTF protocols can outperform RTF protocols in some respects, and may prove to offer more robust solutions in difficult to read scenarios, for example, high traffic density environments. During the early stages of the commercial deployment of UHF RFID, these TTF implementations indeed had a clear edge over the RTF offerings that were available at the time, both in terms of performance as well as cost and robustness in practical settings. The inability of the standards community to find an acceptable solution for the deployment of UHF RFID in regulatory jurisdictions similar to ETSI, at some stage created a situation where TTF alternatives came close to becoming a de facto standard for such jurisdictions. Over the past three to five years, however, it would seem that a combination of technological advances as well as more effective utilization of available frequency spectrum by proposed standards, have turned the tables and have put RTF protocols firmly back in a dominant position. While the inherent benefits of TTF-based protocols, most notably

266

RFID Systems

simplicity of implementation, very fast response time by tags and very little spectrum requirement for readers, still stand firm, it appears as if RTF protocols are now performing sufficiently well in these respects as a more flexible approach to a wide variety of applications. The proposed TOTAL addition to ISO18000-6C represents a last-ditch effort by a TTF protocol to become part of the global standard, although the philosophy of TTF was adapted in the process. The restrictions placed on TOTAL will, however, remove some of the benefits that IP-X, as strongest TTF contender, still offers over the existing ISO18000-6 protocols, and may result in TOTAL finding limited practical adoption. It is therefore likely that TTF standards will be mostly restricted to niche applications where the specific functional advantages offered by such protocols can outweigh the disadvantages associated with non-standardized or proprietary technologies.

Problems 1. Discuss the applicability of passive LF (low frequency), passive HF (high frequency), passive UHF as well as active UHF RFID technology to solve the following automated identification problems: • Reading bolus tags that are swallowed by livestock and that remains in the stomach of the animal for its entire lifetime. • Bulk reading of tags embedded into the caps of bottles filled with pharmaceutical products, typically containing liquids and being stacked multiple deep and multiple high. • Identifying tagged vehicles moving through lanes at a multiple-lane customs gate, requiring the verification of the specific lane that each vehicle moved through. • Identifying tagged shipping containers at a depot, requiring the implementation of an electronic stock-take of containers at the press of a button. 2. Identify the different limiting factors that determine the maximum read range of a passive RFID transponder, and describe the circumstances that will cause each of these factors to become the dominant determinant of read range. 3. Discuss the different criteria that a passive UHF RFID air-interface protocol should comply with, and describe the differences between these criteria for transportation (vehicle tagging) versus goods supply chain (item tagging) applications. 4. Consider an application of passive UHF RFID in the field of open-lane tolling of vehicles traveling on a highway. In ideal circumstances it has been determined that a reader can read tags at a distance of 10 m. The maximum speed of travel at which tags must be read is 220 km/h. Assume that the reader antennas are mounted 6 m from the road surface and are facing directly downward with a viewing angle of 60◦ . Determine the minimum period of time that a reader will have to identify a transponder on a passing vehicle. Discuss the implication of these results on the requirements for the air-interface protocol. 5. Consider an application of passive UHF RFID in the field of item level tagging of goods in a warehouse. Trolleys are stacked with up to 24 bins, each containing 4 manufactured items. The trolleys are moving through dock doors that are 3 m in width, and the UHF RFID readers installed at the dock doors must accurately detect

Comparison of TTF and RTF UHF RFID Protocols

267

all tagged items on the trolleys. If the readers employ the IP-X V4 air-interface protocol and the reader beams can energize tags over a minimum width of 1m in the doorway, calculate the maximum speed at which trolleys may be allowed to move through the dock doors to ensure that all items are read. 6. Discuss the benefits and limitations of the Supertag Switch-off protocol, compared to the IP-X free-running protocol, in terms of the saturation levels of the protocol as well as the time required to identify all tags in a population. 7. Discuss the inherent benefits and drawbacks of RTF versus TTF protocols, taking into account regulatory requirements in different jurisdictions, interference between different readers, maximum number of tags in a population that can be handled, as well as maximum allowed speed of travel of tags. 8. Describe the motivation for the use of TOTAL (tag only talks after listening) variation in the use of the TTF free-running protocol, in applications that require the mixing of RTF and TTF tags in the same population. Discuss how an optimal waiting time for the TOTAL protocol can be arrived at. Also discuss the implications of this waiting time on the expected performance level of the system employing the TOTAL TTF protocol, compared to a TTF protocol that does not have to employ the TOTAL approach. 9. A UHF RFID application requires the generation of user-defined information to be written to and read from transponders. The designer of the system must make recommendations regarding the required read scenarios for the writing of user-defined data to transponders, the reading back of only the transponder ID, and the reading back of the transponder ID plus user defined data. Discuss the different factors that will have to be considered in arriving at a reliable system design, and compare these considerations for a system using an RTF versus a system using a TTF air-interface protocol. 10. Discuss the impact that progress in electronic technology has had on the performance that can be achieved by UHF RFID systems employing different types of air-interface protocols, and compare the impact of these factors on RTF and TTF based systems.

References [1] Van Eeden, H.L. (2004) Passive UHF RFID systems, Ph.D. thesis, Northwest University, Potchefstroom, South Africa. [2] Atkins, R.C., Marais, M.A. and Smit, H. van Z. (1999) Identification System, International application published under the Patent Cooperation Treaty, International Publication Number WO 99/26081, International Publication Date, 27 May. [3] Van Eeden, H.L. (2000) Free running RF identification system with increasing average inter transmission intervals, United States Patent, Patent Number 6,154,136, date of patent: November 28, 2000. [4] Naud´e, C.C. and Marais, H.J. (2008) Degradation in read performance with an increase in the number of user data pages, internal research report, Faculty of Engineering, Northwest University. [5] Dobkin, D. (2008) The RF in RFID: Passive UHF RFID in Practice. Oxford: Elsevier Inc. [6] Naud´e, C.C. (2008) Comparison of EPCglobal and iPX protocols with respect to movement speed of tags and different data link rates using the CISC RFID ASD Simulator, internal research report, Faculty of Engineering, Northwest University. [7] Curty, J-P., Declercq, M., Dehollain, C., and Joehl, N. (2007) Design and optimization of UHF RFID systems, in Springer Science & Business Media. New York: Springer.

268

RFID Systems

[8] Carbunar, B., Ramanathan, M., Koyuturk, K.M., Jagannathan, S. and Grama, A. (2009) Efficient tag detection in RFID systems, Journal of Parallel and Distributed Computing, 69: 180– 196. [9] Shih, D., Sun, P., Yen, D. and Huang, S. (2006) Taxonomy and survey of RFID anti-collision protocols, Computer Communications, 29: 2150– 2166. [10] Lee, H. and Kim, J. (2006) QT-CBP: a new RFID tag anti-collision algorithm using collision bit positioning, Lecture Notes in Computer Science, 4097: 591– 600. [11] Chung, H.B., Mo, H., Kim, N. and Pyo, C. (2007) An advanced RFID system to avoid collision of RFID reader, using channel holder and dual sensitivities, Microwave and Optical Technology Letters, 49(11): 2643– 2647. [12] Available at: http://www rfida.com/weblog/labels/standard.htm: RFID Barriers to Pharma Adoption: Cost Benefits Standards (accessed April 19, 2007). [13] Available at: http://www.ipico.com/index.cfm?id=5894. [14] The Internet of Things. Available at: http://www rfid-weblog.com/50226711/the internet of things.php. [15] More Than 5 Cents, January 12, 2007. Available at: http://www rfid-asia.info/2007/01/more-than5-cents.htm. [16] Yan, Lu, and Zhang, Yan (eds.) (2008) The Internet of Things: From RFID to the Next Generation Pervasive Networks. New York: Auerbach Publications, Taylor & Francis Group. [17] Available at: http://www.scdigest.com/ASSETS/ON TARGET/09-01-27-2.PHP?cid=2201&ctype= content, “RFID News: The five-cent tag is here, the five cent tag is here! Well, almost”, January 27, 2009. [18] Marais, H.J. (2009) Characterisation of a UHF fading channel for an RFID-based EVI system, Master’s thesis, Northwest University. [19] Available at: http://www.magellan-rfid.com/pjm-technology/standards. [20] EPC Radio Frequency Identity Protocols, Class-1 Generation-2 UHF RFID, Protocol for Communications at 860– 960MHz, Version 1.0.9, January 2005. [21] Available at: http://www.alibaba.com/showroom/Rfid Dual Frequency.html. [22] Available at: http://www.emmicroelectronic.com/products, EM-Microelectronic – Marin SA data sheets for EM4022 integrated circuit. [23] Available at: http://www.emmicroelectronic.com/products, EM-Microelectronic – Marin SA data sheets for EM4222, EM4122 and EM4123 integrated circuits.

Part Three Reader Infrastructure Networking

10 Integrating RFID Readers in Enterprise IT1 Christian Floerkemeier and Sanjay Sarma Massachusetts Institute of Technology

With the increasing adoption of RFID, certain deployment challenges are coming into focus. The management and control of a large number of RFID readers are becoming an issue from a network administration perspective. This includes monitoring of the health of RFID readers and maintaining a consistent configuration across all RFID readers. The shared nature of the wireless medium may require coordination among the RFID readers to minimize interference and ensure compliance with local radio regulations. Readers also do not operate independently but are sometimes triggered by external sensors that need to be configured. RFID readers also create significant data volumes. The data captured by the RFID readers needs to be filtered and aggregated due to the presence of redundant and unwanted data. Most importantly, the captured RFID data has to be interpreted in an application context to make the data capture meaningful in the first place. There are today a number of different (partly proprietary) solutions to address these challenges. This chapter presents a taxonomy of the corresponding system services and system architectures. We also discuss how the system architectures relate to emerging standards and how these standards could possibly be enhanced to support future RFID deployments. We aim to provide a comprehensive understanding of common industry practices and thus facilitate the deployment of RFID systems. The chapter is organized as follows. Section 10.1 discusses related work. Section 10.2 presents a number of different services and components that are commonly found in RFID deployments. In Section 10.3, we discuss the services offered by today’s RFID 1

This chapter is based on “An Overview of RFID System Interfaces and Reader Protocols,” by C. Floerkemeier and S. Sarma which appeared in IEEE International Conference on RFID, 2008.  2008 IEEE.

RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

272

RFID Systems

reader products. Section 10.4 presents different system architectures that we identified. In Section 10.5, we discuss EPCglobal specifications that aim to standardize system interfaces. Before we conclude in Section 10.8, Section 10.6 discusses the adoption of today’s EPCglobal reader protocol standards and Section 10.7 presents potential enhancements to existing specifications.

10.1 Related Work There are a number of previous publications that discuss requirements for an RFID infrastructure and propose system architectures [1, 2, 3, 4]. The work presented in this chapter differs from previous work in that we do not propose a particular system architecture but compare different system architectures. We also focus in particular on emerging standards in our analysis. Within the EPCglobal community, technology vendors and end users that deploy RFID technology have been working jointly on the development of specifications that standardize the interfaces between RFID tags, readers and enterprise IT systems. The Architecture Framework published by EPCglobal [5] provides a comprehensive overview of the EPCglobal standards. It shows how the different interface standards are related and outlines the principles that have guided the design of the standards. The EPCglobal Architecture Framework does not dictate a particular system architecture, but leaves this to implementers who can choose the system architecture that is most appropriate for their deployments. The analysis presented in this chapter is complementary to the EPCglobal Architecture Framework because it provides insights into which system architectures are actually deployed. The Architecture Framework document distinguishes a number of different roles, such as RFID Reader, Filtering & Collection and Capture Application, each of which implements a set of different services. Rather than classifying the services as roles which are implemented on a particular device as in the EPCglobal Architecture Framework , we categorize them as base, configuration, monitoring and data processing services. In WLAN access point deployments, the management, monitoring, and control of a large number of wireless access points also represent a challenge. In the WLAN domain, there are a number of different approaches to administer WLAN access points as discussed in [6]. This chapter shows that today’s RFID system architectures have some similarities with the WLAN access point architectures. However, the chapter also shows that RFID systems offer a number of services that do not have equivalents in the WLAN domain, such as application-dependent data processing. In WLAN, the need for central control stems from wireless node mobility and network security, neither of which apply in the same way to RFID. We also leverage some of the concepts introduced in the WLAN domain, such as the distinction of different service sets.

10.2 RFID System Services In this section, we discuss services provided by different components within an RFID system. This includes data and device management, control, and “over-the-air” services. In the remainder of the chapter, we will show how these services are provided by different

Integrating RFID Readers in Enterprise IT

273

Table 10.1 Base service set (BSS). Service

Description

Transponder Singulation

Collects the identification numbers (ID) of (selected) transponders in range

Transponder ID Programming

Writes identification numbers to transponders

Transponder Memory Access

Reads from and writes to the general purpose memory on a transponder

Transponder Deactivation

Disables the transponder for privacy reasons

Monitoring Services Data Processing Services Base Services Configuration Services time

Figure 10.1 Timeline of RFID operation featuring base, configuration, monitoring and data processing services.

entities in different RFID system architectures. Our analysis of the RFID system services begins with the base service set (BSS) specified in the air interface communication standards. These air interface standards that specify the “over-the-air” interface between one or more transponders and a single reader [7] define the services listed in Table 10.1. Base services are in practice triggered in a number of different ways. These include external sensors, external applications, timers on the readers, the detection of certain tags, or humans who press a button on a handheld. The base service set (BSS) is supported by a number of additional services that configure, control and monitor the system components that carry out these services and process the data captured. We distinguish different service sets according to different phases of the RFID operations (Figure 10.1): Configuration services are executed before any base services are executed, monitoring services are running while base services are executed and data processing services are executed once the base services have returned captured data. In case of memory access and deactivation, the processing of the captured tag IDS can trigger subsequent base services such as tag memory access. Prior to any data capture, the networking and radio module embedded in a reader needs to be configured appropriately. The configuration service set (CSS) is responsible for setting network parameters, such as IP addresses, but also RF parameters such as transmitter power and frequency channel and air interface protocol-specific parameters, such as timing and coding parameters (Table 10.2). The configuration phase might also comprise specifying which base reader services such as tag identification or tag memory access are executed upon the appropriate triggers. Readers can also be configured to only use certain antennas and select a particular tag population over the air interface so that unwanted data does not need to be filtered out post-capture (Figure 10.2).

274

RFID Systems

Table 10.2 Configuration Service Set (CSS). Service

Description

Network Interface Configuration

Discovers and sets reader networking parameters and identity, for example, the IP address

Firmware Management

Distributes and manages firmware version on readers

Antenna, Tag Population & Memory Selection

Specifies reader antennas and tag population to be inventoried. In case of tag memory access, specifies memory fields to be accessed

Base Service Set Scheduling

Sets how different BSS services, such as tag inventory, access, and deactivation, are triggered and stopped

RF Transmitter Configuration

Sets transmission channel, hop sequence, transmitter power for readers

Air Interface Protocol-Specific Configuration

Configures timing, coding and modulation parameter of a specific air interface protocol on the readers

Inventory Process Reader

Tag

Select

Query

Ack

RN16

PC + EPC + CRC16

Figure 10.2 “Select” command defined in the EPCglobal UHF Class 1 Generation 2 Protocol (ISO 18000-6C) [7]. A particular tag population is selected with the “Select” command before the inventory process is initiated with the “query” command. Only tags matching the mask of the “Select” command, reply with a random number (RN16) and with their unique ID (EPC).

Parameters characterizing the network interface or regulatory region are configured once and are unlikely to change during the operation of the RFID reader. However, there are also some configuration parameters that will be changed frequently depending on the type of application. Examples include transmitter power adaptation for distance estimates, transmission channel changes for interference avoidance and tag population selection to avoid reading a “parked RFID tag” continuously. The result is that the frequency at which the reader configuration needs to be modified and the resulting coupling between base and configuration services are use case dependent. On the one hand, there are those applications that are unlikely to require changes to the initial configuration of the reader (Figure 10.1). After the initial configuration, the reader executes base services and asynchronously notifies the data consumers about tag data captured. Such applications allow for the separation of the reader control and data processing. On the other hand, some applications require tight coupling between base service execution and configuration services with frequent updates to the configuration (Figure 10.3). The data processing services (DPSS) include services that clean the data captured by filtering out tag IDs of no interest to applications and compute aggregates over the tag data captured (Table 10.3). This includes aggregates in the time domain, where entry and exit

Integrating RFID Readers in Enterprise IT

275

Monitoring Services Data Processing Services Base Services Configuration Services time

Figure 10.3 Timeline of RFID operation featuring base, configuration, monitoring and data processing services with frequent changes to the reader configuration.

Table 10.3 Data Processing Service Set (DPSS). Service

Description

Filtering

Removes unwanted tag identifiers from the set of tag identifiers captured, for example, based on the product type or manufacturer encoded in the identifier

Aggregation

Computes aggregates in the time domain (entry/exit events) and the space domain (across reader antennas and readers) and generates the corresponding “super”-events

Identifier Translation

Translates between different representation of the identifier such as from the raw tag object identifier in hexadecimal format to the EPC in URN notation

Persistent Storage

Stores RFID data captured for future application requests

Reliable Messaging

Allows RFID data to be delivered reliably in the presence of software component, system and network failures

Location/Movement Estimation

Detects false positive reads of far-away tags that are outside the “typical” read range and estimates the direction of movement

Application Logic Execution

Interprets the RFID data captured in an application context and generates the corresponding application events, for example, whether a shipment is complete

of tags in the read range are determined, and aggregates in the space domain, where the data captured across multiple reader antennas or even readers is computed (Figure 10.4). Since there is frequently some uncertainty about the true location and movement of an RFID transponder relative to the reader, there are additional services that eliminate socalled “false positive” reads and that possibly even estimate the movement of RFID transponders. Other services include tag data translation and persistent storage. Messages exchanged between different services in an RFID system travel over a number of communication nodes. This means that some messages may be lost in transit. Additionally, it is possible that either the recipient’s or the sender’s system will fail while a message is in transit, leaving the overall RFID system in a state of confusion as to whether a given message carrying tag data has been processed or not. Reliable messaging protocols provide guaranteed end-to-end delivery of RFID data. Reliable messaging refers to the ability to deliver a message once and only once to its intended receiver, to deliver messages in order, and to inform the sender and receiver about the failure to deliver a message.

276

RFID Systems

Dock Door Dock Door Reader 1 EPC Timestamp 61.43.10 10:28.22 11.49.40 10:28.23 61.43.15 10:28.24 61.43.10 10:28.24 58.49.20 10:40.12 58.49.28 10:40.12 11.49.40 10:40.13 58.49.28 10:40.13

Reader 2 EPC Timestamp 61.43.12 10:28.23 61.43.15 10:28.25 11.49.40 10:28.25 61.43.11 10:28.25 58.49.25 10:40.11 58.49.25 10:40.12 58.49.20 10:40.13 11.49.40 10:40.14

Filter & Aggregate

Logical Reader EPC Timestamp 61.43.10 10:28.22 61.43.12 10:28.23 61.43.15 10:28.24 61.43.11 10:28.25 58.49.25 10:40.11 58.49.20 10:40.12 58.49.28 10:40.12

EPC example format: CompanyPrefix.ItemReference.SerialNumber, for example, 61.64.28

Figure 10.4 Filtering and aggregation of RFID data: data from two dock door readers are combined, duplicate EPCs are eliminated, EPC 11.49.40 is filtered out, and quantities of product categories are calculated. Eliminated “reads” are shown in grey.

Table 10.4 Monitoring Service Set (MSS). Service

Description

Network Connection Monitoring

Checks network connection between different RFID system components

RF Environment Monitoring

Checks RF noise and interference levels

Reader Monitoring

Checks that the reader is up and running and executing BSS as configured

The DPSS also includes services that interpret the RFID data captured in an application context to generate the corresponding application events. For a supply chain application, this might include matching the detected tag identifiers against a list of identifiers in an electronic advance shipping notice. The result of the data interpretation can be the generation of a business event such as “Shipment complete” to an enterprise resource planning system or a immediate feedback to a local staff via a display. Application logic is typically executed after other DPSS services such as filtering, aggregation and tag identifier translation have preprocessed the captured RFID data. While filtering, aggregation and tag data translation functionality typically use predefined operators, data interpretation often relies on custom-developed application logic that processes the incoming RFID data. This results from the significantly broader scope of this service when compared to aggregation or filtering. However, there have been a number of (commercial) efforts to define standard workflows for typical RFID applications such as dock door receiving. While standardized workflows can reduce the amount of

Integrating RFID Readers in Enterprise IT

277

custom application software development required for each deployment, variations from the standardized workflow in real-world processes typically still result in customization and additional software development. It is worthwhile mentioning that the “filtering” can also be performed by limiting data capture to a subset of reader and reader antennas in the first place and by selecting a specific tag population (Table 10.2 and Figure 10.2). The filtering is then effectively carried out over the air interface by configuring the reader appropriately. In ISO 180006C, this is achieved by executing one or more “Select” commands before an inventory round is initiated with a “Query” command. Even aggregation in the time domain can be performed over the air interface using advanced features in air interface protocols. In ISO 18000-6C, there are a number of different inventory flags that allow the user to minimize redundant identifications of the same tag. In many applications, RFID tags are still identified multiple times while they are in the read range due to multipath effect. As a result entry/exit aggregates need to be computed in software. Since tags might miss a “Select” command and only receive the subsequent “Query” command, “filtering” is typically also performed in software. Monitoring services (MSS) observe the health of the reader, the RF environment and the network connection to the reader (Table 10.4). This is carried out via heartbeat messages exchanged between reader and monitor to detect network failures and via the monitoring of antenna status, memory overflows and reboot alarms. The monitoring services also receive information about the current RF noise floor and possible RF interference.

10.3 Reader Capabilities Before we present different RFID system architectures that provide the services listed in the previous sections, we discuss the different RFID reader categories available on the market today. This is important because system architectures depend heavily on the type of readers deployed. A typical reader is comprised of a radio module, a general purpose computing module, a network interface, and general input/output pins. The general purpose computing unit can be a low-end microcontroller or an embedded processor with significant computing resources. The general input/output pins are used to interface with local sensors and actuators. All readers provide the base service set mentioned earlier. The majority of RFID readers currently available provide some limited data processing services (DPSS), such as pre-defined filters and aggregates over the RFID data captured. This includes the post-capture elimination of redundant reads and the accumulation of tag reporting across antennas [8]. Readers also compute entry/exit aggregates [10]. Some RFID reader products also provide limited persistent storage space so no data is lost during a communication failure with the backend IT systems. RFID readers with significant computing resources execute application code on the reader platform [8, 9, 11]. The result is that the RFID reader can independently control all local interaction via with sensors and displays. The reader only transmits the application dependent high-level events that result from the data processing. The reader verifies, for example, a shipment against an advance shipping notice and sends a “shipment complete” event to the enterprise resource planning system. Some reader vendors only allow preferred partners to execute application code on the RFID reader.

278

RFID Systems

Most RFID readers allow users to configure RF transmitter settings, network interface, and antenna and tag selection parameters (Table 10.2). Base service scheduling without network access, such as the immediate writing to memory upon seeing a particular tag ID, is typically only found on RFID readers that allow users to run custom application code on the reader or those that support the EPCglobal Low Level Reader Protocol.

10.4 RFID System Architecture Taxonomy In this section, we present different RFID system architectures that are currently deployed throughout the industry. The analysis is based on interviews with companies which have installed RFID systems. Based on our analysis, we distinguish two different architecture types: An autonomous and a centralized architecture (Figures 10.5 and 10.6). In practice, there are also a number of hybrid architectures that feature elements of both architecture types. The deployment diagrams that illustrate these architecture types have a number of different dimensions (Figure 10.5). Each box in the deployment diagram represents a separate device. Each device has a number of services which are grouped into base, configuration, monitoring, and data processing services. The deployment diagrams also show the communication link type (LAN/WAN) between different devices and group devices which are hosted within the same facility such as a store or distribution center.

Figure 10.5 Centralized Architecture (with and without controller).

Integrating RFID Readers in Enterprise IT

Enterprise Network/System Management Configuration (CSS) • Network Interface • Firmware Mgmt. • Antenna & Tag Population Selection • BSS Scheduling • RF Transmitter • Air Interface Prot.

Monitoring (MSS) • Network Connection • RF Environment • Reader Health

279

Other Enterprise IT Systems/User Interface

WAN

WAN Facility

Reader

Reader

Data Processing (DPSS) • Filtering • Aggregation • Identifier Translation • Location/Movement Est. • Persistent Storage • Reliable Messaging • Application Logic Exec.

Data Processing (DPSS) • Filtering • Aggregation • Identifier Translation • Location/Movement Est. • Persistent Storage • Reliable Messaging • Application Logic Exec.

BSS • Tag Singulation • Tag ID Programming • Tag Memory Access • Tag Deactivation

BSS • Tag Singulation • Tag ID Programming • Tag Memory Access • Tag Deactivation

Figure 10.6 Autonomous Architecture.

Figure 10.5 shows an architecture with a dedicated controlling device at each facility, where one or more RFID readers are deployed. We call this the centralized architecture because a central device provides CSS, MSS and the majority of the DPSS services. Existing enterprise IT monitoring systems do not monitor reader devices directly. Figure 10.5 presents two variations of this architecture type. Figure 10.5(a) features a separate application server2 and a controller. The controller provides the CSS and MSS services and application-agnostic DPSS services such as persistent storage of tag reads, identifier 2

The term “Application Server” refers here to a server that provides a runtime environment for the application. While the term “application server” is often associated with enterprise-class servers that implement the Java Enterprise Edition or Microsoft.NET framework, application servers hosted on-site in RFID systems often feature only a more lightweight execution environment. In an RFID context, these servers are often also referred to as “Edge Servers.”

280

RFID Systems

translation and aggregation across multiple readers. The application server hosts customized application software that processes the captured RFID data in a business context. In Figure 10.5(b), these data processing services are deployed on the same device. In both cases, the RFID readers provide base services, such as tag identification and memory access, and limited data processing services. The latter includes the elimination of redundant reads by computing entry/exit events and the aggregation of tag reads across different antennas. In the autonomous architecture, there is no local controller, but the readers operate autonomously once configured appropriately (Figure 10.6). Extensive data processing takes place on the RFID readers themselves, where a custom application code processes the captured tag data. The readers send locally computed business events such as “shipment complete” to the enterprise information systems. Before the captured RFID data is processed in a business context, the data is typically filtered and aggregated and tag identifiers are translated. To deal with network and system failures, there is a need to provide reliable messaging and persistent storage services. The readers are monitored and also configured via enterprise system and network management tools. Each of the two architecture types has its own strengths and weaknesses with respect to performance, ease of maintenance and cost. The most suitable system architecture is thus dependent on the specific application and enterprise IT organization. Installations with hundreds of readers in the same facility typically favor a centralized architecture. The deployment of isolated readers in remote locations or in applications with significant local interactions with staff benefit from the autonomous architecture. In practice, many hybrids of the architectures types presented here are typically deployed. This includes autonomous architectures for readers in remote locations and centralized architectures in facilities with a large number of RFID readers deployed by the same organization. There are also cases where the RFID readers are monitored remotely, but the data processing is carried out on a local server/controller.

10.5 EPCglobal Standards The EPCglobal Architecture Framework defines a number of roles: RFID reader, filtering&collection, reader management and EPCIS capture applications [5]. In the “centralized with controller” architecture we presented above, each of the roles maps to one individual device in the system architecture except for the controller that implements both the filtering&collection and reader management role (Figure 10.7(a)). In the variation of the centralized architecture without a separate controller device (Figure 10.7(b)), the filtering&collection, reader management and EPCIS capture application role are combined on the same device. In the autonomous architecture, the RFID reader, filtering&collection and EPCIS capture applications roles of the EPCglobal Architecture Framework are implemented on the RFID reader (Figure 10.7(c)). Each of the different roles in the EPCglobal architecture framework is associated with one or more software specifications that standardize interfaces. In the following subsections, we discuss each of these specifications in the context of the service taxonomy presented earlier.

Integrating RFID Readers in Enterprise IT

281

Other Enterprise IT Systems/User Interface

Other Enterprise IT Systems/User Interface

WAN

WAN

Facility

Facility Application Server

Application Server

EPCIS Capture Application Role

EPCIS Capture Application Role

LAN Controller Appliance

Filtering & Collection Role

Filtering & Collection Role

Reader Mngt. Role

Reader Mngt. Role

LAN Reader

RFID Reader Role

LAN

Reader

Reader

Reader

RFID Reader Role

RFID Reader Role

(a) Centralized with Controller

RFID Reader Role

(b) Centralized without Controller

Enterprise Network/System Management

Other Enterprise IT Systems/User Interface

Reader Management Role

Facility WAN

Reader

Reader

EPCIS Capture Application Role

EPCIS Capture Application Role

Filtering & Collection Role

Filtering & Collection Role

RFID Reader Role

RFID Reader Role

(c) Autonomous

Figure 10.7 Centralized and Autonomous Architectures with associated EPCglobal Architecture Framework Roles [5].

282

RFID Systems

10.5.1 Discovery, Configuration and Initialization (DCI) and Reader Management (RM) The EPCglobal DCI specification [12] supports the CSS services of network interface configuration and firmware management (Figure 10.8). The EPCglobal RM protocol [13] supports MSS services3 (Figure 10.8). RM does not reveal any air interface protocol specific statistics, but reports noise levels, transmitter power and failed tag memory access and deactivation operations. RM supports the SNMP protocol and thus facilitates the integration of RFID readers into existing enterprise system monitoring tools which rely on the SNMP protocol.

10.5.2 Low Level Reader Protocol (LLRP)

ALE

RP

LLRP

RM

DCI

In the LLRP specification [14], there is extensive support for the CSS and MSS services listed in Table 10.2 and 10.4. This includes the configuration of air interface parameters, RF transmitter settings, base service set scheduling, and antenna, tag population and tag memory selection (Figure 10.8). The LLRP specification also provides limited data processing capabilities. The Accumulation of TagReportData in the LLRP specification allows for the collection of tag data across multiple reader antennas. The computation of entry and exit event aggregates is not supported. There is no support for the software filtering of tag IDs, but clients can specify the target tag populations that are identified over the air interface using the “Select” command defined in ISO 18000-6C.

