Smart Tag Detection Techniques for Chipless RFID Systems · The core theme of the thesis is the...

Smart Tag Detection Techniques forChipless RFID Systems

Chamath M Divarathne

A thesis submitted in total fulfillment of the requirements of thedegree of

Doctor of Philosophy

Department of Electrical and Computer Systems EngineeringMONASH UNIVERSITY

AUSTRALIA

November 2015

Copyright c© 2015 Chamath M. Divarathne

All rights reserved. No part of the publication may be reproduced in any formby print, photo print, microfilm or any other means without written permis-sion from the author.

Copyright Notices

Notice 1

Under the Copyright Act 1968, this thesis must be used only under the normalconditions of scholarly fair dealing. In particular no results or conclusionsshould be extracted from it, nor should it be copied or closely paraphrasedin whole or in part without the written consent of the author. Proper writtenacknowledgement should be made for any assistance obtained from this thesis.

Notice 2

I certify that I have made all reasonable efforts to secure copyright permissionsfor third-party content included in this thesis and have not knowingly addedcopyright content to my work without the owner′s permission.

iii

Declaration

In accordance with Monash University Doctorate Regulation 17.2 Doctor of

Philosophy and Master of Philosophy (MPhil) regulations the following decla-

rations are made:

I hereby declare that this thesis contains no material which has been accepted

for the award of any other degree or diploma at any university or equivalent

institution and that, to the best of my knowledge and belief, this thesis contains

no material previously published or written by another person, except where

due reference is made in the text of the thesis.

The core theme of the thesis is the development of smart detection techniques

for chipless RFID tags. The ideas, development and writing up of all the work

in the thesis were the principal responsibility of myself, the candidate, working

within the Department of Electrical and Computer Systems Engineering under

the supervision of Associate Professor Nemai Karmakar and Professor Jamie

Evans.

Signed: Chamath Divarathne

Date: November 2015

v

Abstract

RADIO Frequency Identification (RFID) is a wireless technology used

to automatically identify objects attached to its tags. Its applications

span in different areas such as inventory control, logistics, security and item

tracking. Vast majority of commercially available RFID tags use Application-

Specific Integrated Circuits (ASICs) to encode and transmit data. This micro-

chip in the RFID tag makes the tag manufacturing process complicated and

expensive compared to optical barcode printing. Researchers have brought the

idea of removing the micro-chip and using chipless techniques to encode data

into tags, allowing them to be passive, printable and low cost. However, chip-

less RFID technologies have still not been able to replace relatively expensive

chipped RFID tags mainly due to less tag bit capacity. Over the last decade,

researchers have mainly focused on improving the chipless RFID tag design

and the RFID reader architecture. However, they were mostly using primitive

signal processing techniques such as moving average or threshold based detec-

tion. The few advanced signal processing techniques reported so far have high

computation complexity, hence not feasible for commercial implementation.

This thesis presents smart tag detection techniques that are computation-

ally feasible and allowing high tag data encoding capacity. Firstly, four dif-

ferent maximum likelihood (ML) based tag detection techniques have been

developed based on the reader architecture and channel knowledge. In ad-

dition, all of them are able to operate based on both the time and frequency

vii

domain data samples of any frequency domain tag. One of the detection tech-

niques jointly detects the channel as well as the tag type without having any

prior channel knowledge or a calibration tag. A fifth tag detection technique

was developed for an existing frequency domain tag reader using the mag-

nitude of the tag response. However, these single input single output (SISO)

based tag detection techniques suffer from high computation complexity. Two

new detection methods have been developed using the likelihood expressions

derived in above techniques to reduce the computation complexity from expo-

nential to linear order. The first method was a suboptimal bit by bit detection

technique (serial reading) and the second method is a fully optimal Trellis tree

based Viterbi decoding technique. Then a novel, multiple input multiple out-

put (MIMO) based chipless RFID system was introduced and a tag detection

technique for the proposed system was developed. Finally a MIMO chipless

tag was designed which includes a broadband equal power divider, monopole

antennas and spiral resonators.

It was found that, the proposed tag detection techniques for SISO systems

provides significantly higher tag reading accuracy over the existing threshold

based detector. In addition, they are capable of operating without a guard-

band which makes the tag data bit capacity to be doubled without compromis-

ing the reading accuracy. Moreover, the effective SNR gain provided by the

proposed techniques can be represented as increasing the tag reading range.

All these benefits were achieved without compromising the low computation

complexity. The MIMO tag with 2 branches is capable of encoding up to 4

times the total bits stored in existing SISO tags.

These smart tag detection techniques are expected to increase the data bit

capacity in chipless RFID tags hence produce commercialized chipless RFID

systems in future.

viii

List of Publications

Journal Papers

1. Divarathne, Chamath; Karmakar, Nemai, ”Advanced Signal Processing Tech-

niques for MIMO based Chipless RFID Systems”, IEEE Transactions on Antennas

and Propagation, (in draft)

2. Divarathne, Chamath; Karmakar, Nemai, ”Smart Tag Detection Techniques for

SISO based Chipless RFID Systems”, Wireless Power Transfer Journal Special Issue

on Chipless Technologies, Cambridge University Press, (under review, submitted on

28th Feb 2015)

Conference Papers

3. Divarathne, Chamath; Karmakar, Nemai, ”An Advanced Tag Detection Tech-

nique for Chipless RFID Systems”, European Microwave Conference (EuMC), 2015

45th, Paris, France, 6-11 Sept. 2015, (accepted)

4. Divarathne, Chamath; Karmakar, Nemai, ”A Smart Tag Detection Technique

for Chipless RFID Readers”, Fourteenth Australian Symposium on Antennas, Syd-

ney, Australia, 18-19 Feb. 2015

5. Divarathne, Chamath; Karmakar, Nemai, ”A Maximum Likelihood Based Tag

Detection Technique for MIMO Chipless RFID Systems”, 2014 IEEE MTT-S In-

ternational Microwave and RF Conference (IMARC 2014), Bangalore, India, pp.5,8,

15-17 Dec. 2014

6. Divarathne, Chamath; Karmakar, Nemai, ”ML detection based SISO Chipless

ix

RFID tag reading,” European Microwave Conference (EuMC), 2014 44th, Rome,

Italy, pp.762,765, 6-9 Oct. 2014

7. Divarathne, Chamath; Karmakar, Nemai, ”A Feasible Detection Technique for

Chipless RFID Systems based on Likelihood”, 2014 Australian Microwave Sym-

posium (AMS2014), 26-27 Jun. 2014.

8. Divarathne, Chamath; Karmakar, Nemai, ”MIMO based chipless RFID sys-

tem,” 2012 IEEE International Conference on RFID-Technologies and Applications

(RFID-TA2012), pp.423,428, 5-7 Nov. 2012.

Poster Presentations

9. Divarathne, Chamath; Evans, Jamie; Karmakar, Nemai, ”ML Detection based

Chipless RFID Tag Reading”, 15th annual Australian Communications Theory Work-

shop (AusCTW2014), 3-5 Feb. 2014.

Books

10. Karmakar, Nemai; Zomorrodi, Mohammad; Divarathne, Chamath,”Advanced

Signal Processing Techniques for Chipless RFID Systems”, John Wiley & Sons,

Inc. (Book proposal accepted on 12th Aug. 2015.)

x

Acknowledgments

Upon completing my journey of PhD, I would like to thank everyone who

stood by me during these long years and gave me encouragement and support

when I needed it the most.

First and foremost I would like to thank A/Prof. Nemai Karmakar, for

his support and guidance throughout my PhD. He has been actively interested

in my work and has always been available to advise me. I am very grateful

for his patience, motivation, enthusiasm, and immense knowledge in chipless

RFID technologies. I particularly want to thank Prof. Jamie Evans for contin-

uous support and guidance given, ever since he became my co-supervisor in

early 2012. I really appreciate his advice on both technical topics as well as

numerous other disciplines, that taken together, make him a great mentor to

me.

This research work is part of a research project funded by an Australian

Research Council (ARC) Linkage Project Grant number DP110105606: Elec-

tronically Controlled Phased Array Antenna for Universal UHF RFID Appli-

cations.

A special thank goes to Ms. Jane Moodie for assistance in improving

essential research communication skills. I also appreciate the advice given by

Ms. Roslyn Rimington during various milestones of the PhD candidature. I

xi

would like to thank the friendly staff members at ECSE including Mr. Geoff

Binns, Ms. Emily Simic and Ms. Maria Scalzo, who happily helped to retain

my little Greek speaking skills. In addition, I should thank Prof. Jean Am-

strong for lending the arbitrary waveform generator (AWG) numerous times

for carrying out experiments.