CSS Network Interface Configuration Firmware Management Antenna & Tag Population Selection Base Service Set Scheduling RF Transmitter Configuration Air Interface Prot. Configuration MSS Network Connection Monitoring RF Environment Monitoring Reader Monitoring DPSS Aggregation Filtering Identifier Translation Reliable Messaging Persistent Storage Location Movement/Estimation Application Logic Execution

Figure 10.8 High-level overview of the services supported by the EPCglobal specifications DCI, RM, LLRP, RP and ALE. 3

In addition to the support for monitoring services, the EPCglobal RM Specification also allows hosts to selectively switch off reader antennas via the ReadPoint.setAdminStatus method. Strictly speaking, the latter represents a configuration service according to our service taxonomy.

Integrating RFID Readers in Enterprise IT

283

Tag Data Reporting Configuration

Device DataProcessing Configuration Monitoring

LLRP Client

Reader

LLRP Impl. Base Services

(a) Single channel

Device

Device Configuration

LLRP Client

Monitoring

LLRP Client

Reader Data Processing

LLRP Impl. Base Services

(b) Multiple reporting channels

Figure 10.9 Simultaneous notifications to multiple data consumer are not supported by today’s LLRP specification.

LLRP is well suited to standardize the reader interface in the centralized architectures. LLRP allows controllers and software deployed on local application servers to schedule base services and to optimize performance by adjusting the RF transmitter settings and air interface parameters. The LLRP specification envisions only a single connection to a client at a time (Figure 10.9(a)). This allows for one device configuring the reader and subsequently for another device to connect to the reader to receive and process the TagReportData notifications. It does not permit sending the TagReportData to multiple data consumers (cf. Figure 10.9(b)). Thus, LLRP by itself cannot be used to support a deployment scenario where applications on the reader process the data captured and remote system monitoring services receive notifications to monitor the health of the reader and connection. However, LLRP can be operated in conjunction with RM. This provides SNMP access to the reader status and statistics on successful and failed base service operations for remote system monitors. LLRP can also generate significant network traffic since there is no way to restrict TagReportData messages to tags entering and exiting the range of the reader when low notification latencies are desired. The latter is not an issue for local area connections, but can be problematic for wide area connections. The binary communication protocol allows for efficient implementations on RFID readers, but requires additional high-level tools for application developers. LLRP features an extension mechanism that allows reader vendors to define their own proprietary extensions that expose custom feature sets. In our opinion, this is essential since RFID reader vendors will only consider substituting their proprietary reader

284

RFID Systems

protocols for a standardized protocol if the latter allows for extensions supporting proprietary features which are not part of the standard feature set. While LLRP in its current version only features support for the ISO 18000-6C air interface, the support for other air interface protocols is envisioned in the future generation of the specification.

10.5.3 Reader Protocol (RP) The EPCglobal RP specification [15] takes a different approach to interface with a reader in two important aspects when compared to LLRP. While LLRP comprises CSS, MSS and limited DPSS services in one protocol, RP focuses on DPSS functionality (and limited CSS services) with the complementary EPCglobal specification RM supporting MSS services (Figure 10.8). The second major difference is that the combination of RP and RM does not allow CSS services to have access to RF parameters and air interface protocol settings of the reader4 – the protocols are often referred to as “air interface protocol unaware.” This represents a deliberate design choice. The benefit of this approach is that application developers are shielded from the details of RFID operation. Software application development is facilitated by providing a standardized high-level RFID data reporting interface. Another benefit of this approach is that there is no need to reach consensus on the configuration and control features to support in a standardized protocol and possibly across different air interface protocols. An HF reader vendor can continue to use his own proprietary reader protocol to configure its readers. The tag data reporting takes place via a “high-level” reader protocol and is identical to the tag data reporting of UHF Gen2 or ISO 18000-6B readers which also use the EPCglobal RP to report captured data. The drawback of this design choice is that the reader cannot be controlled and configured to the extent possible with today’s proprietary reader protocols or with the recently released LLRP. It is not possible to select a particular air interface protocol or an interference avoidance feature such as dense reader mode remotely. It is also not possible to select a particular frequency channel to minimize reader collisions. All of these represent standard features offered by today’s vendor-specific reader protocols. Since access to these features is necessary at least during the initial deployment, RP requires the use of another vendor-specific protocol to configure the reader appropriately (Figure 10.10). In some applications, the configuration via another “air interface aware” (proprietary) protocol is only required at the time of deployment with no changes during operation. In other applications, RF interference or changes in tag populations require frequent changes to the reader configuration to maintain optimum performance. This results in frequent use of the additional “air interface aware” (proprietary) protocols. RP supports DPSS services that are not included in the LLRP specification. RP allows for the computation of entry and exit events, different tag ID representations and tag report notifications to one or more data consumers. RP also provides an XML message binding that facilitates software application development. LLRP relies on a binary protocol that requires fewer resources on RFID readers. RP supports reliable messaging for the asynchronous notification channels. 4

RP only allows for the CSS services of antenna and tag population selection and limited base service scheduling. Via RP, clients can schedule inventory ID operations, but memory access operations cannot be scheduled and need to be triggered remotely.

Integrating RFID Readers in Enterprise IT

285

Tag Data Reporting Configuration

Device Data Processing Configuration Monitoring

Client

Reader High Level RP

Proprietary RP Base Services

Figure 10.10 High-level reader protocols such as EPCglobal RP and ALE require an additional (proprietary) protocol to configure the reader.

Nearly all of the DPSS services in RP are optional in the specification. This design choice seems to be a direct result of the heterogeneity of the reader landscape. Reader devices with significant computing resources were envisioned to provide the majority of the optional features. Low-end reader devices would only support the mandatory features and possibly a small number of the optional features. However, this design choice also makes application development with RP a challenge. Different reader types will likely support a different combination of optional features.

10.5.4 Application Level Event (ALE) The EPCglobal ALE 1.0/1.1 [16, 17] specifications provide a standardized interface to application-agnostic DPSS services. ALE also supports the selection of readers, reader antennas and tag populations, which represents a CSS service in our taxonomy. ALE 1.0 comprises a feature set similar to the EPCglobal RP specification: filtering, aggregation, tag identifier translation, buffering of messages and notifications to multiple consumers, but the actual implementation is different. ALE relies on a web service (SOAP) transport protocol and provides a convenient “subscribe” mechanism where clients can register their standing queries/notifications with a single command. In RP, this requires a sequence of individual messages. ALE 1.0 does not support tag user memory. ALE 1.1 represents a major advance over ALE 1.0. ALE 1.1 supports reading and writing to user memory on tags. ALE 1.1 also provides new interfaces for defining tag memory fields, for reader/reader antenna to location mapping and for access control. In LLRP and ALE 1.1, the primary interaction between client and implementation for both tag identification and memory access are similar. For tag identification, the

286

RFID Systems

ALE/LLRP client provides the implementation with a specification (ECSpec in ALE/ ROSpec in LLRP) that defines boundary conditions, channels to be used and the desired content and structure of the asynchronous reports (Figure 10.11). The implementation executes this specification, captures the RFID data and responds by returning the information in the reports as requested (Figure 10.12). For the tag memory access, the primary interaction sequence is also similar for ALE 1.1 and LLRP. The client transmits a specification (ECSpec for reading/CCSpec for writing in ALE/AccessSpec in LLRP) (Figure 10.13). The ALE/LLRP implementations respond by carrying out the memory access operations on the tags and return reports that describe which memory access operations were performed. While the basic interaction is similar, there are a number of conceptual differences between the LLRP and ALE 1.1 specification. LLRP allows the client to optionally specify air interface protocol and RF transmitter settings in ROSpec and AccessSpec (Figure 10.11). The only mandatory air interface protocol parameter is the air protocol to be used. The ALE specification abstracts from these low level settings and does not expose them. The ALE 1.1 specification uses “high-level” representations for tag identifiers, such as the URI representations of EPCs defined in the EPCglobal Tag Data Standard [18]. In LLRP, EPCs are represented as bit arrays (Figure 10.12). ALE 1.1 includes the “logical reader API” which decouples the identity of reader devices and antennas from the names of the channels used in ALE subscriptions and reports. This permits the replacement of an RFID reader or a change of networking parameters without the need to update the application software. As mentioned earlier, ALE also provides additional data processing services, such as count and entry/exit aggregates, which are not available in LLRP (Figure 10.12). For tag user memory, ALE 1.1 provides predefined memory field names for elements specified in the EPCglobal Tag Data Standard [18] and support for ISO 15962, both of which facilitate the programming of memory access operations. In LLRP, tag user memory needs to be addressed using memory banks, pointers and length. Masks need to be specified as bit arrays (Figure 10.13). An ALE implementation can also service multiple data consumers simultaneously and disseminate captured tag data to them. LLRP is a network protocol that supports only a single established connection at a time. It does not support multiple clients with different “subscriptions.” The main use case of the EPCglobal ALE specification is on controllers and application servers in the centralized architectures as an application agnostic interface (Figure 10.14), where ALE exposes all those DPSS services that are common across different RFID applications such as filtering, aggregation and tag identifier translation. The ALE interface also allows clients to select readers, reader antennas and tag populations and define tag memory operations. ALE does not provide access to RF transmitter and air protocol settings. At the interface to software and hardware controllers, the control and data processing can often be separated and it is appropriate to abstract from the RFID operational settings. There is no need to expose air protocol settings because the controller implementation is responsible for RFID performance optimization. ALE can also be deployed on an RFID reader as a “high-level” reader protocol with the similar benefits and weaknesses as mentioned in the subsection on RP. ALE provides a high-level starting point that is well suited for writing application logic, freeing the developer from the kind of low-level programming that would be necessary to code directly to LLRP. The application developer benefits in particular from the high-level tag

Integrating RFID Readers in Enterprise IT

287

http://example.com/trigger1 20000 http://example.com/trigger2 3000

Start/Stop Conditions Both LLRP and ALE 1.1 provide a number of different start and stop trigger mechanisms.

dock_1

Channels In ALE 1.1, multiple physical readers and reader antennas can be mapped to a single logical reader.

urn:epc:pat:sgtin-96:X.X.X.*

Report Formatting Both LLRP and ALE 1.1 allow the subscriber to define the content of the asynchronous tag reports. ALE supports a number of different aggregates: entry/exits, count and grouping. ALE 1.1 is air protocol unaware and does not expose air protocol or RF settings.



(a) ALE1.1 ... Periodic ...... ........

Start/Stop Conditions Both LLRP and ALE 1.1 provide a number of different start and stop trigger mechanisms.

1 2 ...

Channels

EPCGlobalClass1Gen2 1 200

Air Interface and RF Configuration

Upon_N_Tags_Or_End_Of_ROSpec ..... true ....

Report Formatting

In LLRP, the AISpec defines which antennas are used

LLRP allows users to specify air interface parameters and RF transmitter settings such as the transmission power.

Both LLRP and ALE 1.1 allow the subscriber to define the content of the asynchronous tag reports.



(b) LLRP

Figure 10.11 Specification of tag identification in ALE 1.1 and LLRP. The LLRP ADD_ROSPEC message is represented in the LLRP LTK XML format (www.llrp.org).

288

RFID Systems





urn:epc:id:sgtin:0614141.112345.3 urn:epc:id:sgtin:0614141.112345.4 0 && done = false) do 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23.

#sort readers by decreasing covered tag count sort(R); candidate := R[1]; R.removeFirst(); candidate_tags := candidate.getCoveredTags(); for each t in candidate_tags do mapped[t] := candidate; for i = 1 to R.size do if R[i].covers(t) then R[i].remove(t); fi

od od done := true;

for i = 1 to mapped.size do if mapped[i] = -1 then done := false; fi od od

334

RFID Systems

12.8 Redundant Reader Elimination Algorithms: A Centralized Heuristic In the following we study several redundant reader elimination algorithms for the optimal tag coverage problem. The algorithms work by identifying redundant readers and provide a mapping between each tag to a single reader. The mapping specifies which reader needs to cover and report each tag. Readers that are not present in any such mapping are redundant and do not need to report any tag. In Section 12.7.2 we have shown the optimal tag coverage problem to be NP-hard even in a centralized setting, where a single host has knowledge about all readers, including all the tags each reader covers. Before studying distributed solutions for the tag coverage problem, we briefly focus on a greedy heuristic in the centralized setting. Our greedy approach is based on the observation that by trying to keep active the readers covering more tags, it is expected that more readers will be detected to be redundant. Algorithm 2 shows the pseudo-code for a central host implementation of the greedy heuristic using an Orca-like [16] syntax. The central host has knowledge of all the readers (line 2), stored as an array of unique reader identifiers. The central host also maintains a mapping between each tag in the system, to one of the readers (line 3). The mapped structure is indexed by tag identifiers, assumed for the sake of simplicity to be contiguous integers ranging from 1 to the total number of tags. The run operation (lines 5–21) implements the greedy approach. Initially, each entry in the mapped structure stores −1 for each tag (line 6), signifying that the tag is not covered (mapped) by any reader. The operation proceeds iteratively, by sorting first all the readers in the system in decreasing order of the number of tags they are currently covering (line 9). The reader with the highest tag cover count is then selected and removed from the list of readers (lines 10–11). The tags covered by this candidate reader (line 12) are mapped (in the mapped structure) to the candidate reader (line 14). All the tags mapped to the candidate reader are then removed from the coverage sets of all the remaining readers in the system (lines 15–17). The operation then scans the mapped structure in search of unmapped tags (lines 18–21). If no tags are left unmapped, the operation stops (line 8). Otherwise, it iterates on the remaining readers, until it runs out of readers or all tags are mapped (line 8). At the end of the operation, all the readers left in the R structure are redundant. The ones selected are a solution and need to be active in order to report the tags mapping to them.

Example 12.8.1 Consider the system shown in Figure 12.4, consisting of 4 readers and 10 tags. Initially, reader R1 covers the most tags, 6. greedy selects R1 to be active, then removes 4 of the tags that are covered by R1 from the coverage set of R2 . While initially R2 was covering 5 tags, at this point it only covers 1. Since R4 covers more tags, 3, it is selected next by greedy. Then, tag T is also removed from R2 ’s coverage, since it was mapped to R4 . Since R2 now covers no tag, reader R3 , covering one tag, is next selected by greedy. Following this step, greedy runs out of unmapped tags, exits the while loop and stops. Since reader R2 was not selected, it is declared to be redundant.

Optimal Tag Coverage and Tag Report Elimination

335

T4 T3

R4

R4 T2

R2

T

R1

R3

T1

R2

R1

R3

(a)

(b)

Figure 12.4 (a) RFID system where greedy provides an optimal tag coverage solution, by detecting reader R2 to be redundant. (b) RFID system where rre provides an optimal tag coverage solution.

12.8.1 Analysis We can prove that greedy satisfies the first condition from Definition 12.7.2. That is, we need to show that for any tag T in the system, greedy assigns a reader. In our RFID system definition from Section 12.7.1 we assume that any tag is covered by at least one reader. Then, the only time when greedy may exit the while loop (line 8) is when it provides a fully assigned mapped structure. Let T be a tag, covered by the readers from the set p(T ). The first time greedy selects one of the readers from p(T ) as a candidate reader, T will be mapped to that reader. If greedy never selects a reader from p(T ), tag T will remain uncovered and the tag count of the readers from p(T ) will be at least one. Then, when greedy runs out of readers with tag count larger than one, it will need to select at least one reader from p(T ). Note, however, that greedy does not satisfy the second condition of Definition 12.7.2. This should certainly be true, otherwise greedy would be a solution to a NP-hard problem. For a counter-example, consider the system in Figure 12.3(b). The central reader, covering 4 tags is selected first by greedy. This is followed then by each of the remaining readers, each covering an additional tag not covered by anyone else. That is, greedy selects all the readers to be active. However, it is easy to see that the central reader is redundant and its presence is not needed to cover and report the tags.

12.9 RRE: A Distributed Solution greedy assumes global knowledge of the RFID system. Since this information may not always be available, we study a distributed and localized alternative. The solution, called rre, assumes no centralized server and does not require direct communication between readers. rre is deterministic in nature, but it relies on randomization to avoid reader

336

RFID Systems

collisions. The randomized reader collision avoidance mechanism used by rre is similar to the one used by RCA (see Section 12.4). We first present rre as an algorithm for the optimal tag coverage problem. Subsequently, we show that the idea behind rre can be generalized to obtain a variant, rre-hc, which provides a solution to the tag reporting problem based on the hop count distance between readers and the sink node. This variant of rre can be used in ad hoc reader networks (see Section 12.2).

12.9.1 RRE The distributed rre algorithm assumes initially that RCA (see Section 12.4) has been previously executed by all readers to identify the tags in their vicinity. Later in this section, we discuss a simple modification to rre to relax this synchronization assumption. In rre, each tag is marked with a value representing the highest number of tags covered by any reader in the tag’s vicinity (that is, any reader whose interrogation zone contains the tag). This is done by each reader writing its tag count on its tags and by requiring tags to overwrite locally stored values only with higher ones. The reader that issues the highest count for a tag, holds the tag. Therefore, rather than a centralized mechanism that chooses the reader with maximum tag coverage, each tag chooses the locally optimal reader in its vicinity. A reader holding none of its covered tags is declared redundant. Furthermore, a reader reports only the tags that it holds. rre consists of two steps. In the first step, each reader attempts to write its tag count (number of covered tags) on each of its covered tags. A tag only stores the highest value seen, along with the identity of the corresponding reader. For this, each reader issues a write command containing its unique reader identifier and its tag count. Similar to RCA, the write operation is performed during O(log ψ) consecutive epochs, once per epoch, where ψ is the total number of readers. During each epoch, the time frame for sending the write request is randomly chosen. As mentioned in Section 12.4, this process ensures w.h.p. that at least one write command issued by each reader is correctly received by all its covered tags. Thus, after O(log ψ) epochs, each tag stores the largest number of tags covered by a reader situated in its vicinity, along with the identity of that reader, called the holder of the tag. In the second step, a reader queries each of the tags in its interrogation zone and reads the identity of the tag’s holder. Each read query issued by a reader for each of its tags is similarly repeated during random time frames for O(log ψ) consecutive time epochs to avoid reader collisions occurring at queried tags. At the end of this step, a reader that holds at least one tag is responsible for monitoring the tag and will have to remain active. However, a reader holding no tag can safely turn off its RFID interface. This is because all the tags covered by that reader are already covered by other readers that will stay active. 12.9.1.1 Implementation Algorithm 3 shows the pseudocode for RRE-TC using an Orca-like [16] syntax. The functionality of a writable tag is shown in operation run of WritableRFIDTag (lines 4–13). When a writable tag receives a write command containing the identifier of the reader issuing the command and its tag count, it saves the values only if the tag count

Optimal Tag Coverage and Tag Report Elimination

337

is larger than the value currently stored. When the command received is a read, the tag returns a packet containing its identifier followed by the reader’s identifier and count value stored locally. The detection of redundant readers is exhibited in operation isRedundant of RFIDReader (lines 18–39). First, a reader selects a random time frame during e consecutive epochs, and sends a broadcast write command containing its identifier and tag count (lines 19–24). Subsequently, it queries each of its covered tags, using a read command, for e consecutive time epochs in order to find the tag’s holder (lines 25–37). Note that after sending a read command, at the chosen time frame, the reader waits either to receive a reply from the queried tag or for the epoch to end (lines 31–35). Example 12.9.1 We illustrate the operation of RRE using the example systems shown in Figure 12.4. For the system shown in Figure 12.4(b), RRE finds an optimal tag coverage solution. Since from all the readers that cover tag T1 , R4 has the highest tag count, R4 will hold T1 . Similarly, tag T2 will be mapped to reader R1 , since R1 covers 3 tags but R2 covers only 2. This effectively makes R2 redundant, since both its tags will be reported by other readers. However, RRE does not find an optimal tag coverage solution for the system shown in Figure 12.4(a). This is because even though R2 is redundant, in RRE, tag T will be mapped to R2 , since R2 covers 5 tags, more than any other reader covering tag T . Algorithm 3 Pseudocode for RRE-TC. Reader and writable tag operation for providing a solution to the tag coverage and optimal tag reporting problem. 1. Object implementation WritableRFIDTag; 2. Rid : integer; #identifier of locking reader 3. count = 0: integer; #count of highest bidder 4. Operation run() 5. guard inQ.first.type = write do 6. if inQ.first.c > count then Rid := inQ.first.rid; 7. 8. count := inQ.first.c; 9. fi; 10. guard inQ.first.type = read do bCast(new packet(Tid ,Rid ,count)); 11. 12. od 13. end 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24.

Object implementation RFIDReader; Rid : integer; #reader identifier tags: array[integer] of integer; #covered tags redundant = true: boolean; #is reader redundant? Operation isRedundant(prefix:integer) #first step of RRE while count++ < e do frame:=getRandom(0,n); sleep(frame); bCast(new packet(write, Rid ,tags.size)); sleep(n-frame-1);

od

338

RFID Systems

#second step of RRE

25. for i in 1..tags.size do 26. while count++ < e do 27. T=getTime(); 28. frame:=getRandom(0,n); 29. sleep(frame); 30. bCast(new packet(read,tags[i])); 31. guard inQ.first.tid = tags[i] do 32. if inQ.rid == Rid then 33. redundant := false; 34. od 35. guard getTime() - T > n do od 36. od 37. od 38. if redundant = true do turnOff(); fi 39. end

12.9.2 RRE-HC We now briefly describe a variant of the distributed redundant reader elimination algorithm, where the tag mapping decision is made based on the distance in hops between readers and the sink node. The variant, named rre-hc, assumes that each reader maintains its hop count distance to the sink node. This could be achieved by each node maintaining a parent node and each node, starting with the sink and advertising its hop count distance to the sink to each of its neighbors. The operation of rre-hc is very similar to the operation of rre. It also runs in two steps. In the first step, a reader issues a write command to each tag in its interrogation zone. The command contains the reader’s identifier and its hop count distance to the sink node. A tag will only overwrite the stored hop count with a strictly smaller value. In the second step, the reader interrogates all its tags to determine the reader identifier that each of them stores. The reader is redundant only if none of its tags stores its identifier. Example 12.9.2 We illustrate the outcome of rre-hc using the reader network from Figure 12.2(b). If reader R1 , with a hop count of 3, is the first to issue a write command, it will hold the tag T2 . If later, R2 issues a similar write command, its identifier and hop count (also 3), will not overwrite the values stored on T2 . However, when later, R5 , with a hop count of 2, issues a write command, its identifier and hop count will overwrite the values stored on T2 . Then, T2 will be reported by R5 and will cross only 2 hops, instead of the 3 it would have crossed if R1 or R2 would have held it. Similarly, tag T1 will also be mapped to and reported by R5 and not by R2 .

12.9.3 Analysis We provide a proof of the accuracy of rre and rre-hc, and analyze their performance. The following theorem establishes the accuracy of the rre variants in the absence of reader failures or arrival of new tags, assuming that readers are synchronized. We elaborate on handling system update and synchronization issues in the next section.

Optimal Tag Coverage and Tag Report Elimination

339

Theorem 12.9.1 When m(R) is defined to be the hop count distance between R and the sink node, rre-hc is an optimal solution to the tag reporting problem for ad hoc reader networks. Proof . We need to show that each covered tag T is mapped to exactly one reader and that reader is the one with the smallest hop count among all the readers covering T . It is straightforward to see that the first condition holds. For the second condition, if the write command from each reader from p(T ) reaches T , T will only store the smallest hop count. With high probability, each write command will reach tag T . This is because a write command is issued in a random chosen slot during e consecutive time frames. Lemma 12.9.2 Complexity of RRE. The total number of epochs in both RRE variants, TRRE is O(γ eβ log ψ).

Proof . The first step of RRE, where each reader sends a write command to all its tags, takes e log ψ epochs. The second step, where readers send queries to each of their tags, takes γ eβ log ψ epochs. Thus, TRRE = O(γ eβ log ψ). Lemma 12.9.3 Write Complexity of RRE. The total number of tag write operations performed by a single run of either variant of RRE in an RFID system is O(ψγ ). Proof . In the worst case, each tag T is written to once by each of the readers in the set p(T ) of readers covering it. This is because from the set p(T ), the reader holding the least number of tags may be the first to issue a write command, followed by second ranked and so on. Thus, in the worst case, each tag may be written to O(ψ) times, which proves the theorem. In the following we study two extensions to the distributed RRE heuristic. For simplicity, we focus only on RRE, since the results can similarly be applied to RRE-HC.

12.9.4 Dependency on RCA We have assumed until now that before running rre, each reader has already executed RCA, detecting all the tags in its interrogation zone. This assumption ensures that on completion of the first step of rre, tags placed in the vicinity of at least two readers store the highest number of tags covered by the readers. However, this may not always be the case and the following example shows how this can affect the correctness of rre. Example 12.9.3 Consider the RFID system from Figure 12.4(b). The count value stored at the completion of rre on tag T2 should be 3, from reader R1 . However, if we assume that initially, readers are not aware of the identity of adjacent tags and RCA needs to be executed just before rre, the following scenario may occur: since R2 only covers two tags, whereas R1 covers three, R2 will complete RCA and also the first step of rre before R1 . Then, R2 , upon identifying itself to be the holder of T2 , will also decide to stay active, despite being redundant.

340

RFID Systems

12.9.4.1 Extension of RRE In order to solve this problem, we require readers to maintain a list of tags held, and to passively listen for tag responses to queries initiated by other readers. The duration of this phase is upper bounded by the time taken by a reader to get an estimate of the maximum number of tags a reader can cover. The listening phase proceeds as follows: assume that a reader R overhears a tag response to a query initiated by another reader and the query content is Rx , Ty , c (see Algorithm 3 line 11). This query indicates that the holder of tag Ty is Rx with a tag count c. Then, if c is larger than its own tag count and Rx = R, reader R removes tag Ty from its list of held tags. When the list is empty, the reader becomes redundant and can be safely turned off. Example 12.9.4 Using the example in Figure 12.4(b), assume R2 has T1 and T2 in its list of held tags on completion of its first step of RRE. During R1 ’s execution of the first step of RRE, R1 chooses at least one time frame during e epochs to write when no other reader is transmitting. When this occurs, R2 overhears the reply of T2 and removes the tag from its coverage list. This is because T2 ’s reply will contain a count higher than its own and a reader identifier different from its own. The situation is identical for tag T1 that will eventually be covered by R4 . Note that this solution may not be applicable to all RFID systems (e.g. UHF).

12.10 Adapting to Topological Changes As mentioned before, readers detected to be redundant by rre can switch off their RFID interface. However, in a real deployment, tags and readers may fail and new components may be randomly deployed. Scenarios where new tags are deployed or existing readers and tags move are particularly important. Thus, in order to adapt to topological changes, readers need to periodically execute rre. Our solution consists of providing a duty-cycle for the RFID interface of readers. The RFID interfaces of readers are periodically activated, allowing them to run rre and discover new tag deployments. Between active intervals, the RFID interface is placed into sleep mode, effectively saving battery power. In the following, we consider each duty cycle to consist of an active interval followed by k sleep intervals. k is a parameter, denoting the ratio of sleep to active intervals. We assume that all active and sleep intervals have the same length and we only consider the case of a single active interval per cycle. Figure 12.5 shows an example duty-cycle with a 1:1 ratio of sleep to active intervals. Unfortunately, a simple duty-cycle for the RFID interface does not handle reader failures or relocations well. This is illustrated in the following example. Example 12.10.1 Consider the system deployment of Figure 12.4(b). Assume that after each reader executes rre once, active reader R1 , fails. Before failing, R1 was holding tags T2 , T3 and T4 , with a count of 3. After R1 fails and when readers R2 , R3 and R4 execute rre again, tags T2 , T3 and T4 will still have a count value of 3. Then, reader R2 will again discover, this time inaccurately, that it is redundant and will not report

Optimal Tag Coverage and Tag Report Elimination

R1

341

Reset Interval

RRE Interval

Reset Interval

RRE Interval

Tr

Trre

Tr

Trre

R2

Reset Interval

RRE Interval

Reset Interval

RRE Interval

Tr

Trre

Tr

Trre

epsilon

Figure 12.5 Example reader duty-cycle with a 1:1 ratio of sleep to active intervals. Each active interval consists of a first, shorter sub-interval for resetting all tags to 0, followed by a longer sub-interval dedicated to running rre. The duty-cycles of two readers shown are off by at most , due to clock drifts.

tag T2 . Then, T2 will not be reported, even though it is still covered by a functioning reader, R2 .

12.10.1 Tag Count Resetting We propose a solution to this problem, which assumes a coarse clock synchronization between readers. Let  be the synchronization error we decide to support. We assume all readers start their active intervals at the same time (within the  error bound). We divide each active interval into two sub-intervals. During the first sub-interval, called the tag reset interval, each reader contacts each of its covered tags and resets their counters to 0. If a tag is covered by multiple readers, it may be reset multiple times. Let Tr denote the length of the tag resetting interval. A reader will attempt to reset its tags only in the first Tr −  section of its tag resetting interval. During the last  section of the sub-interval, a reader will not issue write commands. During the second sub-interval, called the rre interval , each reader executes rre. The length of the rre interval, Trre needs to be at least O(γ log β log ψ), to allow rre to complete. For any reader, at the beginning of each of its rre sub-intervals, all its covered tags have their counters set to 0. It is straightforward to see that if no two reader clocks drift by more than , the silent period between the tag reset and the rre sub-intervals ensures that no resetting command is issued after a reader has written its tag count on any of its covered tags. A tag can then be held and reported by different readers after different active intervals. Then, even if a reader fails, the subset of its tags covered by other readers will still be reported. Note that a reader does not have to always report all its covered tags. Instead, for slowly changing topologies, a reader can implement the following strategy. The reader reports all covered tags after its first active interval but subsequently reports only changes in coverage. A change consists in the detection of a new tag or the disappearance of a previously covered tag.

342

RFID Systems

12.11 The Layered Elimination Optimization (LEO) As shown in Lemma 12.9.3, the distributed rre heuristic requires O(ψγ ) write operation for a single execution by all readers in an RFID system. Writing rates and ranges of writable tags are smaller than reading ranges and rates. This implies that in order to write to tags in their interrogation zones, readers need to use higher transmission power levels, thus consuming more energy. It is therefore important to reduce the number of write operations required to solve the tag coverage and reporting problems. In this section we describe leo, a Layered Elimination Optimization proposed in [18]. leo attempts to extend rre in order to reduce the number of write operations required to solve the optimal tag coverage and reporting problems. The idea behind leo is that while each tag needs to be reported by one reader, it does not matter which reader performs this operation. Instead, the reader that ends up holding and reporting any tag could be the first reader that attempts to write to that tag. Then, leo functions in the following manner. Each tag stores a single value, the identifier of the reader holding it. The initial value for that identifier is NULL for any tag. Before a reader writes its identifier on any of its covered tags, it first reads the reader identifier stored on that tag. The reader then proceeds to write its identifier on the tag, only if the reported reader identifier is NULL.

12.11.1 Implementation Algorithm 4 shows the pseudocode for leo using Orca-like [16] syntax. The pseudocode simplifies the presentation of read and write operations performed by a reader, hiding the details of choosing a random time frame for transmission, during e consecutive epochs. Specifically, the read operation used in the pseudocode specifies the tag identifier and returns the value stored on the tag. Similarly, the write operation specifies the destination tag and the value to be written. The differences from rre are the following. When a tag receives a write command, it records the specified reader identifier only if the stored reader identifier is NULL (lines 3–6). When a tag receives a read command, it returns only its identifier and the reader identifier locally stored (lines 7–9). The operation of an RFID reader differs from rre in the following fashion. For each tag covered, the reader first reads the tags and only attempts to write its identifier if the identifier stored on the tag is NULL (lines 13–17). Afterward, the reader decides to be redundant only if its identifier is not stored on any of its covered tags (lines 18–22). Algorithm 4 Pseudocode for leo. Reader and tag operation for reducing the number of operations performed to provide a solution for the tag coverage and reporting problems. 1. Object implementation WritableRFIDTag; 2. Operation run() 3. guard inQ.first.type = write do 4. if Rid = NULL then 5. Rid := inQ.first.rid; 6. fi; 7. guard inQ.first.type = read do bCast(new packet(Tid ,Rid )); 8. 9. od 10. end

Optimal Tag Coverage and Tag Report Elimination

343

11. Object implementation RFIDReader; 12. Operation isRedundant(prefix:integer) 13. for i in 1..tags.size do 14. if read(tags[i]) = NULL then; 15. write(tags[i],Rid); 16. fi 17. od 18. for i in 1..tags.size do 19. if read(tags[i]) = Rid then; 20. redundant := false; 21. fi 22. od 23. if redundant = true do turnOff(); fi 24. end

12.11.2 Analysis Similar to greedy and rre, leo provides a non-optimal solution to the tag coverage problem. However, note that leo significantly reduces the number of write operations when compared to rre. Specifically, the number of write operations is γ , that is, each tag is written only once. While this improvement is desirable, we note that leo introduces several problems when compared to rre. First, by greedily assigning tags to the readers holding most tags, rre may require fewer nodes to stay active. This is because readers-assigned tags need to be active. By giving preference to readers assigned a higher number of tags, rre is likely to require fewer readers to stay active than a random assignment algorithm as the one proposed by leo. Second, since in leo a tag can only be written once, the algorithm becomes vulnerable when readers fail. If a reader fails, the tags it was assigned will not be overwritten by other readers. Such tags will be subsequently ignored. Third, the test-and-write operation proposed in leo introduces race conditions. If two readers detect a tag that is un-assigned at roughly the same time, both will attempt to mark the tag by issuing write commands. Even though a single write command will succeed, both readers will believe that they are holding the tag. This can lead to both discovering fewer redundant readers and to generating duplicate tag reports. As a final note, the advantage of leo lies in the fact that each tag is written to only once, instead of multiple times as it is likely to happen in rre. Note that a similar effect could be achieved in rre if readers that have detected more tags choose earlier slots for transmission when writing. This approach can effectively prevent readers with a lower tag count from issuing write commands before heavier readers.

12.12 Related Work 12.12.1 Coverage Problems in WSNs Coverage in sensor networks is a well-covered research topic. Tian and Georganas [19] were the first to propose an algorithm for detecting sensors whose coverage area is

344

RFID Systems

completely covered by other sensors. In their solution, a sensor turns itself off only when each sector of its coverage disk is covered by another sensor. Zhang and Hou [20] turned off “redundant” sensors by using bitmaps indicating the coverage redundancy of small zones belonging to the coverage disks of sensors. For the interested reader we also recommend the papers of Slijepcevic and Potkonjak [21], Ye et al. [22] and Gallais et al. [23], which study various aspects and approaches addressing the coverage problem in sensor networks. For further reading on the approach briefly covered in Section 12.5, we direct the reader to the work of Carbunar et al. [24].

12.12.2 Collisions in RFID Systems The reader-collision problem in RFID systems, which is briefly studied in Section 12.4, was first documented in [14]. The solution proposed, of allocating different frequencies to interfering readers, is centralized. A simple decentralized version, where readers listen for collisions and use randomized back-off when detecting one, is discussed. For further reading on this topic, including various centralized and decentralized, time and frequency division approaches for solving this problem, we direct the reader to works by Waldrop et al. [25], Ho et al. [26] Zhou et al. [27], Deolalikar et al. [28] and Birari and Iyer [29]. For more challenges in managing RFID data we recommend the work of Derakhshan et al. [30]. Moreover, Medium Access Control protocols for wired and wireless networks share several details with the reader collision avoidance algorithm. For the interested reader we recommend reading the ALOHA [31], MACA [32] and CSMA/CD [33] MAC protocols.

12.13 Conclusion In this chapter we have studied two important problems in wireless RFID systems. The first problem consists of finding the minimal set of readers that cover all the tags in the system. This problem relates to extending the lifetime of the reader network by detecting and temporarily disabling the RFID interfaces of redundant readers. The second problem consists of determining the optimal way to report tags. We defined coverage in terms of discrete sets of points, tags, and proved that the optimization version of the first problem is NP-hard. For both problems we studied first a centralized heuristic, then distributed and localized algorithms. The distributed algorithms are based on a randomized querying technique, that ensures, w.h.p., the accurate receipt of reader queries by tags. Since writing rates and ranges are smaller than reading rates and ranges, we have also discussed a layered elimination optimization technique and studied its properties. Future research direction may consists of: (i) investigating in detail the energy cost of the read and write operations invoked by RRE; (ii) determining the difference in range for read and write operations; and (iii) developing algorithms that further reduce the number of tag (read/write) operations.

Optimal Tag Coverage and Tag Report Elimination

T3 T2

R2

R3

345

T6

R4

T8

R5

R1 T1

T4

T5

T7

Figure 12.6 Difficulty of consistently breaking ties. The optimal solution keeps only R2 and R4 active. However, in a scenario where R2 , R3 , and R4 , each covering 4 tags, hold a different set of tags, all of them will have to be active.

Problems 1. How and why does RCA differ from TWA? How does RCA avoid collisions and ensure that reader transmissions are received by tags? 2. What is the difference between the optimal tag coverage and the tag reporting problems? 3. Show that RRE does not provide an optimal tag coverage solution for the RFID system shown in Figure 12.4(a). 4. Use the pseudocode of Algorithm 3 to implement the RRE-HC algorithm, assuming that each reader has a local variable storing its distance in hops to the sink host. Then modify the Algorithm 2 to provide a centralized, greedy version of the RRE-HC algorithm. 5. Show that in a network where r nodes can provide full coverage (cover all the tags) a worst case run of RRE can require O(r) nodes to be active, where the constant hidden by the big-oh notation is larger than 1. Hint: use the example system from Figure 12.6 to find the optimum tag coverage solution and determine the worst case RRE result. NB The solutions are provided on the book’s website.

References [1] Bednarz, A. (2002) Wireless technology reshapes retailers. Network World, 12 August. [2] Zhang, T., Ouyang, Y., and He, Y. (2008) Traceable air baggage handling system based on RFID tags in the airport, JTAER, 3(1): 106–115. [3] SkyeTek. (2004) Available at: http://www.skyetek.com/Portals/0/Documents/Products/SkyeModule M1Mini DataSheet.pdf, January. [4] Crossbow Technology Inc. RFID and Asset Tracking. Available at: http://www.xbow.com/.

346

RFID Systems

[5] Baracoda An efficient way to add RFID reader/encoder to Bluetooth PDA and mobile phones. Available at: http://www.baracoda.com/baracoda/products/p 21.html. [6] Wireless Dynamics Wireless Dynamics Inc. announces the mini-SDiD. Available at: http://www.wdi.ca. [7] Mobile Magazine. Nokia 5140 RFID Reader. Available at: http://www.mobilemag.com/content/100/104/ C2607. [8] Carbunar, B., Ramanathan, M.K., Koyut¨urk, M., Jagannathan, S., and Grama, A. (2009) Efficient tag detection in RFID systems, J. Parallel Distrib. Comput., 69(2): 180– 196. [9] PGS Electronics Wireless RFID Systems. Available at: http://www.pgselectronics.com/PGSRFID.htm. [10] O’Connor, M.C. RFID Journal . Reva Announces RFID Network Design. Available at: http://www rfidjournal.com/article/articleview/1638/1/1/. [11] Sarma, S.E., Weis, S.A., and Engels, D.W. (2003) RFID systems and security and privacy implications, in CHES ’02 . Berlin: Springer-Verlag, pp. 454– 469. [12] Micic, A., Nayak, A., Simplot-Ryl, D., and Stojmenovi´c, I. (2005) A hybrid randomized protocol for RFID tag identification, in Proceedings of the IEEE International Workshop on Next Generation Wireless Networks (WoNGeN). [13] Pupunwiwat, P., and Stantic, B. (2009) Unified q-ary tree for RFID tag anti-collision resolution, in ADC . [14] Engels, D.W., and Sarma, S.E. (2002) The reader collision problem, in IEEE International Conference on Systems, Man and Cybernetics. [15] Carbunar, B., Ramanathan, M.K., Koyuturk, M., Hoffmann, C., and Grama, A. (2005) Redundant reader elimination in RFID systems, in 2nd Annual IEEE Communications Society Conference on Sensor and Ad Hoc Communications and Networks (SECON’05). [16] Bal, H.E., Bhoedjang, R., Hofman, R., Jacobs, C., Langendoen, K., Ruhl, T., and Kaashoek, M.F. (1998) Performance evaluation of the Orca shared-object system. ACM Trans. Comput. Syst., 16(1): 1–40. [17] Fowler, R., Paterson, M., and Tanimoto, S. (1981) Optimal packing and covering in the plane are NP complete. Information Processing Letters, 12(3): 133– 137. [18] Hsu, C.-H., Chen, Y.-M., and Yang, C.-T. (2007) A layered optimization approach for redundant reader elimination in wireless RFID networks, in APSCC ’07: Proceedings of the the 2nd IEEE Asia-Pacific Service Computing Conference. [19] Tian, D., and Georganas, N.D. (2002) A coverage-preserving node scheduling scheme for large wireless sensor networks, in Proceedings of the 1st ACM WSNA, ACM Press, pp. 32–41. [20] Zhang, H., and Hou, J. (2004) Maintaining coverage and connectivity in large sensor networks, in International Workshop on Theoretical and Algorithmic Aspects of Sensor, Ad hoc Wireless and Peer-to-Peer Networks. [21] Slijepcevic, S., and Potkonjak, M. (2001) Power efficient organization of wireless sensor networks, in IEEE ICC . [22] Ye, F., Zhong, G., Lu, S., and Zhang, L. (2003) PEAS: a robust energy conserving protocol for long-lived sensor networks. In 23rd IEEE ICDCS . [23] Antoine Gallais, Jean Carle, David Simplot-Ryl, and Ivan Stojmenovi´c (2008) Localized sensor area coverage with low communication overhead, IEEE Trans. Mob. Comput., 7(5). [24] Carbunar, B., Grama, A., Vitek, J., and Carbunar, O. (2006) Redundancy and coverage detection in sensor networks, ACM Trans. Sen. Netw ., 2(1). [25] Waldrop, J., Engels, D.W., and Sarma, S.E. (2003) Colorwave: a MAC for RFID reader networks, in Wireless Communications and Networking (WCNC). [26] Ho, J., Engels, D.W., and Sarma, S.E. (2006) HIQ: a hierarchical q-learning algorithm to solve the reader collision problem, in Proceedings of the Applications and the Internet Workshops. [27] Zhou, Z., Gupta, H., Das, S.R., and Zhu, X. (2007) Slotted scheduled tag access in multi-reader RFID systems, in Proceedings of ICNP . [28] Deolalikar, V., Mesarina, M., Recker, J., and Pradhan, S. (2006) Perturbative time and frequency allocations for RFID reader networks, in Proceedings of Emerging Directions in Embedded and Ubiquitous Computing. [29] Birari, S.M., and Iyer, S. (2005) Pulse: A MAC protocol for RFID networks, in EUC Workshops. [30] Derakhshan, R., Orlowska, M.E., and Li, X. (2007) RFID data management: Challenges and opportunities, in Proceedings of IEEE RFID.

Optimal Tag Coverage and Tag Report Elimination

347

[31] Abramson, N. (1970) The ALOHA system: Another alternative for computer communications, in AFIPS Conf. Proc., Fall Joint Computer Conf . [32] Karn, P. (1990) MACA a new channel access method for packet radio, in ARRL/CRRL Amateur Radio 9th Computer Networking Conf . [33] Carrier sense multiple access with collision detection (CSMA/CD) access method and physical layer Specifications, Institute of Electrical and Electronics Engineers, 1996.

13 Delay/Disruption-Tolerant Mobile RFID Networks: Challenges and Opportunities Hongyi Wu and Zhipeng Yang The Center for Advanced Computer Studies, University of Louisiana at Lafayette

13.1 Motivation Wireless sensor networks have been widely employed in wildlife research, where sensors are attached to animals of interest for pervasive information gathering [1]. While the effectiveness and efficiency of such sensor networks have been demonstrated in multiple research projects, they are limited in applications for large animals only. Recent studies (to be discussed next) have revealed that battery-powered sensors have fundamental limitations that hinder their use in tracking small wildlife. Radio Frequency IDentification (RFID) has received growing attention lately, as a result of reduced tag costs and sizes as well as expanded reader communication ranges. This chapter presents the latest research that deals with a Featherlight Information Network with Delay-Endurable RFID Support (FINDERS), composed of passive RFID tags which are ultra light, durable, and flexible, without power supply for long-lasting applications [2]. It investigates the use of RFID gear for wireless network construction, aiming to find events of interest and gather aggregate information. Like modern smart sensors [1], RFID tags and readers together may constitute a distributed sensor system for pervasive information gathering, subject to different data acquisition, delivery, and storage approaches. Unlike typical active sensors that rely on battery power supply and indispensably require sturdy casing for their protection, however, a passive RFID tag is particularly suitable for environments with strict weight constraints, for example, in the studies of small wildlife, RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

350

RFID Systems

where the weight of the sensor must be under 5% of the weight of the animal to avoid hindering the wildlife movement or the welfare of the animals themselves. This requires very small sensors to be employed, which is in sharp contrast to other projects that target large animals including zebras [3], whales [4], and deers [5], and human beings [6]. Earlier investigation has revealed that most of these small animals cannot carry any active devices (such as GPS receivers or the smallest Crossbow sensors) with battery and casing, since overweight sensors often lead to high mortality rates of the animals being studied. Similar weight constraint also applies in many other fields of scientific studies and industrial applications. While efforts have been made to develop miniature sensors in laboratories, the lowest weight of any active sensor that aims to achieve a given communication range and lifetime is bounded inevitably by its battery and casing, where the latter must be heavy-duty for the protection of the power source and powered electronic circuits under harsh environments. This becomes a key hurdle that limits the applicability of active sensors in various applications with strict weight constraints, not to mention extra hassle involved in ensuring adequate battery power. The proposed FINDERS system intends to address such fundamental challenges in wireless sensor networks by leveraging RFID and modern computer network technologies to establish a featherlight information network for pervasive data gathering under strict weight constraint.

13.2 Overview of FINDERS In this section, we outline the overall architecture of FINDERS, including its constituent components and basic communication schemes. The details of its network protocols are omitted, as it is still an open issue to design the most appropriate protocols for FINDERS. FINDERS consists of two types of nodes, RFID readers and tags, as illustrated in Figure 13.1. The readers are deployed at strategic locations according to specific applications. For example, in wildlife and biological studies, the readers can be set up at natural “choke points” where the animals have to move past or through, because of significant

IR 3

IR 1

GR 2

Tag 4 IR 2 Tag 3 Tag 1

GR 2

IR 4

Tag 2

Figure 13.1 An overview of FINDERS.

Tag 5

Delay/Disruption-Tolerant Mobile RFID Networks: Challenges and Opportunities

351

movement barriers otherwise. Since FINDERS is often deployed under harsh and complex field environments, it is not practically viable for most readers to establish reliable connections to communicate with each other or to access the backbone networks. As a result, they become isolated readers, or IRs (see IRs 1-4 in Figure 13.1). Power supply is challenging for the IRs too. Most of them rely on solar panels and high capacity batteries, which can be recharged/replaced but will post a limit on the daily duty circle of the readers. Only a few readers at convenient locations are equipped with a reliable power source and network connections, and they are dubbed gateway readers (GRs; see GRs 1 and 2 in Figure 13.1). GRs serve as the gateways to deliver data from FINDERS to the destination (e.g. a data server). The readers have relatively large storage space compared with tags, and each of them maintains a data queue. While all readers are fixed, the tags are attached to moving targets and thus become mobile (see Tags 1-8 in Figure 13.1). Many off-the-shelf passive tags that are light and durable and have sufficient reading/writing ranges can be employed in FINDERS. For example, we have adopted the Alien passive RFID system for this research, which consists of ALR-9900 reader and ALN9540 SquiggleT M (passive) tags. The ALR-9900 reader is a powerful device that possesses 64 MB RAM and 64 MB flash memory and supports up to 4 antennae and 50 communication channels. Being very thin and light, an ALN-9540 tag measures 8.15 × 94.8 × 0.05 mm and weights less than one gram. Its memory can hold up to 20 bytes of data. With such small storage space, the tag maintains a simple buffer that can be read and written in one operation, without queuing. The tag is well engineered and packaged, exhibiting excellent survival skills under such harsh environments as underwater, underground tunnels, and extreme temperatures ranging from −25◦ C to +50◦ C. Our lab and field experiments have revealed that the Alien system can achieve a reading/writing distance of some 20 ft. Since the readers are static and most of them are isolated from each other, the communications in FINDERS rely on the mobility of the tags to establish time-varying opportunistic links with nearby readers, forming a delay-tolerant network (DTN; [7]) for data delivery. For example, in the scenario shown in Figure 13.1, Tags 1 and 3 are within the reading/writing range of IR 2. Thus IR 2 may read the data from Tag 1 and write them into Tag 3. When Tag 3 passes by IR 4, it unloads its data to the latter, which in turn writes the data into Tag 5 when it comes into the writing range. Tag 5’s trajectory passes through GR 2. The data can thus be delivered to the destination via this gateway. In general, FINDERS may serve to find events of interest and to gather aggregate system information through sporadic wireless communications between its constituent passive tags and readers.

13.3 General Feasibility Study While the featherlight weight of passive RFID tags provides unique opportunities to support various applications with strict weight constraints, the capacity of FINDERS is unsurprisingly low compared with many other data networks, due to its extremely limited network resources. To understand its capacity and accordingly the applications that it can support, we first carried out a feasibility study of the FINDERS system based on a trivial communication scheme. This feasibility study aims to gain insights into the basics of FINDERS and reveal its general capacity, without consideration of specific protocols and network environments.

352

RFID Systems

To this end, we need to investigate why a data delivery effort may fail and where the data packet may be dropped. First, since the GRs connect to the conventional networks, they usually have sufficient communication bandwidth (for our target applications) and can safely deliver the received data to end users. The tags keep data packets in their nonvolatile memory. As long as they are not falsely overwritten, the data will not be lost (if we ignore any physical failures). On the other hand, each IR maintains a data queue, which may overflow when packet arrival is too high. Therefore, we focus on the queuing behavior of the IRs in our feasibility study. 1. Queuing Model: We consider a general queuing model (i.e. a G/G/1 queue) for each IR. While it is difficult to derive solutions for the G/G/1 queue without a detailed knowledge of packet arrival and service (which depend heavily on the data transmission protocol), we intend to obtain an estimation of the network capacity (i.e. the amount of data that can be handled by the system) based on the condition of queue stability. More specifically, the arrival rate must be lower than the service rate in order to keep the queue stable. This study only requires to discover the average data service rate, to be discussed next. 2. Average Data Service Rate: A data packet is served (i.e. sent out from the IR’s queue) when the IR meets a suitable tag. Thus, it is critical to understand how often the meeting between the IR and a suitable tag can happen. Consider a FINDERS with K IRs, J GRs, and T tags, where each tag has a capacity of m, that is, the tag can store up to m data packets. For simplicity, we assume that a tag can receive data (i.e. becomes a suitable tag) only if it is empty (while more sophisticated schemes for data transmission will be developed in the future). Moreover, we consider that Tag i moves by following a randomly chosen path. For analytic tractability, the time to travel through such a path is assumed to be a random variable under exponential distribution with the rate of λi . This assumption, thought not always true under all network environments, greatly reduces the analytic complexity. Note that our feasibility study does not intend to provide an accurate analysis, but instead to gain a general observation of the capacity of FINDERS. Besides, we assume Tag i has a long-term statistical probability of γikI to meet IR k and the probability of γijG to meet GR j . It is easy to verify that the time intervals to meet IR k and GR j are exponentially distributed with the rate of λIik and K I G G I I I λG k=1 λik , ij , respectively, where λik = λi γik and λij = λi γij . In addition, let λ i = J G which is the total rate for Tag i going to any IRs, and similarly λG = λ j =1 ij . i Based on above assumptions, a Markovian model (see Figure 13.2) is established to analyze the meeting events between Tag i and the readers, and accordingly the data service rate of the IR’s queue. The model consists of K + 2 states. State Sk (1 ≤ k ≤ K) is the state that IR k can transmit m packets to Tag i (i.e. the tag is empty). State SN is for the state where Tag i meets any IRs, but it is not empty and thus cannot receive data packets. State SG indicates that Tag i meets a GR and delivers the data packets it carries. Once the packets are delivered to the GR, they are removed from the tag, which thus becomes empty and ready to receive new data packets. Let’s denote Pk , PN , and PG the steady state probabilities of States Sk , SN , and SG , respective. Based on the Markovian model, we can derive the following state

Delay/Disruption-Tolerant Mobile RFID Networks: Challenges and Opportunities

353

SG

lG i

lG i

li1I

likI lG i

I liK lG i

Sk

S1

SK

lIi

I

li

I

li

SN

Markovian model for Tag i.

Figure 13.2

equations:  I I Pk (λG  i + λi ) = λik PG , ∀1 ≤ k ≤ K,       PN λG = K λI , k=1 i i  I I  + K PG λi = PN λG  k=1 (λik Pk ), i     K k=1 Pk + PN + PG = 1. λG

λI λG

(13.1)

λI

2

ik i i i To solve them, we have PG = λI +λ G , Pk = (λG +λI )2 , and PN = (λG +λI )2 . To facilitate i i i i i i the understanding of FINDERS’ general capacity, we further simplify the above results by letting γikI = γijG = γ (i.e. Tag i has equal probability of meeting all readers), and J accordingly λIik = λG ij = λ. Then we arrive at Pk = (K+J )2 , which is the steady state probability that Tag i becomes empty and arrives at an IR. With an overall rate of (K + J )λ for Tag i to visit any readers, it has a rate of (K + J )λPk to meet IR k when it is empty. Assume all tags are independent, the aggregated arrival rate of empty tags at an IR is T (K + J )λPk . Since each tag can carry m packets, we have the service J rate of an IR’s queue, µ = J +K λmT . 3. Estimated Capacity: To ensure a stable queue at the IR, the service rate must be greater than the arrival rate. Therefore, the entire FINDERS with K IRs can accept no K more than JJ+K λmT packets, or

A≤

JK λmT , J +K

(13.2)

which can serve as an estimation of the network capacity. It shows that, under typical scenarios, FINDERS can easily achieve a data rate of multiple kbps or higher

354

RFID Systems

2500

Maximum Throughput (bps)

2000

1500

1000

500

0

0

50

100

150

200

250

300

350

400

Number of GRs (J)

Figure 13.3

Throughput vs. J .

(see Figures 13.3 & 13.4 where by default a tag can hold 96 bits, K = 5, m = 1, T = 100, λ = 0.05, and J = 2), which is sufficient to support a wide range of applications, such as wildlife tracking and lightweight monitoring. The capacity increases linearly with m, λ, and T , and nonlinearly with J and K. In addition to the overall network capacity, we also have several interesting findings on the data generation in individual events. In general, new data are fed into the IR’s queue only when it meets a tag. Without loss of generality, let’s assume α packets arrive m at the IR when such meeting event occurs. Thus, we have αλT ≤ µ, or α ≤ JJ+K . In Jm other words, when a tag meets an IR, no more than J +K packets can be generated and fed into the IR’s queue. This result provides a guideline for traffic control in FINDERS and leads to two interesting observations. First, α is not related to the number of tags. This is because the tags contribute to both data arrival and service of individual IR. Second, increasing the number of IRs, K, results in a lower α. Though this is a little anti-intuitive (as the IRs help data relaying), it can be explained below. The IRs directly contribute to the network traffic load (when they meet tags), but do not directly consume the data (which can be successfully delivered only when they arrive at the GRs but not IRs). Thus, more IRs decrease the maximum allowed α. However, a larger K does help improve the overall network capacity, because they relay and buffer data packets, thus increasing the opportunity to deliver data to GRs. Additionally, a sufficiently large K is usually required by the applications, in order to achieve the needed coverage and granularity in data acquisition.

Delay/Disruption-Tolerant Mobile RFID Networks: Challenges and Opportunities

7

355

× 104

Maximum Throughput (bps)

6

5

4

3

2

1

0

0

20

40

60

80

100

Tag Capacity (m)

Figure 13.4 Throughput vs. m.

13.4 Unique Challenges and Tactics The above analysis based on a trivial communication scheme has shown that FINDERS, though with extremely limited resource, can provide sufficient capacity and effectively support our target applications. However, to design the most appropriate protocols for FINDERS still remains an open problem. FINDERS is a very unique wireless information network that distinguishes itself from conventional communication networks, sensor networks, and delay-tolerant networks by the following characteristics: • Nodal heterogeneity: FINDERS consists of two very different types of nodes. A reader is a static and powerful device, with large storage, high computing power, and longlasting (but still limited) battery power. On the other hand, the passive tag can be mobile and has extremely limited resource. • Asymmetric communication: The communication in FINDERS can be established between a tag and a reader only, but not tags to tags or readers to readers. The communication must be initiated by a reader, in contrast to symmetric transmissions in most conventional networks. In addition, reading and writing ranges are usually different, according to our experiments. • Nodal mobility: All readers are static, while the passive tags can be attached to moving objects with various mobility patterns, leading to a dynamic network topology. This is similar to that of a mobile ad hoc network but with very frequent partitions.

356

RFID Systems

• Intermittent connectivity: The connectivity of FINDERS is very low and intermittent, forming a sparse network where a tag is connected to a reader only occasionally. In fact, FINDERS is a special DTN with unique communication and storage constraints. • Intermittent computation: Besides connectivity, the computation at the tag is also intermittent. It is available only when the tag is powered up by a nearby reader. Thus, such continuous functions required by many protocols as counters and timers cannot be implemented here. • Critical network resource: The fundamental philosophy in DTN is to trade storage for sporadic connectivity. The DTN node is assumed to have sufficient storage space, which can hold a large volume of data to alleviate the needs of immediate transmission. This, however, is no longer valid in FINDERS, because the buffer space of the tags (the main vehicle for data transportation) is so limited that it may become the critical network resource and communication bottleneck. • Delay tolerability: Data delivery delay in FINDERS is potentially high, due to loose connectivity and extremely limited network resources. However, such delay, though not desirable, is usually tolerable by the applications which aim at pervasive information gathering from a statistical perspective. • Fault tolerability: Redundancy may exist in FINDERS during data acquisition and delivery. Thus, a data packet may be lost without degrading information gathering performance. At the same time, there are critical needs in data filtering and aggregation at IRs and GRs, due to the likely existence of multiple copies of the same data. The above characteristics make the development of FINDERS a very unique, interesting, and challenging problem, calling for effective solutions to overhaul the data acquisition and delivery schemes in such a featherlight information network with extremely limited resources. While various sensor networks [3, 1, 4, 6] have been investigated and deployed, data collection, storage and communication solutions devised therein are not directly applicable to FINDERS. Few earlier studies have ever been conducted on information networks composed of RFID gear with sporadic communication between moving tags and nearby readers. In particular, two sets of design issues specific to FINDERS must be addressed: 1. Effective and Reliable Data Delivery: FINDERS faces unprecedented challenges in communication and networking. First, the sporadically available wireless links render it impossible to form a well-connected mesh network for end-to-end communications, which are the basis of mainstream sensor network technologies [1]. Moreover, the unique asymmetric communication paradigm, the intermittent computation capability and the extremely small buffer size of a passive tag overthrow the fundamental principle of DTN (where ample buffering is employed under the intermittent connectivity to alleviate the needs of immediate transmission), resulting in inefficiency if the existing DTN protocols are employed in FINDERS. To this end, a series of interesting problems related to data delivery need to be investigated: • Routing metrics: Routing metrics are usually adopted by the routing protocol as an indicator of the available resource of a given path or link. For example, path length and end-to-end delay are popularly used in conventional networks. These metrics, however, do not reflect the unique network resource of FINDERS, and thus may

Delay/Disruption-Tolerant Mobile RFID Networks: Challenges and Opportunities

357

lead to poor network performance or even failure if used for routing. New routing metrics must be explored, possibly based on meeting probabilities between tags and readers and available memory of the tags. Maintaining and updating such metrics must be simple and consume insignificant storage space. • Duplication control: In FINDERS and many other DTN networks, replication is necessary during data delivery for achieving a given success ratio. However, replication increases the overhead, and worse yet, excessive replication may even degrade the delivery ratio due to frequent buffer overflow. Duplication control thus becomes a key issue to be tackled in FINDERS. It reaches a system-wide optimization problem. While the data delivery schemes in FINDERS are (and have to be) much simpler compared with the protocols for other wireless or wired networks, such optimization is non-trivial and its performance largely determines the effectiveness and feasibility of FINDERS. • Queue management: Each reader maintains a data queue. Given the limited communication resource (due to sporadic connectivity and extremely limited storage for tags) and the relatively high loading factor (due to newly acquired and duplicated data), it is an overriding design issue in FINDERS to differentiate the packets in the queue by a simple and efficient parameter which signifies their importance and determines which packet to transmit if a communication opportunity becomes available or which packet to drop if the queue is full. Novel approaches need to be explored. For example, the prioritization parameter may be based on the probability that at least one copy of given information can be delivered to the GRs. Note that the goal of such queue management is to facilitate efficient communication in the network with redundancy, in contrast to the QoS-aware algorithms which aim to prioritize raw information. • Protocol design: Routing metrics, data duplication, and queue management are not independent. For instance, data duplication will clearly affect queue management, and at the same time, itself is also influenced by routing. Built on these three basic components, the data delivery protocol needs to be carefully devised by synergizing the interaction among them, in order to achieve efficient network resource utilization. • Protocol optimization: Regardless of the protocol yet to be developed, several key constraints and enhancements must be investigated and explored to optimize network performance, including traffic engineering, power optimization, clustering, erasure coding and network coding, and the employment of heterogeneous types of tags with different levels of resources. Each of these issues becomes an interesting problem in the unique context of a FINDERS system. 2. Efficient Data Acquisition, Processing, and Storage: Besides the above networking issues, data acquisition, processing, and storage in FINDERS must be carefully handled to achieve high efficiency. The related research problems are outlined below: • Data acquisition: Various types of application-specific data may be generated in FINDERS. Some of them can be readily acquired. For example, a passive sensing unit coupled with an RFID tag may generate a reading when the tag is powered up, and the output can be written into the tag’s memory accordingly. However, the acquisition of data is not always straightforward. For example, the location information is important to many applications for tracking moving objects. While the RFID reader may be equipped with a GPS receiver and thus provide coarse

358

RFID Systems

location information about a tag when the tag moves into its reading range, this solution results in errors of some 20 ft or higher. To achieve a finer degree of localization is nontrivial. • Data compression/aggregation: The data delivery in FINDERS relies on the transportation of tags. Given the extremely limited memory of each tag, the data format must be carefully designed by considering the tradeoff between data accuracy and storage efficiency. As typical data compression algorithms do not work well here due to the small data size, novel schemes will be sought for compact data storage. For example, effective models will be explored to represent raw data, by storing only a few characteristic parameters of a given object trajectory. • Data reading/writing: With potentially high mobility of the targets being studied and the short radio range in RFID, the communications between tags and readers should be completed as fast as possible. However, this problem becomes challenging due to the considerable delay in tag arbitration, reading, and writing, especially under high tag density and harsh communication environments. A unique niche in FINDERS is that the tags are not equally important. During tag arbitration, it is desirable to minimize the collision probability of those tags which carry important (fresh) data or are good vehicles for data transportation. After arbitration, the tags should be designated to appropriate states and given priority for successful communications (especially for those with short sojourn time with the reader), avoiding unnecessary overhead due to repeated reads/writes. This can be very useful to expedite the arbitration of tags.