I want to thank present and past members of the MMARS lab includ-

ing Dr. Randika Koswatta, Dr. Uditha Bandara, Dr. Vajira Amaratunga and

Dr. A.K.M. Azad, Dr. Prasanna Kalansuriya, Dr. Shivali Bansal, Dr. Aminul

Islam, Dr. Shakil Buiyan, Dr. Mohammad Zomorrodi, Dr. Emran Amin, Anee

Azim, Wan Wan Zamri, Yixian Yap, Shuvashis Dey, Muhsiul Hassan, Anushka

Bibile, Sika Shrestha and Arif Shahriar for being the ultimate lab neighbours,

providing a great work environment, and for their help and chats.

I would like to express my appreciation to my friends, Dr. Dayan Han-

dapangoda, Kashyapa Sirinanda, Dinuka Karunarathne, Dr. Madara Maras-

inghe, Dr. Tharaka Samarasinghe, Nalika Dona, Gayan Samarasekara, Imesha

Samaratunga, Dr. Saman Atapattu, Stewart Coad and last but certainly not

least Anuradha Madugalla for their feedback, faith and support.

My heartfelt appreciation goes to my parents, siblings and other family

members without whom I will not be able to achieve what I have gained till

date. Their support and guidance have been the main pillars of my successful

achievements.

xii

Dedicated to

my family and all noble friends

. . .

xiii

Contents

1 Introduction 11.1 RFID Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.1.1 SISO based RFID systems . . . . . . . . . . . . . . . . . . 31.1.2 MIMO based RFID systems . . . . . . . . . . . . . . . . . 4

1.2 Research aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.3 Original contributions . . . . . . . . . . . . . . . . . . . . . . . . 61.4 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 RFID Systems 112.1 Introduction to RFID systems . . . . . . . . . . . . . . . . . . . . 112.2 Chipless RFID tag types . . . . . . . . . . . . . . . . . . . . . . . 12

2.2.1 Time domain based tags . . . . . . . . . . . . . . . . . . . 122.2.2 Frequency domain based tags . . . . . . . . . . . . . . . . 14

2.3 MIMO based chipped RFID systems . . . . . . . . . . . . . . . . 172.4 Review of chipless RFID tag detection techniques . . . . . . . . 212.5 Maximum likelihood detection techniques . . . . . . . . . . . . 222.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3 Chipless RFID Tag Design 273.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273.2 SISO tag design . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.2.1 Tag design and fabrication . . . . . . . . . . . . . . . . . . 283.2.2 Experimental setup . . . . . . . . . . . . . . . . . . . . . . 29

3.3 MIMO tag design . . . . . . . . . . . . . . . . . . . . . . . . . . . 313.3.1 Power divider design . . . . . . . . . . . . . . . . . . . . 323.3.2 Monopole antenna design . . . . . . . . . . . . . . . . . . 333.3.3 Spiral resonator design . . . . . . . . . . . . . . . . . . . . 373.3.4 Experimental setup . . . . . . . . . . . . . . . . . . . . . . 39

3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

4 ML Detection Techniques for SISO Chipless RFID Tags 434.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

xv

4.2 System models - Time domain . . . . . . . . . . . . . . . . . . . . 444.2.1 System model I - Real signals . . . . . . . . . . . . . . . . 454.2.2 System model II - Complex signals . . . . . . . . . . . . . 494.2.3 System model III - Channel with a known distribution . 524.2.4 System model IV - Unknown channel . . . . . . . . . . . 554.2.5 Joint optimization of h and tag type . . . . . . . . . . . . 57

4.3 System models - Frequency domain . . . . . . . . . . . . . . . . 594.3.1 System models I - IV . . . . . . . . . . . . . . . . . . . . . 594.3.2 System model V - Power magnitudes . . . . . . . . . . . 62

4.4 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 654.5 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . 694.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

4.6.1 System model I . . . . . . . . . . . . . . . . . . . . . . . . 734.6.2 System model II . . . . . . . . . . . . . . . . . . . . . . . . 774.6.3 System model III . . . . . . . . . . . . . . . . . . . . . . . 824.6.4 System model IV . . . . . . . . . . . . . . . . . . . . . . . 864.6.5 System model V . . . . . . . . . . . . . . . . . . . . . . . . 95

4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

5 Computationally Feasible Tag Detection Techniques 995.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 995.2 Bit by bit detection method . . . . . . . . . . . . . . . . . . . . . 1015.3 Trellis tree based Viterbi decoding . . . . . . . . . . . . . . . . . 105

5.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 1055.3.2 Signal model . . . . . . . . . . . . . . . . . . . . . . . . . 106

5.4 Simulation setup . . . . . . . . . . . . . . . . . . . . . . . . . . . 1125.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

5.5.1 Detection error rate (DER) . . . . . . . . . . . . . . . . . . 1165.5.2 Computation time . . . . . . . . . . . . . . . . . . . . . . 117

5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

6 Signal Processing for MIMO based Chipless RFID Systems 1196.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1196.2 MIMO decomposing techniques . . . . . . . . . . . . . . . . . . 1226.3 Tag detection in MIMO . . . . . . . . . . . . . . . . . . . . . . . . 1256.4 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . 1286.5 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

6.5.1 Method 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1316.5.2 Method 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

6.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1346.6.1 Method 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1346.6.2 Method 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

xvi

6.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

7 Conclusion 1497.1 Fulfilling the goals of the thesis . . . . . . . . . . . . . . . . . . . 1497.2 Limitations of the proposed system . . . . . . . . . . . . . . . . . 1537.3 Potential applications . . . . . . . . . . . . . . . . . . . . . . . . . 1557.4 Future work and open issues . . . . . . . . . . . . . . . . . . . . 156

xvii

List of Figures

1.1 A typical RFID system . . . . . . . . . . . . . . . . . . . . . . . . 21.2 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.1 Comparison of different RFID systems . . . . . . . . . . . . . . . 122.2 Classification of chipless RFID tags . . . . . . . . . . . . . . . . . 132.3 Structure of a multiresonator based chipless RFID tag . . . . . . 16

3.1 Tag types used in SISO and MIMO detection algorithm devel-opments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2 Printed Tag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 293.3 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . 303.4 Magnitude of the tag response for tag [1111] . . . . . . . . . . . 323.5 T-junction power divider (left: CST design, right: fabricated

power divider) TLX8 substrate with εr=2.4, tanδ=0.004 and h=0.5mm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.6 S-parameters of the power divider . . . . . . . . . . . . . . . . . 343.7 Monopole antenna (left: CST design, right: fabricated monopole) 353.8 Return loss of the monopole antenna . . . . . . . . . . . . . . . . 353.9 Simulated radiation pattern of the monopole antenna . . . . . . 363.10 Realized gain of the monopole antenna . . . . . . . . . . . . . . 373.11 Spiral resonators (top: CST design, bottom: fabrication) . . . . . 373.12 CST generated resonator response . . . . . . . . . . . . . . . . . 383.13 Fabricated MIMO tag . . . . . . . . . . . . . . . . . . . . . . . . . 383.14 MIMO tag experiment . . . . . . . . . . . . . . . . . . . . . . . . 393.15 Tag response for [1010] . . . . . . . . . . . . . . . . . . . . . . . . 40