13.5 Related Work FINDERS is related to two emerging technologies: RFID and DTN. RFID has achieved increasingly widespread adoption recently, with several interesting research problems being explored in the past few years. Tag arbitration has been extensively studied, yielding solutions fall into two categories, slotted ALOHA [8, 9, 10, 11], and tree splitting algorithms [12, 13, 10, 14]. The problem of counting RFID tags by using statistic approaches was studied in [15, 16]. Meanwhile, Tan et al. [17] and Sheng et al. [18] aim to discover missing tags and tag popularity, respectively. A load balancing scheme is introduced in Dong et al. [19] to distribute the communication load evenly among readers. In addition, several algorithms have been developed to locate the tags, given that object tracking and localization is one of the most important applications of the RFID system. For example, Hahnel et al. [20] propose to deploy reference tags inside a building and equip the robot with an RFID reader, which scans nearby reference tags to obtain location information. Yamano et al. [21] study the same problem by employing the support vector machine scheme to reduce errors. A location sensing prototype system, called LANDARC (LocAtioN iDentification based on dynaMic Active Rfid Calibration), is presented in Linel et al. [22], where a k-nearest neighbor approach is taken to estimate the location of the target tag. In Wang et al. [23], we have proposed two RFID-based 3-D positioning schemes, namely, the active scheme and the passive scheme, to localize the reader and the tag, respectively, in a three-dimensional space. All of the RFID localization schemes developed so far require reference tags, and thus are difficult to be applied in FINDERS, because deploying reference tags with known locations in the harsh environment is very costly, if not impossible.

Delay/Disruption-Tolerant Mobile RFID Networks: Challenges and Opportunities

359

DTN is a sporadically connected network with frequent partitions [7, 24]. Originally developed for deep space communication in high-delay environments, DTN technology has been recently introduced into wireless sensor networks [25, 26, 27, 28, 29, 30, 4, 31, 6, 32, 33, 34, 35] and mobile ad hoc networks [36, 37, 38, 39, 40, 41, 42, 43, 44, 45]. An in-depth investigation into existing delay-tolerant data delivery protocols can be found in [6, 46]. Almost all of these studies assume sufficient storage space at the DTN nodes, which can hold a large volume of data to alleviate the needs of immediate transmission. But this fundamental philosophy of paying storage for sporadic connectivity is no longer valid in FINDERS, because the available memory could be extremely limited at the tags. Several DTN-based network architectures most relevant to FINDERS are briefed below. Infostation is proposed as a complement to the cellular network, to provide high speed wireless access in isolated coverage areas [38]. It aims to support information distribution, instead of data acquisition and delivery. Data MULE [30] is proposed to collect data in sparse sensor networks, where a mobile entity (called a data mule) travels in the network, receives data from its nearby sensors, and delivers them to sinks. The data mule has sufficient storage space and battery power, and its mobility is controlled in a way to reach efficient data transportation. Similarly, Message Ferrying [45] intends to exploit non-random nodal mobility to help deliver data in partitioned mobile ad hoc networks. Those earlier DTN systems are in sharp contrast to FINDERS, where a tag moves at will (dictated by the object to which it is tagged) and usually possesses intermittent computing capability and only some tens of bytes of memory for user data storage.

13.6 Conclusion We have introduced a Featherlight Information Network with Delay-Endurable RFID Support (FINDERS), composed of passive RFID tags which are ultra light, durable, and flexible, without power supply for long-lasting applications under strict weight constraints and harsh environments. It expands the use of RFID gear for wireless network construction, aiming to find events of interest and gather aggregate information. We have conducted analysis based on a trivial communication scheme to show that FINDERS, though with extremely limited resource, can provide sufficient capacity and effectively support our target applications. At the same time, it remains an open issue to design most appropriate protocols for FINDERS, which faces unprecedented challenges in communication and networking, due to its sporadic wireless links, unique asymmetric communication paradigm, intermittent computation capability, and extremely small memory of tags. Several issues in protocol design including routing metrics, queue management, and duplication control must be carefully addressed. Furthermore, testbed implementation and experiments need to be carried out to investigate practical problems due to unreliable and asymmetric RFID communications and develop corresponding solutions.

Problems 1. Please develop a simple simulation program for the network described in Section 13.3. Use the same parameters to produce simulation results and compare with the analytic

360

RFID Systems

results given in Figures 13.3 & 13.4. Identify the factors that lead to the difference between simulation and analytic results. 2. Vary K, m, T , λ, and J to observe their impacts on network performance and explain why you have such observations. 3. Propose five techniques that can improve the capacity of the delay/disruption-tolerant mobile RFID networks. 4. Experiment the techniques proposed by you and/or discussed in the class in your simulation program. Explain why or why not they are effective. NB The solutions are not provided on the book’s website.

References [1] Akyildiz, I., Su, W., and Sankarasubramaniam, Y. (2002) A survey on sensor networks, IEEE Comm. Magazine, 40(8): 102– 114. [2] Yang, Z., and Wu, H. (2009) Featherlight information network with delay-endurable RFID support (FINDERS), in Proc. of 6th Annual IEEE Communications Society Conference on Sensor, Mesh and Ad Hoc Communications and Networks (SECON), pp. 1–9. [3] n.d.a. Available at: http://www.princeton.edu/ mrm/zebranet.html. [4] Small, T., and Haas, Z.J. (2003) The shared wireless infostation model – a new ad hoc networking paradigm (or where there is a whale, there is a way), in Proc. of ACM International Symposium on Mobile Ad Hoc Networking and Computing (MOBIHOC), pp. 233– 244. [5] n.d.b. Available at: http://www.wu.ece.ufl.edu/projects/DeerNet/DeerNet.html. [6] Wang, Y. and Wu, H. (2006) DFT-MSN: The delay fault tolerant mobile sensor network for pervasive information gathering, in Proc. of IEEE Conference on Computer Communications (INFOCOM), pp. 1–12. [7] Cerf, V., Burleigh, S., Hooke, A., Torgerson, L., Durst, R., Scott, K., Fall, K., and Weiss, H. (2004) Delay tolerant network architecture. Available at: www.draft-irtf-dtnrg-arch-02.txt. [8] Ali, K., Hassanein, H., and Taha, AEM. (2007) RFID anti-collision protocol for dense passive tag environments, in Proc. of IEEE Conference on Local Computer Networks (LCN), pp. 819– 824. [9] Jacomet, M., Ehrsam, A., and Gehrig, U. (1999) Contactless identification device with anti-collision algorithm, in Proc. of International Technical Conference on Circuits, Systems, Computers and Communications (CCNC). [10] Myung, J. and Lee, W. (2006) Adaptive splitting protocols for RFID tag collision arbitration, in Proc. of ACM International Symposium on Mobile Ad hoc Networking and Computing (MOBIHOC), pp. 202–213. [11] Vogt, H. (2002) Efficient object identification with passive RFID tags, in Proc. of International Conference on Pervasive Computing. [12] Hush, D.R., and Wood, C. (1998) Analysis of tree algorithms for RFID arbitration, in Proc. of IEEE International Symposium on Information Theory, p. 107. [13] Law, C., Lee, K., and Siu, K.Y. (2000) Efficient memoryless protocol for tag identification, in Proc. of 4th International Workshop on Discrete Algorithms and Methods for Mobile Computing and Communications, pp. 75–84. [14] Namboodiri, V., and Gao, L. (2007) Energy-aware tag anti-collision protocols for RFID systems, in Proc. of IEEE International Conference on Pervasive Computing and Communications (PERCOM), pp. 23–36. [15] Kodialam, M., and Nandagopal, T. (2006) Fast and reliable estimation schemes in RFID systems, in Proc. of ACM International Symposium on Mobile Ad hoc Networking and Computing (MOBIHOC), pp. 322–333. [16] Qian, C., Ngan, H.L., and Liu, Y. (2008) Cardinality estimation for large-scale RFID systems, in Proc. of IEEE International Conference on Pervasive Computing and Communications (PERCOM). [17] Tan, C.C., Sheng, B., and Li, Q. (2008) How to monitor for missing RFID tags, in Proc. of International Conference on Distributed Computing Systems (ICDCS).

Delay/Disruption-Tolerant Mobile RFID Networks: Challenges and Opportunities

361

[18] Sheng, B., Tan, C.C., Li, Q., and Mao, W. (2008) Finding popular categories for RFID tags, in Proc. of ACM International Symposium on Mobile Ad Hoc Networking and Computing (Mobihoc). [19] Dong, Q., Shukla, A., Shrivastava, V., Agrawal, D., and Banerjee, S. (2007) Load balancing in largescale RFID systems, in Proc. of IEEE Conference on Computer Communications (INFOCOM), pp. 2281– 2285. [20] Hahnel, D., Burgard, W., Fox, D., Fishkin, K., and Philipose, M. (2004) Mapping and localization with RFID technology, in Proc. of IEEE International Conference on Robotics and Automation, pp. 1015– 1020. [21] Yamano, K., Tanaka, K., Hirayama, M., Kondo, E., Kimuro, Y., and Matsumoto, M. (2004) Selflocalization of mobile robots with RFID system by using support vector machine, in Proc. of IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 3756– 3761. [22] Linel, M. Ni, Yunhao Liu, Y.C.L., and Patil, A.P. (2003) Landmarc: Indoor location sensing using active RFID, in Proc. of IEEE Conference on Pervasive Computing and Communications, pp. 407– 415. [23] Wang, C., Wu, H., and Tzeng, N.F. (2007) RFID-based 3-D positioning schemes, in Proc. of IEEE Conference on Computer Communications (INFOCOM), pp. 1235– 1243. [24] Fall, K. (2003) A delay-tolerant network architecture for challenged internets, in Proc. of ACM SIGCOMM Conference, pp. 27–34. [25] n.d.c. Available at: http://www.cens.ucla.edu/. [26] n.d.d. Available at: http://down.dsg.cs.tcd.ie/sendt/. [27] n.d.e. Available at: http://www.sics.se/cna/dtnsn/index.html. [28] Ho, M. and Fall, K. (2004) Poster: delay tolerant networking for sensor networks, in Proc. of IEEE Conference on Sensor and Ad Hoc Communications and Networks. [29] Mainwaring, A., Polastre, J., Szewczyk, R., Culler, D., and Anderson, J. (2002) Wireless sensor networks for habitat monitoring, in Proc. of ACM International Workshop on Wireless Sensor Networks and Applications (WSNA), pp. 88–97. [30] Shah, R.C., Roy, S., Jain, S., and Brunette, W. (2003) Data MULEs: modeling a three-tier architecture for sparse sensor networks, in Proc. of The First International Workshop on Sensor Network Protocols and Applications, pp. 30– 41. [31] Small, T., and Haas, Z.J. (2005) Resource and performance tradeoffs in delay-tolerant wireless networks, in Proc. of ACM SIGCOMM Workshop on Delay Tolerant Networking and Related Topics, pp. 260–267. [32] Wang, Y. and Wu, H. (2007) Delay/fault-tolerant mobile sensor network (DFT-MSN): a new paradigm for pervasive information gathering, IEEE Transactions on Mobile Computing 6(9): 1021– 1034. [33] Wang, Y., Lin, F., and Wu, H. (2005) Poster: efficient data transmission in delay fault tolerant mobile sensor networks (DFT-MSN), in Proc. of IEEE International Conference on Network Protocols (ICNP’05). [34] Wang, Y., Wu, H., Lin, F., and Tzeng, N.F. (2008) Cross-layer protocol design and optimization for delay/fault-tolerant mobile sensor networks, IEEE Journal on Selected Areas in Communications (JSAC), Special Issue on Delay and Disruption Tolerant Wireless Communication. A preliminary version was presented in IEEE ICDCS’07. [35] Wu, H., Wang, Y., Dang, H., and Lin, F. (2007) Analytic, simulation, and empirical evaluation of delay/ fault-tolerant mobile sensor networks, IEEE Transactions on Wireless Communications 6(9): 3287– 3296. [36] Burns, B., Brian, O.B., and Levine, N. (2005) MV routing and capacity building in disruption tolerant networks, in Proc. of IEEE Conference on Computer Communications (INFOCOM), pp. 398 – 408. [37] Ghosh, J., Philip, J., and Qiao, C. (2007) Sociological orbit aware location approximation and routing (SOLAR) in MANET, Ad Hoc Networks 5(2): 189– 209. [38] Goodman, D.J., Borras, J., Mandayam, N., and Yates, R. (1997) Infostations: A new system for data and messaging services, in Proc. of IEEE Vehicular Technology Conference, pp. 969–973. [39] Hui, P., Chaintreau, A., Scott, J., Gass, R., Crowcroft, J., and Diot, C. (2005) Pocket switched networks and human mobility in conference environments, in Proc. of ACM SIGCOMM Workshop on Delay Tolerant Networking and Related Topics (WDTN), pp. 244–251. [40] LeBrun, J., Chuah, C.N., and Ghosal, D. (2005) Knowledge based opportunistic forwarding in vehicular wireless ad hoc networks, in Proc. of IEEE Vehicular Technology Conference (VTC) 2005 Spring, pp. 1–5. [41] Lindgren, A., Doria, A., and Schelen, O. (2004) Probabilistic routing in intermittently connected networks, in Proc. of The First International Workshop on Service Assurance with Partial and Intermittent Resources (SAPIR 2004). [42] Musolesi, M., Hailes, S., and Mascolo, C. (2005) Adaptive routing for intermittently connected mobile ad hoc networks, in Proc. of IEEE 6th International Symposium on a World of Wireless, Mobile and Multimedia Networks (WOWMOM), pp. 1–7.

362

RFID Systems

[43] Spyropoulos, T., Psounis, K., and Raghavendra, C.S. (2005) Spray and wait: an efficient routing scheme for intermittently connected mobile networks, in Proc. of ACM SIGCOMM Workshop on Delay Tolerant Networking and Related Topics, pp. 252–259. [44] Wang, Y., Jain, S., Martonosi, M., and Fall, K. (2005) Erasure-coding based routing for opportunistic networks, in Proc. of ACM SIGCOMM Workshop on Delay Tolerant Networking and Related Topics, pp. 229–236. [45] Zhao, W., Ammar, M., and Zegura, E. (2004) A message ferrying approach for data delivery in sparse mobile ad hoc networks, in Proc. of the 5th ACM International Symposium on Mobile Ad Hoc Networking and Computing (MOBIHOC), pp. 187–198. [46] Wang, Y., Dang, H., and Wu, H. (2007) A survey on analytic studies of delay-tolerant mobile sensor networks, Wireless Communications and Mobile Computing (WCMC) Special Issue on Disruption Tolerant Networking for Mobile or Sensor Networks, 7(10): 1197– 1208.

Part Four Addressing Other Challenges in RFID Systems

14 Improving Read Ranges and Read Rates for Passive RFID Systems Zhiguang Fan1,2 , Fazhong Shen1 , Jianhua Shen1 , and Lixin Ran1 1

Zhejiang University, China

2 Tektronix

(China) Co., Ltd

14.1 Introduction Like other wireless communications or RADAR systems, radio frequency identification (RFID) systems also rely on electromagnetic (EM) waves to work. Similarly, antennas are used to transmit inquiry energy to and sense modulated responses from tags. According to Maxwell’s theory, in the space surrounding an antenna, the distribution of the EM field depends on the distance to the antenna, r. The region immediately surrounding the antenna, estimated by r ≤ λ/6, where λ denotes the wavelength of EM field, is called the nearfield zone, where reactive fields dominate. Low frequency (LF) and high frequency (HF) RFID systems work in this range with standard frequencies of 125 KHz and 13.56 MHz, respectively. The outer region next to the near-field is called the far-field zone, where radiating fields dominate. Ultra-high frequency (UHF) and microwave RFID systems work in this range, whose typical frequencies are 915 MHz and 2.4 GHz, respectively. The field strength decays linearly with 1/r in the far-field zone but exponentially with 1/r 2 or even 1/r 3 in the near-field zone, which is the reason why a UHF or microwave RFID system generally has a longer operational distance than a LF or HF RFID system. From the tag’s point of view, RFID systems can be classified into passive and active. In an active RFID system, the tag has its own power supply, and therefore the communication between the reader and the tag behaves much like a traditional wireless system. However, RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

366

RFID Systems

in a passive RFID system, the tag needs to collect enough energy from the EM fields radiated by the reader to power up itself, and then communicates with the reader by changing its own backscatter cross-section, which makes it a more complex process than in an active RFID system. In this chapter, we discuss “read range” and “read rate” issues of passive RFID systems working in far-field ranges. In EPC’s terminologies [1], the “read range” refers to the distance at which a reader can communicate with a tag. A more complicated definition is the maximum distance between a reader and a tag in which the radiation field from the reader is strong enough to power up the tag and the backscatter signal from the tag is strong enough to be detected correctly by the reader.

From this definition, we can simply conclude that the read range of an RFID system is affected not only by the performance of the reader, but also by that of the tag to be detected. Also in EPC’s terminologies, the “read rate” refers to the maximum rate at which data can be read from a tag, generally expressed in bits-per-second (bps). However, in this chapter, the “read rate” is redefined as the statistical ratio of the quantity of tags successfully read versus that of all the tags. The read range and the read rate characterize the performance of RFID systems from different points of view, and are actually closely related to each other. In most cases, the attempt to improve the read range of an RFID system will also improve the read rate since they are both related to the signal-to-noise ratio (SNR) of the backscatter signals received by the reader. Theoretically, for each read of a tag, the readability is determined by the SNR, and the read range is the maximum distance where the SNR starts to decrease below a specified threshold value. However, in a real application environment, it frequently occurs that not every tag inside the read range can be detected, yielding a read rate lower than 100 %. There are many reasons for this, for example, the polarization and gain variation of a tag’s antenna with different tag orientations, the non-homogeneous field distribution in closed spaces, the existence of various obstacles, different background materials for tags, interference from other wireless systems, read collisions, and so on. The issue of low read rate is an open technical challenge and has plagued the passive RFID industry since its inception. This chapter is arranged as follows. First, we focus on in-depth studies on the working principles of passive RFID systems in the far-field range. The signal descriptions and formulations of SNR and read range are theoretically presented. Based on those equations, we further discuss some issues regarding read range and read rate. Apart from some conclusions that can be drawn directly from those equations, other research efforts related to improving read rates and read ranges are also introduced. At the end of this chapter, two examples are presented to illustrate cost-effective designs of long-range RFID readers utilizing off-the-shelf chips that were originally designed for wireless communications.

14.2 Signal Descriptions and Formulations for Passive Backscatter RFID Systems It is well known that RFID systems began their explosive development only in the last decade. However, RFID actually has a long history of more than half a century [3]. Passive backscatter RFID readers are interrogators for tags, or transponders, based on

Read Ranges and Read Rates for Passive RFID Systems

Reader

367

Forward Signal Antenna

Antenna

Backscatter Signal

Transmitter

Tag LO Circulator

Clutter Microchip

Receiver Obstacles

Figure 14.1 Basic architecture for a passive, far-field RFID system. Reproduced courtesy of  2007 EMW Publishing.

the principles of wireless power transmission [4] and backscatter communication [5–7]. The reader transmits electromagnetic waves into the air and the tag draws energy from radiation fields. The electromagnetic waves are partially backscattered and modulated by the tag, through which the reader remotely retrieves the tag’s information. The basic architecture of such a passive RFID system is shown in Figure 14.1, which consists of a reader, a tag and surrounding obstacles.

14.2.1 Signal Descriptions Generally, the tag modulates the backscatter waves by toggling the termination impedance of the tag’s antenna, yielding amplitude shift keying (ASK) or phase shift keying (PSK) modulations [8]. The reader in Figure 14.1 is composed of a local oscillator (LO), a transmitter, a receiver and a receive-transmit duplex antenna plus a circulator as duplexer. A two-antenna solution is also commonly utilized, where the circulator is replaced by two directivity-optimized antennas to isolate the receiver and transmitter channels. However, in practice, the receive-transmit isolation hardly can be ideal. The limited isolation characteristic of the circulator and the impedance mismatch of the transmitter antenna, and the coupling between the transmitter and receiver antennas inevitably result in the leakage of transmitted power into the receiver. The leakage may seriously degrade the performance of the receiver since it is possibly much stronger than the backscatter signal when the tag is far away from the reader. To begin with, we define a power transmission coefficient α and a power leakage coefficient β to characterize the insertion loss and the isolation of the duplexer, respectively. Then, if PT denotes the signal power feeding into the transmit antenna, βPT will be the power leaking into the receiver, and if PB denotes the backscatter power received by the reader antenna, αPB will be the power entering the receiver. Then, we define ξ that is equal to α/β and call ξ the receive-transmit isolation coefficient of the reader. According to the Friis equation, the power received by the tag antenna, denoted as PR , is given by  c 2 PR = PT GT GR (14.1) 2ωr

368

RFID Systems

where c is the speed of light in free space, ω is the angular frequency of EM wave, r is the distance between the reader and the tag, GR is the gain of the reader antenna and GT is the gain of the tag antenna. A portion of PR is used by the tag to power up its own microchip circuits and the remaining power is reflected back to the tag antenna by the microchip through toggling the antenna’s termination impedance in terms of the stored data in the tag, that is, the ID, and then re-radiated into air to make the backscatter modulation. As mentioned above, there are two kinds of impedance modulations, that is, ASK and PSK. In the case of the ASK tag, the power reflection coefficient typically takes a value of “0”/“” for a data bit of “0”/“1,” respectively; in the case of PSK tag, it takes the same value of  for both data bits of “0” and “1.” Here, we assume that data bits of “0” and “1” have the same transmission probability. Let PA denote the time-averaging absorption power of the tag. Then, in the case of an ASK tag, we can get    PA = 1 −  2 PR ; (14.2a) in the case of a PSK tag, we can get PA = (1 − )PR .

(14.2b)

As can be seen in Figure 14.1, the electromagnetic wave signals picked up by the reader antenna consist of two components: one is the backscatter signal from the tag and the other is the clutter that is the sum of the backscatter signals from surrounding obstacles. The backscatter signal from the tag is indeed composed of the modulated backscatter signal due to the tag’s impedance modulation, as mentioned above, and the unmodulated backscatter signal due to the induced current on the surface of the tag antenna [9]. Both the backscatter signal and the clutter enter the receiver via the circulator. In addition, the leakage signal between the transmitter and receiver ports of the circulator also enters the receiver. These input signals can be summed up as two components: one is the unmodulated signal XU (t) that is the sum of the clutter from the obstacles, the structural backscatter signal from the tag antenna and the receiver-transmit leakage signal, which all act as harmful interferences to the receiver; the other is the modulated backscatter signal XM (t) that conveys the tag’s data information. For a passive backscatter RFID reader, XU (t) is usually much stronger than XM (t), which makes it barely possible for the reader’s receiver to retrieve from the received signals a LO signal XS (t) being synchronized with XM (t). In other words, XS (t) indeed will mainly track the phase of XU (t), but not that of XM (t). Thus, a non-coherent demodulation scheme needs to be adopted for the receiver design. In practice, the leakage component generally dominates in XU (t). Thus, we can ignore other components and approximate XU (t) as XU (t) = AU sin(ωt + θU ),

(14.3a)

where θU denotes the signal phase and AU denotes the signal amplitude that is given by AU =

2βR0 PT ,

(14.3b)

where R0 is the input resistance of the receiver. The modulated backscatter signal XM (t) can be expressed as XM (t) = AM [S(t)] sin(ωt + θM0 + θM [S(t)]), (14.4)

Read Ranges and Read Rates for Passive RFID Systems

369

where S(t) denotes the tag’s binary data sequence of “0”/“1,” θM0 denotes the unmodulated part of signal phase, θM [S(t)] denotes the part of signal phase modulated by S(t) and AM [S(t)] denotes the signal amplitude modulated by S(t). In the case of ASK modulation, AM [S(t)] takes a value of “0”/“AM ” while the tag transmits a data bit of “0”/“1,” respectively, and θM [S(t)] takes a constant value of θM . In the case of PSK modulation, θM [S(t)] takes a value of −θM /θM while the tag transmits a data bit of “0”/“1,” respectively, and AM [S(t)] takes a constant value of “AM .” Using the Friis equation again and Equation (14.1), AM can be given by AM =

 c 2 2ωr

2αR0 PT GT GR .

(14.5)

The LO signal XS (t) of the receiver can be expressed as XS (t) = AS sin(ωt + θS ),

(14.6)

where θS denotes the signal phase and AS denotes the signal amplitude. In practice, XU (t), XM (t), and XS (t) always have phase and amplitude noises. Those signals can be regarded as stochastic processes and their phases θU , θM0 , and θS as time-dependent random variables to denote their phase noises that mainly originate from the LO. The amplitude noises are usually much smaller than the phase noises. In order to simplify the analysis in the later section, we will ignore the former’s effects and denote these signals’ amplitudes as constant quantities.

14.2.2 SNR and Read Range Formulation Based on the analysis in Section 14.2.1, a block diagram for realizing a passive, far-field RFID reader is given in Figure 14.2. In the RF front-end, the most important component is the dual-channel (i.e. I channel and Q channel) non-coherent quadrature demodulator,  which consists of an in-phase power splitter, a π 2-phase-shifting power splitter and two mixers. The baseband for an RFID system is generally realized with digital circuits, generally having an analog-digital-converter (ADC), a digital signal processor (DSP), a central processing unit (CPU), and so on. The read range is actually determined by two conditions: (1) the tag could draw enough energy from incident electromagnetic waves to power up itself; and (2) the modulated backscatter signal from the tag should be strong enough so that the SNR of the demodulation output signal in the reader’s receiver meets the user-specified value. For Case (1): it is obvious that increasing the transmitting power PT of the reader will directly improve the absorption power PA of the tag. However, in practice, PT is generally constrained by regional regulations to be no larger than a specified power value PEIRP (EIRP: Effective Isotropic Radiated Power). For example, the PEIRP is 4 watt within UHF ISM band (902–928MHz) in the United States. Let PT take the maximum allowed value, that is, PT = PEIRP GR . Then, using Equations (14.1), (14.2a) and (14.2b), in the ASK case, we get  c 2    PA = 1 −  2 PEIRP GT 2ωr

(14.7a)

370

RFID Systems

RF Front-end Section

Antenna

Power Amplifier

Digital Baseband Section

CPU

Driver Amplifier Local Oscillator

90°

Circulator

RF Power Band-pass Splitter Filter

DSP + − × /

p/2-Phase-Shifting Power Splitter I

Mixers

0101

Baseband Band-pass Filters

ADC Q

Figure 14.2

Reader architecture. Reproduced courtesy of  2007 EMW Publishing.

And in the PSK case, we get PA = (1 − )PEIRP GT

 c 2 . 2ωr

(14.7b)

Example 14.2.1 Calculating the power absorbed by an ASK tag. Consider an ASK tag working at 2.5 GHz, and a reader radiating a PEIRP power of 20 dBm. Assume the tag’s antenna has a gain of 3 dBi and  = 1. Utilizing Equation (14.7a), we find that this tag will receive a power of −40.4 dBm when it is 10 m away from the reader’s antenna. In order to power up the tag, PA must be no less than a specific tag turn-on threshold power PTH , that is, PA ≥ PTH , which gives an upper limit of operational distance and is called the tag-determined read range, denoted by rTAG . Then, using Equations (14.7a) and (14.7b), for the ASK tag, we get    1 −  2 PEIRP GT c rTAG = (14.8a) 2ω PTH And for the PSK tag, we get  rTAG

c = 2ω

(1 − )PEIRP GT . PTH

(14.8b)

Read Ranges and Read Rates for Passive RFID Systems

371

Example 14.2.2 Calculating the tag-determined read range rTAG for an ASK tag. Consider the same case in Example 14.2.1. If this tag includes a chip having a turn-on threshold power PTH of −15 dBm, we see that the tag will not work because its absorption power PA , that is, −40.4 dBm, is much lower than PTH . In this case, the tag-determined read range rTAG of such a tag can be calculated through using Equation (14.8a) as below:  c (1 − /2)PT GT r= 2ω PTH 3 × 108 1 − 1/2 × 10(20+3+15)/2/10 2 × 2π × 2.5 × 109 = 0.54 meters =

Case (2): assume the baseband band-pass filter utilized in the receiver has a sharp frequency selectivity with low-end and high-end cut-off frequencies of fL and fH , respectively. Referring to Figure 14.2 and using Equations (14.3a), (14.4) and (14.6), we can express the demodulation output signals of I and Q channels as  ! XI (t) = kD AS AU cos(θU − θS ) + kD AS AM [S(t)] cos θM0 − θS + θM [S(t)] + n0 (t), 

!

(14.9a)

XQ (t) = kD AS AU sin(θU − θS ) + kD AS AM [S(t)] sin θM0 − θS + θM [S(t)] + n0 (t), (14.9b) respectively, where θU , θM0 and θS are all time-dependent random variables to denote the phase noise of the LO, n0 (t) denotes the internal thermal noise of the receiver and kD denotes the transfer coefficient of the receiver that takes into consideration the total loss or gain of the RF band-pass filter, the power splitter, the mixer and the baseband bandpass filter. In practice, XU (t) is usually much larger than both XM (t) and n0 (t), which means that the SNR of the demodulation output signal is mainly determined by the phase noise of XU (t) and then we reasonably can ignore the noise contributions of n0 (t) and (θM0 − θS ) terms in Equations (14.9a) and (14.9b). In addition, since XU (t) and XS (t) originate from the same LO, we can assume that their phase noise can be characterized with the same stochastic process except for a time delay. Then, we can rewrite Equations (14.9a) and (14.9b) as  ! XI (t) ≈ kD AS AU cos ωt + θP (t + t) − θP (t)  ! (14.10a) + kD AS AM [S(t)] cos θM0S + θM [S(t)]  ! XI (t) ≈ kD AS AU sin ωt + θP (t + t) − θP (t)  ! (14.10b) + kD AS AM [S(t)] sin θM0S + θM [S(t)] respectively, where θP (t)is a stochastic process characterizing the phase noise of the LO, t denotes the time delay between XU (t) and XS (t), and θM0S denotes the phase

372

RFID Systems

difference between XM (t) and XS (t). Then, the SNRs of XI (t) and XQ (t) can be given in the frequency domain by " fH #  !$ fL S AM [S(t)] cos θM0S + θM [S(t)] df SNR I = " f # (14.11a)  !$ H fL S AU cos ωt + θP (t + t) − θP (t) df " fH #  !$ fL S AM [S(t)] sin θM0S + θM [S(t)] df SNR Q = " f # (14.11b)  !$ H fL S AU sin ωt + θP (t + t) − θP (t) df respectively, where S{ } denotes the operator of calculating the power density spectrum (PDS) of a stochastic process, that is, the Fourier transform of the auto-correlation function of a stochastic process. It is obvious that the SNRs of both XI (t) and XQ (t) strongly depend on the unknown θM0S and t. In other words, there is an inherent uncertainty in terms of the SNR performance of an uncorrelated receiver. Thus, two channels could have different SNRs and the reader’s baseband should choose the one having a higherSNR to retrieve the tag’s ID information.  In the worst case where θM0S takes a value of π 4, ωt takes the values of (2n + 1)π 2 in Equation (14.11a) or nπ in Equation (14.11b), where n is an integer number, and the stochastic processes θP (t) and θP (t + t) are assumed to be uncorrelated to each other, considering that θP (t) and θP (t + t) are mostly taking values near to zero for a practical LO having a low phase noise, the minimal achievable SNR can be approximately expressed as " fH #   !$ fL S AM [S(t)] cos π 4 + θM [S(t)] df SNR MIN ≈ , (14.12) "f # $ 2 fLH S AU θP (t) df which gives a lower boundary of the SNRs of the demodulation output signals. Here, we ignore the correlation effect between θP (t) and θP (t − t), which indeed would dramatically reduce the demodulation output noises [9]. In order to characterize the correlation effect, we define a phase noise improvement factor ψ, which can be achieved experimentally as shown in the later section. Then, we can rewrite Equation (14.12) as " fH #   !$ fL S AM [S(t)] cos π 4 + θM [S(t)] df SNR MIN = . (14.13) "f # $ ψ fLH S AU θP (t) df Now, substituting Equation (14.3b) and (14.5) into Equation (14.13), in the case of ASK tag, we get " fH # $ ξ G2T G2R  c 4 fL S S(t) df SNR MIN = (14.14a) " fH # $ 2ψ 2ωr S θP (t) df fL

in the case of PSK tag, we get SNR MIN

ξ G2T G2R  c 4 = ψ 2ωr

!$ " fH #   fL S cos π 4 + θM [S(t)] df $ " fH # fL S θP (t) df

(14.14b)

Read Ranges and Read Rates for Passive RFID Systems

373

Example 14.2.3 Calculating the SNRMIN for the reader Consider a passive RFID system composed of an ASK tag and a reader working at 950 MHz. Assume the gains of the tag and reader antennas are 3 dBi and 15 dBi, respectively, and the isolation of the circulator is 20 dB and  = 1. The SNRMIN can " fH be calculated " fH by Equation (14.14a). In the calculation, the integration items fL S{S(t)} df , fL S{θP (t)} df and the improvement factor" ψ can be obtained from f experimental data. As measured in Section 14.5, when fLH S{S(t)} df = −6.8 dB, " fH fL S{θP (t)} df = −39.9 dB and ψ = −37.5 dB and the tag is 10 meters far from the reader, the SNRMIN is " fH ξ G2T G2R  c 4 fL S{S(t)} df SNR MIN = " fH 2ψ 2ωr fL S{θP (t)} df = (20 + 3 × 2 + 15 × 2 − 6.8 + 39.9 + 37.5) % 4 & 3 × 108 /2 = 19.6 db +10 lg 2 × 2π × 0.95 × 109 × 10 "f "f The method for calculating fLH S{S(t)} df , fLH S{θP (t)} df and ψ from experimental data can be found in Section 14.5. In Equations (14.14a) and (14.14b), the integral items in the denominators denote the single-sideband in-band phase noise power of the LO (a dimensionless quantity normalized to the carrier power), expressed as PPN , which can readily be obtained by means of numerical integral calculation on the known LO phase noise data or direct measurement with a spectrum analyzer (SPA) device; the integral items in the numerators denote the single-sideband in-band signal power of the tag’s binary data sequence (also treated as a dimensionless quantity), expressed as PDATA , which can be derived according to specific data coding schemes such as unipolar/bipolar coding, return-to-zero/non-return-tozero (RZ/NRZ) coding, Manchester coding, Miller coding, FM0 (bi-phase-space) coding, etc. [8, 10]. Then the read range can be calculated through requiring the SNR MIN be no lower than a user-specified SNR USER , that is, SNR MIN ≥ SNR USER , which is called the reader-determined read range, denoted by rREADER . Then, for the ASK tag, using Equation (14.14a), we get rREADER

c = 2ω



$ " fH #

1/4 ξ G2T G2R fL S S(t) df # $ " 2ψSNR USER fH S θP (t) df

(14.15a)

fL

and for the PSK tag, using Equation (14.14b), we get rREADER

c = 2ω



ξ G2T G2R ψSNR USER

" fH #  !$ 1/4 fL S cos π/4 + θM [S(t)] df " fH # $ fL S θP (t) df

(14.15b)

374

RFID Systems

Example 14.2.4 Calculating the reader-determined read range rREADER . Consider the same case as in Example 14.2.3. Assume the reader requires a minimum SNR of 12 dB to correctly decode the tag data. Then, a reader-determined read range of 11.95 meters can be achieved by calculating Equation (14.15a). Now we have theoretically obtained the tag-determining read range rTAG , the readerdetermining read range rREADER and the minimal achievable SNR MIN of the demodulation output signal in the receiver for both ASK and PSK cases. In the next sections, based on those equations, we will investigate how to improve the read range and the read rate for a passive RFID system.

14.3 Improving the Read Range of a Passive RFID System As discussed in last section, the read range of the passive RFID system is actually determined by the relation between rTAG and rREADER , which also gives two kinds of RFID systems. One is called a tag-determined passive RFID system, where the tag-determined read range is smaller than the reader-determined one, shown as rTAG < rREADER . This situation happens when PA ≥ PT H cannot be met earlier than SNR MIN ≥ SNR USER while increasing the operational distance. The other is called a reader-determined system, which is just the opposite. So, it is obvious that we need to improve both rTAG and rREADER to extend the read range for an RFID system. First, we begin with a discussion about what can be done on the reader’s side. We assume the reader works at a transmitting power PEIRP of 36 dBm and an operating frequency ω of 915 MHz. Then, based on Equation (14.15a) for rREADER , in order to improve the read range, some key design considerations can be presented as follows: • A larger reader antenna gain GR has many advantages: (1) A larger GR will efficiently improve the SNR of the demodulation output signal of the reader and consequently the read range. As shown in Figure 14.3 and Figure 14.4, using Example 14.2.3, the SNR MIN increases log-linearly with the GR . With the increase of the SNR MIN , the read range is also improved. (2) A larger GR can reduce the interference between multiple readers, especially in dense-multiple-reader RFID systems. A larger GR generally means a higher antenna directionality, which can help to focus the electromagnetic energy in a specified area and decrease the inferences among readers. (3) For a given PEIRP , a larger GR requires a lower output power from the power amplifier (PA), which saves the cost of using a high-power RF amplifier and alleviates the leakage power into the receiver. In addition, it should be noted that simply increasing the transmitting power PT of the reader would not improve the receiver’s SNR performance. The reason is that the increase of PT simultaneously increases both the modulated signal power and the noise power according to Equation (14.13). • A large receive-transmit isolation coefficient ξ helps to realize a long read range due to its important effect on the SNR of the demodulation output signal, as shown in Figure 14.5. Generally, the isolation coefficients of commercial circulators vary from 15 to 25 dB. For example, a circulator MAFRIN0461, designed for UHF RFID systems by MIA-COM Company, has an isolation coefficient of 22 dB. In the case of using a circulator as duplexer, the receive-transmit isolation coefficient of the reader is

Read Ranges and Read Rates for Passive RFID Systems

375

40 35 30

SNRMIN (dB)

25 20 15 10 5 0 −5 −10 −15 −20 0

2

4

6

8

10

12

14

16

18

20

Gain of the reader antenna GR (dBi)

Figure 14.3 The calculated SNRMI N vs. the gain GR of the reader antenna when an ASK tag is 10 meters away. Note: In the calculation, GT = 2.15 dBi, ξ = 20 dB, PDATA = −6.8 dB, PPN = −39.9 dB, ψ = −37.5 dB, r = 10 m and  = 1 are assumed. 30 27

Read range (m)

24 21 18 15 12 9 6 3 0

0

2

4

6

8

10

12

14

16

18

20

Gain of the reader antenna GR (dBi)

Figure 14.4 The reader-determined read range vs. the reader antenna gain GR (for ASK tag). Note: In the calculation, GT = 2.15 dBi, ξ = 20 dB, PDATA = −6.8 dB, PPN = −39.9 dB, ψ = −37.5 dB, SNR USER = 12 dB and  = 1 are assumed.

mainly limited by the circulator’s performance. A circulator having both a low inherent leakage and a low insertion loss are preferred for the reader. In addition, since the mismatch reflection signal from the reader antenna also enters the receiver and consequently deteriorates the circulator’s isolation performance, great attention should be paid to the impedance matching between the reader antenna and the circulator. For the two-antenna solution (Bi-static), the isolation between two antennas should also be optimized through antenna design, antenna placement, and antenna orientation.

376

RFID Systems

20 18

Read range (m)

16 14 12 10 8 6 4 2 0 12

14

16

18

20

22

24

Isolation coefficient ξ (dB)

Figure 14.5 Reader-determined read range vs. receiver-transmitter isolation coefficient ξ (for ASK tag). Note: In the calculation, GT = 2.15 dBi, GR = 13 dBi, PDATA = −6.8 dB, PPN = −39.9 dB, ψ = −37.5 dB, SNR USER = 12 dB and  = 1 are assumed.

• A low phase noise LO is strongly recommended. The phase noise is more severe than the internal thermal noise of the receiver, and has a more significant impact on the demodulation output of the reader. Incidentally, considering that the phase noise is concentrated within a narrow band surrounding the carrier frequency and consequently the receiver noise is mainly of low frequency component, the data coding and modulation schemes adopted in a tag should be capable of suppressing the DC and low frequency components in its data signal as much as possible. That would facilitate utilizing band-pass filters in the receiver to suppress harmful noises and simultaneously preserve desired data signals, which consequently can relax the requirement of the phase noise of the LO. Example 14.3.1 Selecting a proper GR for an RFID reader Generally, GR can be chosen from 2 dBi (for example, a standard dipole antenna) to 20 dBi (for example, an aperture antenna or an antenna array), or even higher. From Figure 14.4, we see that the larger GR , the longer read range. However, a larger GR also means a higher reader directivity, that is, less angular coverage for the tags. On the other hand, for a specified PT , the larger GR , the larger PEIRP (PEIRP = PT GR ). However, the PEIRP needs to conform to RFID regulations. For example, Part 15 of FCC of USA advises that the PEIRP should not exceed 36 dBm. Therefore, the selection of GR is also limited. A high gain antenna generally costs more than a low gain one. As for a tag-determined RFID system, a large GR probably means a low cost-efficiency. In this case, the read range is equal to the rTAG , but not the rREADER . However, increasing GR and simultaneously keeping PEIRP equal to the power limit set by the regulation actually cannot improve the rTAG . As an example, using the same conditions as those in calculating the curve in Figure 14.4, Equation (14.8a) and PTH = −15 dBm, we can get the rTAG as 8.3 meters.

Read Ranges and Read Rates for Passive RFID Systems

377

Figure 14.4 shows that the rREADER is larger than 8.3 meters with GR = 12 dBi. So, it is unnecessary to select a GR larger than 12 dBi. The aforementioned comments are focused on the optimization designs of RFID readers. However, in a tag-determined RFID system, small improvements in tag performance will greatly improve the read range. Thus, let’s continue our discussions to explore the potential improvements that can be achieved at the tag side. We begin by assuming the tag works with a specified reader having a transmitting power of PEIRP and a specified operating frequency of ω. Then, based on Equation (14.8) for rTAG , in order to improve the read range, we provide the following design considerations: • A larger tag antenna gain GT will efficiently increase the tag received power PR and consequently the read range. Of course, a larger GT also constrains the tag “visible” to the reader only in a narrower viewing angle. So, it is required to weigh the pros and the cons while adopting higher gain tag antennas in a specific application. • A smaller tag power reflection coefficient  will help to increase the tag absorption power PA and consequently the tag can get enough energy to work at a longer read range. However, from Equation (14.15), it is obvious that a smaller  decreases the performance of rREADER at the same time. Thus, in order to get the optimum read range for an RFID system, a compromise needs to be reached between rREADER and rTAG through tuning the reflection coefficient  of the tag. As seen in Figure 14.6 and Figure 14.7, the power absorbed by the tag decreases with the increase of reflection coefficient, while the SNR performance of the reader is to the contrary. Then the power reflection coefficient can be optimized based on those two figures. • An appropriate data coding scheme should be chosen for a tag to improve the read range. It is well known that different coding schemes have different power density spectrums. In addition, the receiver baseband filter of the reader always has a limited pass-band in order to suppress DC, low and high frequency noises. So, the value of −10 −11 −12

PA (dBm)

−13 −14 −15 −16 −17 −18 −19 −20

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Power reflection coefficient Γ

Figure 14.6 The power absorbed by the tag PA vs. the ASK tag’s reflection coefficient . Note: In the calculation, r = 10 m, PEIRP = 36 dBm, PTH = −15 dBm and GT = 2.15 dBi are assumed.

378

RFID Systems

20 18 16

SNR MIN (dB)

14 12 10 8 6 4 2 0

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Power reflection coefficient Γ

Figure 14.7 The SNR of the reader vs. the ASK tag’s reflection coefficient . Note: In the calculation, GT = 2.15 dBi, GR = 13 dBi, PDATA = −6.8 dB, PPN = −39.9 dB, ψ = −37.5 dB, r = 10 m and ξ = 20 dB are assumed.

the numerator of Equation (14.13), which denotes the single-sideband in-band signal power of the tag’s demodulated data sequence, will vary with the coding scheme, which consequently impacts the SNR MIN of the demodulation output signal and thus the read range. So, it is worthwhile comparing different tag data coding schemes while optimizing an RFID system to extend its read range. After presenting our considerations for improving the read range, here we also want to include a brief discussion about other RFID researches found in the literature. In the reference [11], through optimizing the quality factors of tag antenna and tag chip, a design trade-off is achieved between the turn-on voltage of tag chip and the backscattered power from the tag, which helps to increase the read range. In reference [12], an approach which reconfigures the tag antenna as a Yagi antenna by adding parasitic elements at appropriate separations can dramatically increase the tag antenna gain and therefore the tag’s read range. But it also increases the tag size, which confines its application. In references [13–17], considerable research has been conducted on several tag antenna types, including covered slot antenna, circular patch antenna and meander antenna. In addition, a compact reader antenna with dual polarizations is investigated in the reference [18]. Example 14.3.2 Calculating the power reflection coefficient  in an RFID system. From the above analysis, we learn that a proper power reflection coefficient  is important for an RFID system. Assume the tag is placed 10 meters away from the reader and the parameters are the same as those in the calculations of Figures 14.6 and 14.7. Consider: (1) enough power should be absorbed by the tag to power up itself, which means PA should be larger than the turn-on threshold (−15 dBm here), consequently  should be less than 0.58; and (2) the SNRMIN in the receiver should be larger than the SNRUSER , that is, SNRMIN >12 dB, which means  should be larger than 0.55. Therefore, we should select a  between 0.55 and 0.58.

Read Ranges and Read Rates for Passive RFID Systems

379

14.4 Improving the Read Rate of a Passive RFID System The read rate of an RFID system is defined as the statistical ratio of tags successfully read versus the number of all the tags, which describes how well the reader can detect all the tags in the read range. In EPC terminology, “readability,” which refers to the ability of a reader to obtain data from a tag, has a similar meaning and can be regarded as the read rate in the single-tag case. The users of passive RFID systems are very concerned about the read rate. The performance of passive RFID systems is far from satisfactory regardless of the hype and high expectations of early stage RFID systems. The problem lies in non-100 % read rate, which has hampered the widespread use of passive RFID systems. For example, Wal-Mart’s RFID plan was reported as a revolution in commercial supply chain. But, according to the report by EPCglobal, during early field trials at Wal-Mart using its top 100 suppliers, only 89 % reading rate was achieved. Therefore, in this section, we will investigate the low read rate issue and present some solutions to improve the read rate of a passive RFID system. The read range and the read rate of a passive RFID system are closely related. Based on the fundamental principles of wireless communication, the probability of reading a tag successfully, that is, the readability, is determined by the signal noise ratio (SNR) of the backscatter signal received by the reader, and the distance where the threshold SNR occurs corresponds to the maximum read range. It can be reasonably deduced that a reader of longer read range will help to improve its read rate of the tag at a specified distance compared with that of short read range. So, the considerations we propose to improve the read range in previous section are also definitely applicable to improving the read rate. Many researchers, RFID system suppliers and RFID users have presented their observations about the low read rate issue. Based on that information, it is generally thought that an RFID system can reach a satisfactory read rate only under ideal experimental conditions, but the read rate may substantially deteriorate in real-world applications. Obviously, the difficulty lies in the complex environment surrounding the tags and the readers. Now let’s examine the environmental factors that can affect the RFID performance. Environmental factors include the atmosphere, the electromagnetic environment and the surroundings and obstacles between tags and readers. The atmospheric factors, such as wind, rain and temperature, need to be considered while trying to achieve a higher readability. For example, Ultra High Frequency (UHF) tags are impacted much more by wind and sun than High Frequency (HF) tags [19]. Temperature can affect the performance of chips and then the read rate. Among many environmental factors, the surroundings and obstacles are the biggest troubles, which can absorb and reflect the energy of electromagnetic waves, and thus cause interference or multi-path fading effects. Obstacles in the environment can exhibit various electromagnetic behaviors depending on their material compositions. Generally, they are electromagnetically transparent, reflecting, and/or absorbing. Metals and water cause the biggest issues. They reflect and/or absorb electromagnetic waves respectively, which means the passive RFID tag has no chance of receiving sufficient operating power. In practical applications, many goods contain metal and/or water, such as canned food, detergents and drinks. Experiments show that the water has a large impact on the performance of tag, especially in the UHF range. It is found that tags perform well while reading through frozen beef, but not when the beef is thawed, illustrating the liquid water problem [20].

380

RFID Systems

When an RFID tag is placed in close proximity to a metallic surface, the metal will detune the tag antenna and lower its performance. Currently, there are several methods to solve this issue. One is to use a foam separation. For example, if tags and the metal surface are separated by a piece of foam of about 8–10 mm thickness, the detuning effect will weaken dramatically. However, one shortcoming of this technique is that it also will increase the thickness of tags. Another method is to use wave-absorbing material as the substrate of tags, but wave-absorbing material is very expensive at present. Alternatively, a better solution is to use the metallic surface as the tag backplane and then make it behave as an integral part of the tag antenna [21]. Local surroundings can create unwanted multipath signals and deteriorate the read rate. The multipath signals alter, modify or disrupt a message when it travels between a reader and a tag, similar to the interference that one experiences with a car radio when driving past an intense RF emission source [20]. The effect of surroundings and obstacles generates the need to study the stack-up orientation of tags. Many tag manufacturers have declared that the tag orientation has little effect on the read range and the readability of their tags. However, there are some disputes about the importance of relative orientation of a tag’s antenna relative to the reader’s antenna. In reference [22], Huang et al . studied the above problem experimentally. They adopted the standard procedure that Wal-Mart mandates its suppliers to follow [22]. The experiment is as following: Twenty cartons of goods are affixed with passive RFID tags and are loaded onto a pallet. The pallet is passed through two reader antennas. The reader was equipped with a middleware (RFID application software) to read the ID information from the tags. The reader antennas form a 3 m-wide channel for pallets to pass through. There are different palletizing methods which cause different orientations of tags with respect to the reader antenna. The authors compared several palletizing methods in such a situation and concluded that the read rate of an RFID system is mainly affected by relative positions of the tags and the reader antennas. Some researchers have managed to solve the low read rate issue from a novel point of view. They improved the RFID read rate/reliability through a systematic error detection approach. An intelligent middleware solution can be implemented to detect missed tags. But a shortcoming is that it requires some prior information (the number of tagged items that are entering the interrogation zone), which might be unavailable in some cases. So they proposed a new solution where an RFID reader works along with a normal weighing machine. The idea is to compare the gross weight of the tagged items against the gross weight (of the same items) stored in a back-end database. If some tags are not read at all, these weights would vary and hence incorrect readings could be identified [19]. Passive RFID technology has begun to be adopted in supply chain applications. Although it would be more cost-efficient if a 100 % read rate could be achieved through setting up only one read point, it is not a realistic expectation at the current maturity level of RFID technology. One solution to improve the read rate is to add a few more read points in the supply chain. For instance, if an RFID system can achieve a 50 % read rate at an individual point, then according to the basic probability theory, it can reach a 97 % read rate (about 1 in 33 products missed) with 5 read points or a 99.9 % read rate (about 1 in 1000 products missed) with 10 read points in the end of the supply chain [23]. Recently, the array processing technology in radar system has been introduced into RFID systems. In fact, an RFID system also can be regarded as a radar. This technology

Read Ranges and Read Rates for Passive RFID Systems

381

has the potential to greatly improve RFID performance, including read rate and read range. A typical array processing technology is the beam forming. In some cases, we need the reader antenna to radiate electromagnetic waves in a specific pattern to decrease interferences among tags. The ordinary antenna’s radiation pattern is fixed. However, the antenna array system could steer the radiation beam to selectively cover targeted transponders and thus reduce reading errors and collisions among tags. This technology exploits the spatial diversity and the polarization diversity of tags. The directional beam also reduces the effects of multi-path fading [24]. A novel technology will always face plenty of problems at the beginning. For instance, the bar code technology also had its own problems during early implementations. At present, the EPC Generation 2 RFID tag has shown a remarkable improvement in its ability to identify multiple items. It also has achieved higher read rates and read speeds compared to the Generation 1 tag. We think that, through researchers’ efforts, it looks hopeful that the low read rate problem for passive RFID will be solved in near future.

14.5 Two Design Examples for RFID System Based on the aforementioned discussions, we built two RFID readers with off-the-shelf components, one working at the frequency of 915 MHz and the other at 2.45 GHz, and set up the systems with commercial UHF RFID tags. Those two readers are shown in Figure 14.8 and Figure 14.9, respectively. Interface From Antenna

RF_IN

Receiver

To Baseband Processor

Local Oscillator

To Antenna

Transmitter From Baseband Processor

RF_OUT

Figure 14.8 The 915 MHz RFID reader prototype.

382

RFID Systems

Power Amplifier

Circulator

Coupler & Detector

Drive Amplifier

Mixer

Figure 14.9

Logic Circuit

The 2.45 GHz RFID reader prototype.

The major components for 915 MHz reader include: the frequency synthesizer AD43607, the modulator LT5568, the demodulator LT5575, the first-stage low noise amplifier LT6600-2.5, the second-stage low noise amplifier LT6231, the baseband filter LT1568, the power amplifier SKY65111-348LF, and the directional coupler and detector DD02-999. This reader design also has a circulator sub-module, which is not shown in Figure 14.8. The major components for 2.45 GHz reader include: the frequency synthesizer AD43600, the power splitter SCN-2-27, the 90-degree phase-shifting power splitter QCN-27, the mixer SYM-36H, the power amplifier PA2455, the directional coupler and detector DD02999, and the circulator RC-SS-CC-2.4-2.5-10WR. This reader design also has a baseband sub-module, which is not shown in Figure 14.9. Most of those chips are originally designed for commercial wireless communications, which makes it feasible to design cost-effective readers to promote RFID applications. Here, we use the 915 MHz reader as an example and present its design details. First, we list some key design parameters as follows: for the tag, its half-wavelength dipole antenna has a gain GT of 2.15 dBi, its input threshold power level PTH is −15 dBm, its tag data rate is 160 kbps, the unipolar FM0 coding is utilized for the tag’s data sequence, the ASK modulation method is utilized for backscatter communication and the power reflection coefficient  takes a value of 1; for the reader, the allowed maximum PEIRP is 36 dBm according to the US regulations, it has a vertical-linear-polarization panel antenna that has a gain GR of 13 dBi, its transmitting power PT is constrained to 23 dBm, the receive-transmit isolation coefficient ξ of the reader is 20 dB, which is mainly limited by the circulator’s performance, its baseband band-pass filter approximately has an ideal

Read Ranges and Read Rates for Passive RFID Systems

383

rectangular transfer function, whose low-end cutoff frequency is 10 kHz and high-end cutoff frequency is 320 kHz. Using Equation (14.8a) and based on the parameters given above, we can calculate that the tag-determining maximum operational distance takes a value of 8.3 m and the reader-determining maximum operational distance takes a value of 11.8 m, which shows that this is a tag-determining RFID system. However, it is not an easy task to evaluate rTAG directly by measuring the time-averaging absorption power PA of the tag because the tag IC is really very tiny and testing fixtures may have a big impact on the tag antenna. However, for the purpose of only judging whether the system is tag-determining or reader-determining, we just conduct two comparative operational distance experiments as illustrated in Figure 14.10 (a) and Figure 14.10 (b), which utilize the receiver-attenuation and the transmitter-attenuation, respectively. We observe that the reader cannot detect the tag’s data signal earlier in the case of transmitter-attenuation than in the case of receiver-attenuation while the operational distance increases. Thus, we can conclude that this is a tag-determining RFID system.

Reader Antenna Transmitter

Tag

r

LO Circulator Demodulation Output Signals Receiver

Attenuator (a) Reader Antenna

Transmitter

Attenuator

Tag

r

LO Circulator Demodulation Output Signals Receiver (b)

Figure 14.10 Comparative operational distance experiments. (a) Receiver-attenuation case. (b) Transmitter-attenuation case. Reproduced courtesy of  2007 EMW Publishing.

384

RFID Systems

In addition, using the transmitter-attenuation, we get a maximum operational distance of 4.1 m while the attenuation takes a value of 6 dB. Then, based on Equation (14.7a), we can get an estimated value of 8.2 m for rTAG , which is close to the above calculated value for rTAG . Before using Equation (14.15a) to calculate the reader-determining maximum operational distance rREADER , we need to know the single-sideband in-band signal power PDATA of the tag’s binary data sequence and the band-limited phase noise power PPN of the LO. First, we manage to figure out the PDS of PDATA . Since each tag has a unique data sequence of finite-length (64 bits for the current tag), one tag’s PDS is generally different from another. Here, instead of directly calculating the PDS of a specific tag, we evaluate it in a statistical-averaging-sense through constructing a random-generated data sequence of infinite length. We assume that this sequence is a wide-sense stationary ergodic stochastic process. Then, we can achieve a good approximate result of its PDS through applying discrete fast Fourier transform (DFFT) to its any segment of finite-length (1024 bits here). Then, we numerically integrate its PSD within the receiver’s pass-band of [10 kHz, 320 kHz], which means that PDATA takes a value of −6.8 dB. Second, we utilize an SPA directly to measure the normalized phase noise power of the LO within the receiver’s bandwidth, which means that PPN takes a value of −39.9 dB. In addition, we need to know the phase noise improvement factor ψ. In order to evaluate ψ, we set up two comparative demodulation output noise experiments as illustrated in Figure 14.11(a) and Figure 14.11(b), respectively. The phase noise of the signal generator is much lower than the LO’s and they are uncorrelated to each other. In addition, considering that a SPA device generally does not support accurate measurement of very-low-frequency signals, we utilize a digital oscilloscope to measure the power of demodulation output noise in terms of root mean squared (RMS) value. Then, through evaluating the ratio of the noise power measured in Figure 14.11(b) to that measured in Figure 14.11(a), we find that ψ takes a value of −37.5 dB. Now, substituting the user-specified SNR USER of 12 dB into Equation (14.15a), we find that the reader-determining maximum operational distance rREADER is 11.8 m, which shows further that this is a tag-determining RFID system. Then, let’s put this RFID system in an open indoor environment and test its SNR performance in practice. Here, in order to characterize the reader’s performance in a practical operational scenario where there is a lot of clutter from surrounding obstacles, we do not choose to conduct the measurements in an anechoic chamber. Figure 14.12 shows the measured SNRs of the reader’s demodulated output for several different distances of 2.0 m, 3.0 m, 4.0 m, 5.0 m, 6.0 m, 7.0 m and 8.4 m together with the calculated SNR MIN using Equation (14.14a). We can get a maximum operational distance of 8.4 m, where the SNR of the demodulated output signal is 17.5 dB, which is well above the user-specified SNR USER of 12 dB. Thus, this RFID system indeed is tag-determining and accordingly the read range takes a value of 8.4 m. In Figure 14.12, we can see some small offsets between the measured SNR and the calculated SNR MIN . In addition, the calculated rTAG is a little less than the measured read range. There mainly are two reasons for the offsets. One is the indoor multi-path effects of electromagnetic wave propagation [25–27] that reduce the accuracy of the free space Friis electromagnetic wave propagation equation that has been utilized in Equations (14.1) and (14.5). The other is the inherent SNR uncertainty of the uncorrelated demodulation scheme as discussed in Section 14.3. However, these calculated results indeed have

Read Ranges and Read Rates for Passive RFID Systems

385

Reader Antenna Non-correlated Signal Generator

Transmitter

LO Circulator Demodulation Output Noise Receiver (a) Reader Antenna Transmitter

LO Circulator Demodulation Output Noise Receiver (b)

Figure 14.11 Comparative demodulation output noise experiments. (a) Uncorrelated case. (b) Correlated case. Reproduced courtesy of  2007 EMW Publishing. 60 55

Calculated SNRMIN Measured SNR

50 SNR (dB)

45 40 35 30 25 20 15 10

1

2

3

4

5

6

7

8

9 10 11 12 13

r (m)

Figure 14.12 Calculated SNRMIN and measured SNR of the reader’s demodulator output signal for several different distances r. Reproduced courtesy of  2007 EMW Publishing.