4.1 RFID System Models . . . . . . . . . . . . . . . . . . . . . . . . . 444.2 Overview of Chipless RFID System . . . . . . . . . . . . . . . . . 454.3 Proposed Signal Models . . . . . . . . . . . . . . . . . . . . . . . 494.4 Flowchart of the MATLAB simulation in conjunction with CST

full-wave EM solver simulation . . . . . . . . . . . . . . . . . . . 664.5 A chipless tag coded with bits [1111] . . . . . . . . . . . . . . . . 694.6 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . 70

xix

4.7 Interrogating signal in time and frequency domain . . . . . . . . 714.8 Tag responses for [1111] with a guard band . . . . . . . . . . . . 724.9 Tag responses for [1111] without a guard band . . . . . . . . . . 734.10 DER vs SNR for 4-bit tag with 60 MHz guard band . . . . . . . . 744.11 DER vs SNR without a guard band between resonator frequencies 754.12 DER vs SNR for ML decoder 1 . . . . . . . . . . . . . . . . . . . 764.13 Real and imaginary samples of the tag response [1111] . . . . . 784.14 Frequency signature of tag type [1111] . . . . . . . . . . . . . . . 794.15 DER vs SNR for ML decoder 2 with the presence of a guard-band 794.16 DER vs SNR for ML decoder 2 without a guard-band . . . . . . 804.17 DER vs SNR for ML decoder 2 . . . . . . . . . . . . . . . . . . . 814.18 DER comparison for 8-bit tags . . . . . . . . . . . . . . . . . . . . 824.19 DER vs SNR for ML decoder 3 with the presence of a guard-band 834.20 DER vs SNR for ML decoder 3 without a guard-band . . . . . . 844.21 DER vs SNR for ML decoder 3 . . . . . . . . . . . . . . . . . . . 854.22 DER vs SNR for ML decoder 4 with the presence of a guard-band 864.23 DER vs SNR for ML decoder 4 without a guard-band . . . . . . 874.24 DER vs SNR for ML decoder 4 . . . . . . . . . . . . . . . . . . . 884.25 Channel Estimation Samples when a guard-band is presented . 894.26 PDF of Channel Estimation when a guard-band is presented . . 904.27 Channel Estimation Samples without a guard-band . . . . . . . 914.28 PDF of Channel Estimation without a guard-band . . . . . . . . 914.29 DER comparison with a guard-band . . . . . . . . . . . . . . . . 924.30 DER comparison without a guard-band . . . . . . . . . . . . . . 954.31 Magnitude of the tag response for tag [1111] . . . . . . . . . . . 964.32 DER vs SNR for likelihood based detector 5 for 21-27 GHz backscat-

tering tag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

5.1 Bit by bit detection for a tag having [1111] . . . . . . . . . . . . . 1025.2 Flowchart of bit by bit detection technique . . . . . . . . . . . . 1035.3 Operation of Trellis tree based Viterbi detection technique . . . 1085.4 Viterbi decoding in a Trellis tree . . . . . . . . . . . . . . . . . . . 1105.5 Flow chart of Trellis tree based Viterbi decoding . . . . . . . . . 1115.6 Flowchart of the MATLAB simulation . . . . . . . . . . . . . . . 1135.7 DER comparison for 10-bit tags . . . . . . . . . . . . . . . . . . . 1165.8 Computation complexity comparison . . . . . . . . . . . . . . . 117

6.1 MIMO based chipless RFID system . . . . . . . . . . . . . . . . . 1206.2 MIMO tag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1216.3 MIMO tag operation overview . . . . . . . . . . . . . . . . . . . 1226.4 MIMO tag experiment . . . . . . . . . . . . . . . . . . . . . . . . 1296.5 Tag response for [1010] . . . . . . . . . . . . . . . . . . . . . . . . 1296.6 Flowchart of the MATLAB simulation . . . . . . . . . . . . . . . 131

xx

6.7 Flowchart of the MATLAB simulation . . . . . . . . . . . . . . . 1346.8 Interrogating signal in time domain . . . . . . . . . . . . . . . . 1356.9 Two sided PSD of the Interrogating Signal . . . . . . . . . . . . . 1366.10 Received Signal at the Tag . . . . . . . . . . . . . . . . . . . . . . 1366.11 Filter response of a spiral resonator . . . . . . . . . . . . . . . . . 1376.12 Two-sided PSD of the tag modulated signals (Tx1 and Tx2) . . . 1386.13 Tag modulated signals (Tx1 and Tx2) in time domain . . . . . . 1386.14 Channel realizations . . . . . . . . . . . . . . . . . . . . . . . . . 1396.15 Received Signals at the two Rx antennas of the Reader . . . . . . 1406.16 Actual and the Estimated Tx1 . . . . . . . . . . . . . . . . . . . . 1406.17 Actual and the Estimated Tx2 . . . . . . . . . . . . . . . . . . . . 1416.18 Combined Tag Response . . . . . . . . . . . . . . . . . . . . . . . 1426.19 Combined Tag Response for 100 iterations . . . . . . . . . . . . . 1426.20 BER of the Proposed System vs SNR . . . . . . . . . . . . . . . . 1436.21 Noise Performance of the Proposed System vs SISO counterpart 1446.22 CST generated tag response for a branch having [1111] tag bits . 1456.23 Comparison of DER performances for 6 bit tags . . . . . . . . . 145

xxi

List of Tables

2.1 Comparison of communication system with a chipless RFID sys-tem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3.1 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . 283.2 Tag types given by resonator combinations . . . . . . . . . . . . 31

4.1 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . 684.2 DER comparison for different detection methods . . . . . . . . . 944.3 Likelihood for each tag type . . . . . . . . . . . . . . . . . . . . . 96

5.1 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . 115

6.1 An Example of a Table . . . . . . . . . . . . . . . . . . . . . . . . 1306.2 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . 1326.3 Simulation Parameters . . . . . . . . . . . . . . . . . . . . . . . . 133

7.1 Technical specifications of Raspberry Pi 2 Model B . . . . . . . . 157

xxiii

Chapter 1

Introduction

ITEM tagging and monitoring have become significant than ever before,

due to the recent emergence of new technologies and their penetration

in mass market. Particularly in mass production sites, automated item tag-

ging can increase the productivity and efficiency which in turn will grow the

company revenue. Optical barcode technology is dominant in the item tag-

ging market at present due to its relatively low implementation cost. How-

ever, there are a number of limitations such as low reading range, line of sight

(LOS) requirement for reading and its inability to identify multiple items si-

multaneously. All these challenges suggest barcodes is not a feasible solution

in automating item tagging. Radio frequency identification (RFID) shades the

light to overcome these limitations towards automating the process, however

with high priced tags. The focus of the research is to make the tag comparably

cheaper to the optical barcodes by removing the microchip, but without loos-

ing functionality. Such tags are called chipless RFID tags. The thesis presents a

few advanced tag detection techniques and high data capacity chipless tag us-

ing likelihood based detection techniques and multiple input multiple output

(MIMO) tag design.

1

2 Introduction

Enterprise Software Solutions

RFID TagReader/ Interrogator

Middleware

Interrogation Signal

Tag Response

Figure 1.1: A typical RFID system

1.1 RFID Systems

RFID is a wireless technology used to automatically identify objects attached to

its tags. RFID technology appears to offer as an alternative to optical barcodes

due to its unique advantages such as larger reading range, non LOS reading,

multiple tag detection and its ability to be able to automate the item identifi-

cation process. A typical RFID system has a reader, a tag and middleware [1]

as shown in Figure 1.1. The RFID reader sends an interrogating signal, which

is an electromagnetic (EM) signal, towards the tag and the tag responds back

with information embedded into it to the reader’s receiver. Then the reader ex-

tracts the information originally encoded by the tag and utilizes middleware

to interface that with the user in a meaningful manner. RFID systems have

applications in number of areas including inventory control, security, logistics

and item tracking. The RFID reader usually transmits an interrogating signal

toward the RFID tag. The tag then modulates the signal with its ID codes and

retransmits or backscatters the modulated signal toward the reader. This pro-

cess is called tag modulation. There are different modulation techniques that

are discussed later in the thesis. The response from the tag is then analyzed to

1.1 RFID Systems 3

detect and uniquely identify the tag using the signal processing capabilities at

the reader. Finally, the middleware integrates the tag identification data with

an enterprise software to facilitate the automation process [2]. RFID systems

have applications in a number of areas including inventory control, security,

logistics and item tracking.

A vast majority of commercially available RFID tags use application-specific

integrated circuits (ASICs) to encode and transmit data. This micro-chip in

the RFID tag makes the tag manufacturing process complicated and expensive

compared to optical barcode printing. Researchers have proposed the idea

of removing the micro-chip and using new data encoding techniques. These

chipless RFID tags can be printed on paper, read non-LOS, fabricated at low

cost and made fully passive without requiring any energy source [3].

A number of RFID systems have been reported in the literature [1, 3–35],

which are based on either chipped or chipless RFID tags. In chipped RFID

systems a reasonable amount of processing is done at the tag with the help of

a micro-chip. However, in chipless RFID systems, the tag is given a minimum,

if not nil, processing capabilities, as the RFID reader takes all the burden of the

signal processing. A brief summary of the evolution of both single input single

output (SISO) and MIMO based RFID systems is presented next.

1.1.1 SISO based RFID systems

Early research on RFID systems was mainly based on systems having a single

antenna at the tag, and also a single antenna at the reader hence named as SISO

systems. Since separate antennas for transmission and reception at the reader

as well as at the tag enhance the system performance [36, 37], researchers pro-

posed multiple antennas at the RFID tags and/or at the RFID readers. It is

noteworthy that there is only one dedicated transmission antenna or receiving

4 Introduction

antenna. These SISO systems can be seen in both chipped and chipless RFID

proposed in literature.