386

RFID Systems

good approximations to the actual measurements, which illustrate the effectiveness of the methodology presented here when designing long read range readers.

14.6 Conclusion This chapter investigates the primary working principles of passive RFID systems in farfield range. The signal descriptions and formulations for signal noise ratio and read range are theoretically presented. Based on these equations, read range and read rate issues are discussed further. Apart from the conclusions drawn directly from those equations, other research efforts related to improving read rate and range are also introduced. At the end of this chapter, two examples are presented to demonstrate cost-effective designs of long-range RFID readers with off-the-shelf wireless communication chips.

Acknowledgements The authors of this chapter would like to thank EMW Publishing for kindly allowing us to reuse some theoretical and experimental results originally published in Progress in Electromagnetics Research, PIER71, 109–127, 2007, referenced as [2] in this chapter.

Problems 1. Consider a PSK tag working at 2.5 GHz, and a reader radiating a power of 20 dBm. Assume the tag’s antenna has a gain of 3 dBi and  = 1/2. If the rTAG is designed to be 10 m, what is the value of the turn-on threshold power of the tag? 2. Consider a passive RFID system composed of an ASK tag and a reader working at 950 MHz. Assume the gain of the tag antenna is 3 dBi, and the user-specified threshold SNRUSER = 12 dBm, calculate the gain of" the reader antenna if the rREADER f is 10 m. (Suppose ξ = 20 dB,  = 1, PDATA = fLH S{S(t)} df = −6.8 dB, PPN = " fH fL S{θP (t)} df = −39.9 dB and ψ = 37.5 dB). 3. Consider a passive RFID system composed of a PSK tag and a reader working at 950 MHz. Assume the gain of the tag antenna is 3 dBi, the gain of the reader antenna is 15 dBi and the isolation efficient of the circulator ξ is 20 dB, find the SNRMIN performance of the system when the tag is placed 10 m far "f " f from the reader. (Suppose  = 1/2, PDATA = fLH S{S(t)} df = −6.8 dB, PPN = fLH S{θP (t)} df = −39.9 dB and ψ = 37.5 dB). 4. Consider an ASK tag working at 915 MHz. Assume the gain of the reader antenna is 13 dBi and the user-specified threshold SNRUSER is 12 dB. If the tag antenna is a half wavelength dipole, find the rREADER while varying the tag" antenna’s orienf tation to the reader antenna. (Suppose ξ = 20 dB,  = 1, PDATA = fLH S{S(t)} df = " fH −6.8 dB, PP N = fL S{θP (t)} df = −39.9 dB and ψ = 40.9 dB). 5. In Section 14.5, a 915 MHz passive RFID system is introduced and a group of specific values for its parameters were designed, calculated and measured. Then, the problem

Read Ranges and Read Rates for Passive RFID Systems

387

for the readers is as follows: figure out the values of the parameters of a 2.45 GHz RFID system, which has a reader-determined read range rREADER = 15 m and a tagdetermined read range rTAG = 10 m.

References [1] RFID Implementation Cookbook. Available at: http://www.epcglobalinc.org/what/ cookbook/. [2] Fan, Z.G., Qiao, S., Huangfu, J.T., and Ran, L.X. (2007) Signal descriptions and formulations for long range UHF RFID reader, Progress In Electromagnetics Research, PIER 71: 109–127. [3] Landt, J. (2005) The history of RFID, IEEE Potentials, 24(4): 8–11. [4] Brown, W.C. (1984) The history of power transmission by radio waves, IEEE Trans. Microwave Theory Tech, 32(9): 1230– 1242. [5] Stockman, H. (1948) Communication by means of reflected power, Proc. IRE , Oct. pp. 1196– 1204. [6] Harrington, R.F. (1964) Theory of loaded scatterers, Proc. Inst. Elect. Eng., 111(4): 617– 623. [7] Koelle, A., Depp, S., and Freyman, R. (1975) Short-range radio-telemetry for electronic identification using modulated backscatter, Proc. IEEE ., 63(8): 1260– 1260. [8] Finkenzeller, K. (2003) RFID Handbook: Radio-Frequency Identification Fundamentals and Applications, 2nd edn. New York: John Wiley & Sons, Ltd. [9] Ruck, G.T., et al . (1970) Radar Cross Section Handbook , vols. 1–2. New York: Plenum. [10] Saunders, W.K. (1990) CW and FM radar, in M. Skolnik (ed.) Radar Handbook . 2nd edn. New York: McGraw-Hill. [11] Lee, J.W., Kwon, H., and Lee, B. (2006) Design consideration of UHF RFID tag for increased reading range, in Proceedings of IEEE MTT-S International Microwave Symposium, San Francisco, USA, June, pp. 1588– 1591. [12] Cheng, C.H., and Murch, R.D. (2007) Antenna modifications for enhancing RFID tag reading range, in Proceedings of 2007 IEEE Antennas and Propagation Society International Symposium, Honolulu, HI, USA, June, pp. 1084– 1087. [13] Rao, K.V.S., Nikitin, P.V., and Lam, S.F. (2005) Antenna design for UHF RFID tags: a review and a practical application, IEEE Trans. Antennas Propagat., 53(12): 3870– 3876. [14] Foster, P.R., and Burberry, R.A. (1999) Antenna problems in RFID systems, in IEE Colloquium on RFID Technology (Ref. No. 1999/123), 3: 31–35. [15] Karthaus, U., and Fischer, M. (2003) Fully integrated passive UHF RFID transponder IC with 16.7-µW minimum RF input power, IEEE Journal of Solid State Circuits, 38(10): 1602– 1608. [16] De Vita, G., and Iannaccone, G. (2005) Design criteria for the RF section of UHF and microwave passive RFID transponders, IEEE Trans. Microwave Theory Tech., 53(9): 2978– 2990. [17] Curty, J.P., et al . (2005) Remotely powered addressable UHF RFID integrated system. IEEE Journal of Solid State Circuits, 40(11): 2193– 2202. [18] Zhang, M., Chen, Y., Jiao, Y., and Zhang, F. (2006) Dual circularly polarized antenna of compact structure for RFID application, Journal of Electromagnetic Waves and Applications, 20(14): 1895– 1902. [19] Potdar, V., Hayati, P., and Chang, E. (2007) Improving RFID read rate reliability by a systematic error detection approach, in Proceedings of the 1st RFID Eurasia Conference, Istanbul, Turkey, pp. 1–5. [20] Clarke, R.H., Twede, D., Tazelaar, J.R., and Boyer, K.K. (2006) Radio Frequency Identification (RFID) performance: the effect of tag orientation and package contents, Packaging Technology and Science, 9: 45–54. [21] Daily, J., and McCann, R. (2007) Improving RFID read rate in metallic tractor-trailer applications, in Proceedings of IEEE Region 5 Technical Conference, Fayetteville, AR, USA, pp. 404– 408. [22] Huang, C.T., Lo, L.W., Wang, W.L., and Chen, H.L. (2008) A study for optimizing the reading rate of RFID tagged cartons in palletizing process, in IEEE International Conference on Industrial Engineering and Engineering Management, Singapore, pp. 1138– 1142. [23] Cook, C., and Brown, M. (2006) Practical Performance Expectations for Smart Packaging, Texas Instruments and RFID4U White Paper, December. [24] Karmakar, N.C., Roy, S.M., and Ikram, M.S. (2008) Development of smart antenna for RFID reader, in IEEE International Conference on RFID, Las Vegas, NV, USA, pp. 65–73.

388

RFID Systems

[25] Kim, D., Ingram, M.A., and Smith, W.W. (2003) Measurements of small-scale fading and path loss for long range RF tags, IEEE Trans Antennas Propagat., 51(8): 1740– 1749. [26] Yarkoni, N., and Blaunstein, N. (2006) Prediction of propagation characteristics in indoor radio communication environments, Progress In Electromagnetics Research, PIER 59: 151–174. [27] Martinez, D., Las-Heras, F., and Ayestaran, R.G. (2007) Fast methods for evaluating the electric field level in 2D-indoor environments, Progress In Electromagnetics Research. 2007; PIER 69: 247– 255.

15 Principles and Techniques of RFID Positioning Yimin Zhang, Xin Li, and Moeness Amin Villanova University

15.1 Introduction Radio frequency identification (RFID) systems, which basically consist of readers and tags, were originally developed for the identification of tagged objects, as the name RFID implies. Recently, precise positioning and tracking of RFID tags or readers have received considerable attention from both academia and industry. For example, finding the position of RFID tags is an important task in various real-time locating systems (RTLS) to locate and track products, assets, and personnel with attached RFID tags in an area covered by the RFID readers, for example, [1, 4, 15, 17, 51, 52, 69, 74, 75, 85]. In other applications, it is desirable for a reader to identify its own position with the assistance of reference tags, for example, [16, 18, 26, 42, 66, 73, 80, 81, 87]. Numerous RFID localization products have been developed for various applications. RFID positioning techniques and applications have been surveyed in [9, 13, 46, 89, 64]. It is pointed out that, depending on applications, the required positioning accuracy may differ. For some applications, such as parcel tracking in a warehouse, accuracy of 1 meter is acceptable and considered sufficient [35], whereas an accuracy of several centimeters is desired to unambiguously identify tagged parcels placed on a conveyer belt [85]. Therefore, it is important to select the appropriate positioning technique that meets the varying positioning accuracy and cost requirements. This chapter provides a comprehensive introduction of the principles and techniques of RFID positioning. The majority of RFID positioning systems are based on the fusion of multiple pieces of relevant information. Examples of such information include range,

RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

390

RFID Systems

direction-of-arrival (DOA), and the propagation characteristics which are obtained from signal strength, time, and/or phase measurements at a single or multiple antenna positions. One of the commonly used approaches to locate an RFID tag is trilateration. This approach determines the tag position by incorporating the range information of an RFID tag estimated at multiple reader antenna positions. Range information can be obtained, for example, through the received signal strength (RSS), round-trip time-of-flight (TOF), timedifference-of-arrival (TDOA), and/or phase-difference-of-arrival (PDOA) of the RFID signals. RSS is an easily measured quantity which provides range information based on the fact that a radio frequency (RF) signal emitted or backscattered from a tag attenuates with a law related to the distance which the RF signal traveled. Passive (including semipassive) RFID systems can use round-trip TOF of a signal transmitted from an RFID reader and backscattered from the RFID tag for the estimation of the round-trip distance between the reader and the tag. In active RFID systems, the estimation of one-way TOF from the tag to the reader is often difficult because it requires precise synchronization between the reader and the tag. A rather practical solution, referred to as TDOA, is to compute the difference of the time-of-arrival (TOA) between multiple reader antennas which receive the same signal transmitted by the RFID tag. Accurate range estimation using time-based techniques, such as TOF and TDOA, may require a wideband signal to be used. Range estimation using PDOA utilizes the different phase delays exhibited by signals with different carrier frequencies when propagating over the same distance between the reader and the tag. Tag localization can also utilize the triangulation technique based on the DOA information observed at multiple reader antenna positions. Array processing techniques that exploit the phase information of signal arrivals observed at multiple collocated antennas can achieve high DOA estimation accuracy. When high positioning accuracy is not required, the use of directional antennas is a low-cost alternative approach to obtain DOA information. RFID tag positions can also be determined at a single reader position by combining range and DOA information. Other RFID tag locating techniques include proximity and radio map matching. The former locates a tag by finding the closest reader antenna, whereas the latter compares the RSS or other signal signatures of a tag with that of reference tags whose positions are known a priori . It is worth noting that a very challenging problem in RFID positioning lies in the effect of complicated wave propagation due to the presence of various obstacles and reflectors in the environment. Walls, human bodies, furniture and supplies that contain metallic and liquid materials, such as partitions, cabinets, bookshelves, water containers, may cause obstruction and reflections of electromagnetic waves. When a signal transmitted/backscattered from an RFID tag arrives at an RFID reader over a multiplicity of paths, it extends the delay profile and results in fluctuation in the RSS as well as the received signal phase [58, 90]. Similar effects can be observed for downlink propagation from a reader to a tag. Multipath propagation alters both signal strengths and phase. As such, RFID positioning techniques based on RSS and/or signal phase may become inaccurate. The localization performance of the TOF- and TDOA-based techniques may also be compromised due to multipath. A number of approaches have been developed to improve the RFID positioning performance in a multipath environment. A class of such approaches is based on the comparison of the RSS, tag count/tag detection rate, and/or distribution probability of RSS

Principles and Techniques of RFID Positioning

391

corresponding to an unknown RFID tag with those measured for multiple known reference tags located in its vicinity [28, 51, 68, 79]. Because tags in the same area and close proximity are likely to be similarly affected by the propagation channel, multipath propagation effect can be mitigated through calibration using the information collected from the reference tags [61]. A well-known example of such approaches is LANDMARC [51]. Another approach is to use ultra-wideband (UWB) signals that allow high-resolution delay profile estimation and thereby achieve effective discrimination of multipath signals. As a result, the range information can be estimated based only on the TOF of the direct path [48, 50, 53]. The use of frequency hopping signals can achieve frequency-domain diversity to combat multipath effects [3]. RFID-based positioning and tracking techniques share similarities with those exploiting other technologies, such as acoustic, ultrasound, vision, infrared, laser, radar, GPS and WLAN (see, for example, [10, 27, 55, 78]). In particular, positioning of active RFID systems bears great similarity to WLAN-based positioning techniques. The concept of a passive RFID system, on the other hand, is analogous to an active radar system for the estimation of tag range and DOA, whereas differences exist as well. An RFID system is usually required to operate in much reduced complexity compared to other systems. Contrasting with radar systems which are typically wideband, non-UWB RFID systems use a much narrower frequency bandwidth. In addition, RFID systems have anti-collision capabilities and, as a result, only one tag needs to be considered at a time. While rangeDoppler radars use the Doppler frequencies associated with moving targets to mitigate clutter, many RFID systems do not benefit from Doppler frequencies for clutter mitigation as they are typically stationary or move at a low-speed. Rather, backscattered signals from passive RFID tags are modulated and thus their spectra are shifted from that of the energizing carrier signal, enabling the suppression of strong carrier presence from the reader and clutter reflection. Non-RFID positioning techniques can be incorporated into an RFID system to enhance its positioning capability, such as coverage extension and positioning accuracy improvement. For example, incorporating GPS and other global navigation satellite system (GNSS) receivers into RFID tags can provide positions in wide outdoor areas. WLAN infrastructure can be used to construct RTLS in indoor environments [20]. Passive RFID and laser range scanner are jointly used to improve the localization accuracy of mobile robots and persons [26]. A laser-activated RFID-based indoor localization system was developed for mobile robots, where a number of laser-activated active tags are placed in the environment as landmarks [87]. A real-time identification and localization system, named LotTrack, uses active RFID and ultrasound technologies to improve tracking visibility for logistics in a wafer fabrication clean-room [74]. The aim of this chapter is to provide a comprehensive introduction of the principles and techniques involved in RFID positioning. We first summarize key principles of information acquisition, and then introduce RFID positioning algorithms and techniques. As depicted in Figure 15.1, an RFID positioning system typically involves two major functional blocks, that is, location sensing and positioning processing. The location sensing block senses the tag location in terms of range and/or DOA using proper location metrics. This function block is discussed in Sections 15.2 and 15.3, respectively, for range and DOA estimation techniques. The objective of the positioning processing block is to find the location of an RFID tag or reader based on the information obtained from the location sensing

392

RFID Systems

Location Sensing Received RF Signals

Positioning Processing

RFID Positioning Information

Location Sensing Range: Section 15.2 DOA: Section 15.3

Figure 15.1

Section 15.4

Block diagram of an RFID positioning system.

block. The positioning processing algorithms, techniques, and applications are addressed in Section 15.4. In Section 15.5, possible measures for the improvement of positioning accuracy are discussed. The chapter is concluded in Section 15.6.

15.2 Tag Range Estimation Techniques Many RFID positioning techniques are based on the range information of tags, evaluated from a single or multiple RFID reader antennas. The accuracy of range estimation, therefore, directly affects the performance of the positioning of RFID tags. In this section, range estimation techniques based on RSS, phase, and time measurements are presented.

15.2.1 RSS-Based Techniques For an active RFID system operating in a free space environment, the signal power received at the reader is expressed as [19]   λ 2 PRX ,reader = PTX ,tag Gtag Greader (15.1) 4πd where PTX ,tag is the transmit power at the active tag, Gtag and Greader are the antenna gain of the tag and the reader, respectively, λ is the wavelength, and d is the range between the tag and reader. For a passive RFID system, the signal is transmitted from the reader and backscattered at the tag. Thus, a round-trip path loss should be considered. The received signal power becomes   λ 4 2 2 PRX ,reader = PTX ,reader ηGtag Greader (15.2) 4πd where PTX ,reader is the transmit power from the reader and η is the power transfer efficiency of the passive tag. The typical value of η is 1/3 or −5 dB [19], but this value may change as technology advances. Equations (15.1) and (15.2) clearly show that the RF signal power decays with d. In the above two equations, which consider free-space propagation, the one-way power

Principles and Techniques of RFID Positioning

393

attenuation is proportional to d 2 , whereas the round-trip power attenuation is proportional to d 4 . The actual attenuation rate varies, however, depending on the environment where the RFID system is deployed. In this case, Equation (15.2) is revised as   λ 2n 2 2 (15.3) PRX ,reader = PTX ,reader ηGtag Greader 4πd which shows that the signal strength is inversely proportional to d 2n , where n is referred to as the path loss exponent. The typical value of n is between 1.6 and 1.8 for line-of-sight (LOS) indoor environments, and between 2 and 6 for outdoor propagation environments [58]. As a result, when the path loss exponent is known or can be estimated for a specific environment, range d can be estimated from RSS measurements. The ratio between the backscattered signal power a reader receives from a passive RFID tag and the power the reader transmits is illustrated in Figure 15.2, where different values of path loss exponent n are considered. The following parameters are used: f = 915 MHz, η = 1/3, Gtag = 1 (or 0 dB), and Greader = 4 (or 6 dB). In practice, tags are expected to have low directivity, whereas a reader antenna gain of about 6 dB is commonly used because FCC regulations require proportional reduction of the transmit power when the transmit antenna gain exceeds 6 dB. The use of RSS for range estimation is simple and handy [5, 18, 25, 30, 31]. Many RFID readers make the RSS information available. Moreover, the tag detection rate or tag count patterns also provide information related to the RSS and thus can be used to estimate the range [68, 79]. That is, the number or rate of successful readings of a tag is associated with the RSS information. The tag detection rate and the tag count patterns can be fused with the RSS measurements to increase the localization reliability [70]. Multiple

−10 n = 1.6 n = 1.8 n=2 n=3 n=4 n=5 n=6

−20 −30 Power Ratio (dB)

−40 −50 −60 −70 −80 −90 −100 −110 −120 10−1

100

101

102

Tag Range (m)

Figure 15.2 Ratio between the received power and the transmitted power of a passive RFID reader for different path loss exponents (f = 915 MHz, η = 1/3, Gtag = 0 dB, and Greader = 6 dB).

394

RFID Systems

measurement results can be used to improve the reliability of RSS-based range estimations, particularly in a multipath environment. A convenient way is to use a frequency hopping method in which a reader sends out bursts at specified intervals in a frequency hopping manner and measures response at the corresponding frequency each time [3]. As such, multiple RSS values evaluated at these frequencies become available and the reliability of range estimation is improved by selecting the highest, average, median values, or by using other combining approaches. The range estimation performance from RSS is, however, not robust, particularly when the RFID system is operated in a complex propagation environment. Obstruction of LOS between the reader and the tag yields additional signal loss, often referred to as shadowing. The effect of shadowing is typically characterized by using the log-normal distribution 2 with variance σsh . In addition, RSS is also sensitive to reflection and scattering from walls, furniture, and various conductive materials in the propagation environment. Such reflection and scattering yield the multipath fading phenomenon. The effect of multipath fading can be described using the Rayleigh distribution or the Ricean distribution. The latter assumes that a dominant direct path is present, whereas the former assumes no dominant path. Various models are available to describe and predict the RSS in different indoor and outdoor propagation environments [58]. For unbiased parameter estimators, the Cramer-Rao lower bound (CRLB) is known to provide the minimum achievable mean square error (MSE) of a set of parameters, given the probability density function (PDF) of random variables involved in the problem [56]. ˆ the CRLB of the range estimation error due to log-normal For the range estimation d, shadowing effect is given, in the form of root mean square error (RMSE), as [24, 57]. ˆ ≥ RMSE(d)

ln 10 σsh d 10 n

(15.4)

It shows that the error increases with the standard deviation σsh of the shadowing, and decreases with the path loss exponent. In addition, the accuracy of RSS-based range estimation degrades as the range d increases. Figure 15.3 shows simulated RSS results to illustrate the significance of path obstruction and reflection for a downlink scenario [90]. The reader transmits a 1 W RF signal from its antenna located at x = 0 m, y = 0.5 m, and z = 0.5 m. The signal power received by a tag at various locations is shown in different gray levels in dBm. In a free space, as depicted in Figure 15.3(a), the received signal power is monotonically attenuated as the range increases. When a metallic cabinet, consisting of perfect conductor surfaces, is placed at the side of the link path, as depicted in Figure 15.3(b), the RSS behind the cabinet is significantly reduced due to path obstruction. On the other hand, when the cabinet is in front of the transmitter, as depicted in Figure 15.3(c), the received power oscillates with observation locations. In the latter two cases, the RSS no longer shows a unique, monotonic relationship with the range, making RSS-based range estimation difficult.

15.2.2 Phase-Based Techniques PDOA-based approaches allow coherent signal processing and, therefore, at least in theory, can improve range estimation performance of passive RFID tags compared to RSS-based techniques [38]. During a time period designated for uplink data transmission,

Principles and Techniques of RFID Positioning

395

1

x

y

y (m)

Reader antenna

z

10

0.8

0

0.6

−10

0.4

−20 −30

0.2

−40 0 0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

−50

x (m)

(a) Free space 1

10

0.8

0

0.6

−10

Reader antenna y

x

y (m)

z

−20

0.4

−30 0.2

−40

0 0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

−50

x (m)

(b) Metallic cabinet at the side of link path 1

y (m)

z Reader antenna y

x

10

0.8

0

0.6

−10 −20

0.4

−30 0.2

−40

0 0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

−50

x (m)

(c) Metallic cabinet in front of the reader

Figure 15.3

Simulated RSS in free space and in the presence of a metallic cabinet.

a reader transmits two continuous-wave (CW) signals which are then backscattered by a tag and received at the reader. The two CW signals propagate over the same distance, but their phase delays are proportional to their respective carrier frequencies. Therefore, a reader can estimate the tag range based on the phase difference observed at the two frequencies. Note that the response time that an IC-based passive tag takes is irrelevant to the phase difference used in PDOA-based range estimation. PDOA-based approaches share the same concept as the dual-frequency radar techniques for range estimation [2, 83]. Consider that an RFID reader transmits two CW signals at frequencies f1 and f2 . Without considering the modulation performed at the RFID tag and the receiver noise, the phase of the uplink signal at frequency fi can be expressed as φi = 4πfi d/c, where

396

RFID Systems

i = 1, 2, c = 3 × 108 m/s is the velocity of RF signal propagation, and d is the range between the reader and tag. Therefore, range d can be estimated from the phase difference observed at the return signal corresponding to the two frequencies. In reality, the phase observation is subject to wrapping, that is, the phase at each frequency is observable only within the range 0 ≤ φi < 2π. As a result, the tag range is estimated as dˆ =

cm cφ + 4π(f2 − f1 ) 2(f2 − f1 )

(15.5)

where 0 ≤ φ = φ2 − φ1 < 2π is the wrapped phase difference observation and m is an unknown integer. The second term in the above expression denotes the range ambiguity due to phase wrapping. Note that, because backscattering modulation changes the signal phase at both carrier frequencies in the same way, Equation (15.5) remains valid when the backscattering modulation is applied. The maximum unambiguous range is dmax = c/ [2|f2 − f1 |]. For example, when f = |f2 − f1 | = 10 MHz, dmax is 15 m. When f = 1 MHz, dmax becomes 150 m. Clearly, a large frequency separation, which is more resistant to noise [45], yields a small value of dmax . The PDOA approaches can be extended to more than two frequencies to provide multiple frequency pairs. The ranges estimated from multiple frequency pairs can be averaged to yield a more robust range estimation against noise and other perturbations [38, 39]. In this case, equal frequency separation is desirable to obtain range estimates with similar variance. Multiple frequencies can also be designed to have unequal separation for robust range estimation over an extended unambiguous tag range [45]. Instead of simultaneous transmission of multiple CW signals, they can also be transmitted in sequential, yielding a frequency hopping implementations [84]. The advantage of PDOA approaches lies in its high accuracy in range estimation and its robustness to the variation of signal strength due to obstructions [18]. On the other hand, the range estimation accuracy is sensitive to the phase distortion caused by multipath propagation. Averaging over multiple frequency pairs or/and multiple estimation results may improve the reliability in a multipath environment.

15.2.3 Time-Based Techniques Measurement of round-trip TOF in a passive RFID system can be used to estimate the tag range as c · (TOP − Tp ) TOF = dˆ = c · 2 2

(15.6)

where TOP is the overall round-trip time delay which includes the round-trip propagation time, TOF, as well as the signal processing time consumed at the tag circuitry, denoted as Tp . The measurement of round-trip TOF or TOP only utilizes the clock at the reader and thus does not require clock synchronization between the reader and the tag. For active RFID tags, on the other hand, measurement of one-way TOA requires that the reader and the tag have precisely synchronized clocks. In many active RFID systems, however, achieving precise synchronization between a reader and a tag is impractical.

Principles and Techniques of RFID Positioning

397

Rather, it is often feasible to synchronously process the data received at multiple readers or reader antennas, and thus the TDOA related to different tag-reader antenna paths can be estimated. The TDOA information obtained from a pair of reader antennas corresponding to the same signal transmitted from an active RFID tag yields hyperbola location trajectories with the foci positioned at the two reader antennas. Thus, by utilizing multiple reader antennas to form multiple antenna pairs, the tag position can be determined as the intersection of the respective hyperbolas. For a conventional narrowband RFID system, immediate application of time-based techniques (e.g. TOA and TDOA) for the localization of RFID tags is often difficult because of the poor time resolution limited by the frequency bandwidth. In addition, timebased techniques may experience additional challenges in the presence of multipath [22, 23, 44, 47, 49, 62]. Nevertheless, time-based range estimation techniques could be promising when sufficient signal bandwidth is available, for example, when the UWB techniques are used [3, 48]. UWB is defined by FCC and ITU-R in terms of a transmission from an antenna for which the emitted signal bandwidth exceeds 500 MHz or 20% of the center frequency. FCC approved license-free use of low-power UWB radio transmission with an enormous bandwidth of 7.5 GHz at the frequency band 3.1–10.6 GHz [21]. Such a wide bandwidth provides an excellent means for wireless positioning due to its high time-domain resolution. Specifically, for a single-path additive white Gaussian noise (AWGN) channel, it was shown that the RMSE of the range estimate dˆ derived from the TOA estimation is lower bounded by [24]: ˆ ≥ √ √c RMSE(d) 2 2π SNRBW

(15.7)

where SNR is the signal-to-noise power ratio and BW is the effective signal bandwidth. Therefore, UWB RFID systems can achieve a high range resolution by utilizing a wide bandwidth, although the SNR is usually low. Note that, unlike RSS-based techniques, the RMSE of the range estimate obtained from time-based approaches is independent of d. UWB signaling can be carrier-based or impulse-based. Both types of UWB RFID systems can provide good immunity against signal distortion and multipath effects. Carrierbased UWB RFID systems can achieve frequency diversity to alleviate the impact of multipath fading [24]. On the other hand, due to high time-domain resolution, impulsebased UWB RFID systems can resolve multipath components to eliminate the effect of reflection and scattering paths.

15.3 DOA Estimation Techniques Some RFID tag positioning techniques are based on the DOA information of the RF signal, observed at multiple reader antenna positions. In these techniques, the positioning performance of RFID tags is affected by the accuracy of DOA estimation. DOA estimation is typically achieved using directional antennas, phased arrays and smart antennas. Utilization of directive beams also helps to enhance the read range and to reduce interference as well as multipath effects.

398

RFID Systems

15.3.1 Directional Antenna A directional antenna can transmit energy to or receive energy from a small angular sector so as to improve the radiation (or reception) efficiency or to mitigate interference. In an RFID system, when a tag enters the area covered by a directional reader antenna, the reader can sense it and thus determine its rough DOA. The accuracy of the DOA depends on the antenna beamwidth. An antenna with a narrower beamwidth yields a higher DOA accuracy.

15.3.2 Phased Array A phased array is a group of antennas in which the relative phases of the respective signals feeding or weighted to the antennas are varied in such a way that an effective radiation/reception pattern of the array is reinforced in the desired direction, whereas low array sensitivity is exhibited in other directions. Some phased arrays steer, without physical movement, the beams to fixed directions (known as the switched beam technique), whereas some can electronically steer the beams to any directions. For example, the bidirectional electronically steerable phased array (BESPA), developed by RF Controls, steers beams with a phase shift network and firmware control algorithm [59].

15.3.3 Smart Antenna Smart antennas, also known as adaptive array antennas or adaptive arrays, are antenna arrays with sophisticated signal processing capability. They can be designed to adaptively steer beams and nulls toward arbitrary directions and to provide high-resolution DOA estimations. The number of antennas and the size of the array aperture are key parameters that determine the capability of an adaptive array. An RF signal generates time delays when it propagates in the space. For a narrowband signal, such time delay can be equivalently considered as a phase delay across multiple antennas. Thus, the measurement of the phase difference between signals received at different array antennas can be used for DOA estimation. Commonly used DOA estimation techniques include maximum likelihood (ML), multiple signal classification (MUSIC), estimation of signal parameters via rotational invariance technique (ESPRIT), minimum variance distortionless response (MVDR), matrix pencil method or one of their derivatives [32, 76]. For a narrowband tag signal arrived from DOA θ observed at an RFID reader equipped with an N -element uniform-linear array (ULA) with inter-element spacing l, the RMSE of the estimated spatial frequency, defined as ω = λl sin(θ ), is lower bounded by the CRLB [67], 6 RMSE(ω) ˆ ≥ (15.8) 3 KN SNR where K is the number of available data snapshots. It shows that the accuracy of DOA estimate can be improved by increasing SNR, the number of snapshots, and the number of array antennas.

Principles and Techniques of RFID Positioning

399

Note that most of the literature considers far-field DOA estimation problems where the unknown object is located at the far field of the antenna array. When the tag is closely placed around a reader, RFID tag positioning and tracking may involve rather complicated near-field DOA estimation problems, where the phase difference between different array antennas is a function of both DOA and the range [77]. For a two-antenna array, however, it is shown that the phase difference is approximately a function of only the DOA, thus simplifying the DOA estimation problem. Another important issue to be taken into account in near-field DOA estimation is the effect of the radiation field pattern, particularly the phase, of the antennas. Because the tag may be viewed by different array antennas from different angles, the observed phase difference should first be compensated by the phase difference in the antenna patterns before it is used for DOA estimation [77]. In the presence of multiple tags, simultaneous estimation of their DOAs, in principle, requires more antennas than the number of tags. However, in practice, multiple tags are often discriminated in the process of collision avoidance and, as a result, only a single tag should be considered at each time. With the use of multiple antennas, nevertheless, it is possible to design a reader to simultaneously resolve multiple tags without collision, yielding faster tag reading and DOA estimations.

15.4 RFID Positioning Techniques Using the estimated range and/or DOA information, a tag or reader can be localized using various techniques. In this section, we provide an overview of different RFID positioning techniques.

15.4.1 Trilateration/Multilateration 15.4.1.1 Principles and Algorithms The trilateration/multilateration method determines the position of a tag or reader using the range information estimated at several spatially separated reference points (reader antennas or reference tags). As previously mentioned, the range can be estimated using RSS-, phase-, or time-based techniques. Specifically, to unambiguously localize a tag in an n-dimensional space, range information from at least n+1 reference points is required. However, in some cases, n reference points may suffice if the range ambiguity can be resolved by other means. For example, bilateration that uses only two reference points yields two intersections in a two-dimensional (2-D) plane [74], and the tag position may be uniquely determined if the other intersection is out of the interested area. Consider a tag positioning problem in a 2-D space as an example. Figure 15.4 shows how the tag position can be estimated using the trilateration method, where the range of the unknown tag to reference points (reader antennas) p1 (x1 , y1 ), p2 (x2 , y2 ), and p3 (x3 , y3 ) are estimated as d1 , d2 and d3, respectively. The location of the unknown tag, denoted as (x, y), can be determined by solving the following three equations (x − xi )2 + (y − yi )2 = di2 ,

i = 1, 2, 3

(15.9)

400

RFID Systems

p1(x1,y1) p2(x2,y2)

d1 d2 p d3 p3(x3,y3)

Figure 15.4

Tag positioning using trilateration. All circles are intersected at the tag position.

As a result, the coordinate of the unknown tag is obtained as  1 (d12 − y12 − x12 )(y2 − y3 ) + (d22 − y22 − x22 )(y3 − y1 ) + (d32 − y32 − x32 )(y1 − y2 )    x = −   2 x1 (y2 − y3 ) + x2 (y3 − y1 ) + x3 (y1 − y2 )   1 (d12 − y12 − x12 )(x2 − x3 ) + (d22 − y22 − x22 )(x3 − x1 ) + (d32 − y32 − x32 )(x1 − x2 )   y = − 2 y1 (x2 − x3 ) + y2 (x3 − x1 ) + y3 (x1 − x2 ) (15.10) In reality, the range estimates would have measurement errors, thus yielding an erroneous estimation of the tag position. One way to improve the positioning accuracy in this case is to use the multilateration method as described below, where more than three reference points are utilized. For M > 3 reference points, M equations can be established similar to Equation (15.9). In this case, it becomes an over-determined problem. For convenience, we express them in a vector form as [71]   x A =b (15.11) y   x1 − x2 y1 − y2   .. .. where A =   is an (M − 1) × 2 matrix, and . . xM−1 − xM yM−1 − yM

 b=

1  2

d22 − d12 + x12 − x22 + y12 − y22 .. . 2 2 2 2 2 2 − dM−1 + xM−1 − xM + yM−1 − yM dM

   is an (M − 1) × 1 vector. The

Principles and Techniques of RFID Positioning

401

least-square solution of Equation (15.11) is given by 

x y



= (AT A)−1 AT b

(15.12)

where (·)T denotes the transpose of a matrix or a vector. 15.4.1.2 Applications SpotON [33, 34] uses RSS to localize long-range active RFID tags in a threedimensional (3-D) space. Multiple receivers collect RSS measurements and use the trilateration/multilateration method to estimate the tag locations. The Local Position Measurement (LPM) method [69] localizes outdoor active tags based on TDOA measurements using at least four synchronized reader antennas with known positions. Positioning accuracy can be improved by increasing the number of reader antennas through the minimization of the weighted mean square error. A Patient Management and Tracking System (PMTS) using the RSS-based trilateration technique is developed in [41]. It is reported that the average accuracy is less than 1 m in an open-space test. UWB-based RFID systems have been developed for RTLS [48, 53]. The Sapphire DART system developed by Multispectral Solutions achieves a 200 m read range with an accuracy of about 0.3 m even in a multipath environment [48]. Recently, development of passive and semi-passive UWB-based RFID techniques was also reported. For example, Martec developed a UWB-based carrierless RFID system, named Passpulse [50], for both passive and semi-passive operations. Passive SAW (Surface Acoustic Wave) ID-tags use TOA measurements to localize a tag [15]. SAW is an electromechanical device constructed of a piezoelectric crystal or ceramic to convert an RF signal to mechanical wave, which has a much smaller wavelength than that of the RF signal and thus is convenient to implement or measure the delay with a miniaturized size. Localization of a SAW tag is achieved by analyzing the round-trip TOF observed at three separate reader antennas through trilateration. Each tag has a fixed code described as its unique impulse response. Thus, a reader interrogates a tag by transmitting the time inverse of the tag-specific impulse response, and the tag then retransmits the correlated signal with a high peak to be easily detected at the reader. A 20-cm position accuracy was reported in [15] for a signal of carrier frequency 2.5 GHz and signal bandwidth of 40 MHz. RSS-based trilateration/multilateration technology can also be used to locate the position of an RFID reader. Using multiple reference tags which are placed at known positions, a mobile reader can localize itself based on the RSS measurements corresponding to two or more tags [18].

15.4.2 Triangulation 15.4.2.1 Principles and Algorithms Triangulation is a process to determine the location of a radio transmitter by measuring the DOA of the received signal from two or more known reference points. As we

402

RFID Systems

p(x,y)

D

a

b p2(x2,y2)

p1(x1,y1)

Figure 15.5 two lines.

Tag positioning using triangulation. The tag location is obtained by intersecting the

discussed in Section 15.3, DOA information can be obtained using directional antenna, phased array, or smart antennas. Figure 15.5 illustrates the basic principle of triangulation. In this example, the DOAs of the tag signal measured at two reference points p1 (x1 , y1 ) and p2 (x2 , y2 ) are respectively α and β with respect to the line determined by the two reference points (observation antennas). The tag can be localized by intersecting the two rays. The coordinate of the tag is thus given by  x = x1 + D cos (α + γ ) (15.13) y = y1 + D sin (α + γ ) sin(β) (x1 sin(α+β)   −1 y1 −y2 tan x1 −x2 .

where D =

− x2 )2 + (y1 − y2 )2 is the distance from the unknown tag to p1

and γ = When there are M > 2 reference points, pi (xi , yi ), i = 1, . . . , M, and the ith reference point has the measured DOA αi with respect to the x-axis, as shown in Figure 15.6, the unknown tag location can be obtained by intersecting the multiple lines, each passing through an observation position with the slope determined by the respective DOA. Thus, the tag location is estimated by solving the following equations ki x − y = ki xi − yi , i = 1, . . . , M

(15.14)

where ki = tan(αi ) refers to the slope of the line passing through pi (xi , yi ). Stacking Equation (15.14) for the M observations yields the following over-determined problem   x C =d (15.15) y     k1 x1 − y1 k1 , −1     .. .. where C =   is an M × 2 matrix, and d =   is an M × 1 . .

kM , −1 kM xM − yM vector. Note that, for the case of |ki | = ∞, that is, αi = 0.5π or 1.5π, Equation (15.14) becomes x = xi , and the corresponding row of C and d should be [1, 0] and [xi ], respectively.

Principles and Techniques of RFID Positioning

403

p(x,y)

aM a1 y

p1(x1,y1) x

a2

pM (xM,yM)

p2(x2,y2)

Figure 15.6 Tag positioning using triangulation with more than two reader antennas. The tag location is obtained by intersecting the multiple lines.

Similar to Equation (15.12), the least-square solution of Equation (15.15) is given by   x = (CT C)−1 CT d (15.16) y The localization accuracy can be improved by incorporating more reference points. 15.4.2.2 Applications The FAST Tag Over-the-Conveyer RFID Tunnel System developed by Accu-Sort uses narrow-beam antennas to locate tags between closely spaced cartons [6]. In [85], DOA information obtained at two separate arrays is used to locate and track a moving tag on a conveyor belt. With the use of two sets of obliquely oriented two-element arrays, tags on the belt can be accurately localized through a triangulation operation. At the expense of higher signal processing complexity, it can be formulated as a near-field DOA estimation problem using a four-element array to yield more accurate positioning estimation [77]. In [4], the triangulation principle is used to locate indoor tags. Each reader uses two directional antennas to identify which side the tag is located. Although the angle estimation based on one reader does not have a high resolution, the use of multiple readers is expected to improve the estimation. In another example, three directional antennas are used to determine the rough area in which the tag is located [37].

15.4.3 Hybrid Direction/Range Methods 15.4.3.1 Principles and Algorithms When both the DOA and range information of a tag are available at a reference point (reader antennas), the location of the tag can be uniquely determined. The principle is illustrated in Figure 15.7, where the DOA of the backscattering signal from the tag is α with respect to the horizontal axis of the reference coordinate and the estimated range is d. The tag is localized by intersecting the incident ray and range curve and thus one can

404

RFID Systems

p(x,y)

d a p1(x1,y1)

Figure 15.7 Tag positioning based on direction and range information. The tag is localized by intersecting the incident ray and the range curve.

get its coordinate as



x = x1 + d cos α, y = y1 + d sin α.

(15.17)

When the range and direction information at multiple reference points pi (xi , yi ), i = 1, . . . , M are available, the tag location can be estimated by solving the following equations  x = xi + di cos αi (15.18) y = yi + di sin αi , i = 1, . . . , M where di is the range from the tag to the reference point pi , and αi is the DOA of the tag measured at reference point pi . Similar to Equation (15.14), stacking the above equation for the M reference points renders the following over-determined problem   x E =f (15.19) y     x1 + d1 cos α1 1 0  y1 + d1 sin α1  0 1      x2 + d2 cos α2  1 0     y2 + d2 sin α2  0 1 is a 2M × 2 matrix, and f =  where E =    is a 2M × 1 vector. . .   . . . .   . . .   1 0 xM + dM cos αM  0 1 yM + dM sin αM The least-square solution of Equation (15.19) is given by   x = (ET E)−1 ET f (15.20) y which results in

   x=

1 M

  y=

1 M

M  i=1 M 

(xi + di cos αi ) (15.21) (yi + di sin αi )

i=1

The localization accuracy can be improved by incorporating more reference points.

Principles and Techniques of RFID Positioning

405

15.4.3.2 Applications Trolley Scan developed an RFID Radar system for identifying and locating passive RFID tags [75]. For 2-D applications, a reader is equipped with three high-gain patch antennas. One of them is used to transmit energizing signal, whereas the other two receive backscattered tag signals to enable range and DOA estimations. For 3-D applications, a third receive antenna is required. In a static situation where tags are relatively stationary, the radar achieves high range and DOA estimation accuracy by taking a long integration time. It is reported that the accuracy of range estimation is less than 0.5 m and DOA accuracy is better than 1 degree, at the maximum coverage range of 100 m. A scheme for real-time 2-D localization of a SAW tag using a single reader antenna was demonstrated in [1]. In this scheme, the range is estimated based on the TOF measurement, and the DOA is implicitly obtained using the angular rotation of the reader’s antenna and performing an operation of complex pattern matching (maximum correlation) between the received tag signal response and the ideal signal response (pattern) saved in the reader. This scheme requires that the tag is normal to that of the LOS between the tag and the reader antenna, and that the tag’s response pattern is known.

15.4.4 Radio Map Matching Methods 15.4.4.1 Principles Radio map matching methods are also known as “scene analysis” approaches. They are composed of two distinctive steps. In the first step, the radio scene information or RF fingerprints in the environment are collected to form a radio map, described as M = {(m1 , p1 ), (m2 , p2 ), · · · , (mNR , pNR )}

(15.22)

where mi denotes the measurement vector corresponding to the ith known position pi = [xi , yi , zi ]T , and NR is the total number of elements in the radio map. The elements of vector mi correspond to the RF fingerprint measurements at multiple reader antennas. The RSS and tag count can be used as the fingerprints [79]. In the second step, unknown tags are localized by matching the measured data corresponding to the unknown tags with an appropriate subset of fingerprints recorded in the radio map. Two major fingerprintingbased matching methods are the k-nearest-neighbor (kNN) and the probabilistic methods. The kNN method [51] uses the fingerprint (say, the RSS measurement) of an unknown tag, recorded as m, to find its k closest matches (mmj , pmj ), j = 1, · · · , k, in the radio map according to mj =

arg min

i∈{1,···Nr },i ∈{m / 1 ,·· mj −1 }

||m − mi ||

(15.23)

ˆ is then obtained as the weighted sum of The estimated location of the unknown tag, p, the positions corresponding to the k nearest neighbors, that is pˆ =

k  j =1

wj pmj

(15.24)

406

RFID Systems

 where wj is the weighting factor for the j th nearest neighbor with kj =1 wj = 1. There are several ways to determine the weighting factors. Specifically, when the uniform weighting scheme is used, that is, all wj ’s take the same value of 1/k, it becomes the arithmetic average of the k nearest neighbors as 1 pmj k k

pˆ =

(15.25)

j =1

The probabilistic methods [14, 26, 37, 52, 60, 70, 73], on the other hand, are to find the location of a tag from multiple possible locations to yield the highest posterior probability. In this case, vector mi describes the joint PDF of the measurement or measurement error, observed at multiple reader antennas, corresponding to the ith reference tag at pi = [xi , yi , zi ]T . In this way, the probabilistic radio map models the distribution of the measurement in different geographical positions. By exploiting the Bayesian rule, location estimation errors can be mitigated for improved tag positioning. Let o = [xo , yo , zo ]T denote the observation of the position p = [x, y, z]T . For any given reference position p, the probability distribution of the observation variable p(o|p), namely, the likelihood function, can be obtained through measurements at multiple readers. By using the Bayesian rule, the posterior probability of the position p is p(p|o) =

p(o|p)p(p) p(o)

(15.26)

where p(p) is the prior probability of being at position p before knowing the value of the observation variable. When a uniform prior distribution is assumed, that is, the distribution of unknown tag does not have any preference for any particular position, the likelihood function completely determines the posterior distribution of the location. Further, the posterior distribution can be used to choose an optimal estimator of the unknown tag. For example, the position of the unknown tag can be estimated as pˆ =

NR 

p(pi |o)pi

(15.27)

i=1

Consequently, the posterior probability is maximized and the squared localization error is minimized. In practice, the RSS measurements often suffer from multipath propagation and shadowing. By taking these factors into account in the pre-stored radio map, their effects on the location estimation can be mitigated. In addition to the RSS, the spatial signatures can also be used into map matching. The primary advantage of map matching methods lies in the corporation of the environment effect, such as NLOS propagation and multipath. However, the radio map should be constructed based on dense reference tags to represent the current environment and should be periodically updated to reflect the environmental dynamics [82]. 15.4.4.2 Applications Radio map matching methods have been widely used for RFID positioning. LANDMARC [51] is a well-known approach that uses several readers and a number of reference

Principles and Techniques of RFID Positioning

407

tags to locate indoor active RFID tags. The RSS of the reference tags are first recorded at each reader. When an unknown tag is present, its RSS is measured and the Euclidean distance relative to the RSS vector of each reference tag is calculated. The unknown tag is localized using the kNN method that takes a weighted sum of the coordinates of the k reference tags with the smallest Euclidean distance. Empirically, the reference tag with the shortest Euclidean distance takes the highest weight. The advantages of using reference tags are multi-fold. First, by using low-cost reference RFID tags, it maintains a low number of expensive RFID readers. Second, the effect of environment dynamics can be mitigated because the unknown tag and the reference tags are subject to similar propagation characteristics. The radio map of the reference tags can be dynamically updated to maintain the accuracy. Accurate positioning requires proper and dense distribution of reference tags and that the RSS of the reference tags is adequately updated. Several variants of LANDMARC have been developed to improve the positioning accuracy and/or to reduce the system complexity. For example, the positioning error can be reduced through the removal of dissimilar reference tags [86, 88]. For the localization of stationary tags, the use of mobile readers for the measurement of the RSS of all tags from different locations is proposed [14]. Further, to reduce the effect of RSS fluctuation and measurement noise on the position estimation, Kalman filtering and probabilistic map matching can be utilized [14]. In the Flexible Localization EXplOits Rfid (FLEXOR) scheme, the area of interest is divided into a number of hexagonal cells [72]. Reference tags are placed at the center as well as on the vertices of each cell. Two localization modes are provided: the region mode finds a cell tag nearest to the unknown tag, whereas the coordinate mode determines the coordinate of the unknown tag through the weighted average of the coordinates of three reference tags, one at the center and two on the vertices of the same cell. It is reported that this scheme reduces computation overhead and provides similar accuracy as LANDMARC. A smart Book-LOCating System (BLOCS) was developed to locate tagged books on bookshelves [65]. The single book mode localizes a tagged book by minimizing the RSS-based Euclidean distance between bookshelf tags and the tagged book, whereas the book list mode routinely provides a list of the bookshelves and the misplaced books to help a librarian to localize all misplaced books. When reference tags can be simultaneously detected by a set of readers, the difference between their actual and RSS-based estimated locations can be used as a correction factor to mitigate the position estimation error of the unknown tag [36]. The use of multi-power level transmission in a LANDMARC-based tag positioning system was proposed in [80]. To locate an unknown tag, the readers start with the lowest power level and gradually increase the transmit power until they receive the response from the unknown tag. Only the reference tags that are activated by the same power level as the unknown tag but not activated by a lower level are selected to estimate the range of the unknown tag. Such range estimations obtained from multiple readers are then used to trilaterate the position of the unknown tag. As we discussed above, the use of a large number of reference tags can improve the localization accuracy of a LANDMARC system. The VIrtual Reference Elimination (VIRE) scheme develops the concepts of virtual reference tags, instead of placing more physical reference tags, to improve the positioning accuracy without increasing the number of actual reference tags [88]. In this scheme, the entire sensing area is divided into a

408

RFID Systems

number of small regions. Each region is centered by a physical reference tag and contains many virtual reference tags whose RSS values corresponding to each reader are determined through interpolation operations.

15.4.5 Proximity 15.4.5.1 Concept An RFID reader has a limited read range and thus can only reach those tags that are located within a limited coverage area around the reader antennas. Therefore, observing whether a tag is within the reach of a reader antenna yields the proximity (or connectivity) information of the tag [63]. While high accuracy of tag positions in this case requires dense deployment of reader antennas with a small coverage area, this approach is easy to implement. The simplest implementation may be the distributed antenna scheme, where a variety of antennas, regardless of being omnidirectional or directional, are distributed in an area of interest. Each antenna senses tags in its respective coverage area. When a tag is sensed by a reader antenna, the tag location is assumed to be the same as this antenna. When a tag is detected by more than one antenna, it is considered that the tag is close to the antenna with the strongest RSS. Alternatively, the tag position can be localized using a weighted average of coordinates of those antennas. As thus, the positioning accuracy is on the order of the size of the antenna coverage area or smaller. In this case, the position of the unknown tag is estimated as pˆ =

NR 

wi pi

(15.28)

i=1

where pi = [xi , yi , zi ]T denotes the coordinate of the ith reader NR antenna, NR is the number wi = 1. The i th weighting of reader antennas. The weighting factors are subject to i=1 factor is set to wi = 0 if the ith antenna cannot sense the unknown tag. 15.4.5.2 Applications Mojix’s STAR system consists of a single STAR receiver and multiple transmitters (known as eNodes) via wired connection [54]. A STAR receiver can manage up to 512 eNodes, which are daisy-chained with STAR receiver to cover a large geographical area or mounted in a facility in an orientation to enable the STAR receiver determining a tag’s location in three dimensions. The eNodes transmit RF signals that interrogate and power up the tags, which then backscatter the signal to a receiver as far as 600 feet (200 m) away. The STAR system can localize a tag by determining which eNodes can activate the tag. Robot search and rescue suffers from hostile conditions encountered after a disaster. A robot equipped with an inertial measurement unit can record its own trajectory. By further utilizing RFID readers in the robots and deploying RFID tags in the area, the robots can use the deployed tags as common references so as to minimize the trajectory error and better coordinate their exploration [42]. An embedded navigation system composed of

Principles and Techniques of RFID Positioning

409

GPS and active RFID is proposed to localize pedestrians [43]. With the use of Kalman filtering, the GPS is used outdoors to adjust errors in position and direction, whereas the RFID is used for indoor localization. RFID readers are placed on fixed positions so that the pedestrian with an active tag can be localized by detecting the ID signal from the tag. A 3-D localization scheme is proposed for locating passive or active RFID tags based only on the connectivity information associated with multiple readers [11, 12]. The estimated position of the unknown tag is obtained by simply averaging the corresponding positions of the virtual landmarks contained in the bounded space. Clearly, the accuracy depends closely on the density of the readers and on the size of virtual landmarks. A localization system using geographical location for wireless sensor and RFID networks is demonstrated in [40]. In this scheme, an RFID reader first obtains its own location information from the messages broadcast by the nearby reference nodes. The reader then reads nearby tags and reports data upward until it arrives at the location server. Thus, the location system can monitor and track RFID tags based on the general purpose geographical location identifier for each region. An RFID-based human-probe positioning system for urban sensing was proposed in [66]. In this system, RFID tags are deployed as landmarks in an urban area. When a person who carries an RFID reader moves into the area, the reader can be self-localized based on the proximity principle.

15.5 Improving Positioning Accuracy As we discussed in Section 15.1, an RFID positioning system, in general, consists of the location sensing and the positioning processing. The former obtains necessary information, such as range and DOA, using various resources transmitted and received at single or multiple antennas. Such information is then fed to the positioning processing block for data fusion to yield the location information of the RFID tag or reader of interest. Therefore, positioning accuracy of an RFID system can be improved from both location sensing and positioning processing perspectives. The selection of appropriate sensing techniques, under the resource limitations and system constraints, is critical in achieving satisfactory range and DOA estimations. For example, RSS-based range estimation techniques are simple but its sensitivity reduces as the range increases, whereas time-based range estimation techniques require a large signal bandwidth, and the accuracy is not sensitive to the range. A high SNR is always beneficial to improve the positioning performance. In addition, improved location sensing can be achieved by collecting more and diversified information in terms of time, frequency, space, and polarization. It is desirable to have over-determined measurement data sets that are more than the minimum requirement: • Time diversity: A simple way to combat noise is to accumulate over a longer time, provided that the reader and tag are stationary [33, 75]. • Frequency diversity: The use of a wider frequency bandwidth provides frequency diversity for enhanced robustness against noise as well as multipath fading. In particular, the use of UWB signals enables the achievement of significant frequency diversity or,

410

RFID Systems

equivalently, time resolution in discriminating reflection paths from the direct path [48, 50, 53]. Using frequency diversity at the same antenna, however, is not effective in combating path obstructions. • Spatial diversity: Multiple spatially separated readers or reader antennas can be exploited to improve the RFID positioning in the presence of path obstruction because the probability that the LOS to all the readers is obstructed is very low. Moderately separated antennas (one wavelength or larger inter-element spacing) are effective in combating multipath fading. The use of more collocated antennas in a smart antenna enables higher degree of processing capability to provide more accurate DOA estimation. The antennas at different locations can be physically placed, or can be synthesized by moving antennas in different positions [29]. • Polarization diversity: The use of antennas with different polarizations is another effective way to combat multipath fading. Compared to spatial diversity, polarization diversity does not have an inter-element spacing requirement because different polarizations experience different propagation characteristics even when the antennas are closed spaced. Better positioning processing can be achieved by fusing the collected data in an optimal or suboptimal way, for example: • Utilizing as much information as collected at the location sensing block: Use all observation data to globally optimize a criterion, such as the mean square of positioning error or the likelihood function, and consider positioning methods that exploit the probabilistic characteristics of the measurement data and the historical status [26, 70, 73]. For example, Bayesian algorithm can be used to maximize the posterior probability of the positioning estimation. A Kalman filter is a convenient approach to exploit the dynamics of an RFID tag or reader, particularly when they are in motion [14]. • Exploiting optimization algorithms to solve over-determined localization problems with noisy observations generally yields performance improvement: Using convex optimization techniques for target and node localization is an important area in radar and wireless network communities [7, 8]. Weighted averaging operations incorporating the reliability of the measurement data or processed results usually improve the positioning accuracy. The collection of more location sensing data as well as advanced positioning processing requires more investment in terms of hardware and/or signal processing capability. The decision is to best trade-off between the affordable system complexity and the required system performance. The presence of various obstacles and reflectors in the environment alters the propagation decay and delay profile and thereby imposes significant challenges to RFID positioning. By using diversified information in terms of time, frequency, space, and polarization, as summarized above, the RFID positioning performance in a multipath environment can be improved. When it is available and feasible, the use of signals with excessive signal bandwidth, such as UWB signals, allows separation and discrimination of multipath signals and thereby enables elimination of the multipath effects. The use of directional antennas can also reduce the significance of multipath effects for performance improvement.

Principles and Techniques of RFID Positioning

411

15.6 Conclusion In this chapter, we have reviewed the principles, algorithms, and techniques of RFID positioning. In most RFID positioning problems, the location information of RFID tags is interested, but some applications involve the localization of RFID readers. An RFID positioning system, in general, consists of the location sensing component and the positioning processing component. The location sensing component provides necessary information, such as range and DOA, which is then fed to the positioning component for data fusion to yield the location information of the RFID tag or reader of interest. Different range and DOA estimation techniques have been introduced. RFID positioning principles and techniques based on trilateration, triangulation, hybrid direction/range, radio map matching, and proximity methods were presented. Potential approaches for improving positioning accuracy were also discussed.

Problems 1. What are the two major functional steps in a typical RFID positioning system? Please briefly describe the function of each step. 2. What are the three important tag range estimation techniques? Briefly describe how to improve the range estimation accuracy for each technique in the presence of multipath propagation. 3. What is the range ambiguity problem in the phase-based range estimation techniques? 4. Why can carrier-based and impulse-based UWB signaling improve positioning accuracy? 5. From the CRLB of the DOA estimation for a narrowband tag signal, what approaches can be used to improve the DOA estimation accuracy? 6. To unambiguously localize a tag in an n-dimension space, what is the required minimum number of reference points from which range information is provided? 7. Which positioning techniques can be used for tag positioning? 8. Consider a two-dimensional positioning scenario in which four reference points are located at p1 (12.2, 3.8), p2 (1.8, 4.9), p3 (7.5, 2.8), and p4 (6.3, 9.2), respectively, and the corresponding tag range estimates are 6.2, 4.5, 2.9, and 4.0. All the units are in meters. Find the position of the tag by using the multilateration method. 9. Consider a positioning scenario in which four reference points are also located at p1 (12.2, 3.8), p2 (1.8, 4.9), p3 (7.5, 2.8), and p4 (6.3, 9.2), respectively, and the corresponding DOAs, relative to the horizontal axis, are 151, 29, 111, and 254 degrees. The units of the reference positions are in meters. Find the position of the tag by using the triangulation method. 10. What is the principle of radio map matching methods for tag positioning? Describe the advantages and disadvantages of this kind of method. 11. What are the two primary approaches to improve positioning accuracy?