However, chipless RFID technologies have still not been able to replace

optical barcoding or chipped RFID tags mainly due to lower data capacity,

reading range and tag reading accuracy. Over the last decade, researchers

have mainly focused on overcoming these challenges by improving the chip-

less RFID tag design and the RFID reader architecture [2, 3, 38–41]. However,

they were mostly using primitive signal processing techniques such as moving

average or threshold based detection [42–44]. The few advanced signal pro-

cessing techniques [45–50] reported so far have high computation complexity,

hence they are not feasible for commercial implementation. It was identified

that, there is a significant research gap for computationally feasible smart tag

detection techniques for chipless RFID systems.

The thesis addresses the research gap and hypothesizes likelihood based

detection techniques for accurate tag detection. The techniques also improve

data capacity by removing the guard-bands between resonant frequencies of

the domain chipless RFID tags. They also increase the read range by requiring

low signal to noise ratio (SNR) in detected signals. Then the tag reading was

made faster with computationally feasible tag detection techniques such as

trellis tree based Viterbi decoding and bit-by-bit suboptimal approaches.

1.1.2 MIMO based RFID systems

On the other hand, researchers have successfully applied multiple input mul-

tiple output (MIMO) techniques in [51] RFID systems using multiple antennas

for transmission or reception. These multiple antennas lead to achieve spe-

cific MIMO advantages such as spatial multiplexing and high diversity gain

[51–53]. Initial research was carried out, deploying more than one receiving

1.2 Research aims 5

antenna at the reader anticipating to take advantage of receiver diversity. The

next development was multiple transmitting antennas on the RFID tag to ex-

plore spatial diversity. These MIMO based RFID systems possess higher data

capacity and improved reading range. However, these MIMO based RFID sys-

tems reported so far are only based on chipped tags [12,36,37,51–54]. The main

challenge for chipless RFID systems is MIMO requires advanced signal pro-

cessing capabilities and the chipless tag is not able to achieve it. As a result,

more advanced signal processing techniques should be investigated for de-

ploying at the RFID reader. The thesis designs and develop tag detection tech-

niques for MIMO based chipless RFID systems overcoming these challenges.

1.2 Research aims

Chipless RFID systems will be commercially feasible only when they can be

fully printable hence reducing the tag fabrication cost to a fraction of a cent.

The main challenges identified for fully printable chipless RFID systems are

the low data capacity, low reading range and low reading accuracy. This re-

search aims to overcome these obstacles and develop advanced yet computa-

tionally feasible tag detection techniques for the chipless RFID systems.

The main objective of this research is to develop computationally feasible

tag detection techniques for chipless RFID systems that:

1. possess a higher reading accuracy compared to existing chipless RFID

systems,

2. allow encoding of higher data capacity by improving the spectral effi-

ciency in frequency based chipless RFID systems,

3. demonstrates the improvement of the reading range, and

4. have the potential to replace optical barcode systems used in mass pro-

duction item tagging.

6 Introduction

In order to achieve these research aims the following sub tasks were fulfilled.

• Four ML based tag detection techniques were developed under different

practical scenarios.

• A channel estimation technique was developed.

• A power magnitude based tag detection technique for an existing reader

was developed.

• Two computationally feasible tag detection techniques were developed.

• A MIMO based chipless RFID system was proposed.

• A SISO and a MIMO tag were developed to verify the proposed tag de-

tection techniques.

Successful completion of these sub tasks led to a number of original contri-

butions which will be explained in the next section.

1.3 Original contributions

During the course of this thesis, the following original contributions to the field

of research were generated:

• Four ML based tag detection techniques (parallel reading) with extremely

low Detection Error Rate (DER).

• A detection technique for an existing chipless RFID reader that uses mag-

nitudes of a frequency domain tag response.

• All of these five detection techniques are compatible with large number

of different frequency domain based tag types currently available.

• Developed a computationally feasible sub-optimal tag detection tech-

nique (serial reading).

• Trellis tree based Viterbi tag detection technique for RFID systems with

a low computation complexity.

• A channel estimation technique for chipless RFID systems.

1.4 Thesis outline 7

• Developed advanced signal processing techniques for MIMO based RFID

system that improves the data bit capacity by a factor of two.

• Designed a MIMO based chipless RFID tag.

The following outcomes are achieved through the above development.

• All proposed detection techniques improves DER compared to existing

threshold based technique.

• All proposed detection techniques improves the tag reading range.

• All proposed detection techniques allows to remove the guard-band presents

between resonance frequencies allowing the data capacity in chipless

RFID tags to be doubled.

• Introduced ONE TIME tag reading philosophy to chipless RFID systems.

To date, the above original contributions to the field of research have gen-

erated (i) one book proposal, (ii) one submitted referred journal papers of high

impact factor and (iii) five referred conference papers (iv) one poster presenta-

tion. A full list of publications can be found on pages ix-x.

1.4 Thesis outline

This section provides a brief description of the chapters presented in this the-

sis. As can be seen in Figure 1.2, the first two chapters gives a brief introduction

to the thesis and a literature survey of the reported work in the field respec-

tively. Chapter 3 involves both SISO and MIMO tag design and experimental

verifications of the proposed tag detection techniques. Chapter 4, 5 and 6 dis-

cuss proposed tag detection techniques for chipless RFID systems. Chapter

4 and 6 presents tag detection techniques for SISO and MIMO based chipless

RFID systems respectively, while chapter 5 presents computationally feasible

tag detection techniques. Finally chapter 7 concludes the thesis and shares the

8 Introduction

Ch.1: Introduction

Ch.2: RFID Systems

Ch.4: ML Detection Techniques for SISO

Chipless RFID Systems

Ch.6: Signal Processing for MIMO Chipless RFID

Systems

Ch.5: Computationally Feasible Tag Detection

Techniques

Ch.7: Conclusion

Research problem and the background

Tag detection

techniques

Conclusion and future directions

Ch.3: Chipless RFID Tag Design

Tag Development

Figure 1.2: Thesis structure

future directions and recommendations. A detailed summary of the contents

presents in each chapter is explained next.

1.4 Thesis outline 9

Chapter 1: Introduction

An introduction to RFID systems and the research aims were presented. Then

the original contributions of the research were listed followed by an outline of

the chapters.

Chapter 2: RFID Systems

The first part of the chapter focuses on two areas in SISO chipless RFID sys-

tems. First it reviews the available chipless RFID tag types and identify the

potential candidate tag types for further investigation. Next, the existing tag

detection techniques for chipless RFID systems, their limitations and areas for

improvements are listed. Last part of the chapter presents the state of the

art MIMO based chipped RFID systems followed by the major challenges for

MIMO based chipless RFID systems.

Chapter 3: Chipless RFID Tag Design

This chapter presents the designing of two chipless RFID tag types for exper-

imental verification of the proposed detection techniques. The first type is a

circular resonator based SISO chipless RFID tag. The tag is printed on a paper

film using a printer with conductive ink. Its performance is verified using mea-

surement data. Then a novel MIMO based chipless RFID tag was designed in

CST the results are presented. The performance of the individual components

of the tag as well as the integrated MIMO tag are verified using measurement

data.

Chapter 4: ML Detection Techniques for SISO based Chipless RFID Sys-

tems

Four likelihood based detection techniques have been presented and their per-

formances have been verified using CST and MALAB simulation. A fifth tag

detection technique is developed for an existing chipless RFID reader and its

performances are verified using empirical measurements. The superior per-

10 Introduction

formances of the proposed tag detection techniques were compared with the

existing detection techniques. The disadvantages of these detection methods

are identified and solutions to them are presented in Chapter 5.

Chapter 5: Computationally Feasible Tag Detection Techniques

Two computationally feasible tag detection technique are introduced in Chap-

ter 5. The first detection technique is a suboptimal bit by bit detection (serial

bit reading) in contrast to detecting the all the tag bits once (parallel bit read-

ing). The next detection technique is based on trellis tree and Viterbi decoding.

This detection method can be incorporated with the proposed tag detection

techniques in both Chapter 4 and 6.

Chapter 6: Signal Processing for MIMO based Chipless RFID System

A MIMO based chipless RFID system is proposed and a MIMO decomposing

technique was used for separating the tag responses in each branch. Next, an

ML based tag detection technique was introduced to detect the tag bits en-

coded in each branch. Its performances were evaluated using MATLAB simu-

lations and then further verified using measurement data.