412

RFID Systems

References [1] Arumugam, D., Ambravaneswaran, V., Modi, A., and Engels, D. (2007) 2D localization using SAW-based RFID systems: a single antenna approach, Int. Journal of Radio Frequency Identification Technology and Applications, 1(4). [2] Ahamd, F,. Amin, M., and Setlur, P. (2006) Through-the-wall target localization using dual-frequency CW radars, Proc. SPIE , 6201, April. [3] Amir, I. and Naim, A. Frequency hopping range estimation with low power consumption, US Patent, 7061428 B1. [4] Amir, I. Dual antenna base station for improved RFID localization, US Patent, 20090027209 A1. [5] Assad, M.A. (2007) A real-time laboratory testbed for evaluating localization performance of WiFi RFID technologies [thesis], Worcester Polytechnic Institute. [6] Accu-Sort Systems, Accu-Sort FAST Tag Over-the-Conveyor (OTC) RFID Tunnel. Available at: http:// www.marktecprods.com/PDF/fast tagotc.pdf. [7] Cheung, K.W., Ma, W.-K., and So, H.C. (2004) Accurate approximation algorithm for TOA-based maximum likelihood mobile location using semidefinite programming, in Proc. IEEE Int. Conf. Acoustics, Speech, and Signal Processing (ICASSP), May, pp. 145– 148. [8] Biswas, P., Liang, T., Toh, K., Wang, T., and Ye, Y. (2006) Semidefinite programming approaches for sensor network localization with noisy distance measurements, IEEE Trans. Automation Science and Engineering, 3(4): 360– 371. [9] Bhatia, A., Mehta, B., and Gupta, R. (2009) Different localization techniques for real time location sensing using passive RFID. Available at: http://filedb.experts-exchange.com/incoming/2009/02 w09/110353/ RFID-DTOA.pdf. [10] Bahl, P. and Padmanabhan, V.N. (2000) RADAR: an in-building RF-based user location and tracking system, in Proc. IEEE InfoCom, Mar, 2, pp. 775– 784. [11] Bouet, M. and Pujolle, G. (2008) 3-D localization schemes of RFID tags with static and mobile readers, in Proc. Int. Federation for Information Processing (IFIP). [12] Bouet, M. and Pujolle, G. (2008) A range-free 3-D localization method for RFID tags based on virtual landmarks, in Proc. Int. Symp. Personal, Indoor and Mobile Radio Communications (PIMRC), Sept. [13] Bouet, M. and Santos, A.L. (2008) RFID tags: Positioning principles and localization techniques, in Proc. IFIP Wireless Days 2008 , Nov. [14] Bekkali, A., Sanson, H., and Matsumoto, M. (2007) RFID indoor positioning based on probabilistic RFID map and Kalman filtering, in Proc. IEEE Conf. Wireless and Mobile Computing, Networking and Communications (WiMob), Oct., pp. 21–21. [15] Bechteler, T., and Yenigun, H. (2003) 2-D localization and identification based on SAW ID-tags at 2.5 GHz, IEEE Trans. Microwave Theory and Techniques, 51(5): 1584– 1590. [16] Chon, H., Jun, S., Jung, H., and An, S. (2004) Using RFID for accurate positioning, Journal of Global Positioning Systems, 3(1–2): 32–39. [17] Contractor, B.V. (2008) Two-dimensional localization of passive UHF RFID tags [thesis], Wright State University. [18] Duron, M. and Bridgelall, R. Reverse infrastructure location system and method, US Patent, 20080157972 A1. [19] Dobkin, D.M. (2008) The RF in RFID. Burlington, MA: Newnes. [20] Ekahau, The case for real time location systems – from technology to business reality, White Paper, 2008. Available at: http://www.integratedsolutionsmag.com/index.php?option=com docman&task=doc view&gid=98. [21] Federal Communications Commission Report and Order, FCC 02-48, Feb. 2002. Available at: http://hraunfoss.fcc.gov/edocs public/attachmatch/FCC-02-48A1.pdf [22] Guvenc, I., Sahinoglu, Z., and Orlik, P.V. (2006) TOA estimation for IR-UWB systems with different transceiver types, IEEE Trans. Microwave Theory and Techniques, 54(4): 1876– 1886. [23] Ge, F.X., Shen, D. X., Peng, Y.N., and Li, V. O. K. (2007) Super-resolution time delay estimation in multipath environments, IEEE Trans. Circuits and Systems- I:Regular Papers, 54(9): 1977– 1986. [24] Gezici, S., Tian, Z., Giannakis, G., Kobayashi, A., Molisch, H., Poor, H., and Sahinoglu, Z. (2005) Localization via ultra-wideband radios: A look at positioning aspects for future sensor networks, IEEE Signal Processing Magazine, 22(4): 70–84.

Principles and Techniques of RFID Positioning

413

[25] Hatami, A., Alavi, B., Pahlavan, K., and Kanaan, M. (2006) A comparative performance evaluation of indoor geolocation technologies, Interdisciplinary Information Sciences, 12: 133– 146. [26] Hahnel, D., Burgard, W., Fox, D., Fishkin, K., and Philipose, M. (2004) Mapping and localization with RFID technology, in Proc. IEEE Int. Conf. Robotics and Automation (ICRA), 1: 1015– 1020. [27] Harter, A., Hopper, A., Steggles, P., Ward, A., and Webster, P. (1999) The anatomy of a contextaware application, in Proc. ACM/IEEE Int. Conf. Mobile Computing and Networking (MobiCom), Aug., pp. 59–68. [28] Huang, X., Janaswamy, R., and Ganz, A. (2006) Scout: outdoor localization using active RFID technology, in Proc. Int. Conf. Broadband Communications, Networks and Systems (BROADNETS), Oct., pp. 1–10. [29] Hinske, S. and Langheinrich, M. (2008) Using a movable RFID antenna to automatically determine the position and orientation of objects on a tabletop, in Proc. European Conf. Smart Sensing and Context, 5279, pp. 14–26. [30] Hatami, A. and Pahlavan, K. (2005) A comparative performance evaluation of RSS-Based positioning algorithms used in WLAN networks, in Proc. IEEE WCNC . [31] Heidari, M. and Pahlavan, K. (2007) Performance evaluation of WiFi RFID localization technologies, in RFID Technology and Applications. Cambridge: Cambridge University Press. [32] Hudson, J.E. (1981) Adaptive Array Principles. London: Peter Peregrinus. [33] Hightower, J., Vakili, C., Borriello, G., and Want, R. (2001) Design and calibration of the SpotON Ad-Hoc location sensing system, University of Washington CSE Technical Report, Aug. Available at: http://www.cs.washington.edu/homes/jeffro/pubs/hightower2001design/hightower2001design.pdf. [34] Hightower, J., Want, R., and Borriello, G. (2000) SpotON: An indoor 3D location sensing technology based on RF signal strength, University of Washington CSE Technical Report, Feb. Available at: http://seattle.intel-research.net/people/jhightower/pubs/hightower2000indoor/hightower2000indoor.pdf. [35] Ingram, S.J., Harmer, D., and Quinlan, M. (2004) Ultra wideband indoor positioning system and their use in emergencies, in Proc. IEEE Position Location and Navigation Symp., Monterey, CA, April, pp. 706– 715. [36] Jin, G., Lu, X., and Park, M. (2006) An indoor localization mechanism using active RFID tag, in Proc. IEEE Int. Conf. Sensor Networks, Ubiquitous, and Trustworthy Computing, June, pp. 40–43. [37] Jia, S.. Sheng, J., and Takase, K. (2008) Improvement of performance of localization ID tag using multiantenna RFID system, in Proc. Annual Conf. Society of Instrument and Control Engineers (SICE). [38] Knox, M. and Bridgelall, R. (2006) Object localization based security using RFID, Int. Patent WO 2006/039119 A1. [39] Knox, M., Bridgelall, R., Duron, M., Knadle, R., and Bender, J. (2006) Angle of position object location system and method, Int. Patent WO 2006/026518 A2. [40] Kim, S., Ko, D., and An, S. (2008) Geographical location based RFID tracking system, in Proc. Int. Symp. World of Wireless, Mobile and Multimedia Networks (WoWMoM), June, pp. 1–3. [41] Kim, D., Kim, J., Kim, S., and Yoo, S. (2008) Design of RFID based the patient management and tracking system in hospital, in Proc. IEEE Annual Int. Conf. Engineering in Medicine and Biology Society, Aug., pp. 1459– 1461. [42] Kleiner, A., Prediger, J., and Nebel, B. (2006) RFID technology-based exploration and SLAM for search and rescue, in Proc. IEEE/RSJ Int. Conf. Intelligent Robots and Systems, pp. 4054– 4059. [43] Kourogi, M., Sakata, N., Okuma, T., and Kurata, T. (2006) Indoor-outdoor pedestrian navigation with an embedded GPS-RFID-self-contained sensor system, in Proc. Int. Conf. Artificial Reality and Telexistence (ICAT), pp. 1310– 1321. [44] Li, J. and Wu, R. (1998) An efficient algorithm for time delay estimation, IEEE Trans. Signal Processing, 46: 2231– 2235. [45] Li, X. Zhang, Y. and Amin, M.G. (2009) Multifrequency-based range estimation of RFID tags, in Proc. IEEE Int. Conf. RFID, April. [46] Miller, L. (2006) Indoor navigation for first responders: a feasibility study, Technical Report, National Institute of Standards and Technology, Feb. [47] Moghaddam, P.P., Amindavar, H., and Kirlin, R.L. (2003) A new time-delay estimation in multipath, IEEE Trans. Signal Processing, 51(5): 1129– 1142. [48] Multispectral Solutions, Sapphire DART Ultra Wideband Precision Asset Location System. Available at: http://www.multispectral.com/pdf/Sapphire Revolution.pdf. [49] Manickam,T.G., Vaccaro, R.J., and Tufts, D.W. (1994) A least-squares algorithm for multipath time-delay estimation, IEEE Trans. Signal Processing, 42(11): 3229– 3233. [50] Muchkaev, A. and Waverly, G. Carrierless RFID system, US patent 7,385,511 B2.

414

RFID Systems

[51] Ni, L., Liu, Y., Lau, Y., and Patil, A. (2004) LANDMARC: indoor location sensing using active RFID, ACM Wireless Networks, 10(6): 701– 710. [52] Oktem, R., Aydin, E., and Cagiltay, N. (2008) An RFID based location finding and tracking with guidance, in Proc. Int. Conf. Wireless Communications, Networking and Mobile Computing (WiCOM), Oct., pp. 1–4. [53] O’Connor, M.C. (2004) FCC certifies Ubisense’s UWB, RFID Journal , Dec. Available at: http://www. rfidjournal.com/article/articleview/1285/1/1/. [54] O’Connor, M.C. (2008) New RFID technology helps Kraft, P&G, Kimberly-Clark go the distance, RFID Journal , April. Available at: http://www rfidjournal.com/article/articleview/4041/. [55] Priyantha, N., Chakraborty, A., and Balakrishnan, H. (2000) The cricket location-support system, in Proc. ACM Int. Conf. Mobile Computing and Networking (MobiCom), Aug., pp. 32–43. [56] Poor, H.V. (1994) An Introduction to Signal Detection and Estimation. New York: Springer-Verlag. [57] Qi, Y. (2004) Wireless geolocation in a non-line-of-sight environment, thesis. Princeton University, Dec. [58] Rappaport, T.S. (2002) Wireless Communications: Principles and Practice, 2nd edn. Upper Saddle River, NJ: Prentice Hall. [59] RF Controls, Bidirectional electronically steerable phased-array antennas for passive UHF RFID systems, RFID Journal , White Paper, Jan. 2009. Available at: http://www rfidjournal.com/whitepapers/ download/260. [60] Roos, T., Myllymaki, P., Tirri, H., Misikangas, P., and Sievanen, J. (2002) A probabilistic approach to WLAN user location estimation, Int. Journal of Wireless Information Networks, 9(3): 155– 164. [61] Sypniewski, J. (2000) The DSP algorithms for locally deployable RF tracking system, in Proc. Int. Conf. Signal Processing with Applications, Oct. [62] Sarac, U., Harmanci, F.K., and Akgul, T. (2008) Experimental analysis of detection and localization of multiple emitters in multipath environments, IEEE Antennas and Propagation Magazine, 50(5): 61–70. [63] Song, J., Haas, C.T., and Caldas, C.H. (2007) A proximity-based method for locating RFID tagged objects, Advanced Engineering Informatics, 21(4): 367–376. [64] Sanpechuda, T. and Kovavisaruch, L. (2008) A review of RFID localization: Applications and techniques, in Proc. Int. Conf. Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (ECTI-CON), 2, pp. 769– 772, May. [65] Sue, K. and Lo, Y. (2007) BLOCS: A smart book-locating system based on RFID in libraries, in Proc. Int. Conf. Service Systems and Service Management, June, pp. 1–6. [66] Suzuki, R., Martins, M., Ishida, Y. Tobe, Y., Konomi, S., and Sezaki, K. (2008) An RFID-based humanprobe positioning system, in Proc. Int. Conf. Networked Sensing Systems (INSS), June, pp. 248–258. [67] Stoica, P. and Nehorai, A. (1989) MUSIC, maximum likelihood, and Cramer-Rao bound, IEEE Trans. Acoustic, Speech, and Signal Processing, 37(5): 720– 741. [68] Schneegans, S., Vorst, P., and Zell, A. (2007) Using RFID snapshots for mobile robot self-localization, in Proc. 3rd European Conf. Mobile Robots (ECMR), Freiburg, Germany, Sept. [69] Stelzer, A., Pourvoyeur, K., and Fischer, A. (2004) Concept and application of LPM-a novel 3-D local position measurement system, IEEE Trans. Microwave Theory and Techniques, 52(12): 2664– 2669. [70] Subramanian, S.P., Sommer, J., Schmitt, S., and Rosenstiel, W. (2008) RIL: Reliable RFID based indoor localization for pedestrians, in Proc. Int. Conf. Software, Telecommunications and Computer Networks (SoftCOM), Sept., pp. 218– 222. [71] Sayed, A.H., Tarughat, A., and Khajehnouri, N. (2005) Network-based wireless location, IEEE Signal Processing Magazine, 22(4): 24– 40. [72] Sue, K., Tsai, C., and Lin, M. (2006) FLEXOR: A flexible localization scheme based on RFID, in Proc. Int. Conf. Information Networking (ICOIN), Jan., pp. 306–316. [73] Schneegans, S., Vorst, P., and Zell, A. (2007) Using RFID snapshots for mobile robot self-localization, in Proc. European Conf. Mobile Robots (ECMR), Germany. [74] Thiesse, F., Fleisch, E., and Dierkes, M. (2006) LotTrack: RFID-based process control in the semiconductor industry, IEEE Pervasive Computing, 5(1): 47–53. [75] Trolley Scan, RFID-radar. Available at: http://rfid-radar.com/introduc.html. [76] Van Trees, H.L. (2002) Optimum Array Processing, Part VI of Detection, Estimation, and Modulation Theory. New York: John Wiley & Sons. Ltd. [77] Wang, J., Amin, M., and Zhang, Y. (2006) Signal and array processing techniques for RFID readers, in Proc. SPIE Symp. Defense and Security, Orlando, FL, April. [78] Want, R., Hopper, A., Falcao, V., and Gibbons, J. (1992) The active badge location system, ACM Trans. Information Systems, Jan., pp. 91–102.

Principles and Techniques of RFID Positioning

415

[79] Wilson, A., Prashanth, D., and Aghajan, H. (2007) Utilizing RFID signaling scheme for localization of stationary objects and speed estimation of mobile objects, in Proc. IEEE Conf. RFID, March, pp. 94–99. [80] Wang, C. Wu, H., and Tzeng, N. (2007) RFID-based 3D positioning schemes, in Proc. IEEE InfoCom, pp. 1235– 1243. [81] Xu, B. and Gang, W. (2006) Random sampling algorithm in RFID indoor location system, in Proc. IEEE Int. Workshop Electronic Design, Test and Applications, Jan. [82] Yin, J., Yang, Q., and Ni, L.M. (2008) Learning adaptive temporal radio maps for signal-strength-based location estimation, IEEE Trans. Mobile Computing, 7(7): 869– 883. [83] Zhang, Y., Amin, M.G., and Ahmad, F. (2008) Time-frequency analysis for the localization of multiple moving targets using dual-frequency radars, IEEE Signal Processing Letters, 15: 777– 780. [84] Zhang, Y., Amin, M.G., and Ahmad, F. (2008) Narrowband frequency-hopping radars for the range estimation of moving and vibrating targets, in SPIE Symposium on Defense and Security Symposium, Orlando, FL, March. [85] Zhang, Y. Amin, M.G., and Kaushik, S. (2007) Localization and tracking of passive RFID tags based on direction estimation, Int. J. Antennas and Propagation, Dec. [86] Zhang, T., Chen, Z., Ouyang, Y., Hao, J., and Xiong, Z. (2008) An improved RFID-based locating algorithm by eliminating diversity of active tags for indoor environment, The Computer Journal Advance Access, Aug. [87] Zhou, Y., Liu, W., and Huang, P. (2007) Laser-activated RFID-based indoor localization system for mobile robots, in Proc. IEEE Int. Conf. Robotics and Automation, pp. 4600– 4605. [88] Zhao, Y., Liu, Y., and Ni, L. (2007) VIRE: active RFID-based localization using virtual reference elimination, in Proc. Int. Conf. Parallel Processing (ICPP). [89] Zhou, J. and Shi, J. (2008) RFID localization algorithms and applications – a review, Journal of Intelligent Manufacturing, DOI:10.1007/s10845-008-0158-5, Aug. [90] Zhang, Y., Yemelyanov, K., Li, X., and Amin, M. (2009) Effect of metallic objects and liquid supplies on RFID links, in Proc. IEEE AP-S Symp., June.

16 Towards Secure and Privacy-Enhanced RFID Systems Heiko Knospe1 and Kerstin Lemke-Rust2 1

Cologne University of Applied Sciences

2

Bonn-Rhein-Sieg University of Applied Sciences

16.1 Introduction RFID is a technology for automated identification over a radio interface and drives many applications in ubiquitous and pervasive computing today. RFID systems are applied in security-sensitive areas such as ticketing, electronic passports, product anti-counterfeiting or entry systems. In this chapter we follow a generic approach to security and privacy of RFID systems. We also call it a generic (or low-level) approach because it considers generic threats and solutions to an RFID system and does not discuss application-level security protocols. This chapter is organized as follows. Section 16.2 provides an introduction to the notions of security and privacy. Section 16.3 classifies RFID transponders with respect to chip area and security functionality. In Section 16.4 we evaluate the attacks to security and privacy that are crucial to generic RFID systems. For each attack, recommendations on possible countermeasures are provided. Section 16.5 presents lightweight cryptography and discusses feasible implementations for RFID tags.

16.2 Security and Privacy Security aims at minimizing risks to the assets of computer systems. Any security design faces the following questions: 1. Which threats apply to the assets of the computer system? RFID Systems: Research Trends and Challenges and Ivan Stojmenovi´c  2010 John Wiley & Sons, Ltd

Edited by Miodrag Boli´c, David Simplot-Ryl,

418

RFID Systems

2. Which security objectives should be addressed to counteract the threats? 3. Which security functions should implement the security objectives? 4. To what extent are the implemented security functions effective in counteracting the threats? 5. What are the remaining vulnerabilities and risks to the assets of the computer system? Therefore, any security problem starts with an analysis of threats. Threats are countered by security objectives. The three most widely used security objectives from [1] are • Confidentiality: To protect and preserve the confidentiality of information means ensuring that it is not disclosed or available to unauthorized entities. • Integrity: To preserve the integrity of information means protecting the accuracy and completeness of information and the methods that are used to process and manage it. • Availability: An asset is available if it is accessible and usable when needed by an authorized entity. Other security objectives exist, for example, authenticity, non-repudiation, accountability. Also privacy forms a special class of security objectives aimed at maintaining the right of self-determination of individuals. Among them are anonymity, pseudonymity, unlinkability, and unobservability. For a detailed terminology on privacy see, for example, [2]. Note that the choice of security objectives for a specific security problem can be subjectively biased. Security mechanisms are chosen in order to support security objectives and to counteract the threats. There are three generic approaches to implement security as • a preventive action; • a detective action; • a reaction. Security mechanisms are not solely a technical problem. They also demand supporting organizational security mechanisms. The effectiveness of security mechanisms is gradually assessed according to the attack potential of an adversary. Attack potential is the combination of expenditure of money, time, equipment, and insider knowledge. Almost every solution to a security problem leaves some remaining risks to the assets. The provider of a security system is then responsible for managing and controlling these remaining risks. Possible approaches are to accept the risk, to avoid the risk, to transfer the risk, or to reduce the risk (ISO 27001 and ISO 27002 2008).

16.3 Classification of RFID Systems RFID transponders and RFID readers that are located in the end-user environment constitute the front-end of an RFID system. Optionally, further components for data and application management are used. If so, the RFID readers are connected to a corresponding back-end system. Two solutions exist in practice:

Towards Secure and Privacy-Enhanced RFID Systems

419

• either the RFID reader has an online connection to the back-end system (online RFID system); or • the RFID reader is offline but contacted in some regular time intervals (offline RFID system). In an online RFID system, the RFID reader can access the back-end system during transaction processing. Because of this, an online RFID system usually facilitates the detection of fraud, for example, the multiple use of tickets. Further, the download of configuration data and software can easily be managed for an online RFID system. In an offline RFID system RFID readers work autonomously. Transactions are downloaded from time to time by maintenance personnel. Only at an inspection, can new configuration data and software be uploaded to a reader. We provide a classification of current RFID systems with particular focus on the implemented security functionality. The security of an RFID system is mainly influenced by the transponders and their chip area. Chip area is needed for control logic, data and possibly program memory or cryptographic coprocessors. Security mechanisms therefore cause additional costs. Since the actual chip area depends on the technology, the number of gate equivalents (GEs) is generally used to measure the hardware complexity. One GE corresponds to the chip area required by a two-input NAND gate. Based on the chip and the security functionality, RFID systems can be roughly classified into three categories: 1. Low-end RFID systems: In this category, we have low-cost tags with typically less than 5000 GEs. Low-end systems are mainly used for automatic identification. Among these, there are one-bit transponders used for anti-theft in stores and tags with a permanent ID but without any security or privacy functions. Other low-end tags provide some elementary functions: a password-protected KILL command, an access password, read, write and lock memory blocks, checksums (e.g. CRC16) and a pseudo random number generator. The prevalent EPC (Electronic Product Code) Class 1 Generation 2 UHF tags (EPCglobal (2008), ISO 18000-6C) belong to this category. 2. Mid-range RFID systems: This category includes read-write transponders of moderate cost with data memory ranging from a few hundred bytes to more than 100 kB non-volatile read-write memory. Mid-range RFID systems are used for various applications such as ticketing, access control, and automotive immobilizers. Usually, these tags implement mutual authentication and enforce an access control for the transponder’s memory. Session keys are established between the RFID transponder and the reader and transmitted data is encrypted and integrity protected over the radio interface. The transponders are usually implemented as a “hard-wired” state machine and often proprietary cryptographic functions are in use. MIFARE Classic (see Section 16.4.8) is the most prominent example in this category. MIFARE is compliant with ISO 14443 (1–3) and provides read-writable data blocks that are organized in multiple cryptographically separated sectors, thereby allowing for multi-applications. 3. High-end RFID systems: These RFID transponders are compliant with ISO 14443 (1–4), contain microprocessor-based smartcard chips and are equipped with a smartcard operating system. Both the chip and its operating system implement security mechanisms. The security mechanisms are application specific and include advanced

420

RFID Systems

cryptographic functions. In the very high-end, we find dual-interface smartcards with crypto-coprocessors allowing for public key cryptography such as digital signature operation. Most current RFID standards (including ISO 14443, ISO 15693 and ISO 18000) focus on the physical characteristics, the radio interface and transmission protocols and hardly specify any security functions. The standardization of RFID security services is work in progress, for example, in the forthcoming standard ISO 29167.

16.4 Attacks on RFID Systems and Appropriate Countermeasures Components that are located in an end-user environment are usually at high risk of attack. This concerns the RFID reader, the RFID transponder and the communication between them. The RFID transponder is usually exposed to a higher risk of attack because it is handed over to all the end-users of an RFID system while RFID reader are usually located in environmentally protected areas. The assets and the attack objectives depend on the concrete RFID application. From a more generic point of view, threats to RFID systems comprise • RFID-enabled theft of assets stored in the RFID transponder, • RFID-enabled theft of identity credentials stored in the RFID transponder, and • RFID enabled vandalism [3]. Security is usually achieved by a combination of technical and organizational security mechanisms. Technical security mechanisms include hardware and software mechanisms of the components and the use of secure cryptographic protocols. Organizational security mechanisms are realized in the environment of the RFID-system and may include guards, alarm systems, auditing, as well as fraud detection in back-end systems. This section aims at providing an overview of security attacks and corresponding risks. As with any information system there are four generic attacks on the communication flow between a sender and a receiver (see Figure 16.1): (a) eavesdropping of messages (see Section 16.4.1), (b) interruption of messages (see Section 16.4.2), (c) manipulation of messages (see Section 16.4.3), and (d) generation of messages (see Section 16.4.4). Afterwards, we discuss a specific attack on a message flow that applies to wireless communication channels: (e) relay of messages (see Section 16.4.5). Attacks can also be directed at the data of an RFID system. Tracking infringes the privacy of human individuals due to the fact that active communication initiated from an adversary’s RFID reader cannot be notified (see Section 16.4.6). Cloning means that an attacker produces duplicates of RFID transponders. It is discussed in Section 16.4.7. Cloning can be achieved by cryptanalytical and implementation attacks as discussed in Section 16.4.8 and Section 16.4.9. There are other highly relevant attack paths that apply to almost any computer security system and also to RFID systems. Among them are:

Towards Secure and Privacy-Enhanced RFID Systems

Receiver

Sender

421

Adversary

Adversary

(a) Eavesdropping

Sender

Receiver

Adversary (c) Manipulation

Figure 16.1

Receiver

Sender

(b) Interruption

Sender

Receiver

Adversary (d) Generation

Illustration of common attacks to the information flow.

• social engineering attacks that spoof human users and administrators in order to thwart system security; • network attacks such as denial-of-service that are conducted via network interfaces on the back-end system and also on the RFID readers in an online RFID system; • protocol attacks that exploit weaknesses of cryptographic protocols used on a communication interface of the RFID system; • software implementation attacks that exploit software vulnerabilities such as buffer overflows and unsecured software updates on the software running in components of an RFID system. These general attacks on computer systems are beyond the scope of this chapter. For an overview on computer security, [4] is recommended. A valuable survey on RFID security and privacy can also be found in [7].

16.4.1 Eavesdropping of Messages The communication between the RFID reader and the RFID transponder is made via the radio interface that is subject to interception by an adversary in the vicinity. Such an interception of the data communication is an eavesdropping attack. Note that eavesdropping is a passive attack. For the RFID-communication two different physical channels are used: • RFID reader to RFID transponder (forward channel ): This channel provides the RFID transponder with energy and is used for transferring data from the reader to the transponder. • RFID transponder to RFID reader (backward channel ): This channel is used by the RFID transponder to send data from the transponder to the reader.

422

RFID Systems

At 13.56 MHz, the signal from the reader to the transponder is about 80 dB stronger [5] than the load modulation signal which is used for communication on the backward channel. Therefore, from an enlarged distance, it is significantly more difficult to observe data sent from the transponder than the data which the RFID reader sends to it. However, eavesdropping on the two-channel RF communication from several meters imposes a real threat, for example, recent work by Hancke [6, 8] practically demonstrated that the two-way communication at 13.56 MHz between an RFID reader and an RFID tag can be intercepted from about 3 meters. It is said that results vary on eavesdropping components and environmental conditions. In a concrete setting a faraway eavesdropper may only be able to monitor the forward channel which is said to be possible from a distance up to about 10 meters [9]. In the frequency range of 868 MHz to 2.45 GHz the eavesdropping range can be enlarged with the use of directional antennas. Under good conditions the forward channel is said to be receivable from distances up to a few hundred meters and the backwardchannel from distances up to a few tens of meters [5]. In order to prevent eavesdropping attacks it is necessary to apply cryptographic encryption of messages (see Sections 16.5.2 and 16.5.3).

16.4.2 Denial-of-Service For RFID systems it is essential that the communication channel between RFID transponders and RFID readers remains fully functional. Denial-of-service attacks aim at interrupting the communication between an RFID reader and an RFID transponder. Attacks apply to both the RFID reader and the RFID transponder. A denial-of-service attack may involve the destruction of RFID transponders so that they do no longer respond to reader requests. Destruction of a transponder can be carried out mechanically or chemically to cut off the antenna from the chip, or by a high electromagnetic field strength at the carrier frequency in order to thermally destroy the transponder [5]. Accordingly, an RFID reader can be subject to vandalism. Another denial-of-service approach is to remove the detachment of RFID tags to valuable goods. Shielding of RFID transponders with aluminium foil has the same effect and may have the advantage that no physical destruction of RFID tags is required. A jamming transmitter is a third approach to denial-of-service attacks. A blocker tag [9] emulates a high number of transponders in the field of the RFID reader. In order to cope with the presence of multiple transponders in the reader’s range an anti-collision mechanism is implemented in an RFID reader. The anti-collision mechanism shrinks the referenced number space until only one tag responds that is afterwards selected for the communication session. A blocker tag, however, emits multiple serial numbers in each referenced number space. Identification of the genuine serial number among emulated serial numbers is thereby prevented. Due to collisions the reader is not able to determine a unique serial number and the reader is stuck with the anti-collision mechanism. Denial-of-service attacks are a risk to anti-theft protection and access control systems. Prevention and detection of denial-of-service attacks are known to be a difficult problem that cannot be solved with cryptography.

Towards Secure and Privacy-Enhanced RFID Systems

423

16.4.3 Manipulation of Messages RFID messages are sent on a radio channel in the near vicinity of an RFID reader and an RFID transponder. Manipulation on this radio channel is in principle possible with a transmitter under the control of the adversary, but may have a high risk resulting in a denial-of-service attack. So far manipulation of messages in transit is not known as a common attack on RFID systems. In order to detect manipulations of messages it is required that sender and receiver use cryptographic checksums (see Section 16.5.4). Hereby, it is necessary that the receiver of a message checks that the cryptographic checksum is correct.

16.4.4 Generation of Messages Generation of messages by an adversary is a common attack to RFID systems. Attack objectives include the impersonation of holders of a genuine RFID transponder. Note that an RFID system is based upon machine-to-machine authentication. An adversary that has gained possession of a genuine RFID transponder can impersonate the genuine owner of the RFID transponder. Most prominent is the replay attack that retransmits a sniffed original RFID message at a later time and possibly in another context. In order to detect unauthorized generation of messages it is required that sender and receiver use cryptographic message authentication codes and message integrity. Hereby, it is necessary that the receiver of a message checks that the message authentication code is correct and the message integrity is fulfilled. Unauthorized transfer of ownership of transponders requires additional measures such as human user authentication based on knowledge or biometrics.

16.4.5 Relay of Messages Authentication with RFID transponders usually assumes that the transponder’