Chapter 7: Conclusion

Chapter 7 reiterate the research objectives and its successful achievements.

Also future directions of the research and possible applications are presented.

Chapter 2

RFID Systems

2.1 Introduction to RFID systems

There are two main types of RFID systems: chipped RFID systems and chipless

RFID systems. The RFID tag used in a chipped RFID system has a microchip so

that the tag is to a certain extent, capable of performing signal processing. The

microchip can be powered using a dedicated battery or by using the power

available from the interrogating signal. However, due to the presence of a

microchip in tag, the implementation cost in mass production industries is as

high as a few tens of cents per tag [55]. To obtain a cheaper technology, re-

searchers have focused on developing cheap chipless RFID systems to satisfy

similar requirements in the absence of a microchip in the tag. In addition, these

tags does not need an energy source and can be printable. The downsides of

this type is that the tag does not have signal processing capacity. Moreover, the

data bit capacity is lower than the chipped type. A summary of a comparison

between these two RFID systems is shown in 2.1. There are different types of

chipless tags available in the reported literature and most of them have a mass

production implementation cost at a fraction of a cent per tag [3,56,57]. A crit-

ical literature review was performed mainly on two areas; chipless RFID tags

and MIMO based chipless RFID systems.

11

12 RFID Systems

RFID System

Chipped RFID Chipless RFID

· Contains a microchip· Have signal processing

capacity· Battery operated /

power excavenging from interrogation

· Large data bit capacity· Costly

· No microchip· No signal processing

capcity· No battery / power

excavenging· Less data capacity· Very cheap

Figure 2.1: Comparison of different RFID systems

2.2 Chipless RFID tag types

A general classification of chipless RFID tags is shown in Figure 2.2. The two

main types of chipless RFID tags are time domain based tags and frequency

domain based tags and they will be discussed in detail. A few hybrid domain

chipless tags are reported. However, they involve complex design and signal

processing to decode data. Therefore they are beyond the scope of the thesis,

hence is not considered in the review.

2.2.1 Time domain based tags

In time domain based tags, a sharp pulse is transmitted from the interrogator

towards the tag and the tag produces multiple echoes of it in a pattern unique

to the tag. Effectively, the interrogating pulse results in a train of pulses re-

flected by the tag which can be used to encode data based on the number of

echoes and the round trip delay. Radio frequency surface acoustic wave (RF-

SAW) tag is the most popular time domain based tags available in the market

2.2 Chipless RFID tag types 13

Chipless RFID tags

Time domain based tags Frequency domain based tags

Surface Acoustic Wave (SAW) tags

Thin-film-transistor circuit (TFTC) tags

Delay line based tags

Chemical tags

Ink-tattoo tags

Planer circuit tags

Capacitively tuned dipole based tags

Space filling curve based tags

Multiresonator based tags

Figure 2.2: Classification of chipless RFID tags

14 RFID Systems

[58–60]. In commercially available SAW RFID systems, a chirped Gaussian EM

pulse is used to excite the SAW tags. The chirped Gaussian pulse is first con-

verted to a SAW using an inter-digital transducer (IDT) before transmitting at

the reader. This SAW propagates across the piezoelectric crystal which will

reflect the wave number of times to form a pulse train with different time de-

lays [61–63]. Then the train of pulses will be converted back to an EM wave

using the IDT, hence identify the encoded data of the tag. [38, 64]. Another

time domain chipless tag is thin-film-transistor circuit (TFTC) tag which are

compact in size and low in power consumption and can also be fabricated at

high speed on low-cost plastic film [39, 65]. Organic transistors are printed on

plastic films. Both RF-SAW and TFTC require special material and complex

fabrication process, hence cannot meet the sub-cent per tag cost. On the other

hand, fully printable delay line based chipless tags are implemented using the

discontinuous microstrip delay lines forming different sections as reported in

[40,66,67]. The tag is excited using a sharp EM pulse and once the tag receives

this pulse, it reflects echoes of the pulse at each discontinuous point along the

microstrip line. The length of the delay line between various discontinuities

determines the time delay between different echoes. Main advantage of this

tag is that it can be printed. However, the physical size of the tag gets bigger

as the number of data bits increases.

2.2.2 Frequency domain based tags

In frequency domain based tags, data is encoded in the frequency spectrum

of the interrogating signal using resonant structures. The presence and the

absence of a set of resonators at predefined frequencies are used to encode

data. It is reported that, frequency domain based tags are usually compact and

high in bit capacity [68]. On the other hand, the spectral efficiency is very less

2.2 Chipless RFID tag types 15

in frequency domain tags [3]. Two different categories of frequency domain

based tags can be identified based on the nature of the tag, namely, chemical

tags and planner circuit tags.

Chemical tags are constructed using a decomposition of nano-metric reso-

nant fibres or special ink. When an EM wave hits on the chemical the fibres or

the ink, it starts resonating at a predefined frequency, hence leaving a signa-

ture on the spectrum of the reflected EM wave. Several nano-metric fibre based

tags have been reported in [69–71]. Ink-tattoo chipless RFID tags are created

by forming a unique electronic ink pattern (tattoo) on the tag [72–74]. Once

the tag is excited with a microwave signal certain parts of the tattoo starts res-

onating at predefined frequencies and hence data can be encoded. However,

both nano-metric and ink-tattoo tags provide unpredictable frequency signa-

tures and hard to repeat the performance. To alleviate these problems, planar

passive microwave resonator on microwave laminates were designed.

Planar circuit tags are designed using the resonating structures fabricated

on a planar microstrip or co-planar waveguide. The first of this kind was de-

veloped by Jalaly [75] using capacitively tuned dipoles which are resonating at

different frequencies. Once the tag is interrogated with a frequency sweep, the

reader examined the frequency spectrum of the backscattered signal for mag-

nitude dips which correspond to the resonating dipoles. However, each of the

dipole length is half of the corresponding wavelength that increases the phys-

ical dimensions of the tag. McVay [76] extends the concept to introduce space-

filling curves whose footprint is much smaller than the wavelength. Multires-

onator based chipless RFID tags were proposed by Preradovic et.al. [77] and

they were reported to be amongst the chipless tags with highest bit capac-

ity [68]. The structure of the multiresonator based chipless tag is illustrated

in Figure 2.3. The tag has three main components namely, a vertically polar-

ized UWB disc-loaded monopole receiving antenna, a multiresonating circuit

16 RFID Systems

Rx Antenna

Tx AntennaMulti-resonating circuit

Figure 2.3: Structure of a multiresonator based chipless RFID tag

consists of cascaded spiral resonators and a horizontally polarized UWB disc-

loaded monopole transmitting antenna [78–80].

Tag has a dedicated transmitting and a receiving antenna which are cross

polarized to minimize the cross coupling. The reader transmits a UWB in-

terrogating signal in the same polarization as that of the receiving antenna of

the tag. The received UWB signal propagates through the cascaded spiral res-

onators, loosing energy at the resonating frequencies. Finally, the resultant sig-

nal is transmitted with the transmitting antenna of the tag towards the reader.

The orthogonal polarization of the receiving antenna of the reader and trans-

mitting antenna of the tag avoids cross talk and maximises the energy of the

retransmitted signal from the tag. The frequency spectrum of the received sig-

nal is examined to see the presence or the absence of the spiral resonators at

different frequencies to decode the tag data. After a close analysis of the re-

ported chipless RFID tags, it was clear that the multi-resonator based chipless

tag stands out from the rest due to following reasons. Firstly, it is reported

as one of the chipless tags with the highest data capacity if not the highest.

Apart from that, it is microwave based and printable on a planar circuit which

makes it a feasible candidate for low-cost chipless RFID applications. As a re-

2.3 MIMO based chipped RFID systems 17

sult, multi-resonator based chipless tags are selected to incorporate with the

proposed MIMO based chipless RFID system. There is no reported literature

available on MIMO based chipless RFID systems except the author’s reported

in [81]. Hence, a comprehensive review was performed on the MIMO based

chipped RFID systems in next section.

2.3 MIMO based chipped RFID systems

The rapid development of RFID devices and their wide use in mass-market

applications have prompted researchers to work mainly on improving RFID

System performances. The performance metrics include interrogation range,

bit-error-rate (BER), data rate, anti-collision and implementation cost. In wire-

less communication, most of these performance metrics have been improved

using Multiple Input Multiple Output (MIMO) antenna technology. As a re-

sult, during the last few years, there has been a significant research-focus on

applying MIMO technology to chipped RFID systems. The most important

MIMO based chipped RFID systems reported in literature are revisited here.

Chizhik [51] introduced the concept of pinhole/keyhole when describing

the capacities of multi-element transmit and receive antennas. The formulation

of a pinhole can be visualized in the following example. Picture, a two-element

transmitting array and a two-element receiving array that are separated by a

screen with a small keyhole/pinhole punched through it. The only way for

the radio wave to propagate is to pass through the pinhole. A hallway or a

tunnel is perhaps a more realistic environment where the pinhole concept can

be experienced. In [36], Joshua has shown that pinhole diversity is available

in a rich scattering environment caused by the modulating backscatter with

multiple RF-tag antennas.

In [37], a UHF chipped RFID system has been investigated with multi-

18 RFID Systems

ple readers where the channel from transmitter to the receiver of the reader,

via the RFID tag was assumed to be a pinhole channel. Both forward (from

reader to tag) and reverse (from tag to reader) links are assumed to be having

a Nakagami-m fading channels [82]. An Mt × 1×Mr pinhole channel has been

investigated further where, there are Mt transmitting reader antennas, one RF-

tag antenna and Mr receiving antennas at the reader. There are two system

configurations analyzed in this work, namely mono-static system with transmit

and receive antennas to be collocated at the reader and bi-static system with

reader transmit and receive antennas to be spaced far apart.

It was observed in [37] that the average reverse-link interrogation range

will be large when monostatic structure is used instead of bistatic structure.

However, in the rich scattering environments bistatic structure retains more

reliable compared to monostatic. In most cases, MIMO system has outper-

formed the Single Input Single Output (SISO) counterpart. For example, 3× 3

MIMO-RFID can achieve 60% extra gain in average reverse-link interrogation

range compared to that of SISO-RFID system. The reasons could be the ex-

ploitation of pinhole diversity and improved SNR received at the reader with

the employment of multiple antennas. In this work the main concern was to

improve the range rather than identifying multiple tags.

In [52] Robert developed a ultra-high frequency (UHF) frontend for MIMO

applications in RFID. Passive chipped RFID tag has been used for the experi-

ment with one transmitting antenna and two receiving antennas at the reader.

Using the measured data, it was claimed that transmit and receive beamform-

ing as well as tag localization can be performed using their frontend.

The gains available for chipped RFID tags using multiple antennas are dis-

cussed in [54]. In conventional MIMO, the environment should have rich scat-

tering to exploit spatial multiplexing. In line with that, one can conclude that

line of sight (LOS) is not in favor of spatial multiplexing. Even though, RFID

2.3 MIMO based chipped RFID systems 19

system channels exhibits LOS propagation, heavy small-scale fading will be

present due to indoor operation, a cluttered reader environment and inho-

mogeneous nature of the tagged objects [54]. They have studied about an

M × L × N dyadic backscatter channel which is a pinhole channel that de-

scribes the backscatter propagation radio channel with M transmitter anten-

nas, L RFID tag antennas and N receiver antennas. This channel was investi-

gated first by Ingram et al. [53] with the use of multiple antennas to exploit

transmit-diversity and spatial multiplexing to increase the range and commu-

nication capacity.

In [54], it was shown that the dyadic backscatter channel has deeper fades

than that of the one-way Rayleigh channel but, improves as more RFID tag

antennas are added. However, pinhole diversity has two advantages over

both conventional coherent diversity combining and non-coherent diversity

combining. Firstly, it changes the channel distribution to have comparatively

less fading. Moreover, diversity gains can be realized in the dyadic backscat-

ter channel using only multiple RFID tag antennas to modulate backscatter.

Hence, no diversity combining is required at the reader, making no changes in

the reader receiver hardware, reader transmitter hardware or signaling scheme.

In fact the actual communication gain in the above dyadic backscatter channel

is due to both pinhole diversity gain and increased scattering aperture. In ad-

dition, if conventional diversity combining techniques such as maximum ratio

combining (MRC) are employed at the reader, even greater gain is achievable.

However, it is important to notice that, antenna correlation at the tag has to be

at a minimum as possible to explore maximum pinhole diversity.

After reviewing the above MIMO based chipped RFID systems, following

system characteristic were identified. They can be used as design guidelines

when developing a MIMO based chipless RFID system.

20 RFID Systems

• Multiple RF Tag Antennas - each antenna of the tag can be used to mod-

ulate the interrogating signal

• Bi-static reader architecture – therefore separate antennas for transmit-

ter and receiver

• Antenna arrays instead of a signal antenna – multiple diversity branches

are available

• Operating Frequency - should be high enough so that a feasible uncor-

related antenna element spacing at the tag can be achievable

• Antennas Configuration - cross-polarized antennas may be used at the

reader to reduce self-interference and at the tag to reduce envelop cor-

relation between the signals scattered from each tag antenna. However,

using cross polarized antennas can lead to detrimental effects of unequal

diversity branch power.

It is evident that, MIMO based chipped RFID systems have a number of

advantages compared to SISO counterpart. Although, they achieve diversity

gains from employing multiple antennas, they utilize a microchip in the tag to

encode data. The microchip incurs high implementation costs as well as the

need to have an energy source to power the microchip. Therefore, they do not

provide a feasible alternative for optical barcoding that is currently utilized in

most mass production item tagging. Chipless RFID systems on the other hand,

do not contain a microchip hence less implementation cost. This is a significant

advantage in mass production item tagging. However, there is no reported lit-

erature on MIMO based chipless RFID system to date except the authors work

in [81]. Thus, there is a need to develop MIMO based chipless RFID systems

that has the potential to replace optical barcode systems used in mass produc-

tion item tagging.

2.4 Review of chipless RFID tag detection techniques 21

2.4 Review of chipless RFID tag detection techniques

A number of tag detection techniques for chipless RFID systems have been

reported. This section summarizes some of the available techniques and com-

pares them in terms of detection accuracy and the computation complexity.

The multi-resonator based chipless RFID system presented in [43] uses a

threshold based tag detection technique where magnitude of the tag response

at resonating frequencies are compared against a threshold level to identify the

tag bits. Similar techniques is applied in the chipless RFID systems proposed

in [41, 44, 83–85]. There have been reported tag detection techniques such as

[86] and [46], based on purely phase information of the received signal. It was

mentioned that information embedded in phase allows to reduce the transmit

power compared to magnitude based detection techniques.

It was reported in [42] that moving average filtering technique was able to

improve the tag detection performance. The chipless RFID systems proposed

in [87] and [50] utilize both magnitude and phase information for decision

making. Utilizing information embedded on both magnitude and phase al-

lows the tag reading to be highly reliable. However, the complete tag detection

algorithm was not reported and amalgamating information available in both

the magnitude and phase is not known. In addition, a detailed analysis on the

performance of the tag detection technique needs to be performed.

The tag detection techniques discussed so far are using simple signal pro-

cessing techniques such as threshold based detection [43] combined with mov-

ing averages [42]. The tag detection proposed in [49] presents an advanced tag

detection technique using a continuous wavelet transform. It has managed to

overcome the difficulty to detect the signal scattered at the tags with the delay

information in the presence of noise. The wavelet transform effectively acts as

a matched filter and one benefit of this technique is that a reference tag is not

required in advance.

22 RFID Systems

There have been other advanced tag detection techniques such as signal

space representation (SSR) reported in [47, 88]. The basic principle is to map

the received signal vector into a point in an N dimensional space. All pos-

sible tag responses are mapped into fixed points in the N dimensional space

and the minimum distance between the received signal point and other fixed

points is calculated to identify the tag bit combination. This is a very accurate

technique and has shown improved performance in successful tag detection.

However, one of the challenges is its exponential computation complexity as

the number of data bits increases. For example, 20 bit tags have about 1 mil-

lion unique combinations and the one million distances need to be calculated

before making a decision.

There have been other techniques reported in [45, 48, 89–99] using tech-

niques similar to mentioned above. The main limitation of the existing tech-

niques is its low tag reading reliability mostly because of the primitive tag de-

tection techniques used. The advanced tag detection techniques have been able

to achieve robust tag reading however suffer from high computation complex-

ity. Therefore, the motivation is to develop advanced tag detection techniques

that can achieve reliable tag reading while having relatively low computation

complexity. A maximum likelihood (ML) detection technique with less com-

putation complexity is proposed in the thesis. The following section reviews

ML techniques for chipless RFID proposed in the thesis.

2.5 Maximum likelihood detection techniques

ML detection is a signal processing technique used in communications sys-

tems to make decisions by observing a received signal, and comparing it with

all the possible combinations. An optimal ML detection technique utilizes all

aspects of the received signal before making a decision. A suboptimal like-

2.5 Maximum likelihood detection techniques 23

lihood detector would use only the main aspects of the received signal such

as the magnitude while discarding the phase. Generally, ML based detection

techniques have a very high accuracy. Therefore, ML detection techniques are

widely used in communication systems. However, the main drawback is high

computation complexity as a result they are not scalable.

There are a few differences on signals available in communication systems

compared to that in a chipless RFID environment. A comparison of the en-

vironment available in a communication system and a typical chipless RFID

system at 2.4 GHz is shown in 2.1.

Table 2.1: Comparison of communication system with a chipless RFID system

Metric Communication system Chipless RFID system

Transmitter & Receiver Separated Co-locatedInterference Inter symbol interference (ISI) Inter resonator interference (IRI)Bandwidth Narrowband (<5 MHz) Broadband ( > 300 MHz)Channel Multipath propagation Strong light of sightPropagation distance Up to many kms < 1 mSpectral efficiency Several bps per Hz ≈ 1 bit per 100 MHz (@ 2.4 GHz)

Signals considered in communication systems are generated based on the

modulated bits. Since bits in communication systems are random, the gener-

ated signals are also random in nature. On the other hand, tag signals have

a limited number of data bits and as a result they can be interrogated multi-

ple times for a higher reading reliability without compromising the reading

time. It is similar to a signal obtained by repeating the same bit sequence in a

communication system.

Even though the transmitter and receiver are physically separated in com-

munication systems, they are co-located in chipless RFID systems. As a re-

sult, the receiver has access to the interrogating signal hence a perfect synchro-

nization can be achievable. In communication systems, modulated signals can

have inter symbol interference (ISI) as a result of channel delay spread. How-

ever, in a frequency domain chipless RFID systems, a bit is modulated using a

24 RFID Systems

resonator and the resonator response can interfere with neighboring resonator

responses and causes inter resonator interference (IRI).

Bandwidth in chipless RFID systems are quite high (> 300 MHz) compared

to that of communications systems. As a result noise level can be expected to

be quite high. However, the chipless RFID systems operates in short-ranges (<

1 m) as a result has a strong line of sight causes the received signal power to be

relatively higher. Another important observations is the extremely low spec-

tral efficiency of chipless RFID systems as shown in Table 2.1. This is mainly

due to the fact that chipless tags has no computation resources and these pas-

sive tag designing is deliberately made to be simple. On the other hand the

number of data bits required to transmit is extremely low compared to that of

a communication system.

The understanding of ML detection techniques and the differences in two

environments was utilized in proposing likelihood based detection techniques

presented in Chapter 4.

2.6 Conclusion

A literature survey was carried out in three main categories. The chapter first

summarized the available chipless RFID tag types and multi-resonator based

chipless tags were identified as a potential tag type for further investigation.

A number of tag designs are presented in Chapter 3. Then, available chipped

MIMO systems were studied and the key take aways were discussed. It was

highlighted the importance of having multiple antennas both at the tag and

the reader. Orthogonal antenna configuration is identified as an important

aspect when designing a chipless RFID system having multiple antennas. In

addition, the operating frequency should be selected such that the uncorre-

lated antenna element spacing is achievable. These findings will be used in

2.6 Conclusion 25

proposing the novel MIMO based chipless RFID system. Finally, the available

tag detection techniques for chipless RFID tag reading was presented and their

advantages and limitations were identified. Maximum likelihood (ML) based

detection techniques used in communication systems were discussed and dif-

ference aspects are compared under a chipless RFID system environment. It

was identified that, IRI caused when the resonating frequencies are close to

each other is similar to ISI in communication systems. Therefore, the tech-

niques used in communication system to mitigate ISI can help in designing

tags which increases data bit capacity in chipless tags. Therefore, the moti-

vation is to remove the guard-band presented between resonance frequencies

and mitigate the interference using signal processing techniques at the reader.

In the next chapter, various chipless RFID tag design and reading methods

will be presented. Then the subsequent chapters will present the proposed

likelihood based detection techniques of these chipless RFID tags and their tag

reading accuracy improvement. In the very last part of the thesis, MIMO sig-

nal processing for chipless RFID will be presented followed by the conclusion

of the thesis in Chapter 8.

Chapter 3

Chipless RFID Tag Design

3.1 Introduction

In this chapter the physical layer developments of the two chipless RFID tag

types, experimental setup and results are presented in microwave and mm

wave domains. The first tag type is a SISO tag having circular resonators that

operates between 21 - 27 GHz. The second type is the novel MIMO tag having

multi resonators that operates around 2.4 GHz. A brief summary of the two tag

types are given in Figure 3.1. The rest of the chapter describes the design of the

tags and experimental verification of the proposed tag detection techniques in

Chapter 4.

3.2 SISO tag design

In this section a circular patch resonator based backscattering chipless RFID

tag design will be discussed. The tags are operating between 21 - 27 GHz fre-

quency range. The tags are fabricated on a thin film paper using the SATO

printer having a conductive ink. Then the tags were read using an existing

chipless RFID reader developed by Kalansuriya at Monash Microwave, An-

tenna, RFID and Sensor (MMARS) laboratory, Monash University.

27

28 Chipless RFID Tag Design

Type : Backscattering tag Frequency : 21-27 GHz Element Type: Circular Resonator No of bits : 4 bits

SISO Tag

Type : Retransmission tag Frequency : 2.2-2.6 GHz Element Type: Spiral resonators with orthogonally polarized disc located monopole No of bits : 3*2 bits

Tag Design

MIMO Tag

Figure 3.1: Tag types used in SISO and MIMO detection algorithm develop-ments

3.2.1 Tag design and fabrication

Figure 3.2 shows the photograph of a 4-bit tag designed at 21-27 GHz ultra

wide band (UWB).The tag is comprised of 4 sets of distinct resonators thought-

fully located so that nearby resonances are not interfered. The tag is designed

to occupy 4 bits in the 21-27 frequency band. The resonance frequencies are

selected as 22.5, 23.5, 24.5 and 25.5 GHz as simulations shows that the 3-dB

bandwidth of each resonator response at this frequency band is about 1 GHz.

The diameter of the circle is inversely proportional to the resonance frequency.

The diameters for the above resonance frequencies were found and displayed

in Table 3.1

Table 3.1: Simulation Parameters

Resonance frequency (GHz) Diameter of the circular patch (mm)

22.5 4.3823.5 4.5924.5 4.8425.5 5.13

The specific frequency resonators are repeated to increase the backscattered

3.2 SISO tag design 29

Figure 3.2: Printed Tag

signals of the individual resonators. The tag is printed in-home using MMARS

laboratory’s SATO printer.This monochrome printer has a resolution of 600

dpi. The substrate is 0.09 mm thick glossy paper with bulk silver coating of

10 µm on top. The design is thermally transferred to the paper substrate and

created the fully printable silver tag on the paper substrate.

3.2.2 Experimental setup

In this section, the experimental setup used to verify the performance of the

tags designed. A chipless RFID reader that reads the magnitude of the tag re-

sponse was used to read the tag designed in previous section. An experiment

was set up as shown in Figure 3.3. The reader transmits a narrow banded si-

nusoid signal using a transmit horn antenna and as shown in the figure the

receiver receives the tag response for that frequency using a receive horn an-

tenna. The magnitude of the received signal is recorded along with the fre-

quency of the signal sent. Similarly, the frequency is swept across the band 21

- 27 GHz and the complete tag response was recorded.


Tag Reader RF Electronics

Horn Antenna Under

Perspex Table

LCD Display Perspex Table

Digital Control & Power Supply Unit

Figure 3.3: Experimental setup

Figure 3.4 shows the tag response ([1111]) recorded by the chipless RFID

reader. It can be seen that the four resonances occur at the designed resonat-

ing frequencies. The presence (ON) and absence (OFF) of resonators can be

selected to form the 16 different tag types for a 4-bit tag as shown in Table 3.2

The circular patches having the same diameters increases the effective radar

cross section hence forms a bigger frequency dip at the resonance frequency.

The 4 resonator sets are selected such that tags with all 16 combinations can be

designed. These 16 tag types were fabricated and used for testing the detection

3.3 MIMO tag design 31

techniques derived in Chapter 4.

Table 3.2: Tag types given by resonator combinations

Resonator CombinationsTag Type f1 f2 f3 f4

[0000] OFF OFF OFF OFF

[0001] OFF OFF OFF ON

[0010] OFF OFF ON OFF

[0011] OFF OFF ON ON

[0100] OFF ON OFF OFF

[0101] OFF ON OFF ON

[0110] OFF ON ON OFF

[0111] OFF ON ON ON

[1000] ON OFF OFF OFF

[1001] ON OFF OFF ON

[1010] ON OFF ON OFF

[1011] ON OFF ON ON

[1100] ON ON OFF OFF

[1101] ON ON OFF ON

[1110] ON ON ON OFF

[1111] ON ON ON ON

3.3 MIMO tag design

One of the original contribution of the proposed MIMO based chipless RFID

system is the MIMO tag design. The main parts of the proposed chipless

MIMO tag are identified as the power divider, the monopole tag antennas

and the multi spiral resonators. These individual components were designed

in computer simulation tool (CST) and their performance were verified. Fol-


0

0.2

0.4

0.6

0.8

1

21 22 23 24 25 26 27

No

rmal

ized

po

we

r m

agin

itu

de

Frequency ( /GHz)

f1 f2f3 f4

Figure 3.4: Magnitude of the tag response for tag [1111]

lowings are the detailed descriptions of the component level design and their

integration to a complete MIMO tag.

3.3.1 Power divider design

The power divider needs to divide the power equally into two branches over a

broadband of 400 MHz centred at 2.4 GHz. Therefore a symmetrical T-junction

power divider was designed. The design parameters were the length and the

width of each transmission line presented in the divider. The designed and

fabricated T-junction power divider is shown in Figure 3.5. The power divider

is designed on TLX8 substrate with relative permittivity (εr) of 2.4, loss tan-

gent (tanδ) of 0.004 and thickness (h) of 0.5 mm. Both the simulated and the

experimental S parameter magnitudes vs frequency for the power divider are

shown in Figure 3.6. It is clear that, both simulated and measured S21 and S31

are similar and close to the required value of -3dB (half power division). The

measured and simulated S11 in dB (return loss) vs frequency is also shown

in the figure. There is a difference in the two curves and this is possibly due

to fabrication defects, where it shifts the operating frequency little higher. As


Figure 3.5: T-junction power divider (left: CST design, right: fabricated powerdivider) TLX8 substrate with εr=2.4, tanδ=0.004 and h=0.5 mm

far as the performance is concerned, the fabricated power divider still has a

bandwidth of over 500 MHz at -15dB centered around 2.4 GHz. So the power

divider performance is acceptable and any further tuning was not attempted.

Once the satisfactory performance of the T-junction power divider is obtained,

the next component that was designed was the monopole antenna.

3.3.2 Monopole antenna design

A monopole antenna was designed to operate at 2.4 GHz with a bandwidth of

400 MHz. The monopole antenna is selected due to its figure of eight bream

radiation pattern and distinct polarisation so that vertical and horizontal an-

tennas can be used to reduce the cross-talk between the transmit and receive

chains off the MIMO chipless RFID tag. A microstrip based antenna with a

circular patch was used for size optimisation. Figure 3.7 shows the monopole

antenna design in CST as well as the fabricated antenna. The design parame-

ters are the radius of the circular patch, the gap between the ground plane and


2 2.5 3−35

−30

−25

−20

−15

−10

−5

0

Frequency /GHz

S P

aram

eter

Mag

nitu

des

/dB

Simulated S11Simulated S21Simulated S31Measured S11Measured S21Measured S31

Figure 3.6: S-parameters of the power divider

the disc edge and the dimension of the feed line.

The performance of the monopole antennas is presented next. Figure 3.8

shows the return loss vs frequency of the antenna obtained from both simu-

lated and measured data. It can be seen that, they both agree well and the


Figure 3.7: Monopole antenna (left: CST design, right: fabricated monopole)

Figure 3.8: Return loss of the monopole antenna

antenna has an operating bandwidth of well over 400 MHz at -10dB.

Figure 3.9 shows the antenna radiation pattern at 2.4 GHz which is a figure-

of-eight omni-directional radiation pattern with the maximum gain of the main


Farfield Realized Gain Abs (Phi=90)

Theta / Degree vs. dB

Figure 3.9: Simulated radiation pattern of the monopole antenna

lobe with 2.7 dB, which is acceptable.

The realized gain of the antenna on the main lobe was analyzed over a

bandwidth of 400 MHz centered around 2.4 GHz and the results are shown in

Figure 3.10. It is clear that, the realized gain was above 2.5 dB over the fre-

quency band. These results verify the successful operation of the monopole

antenna at a bandwidth of 400 MHz centered around 2.4 GHz. After obtain-

ing satisfactory performance from the designed antenna, the spiral resonators

were designed.


Figure 3.10: Realized gain of the monopole antenna

Figure 3.11: Spiral resonators (top: CST design, bottom: fabrication)

3.3.3 Spiral resonator design

A set of spiral resonators were designed to operate at the resonance frequencies

given by Table 5.1. Design parameters are the transmission line lengths, widths

and the gap between the micro-strip lines. Figure 3.11 shows both the CST

designs on spiral resonators and one of the fabricated resonators.

A spiral resonator response obtained from CST simulations is shown in Fig-


Figure 3.12: CST generated resonator response

Figure 3.13: Fabricated MIMO tag

ure 3.12. It can clearly see that both the amplitude and the phase contain the

data and existing techniques mostly rely on only one of them. However the

proposed detection techniques use the information available in both ampli-

tude and the phase. A photograph of the fabricated MIMO tag with all ’0’ bits

(no multi-resonator) in branch connected to Tx1 and all ’1’ bits in the branch

connected Tx2 is shown in Figure 3.13.


Figure 3.14: MIMO tag experiment

3.3.4 Experimental setup

The schematic of the experiment conducted to measure the MIMO tag re-

sponse using an arbitrary waveform generator (AWG) and an oscilloscope

with a high sampling rate is shown in Figure 3.14. However, antennas were

replaced using cables as the sole purpose of this work is to verify the validity

of the ML based detection method.

Figure 3.15 shows the CST generated tag response and the measured tag re-

sponse for tag bits [1010]. S21 measurements across the resonator were recorded

using a performance network analyzer and converted from log scale to linear

scale for comparison with the CST generated response. It can be seen that they

are closely matched.

So far the design of a MIMO based chipless RFID tag was discussed. Its in-

dividual component performances were analyzed using the CST simulated re-

sults and experimental results. It is clear in Figure 3.15 there are some anoma-

lies between the theoretical and experimental. This may be because of the

coupling between the two branches in the MIMO tag. However, the proposed


0

0.2

0.4

0.6

0.8

1

2.1 2.2 2.3 2.4 2.5 2.6 2.7

No

rmal

ize

d A

mp

litu

de

Frequency ( GHz)

CST generated tagresponse

Experimental tagresponse

Figure 3.15: Tag response for [1010]

MIMO based chipless RFID system has an inbuilt mechanism to overcome this

challenge. Any coupling between the antenna can be incorporated to the 2x2

MIMO channel matrix. This can be performed using a calibration tag. Now

the system is free from any coupling between the two branches and with the

known channel, it is possible to perform tag detection.

3.4 Conclusion

After analyzing the simulations, it is noteworthy to pinpoint that, even though

there are only two transmitting branches presented in the RFID tag considered,

it is theoretically possible to add more branches and still recover the transmit-

ted signals given that, the number of receiving antennas in the reader is larger

or equal to the number of transmitting branches in the tag. Hence, without

increasing the bandwidth, the bit capacity can be further increased using the

same frequency resonators compared with having only one branch at the tag.

However, it is required to evaluate the effect of mutual coupling between an-

tennas with higher number of transmitting branches in the tag.

In the RFID tag proposed, there is only one receiving antenna through

3.4 Conclusion 41

which, the received signal will be divided into two equal components. The

proposed concept can be extended to having a dedicated receiving antenna

for each component, hence increasing the effective signal-to-noise ratio (SNR)

at each branch. Therefore, with multiple dedicated transmitting and receiv-

ing antennas on the tag can further improve the performances. In addition,

the concept can be further extended to multiple tag detection if each branch is

considered as a separate tag.

Furthermore, the use of IQ modulation/demodulation allows an extra de-

gree of freedom to increase the bit capacity. Since the baseband signal consid-

ered is complex it is possible to have asymmetric frequency response in pos-

itive and negative frequencies. Therefore, the eligible frequency band in the

passband centered around the RF carrier doubles, allowing more resonators to

be placed in the tag, without increasing the sampling rate of the ADC at the

receiving end of the reader. After analyzing the above results, it can be con-

cluded that, MIMO is a competitive candidate for improving reliability or the

bit capacity of a resonator based chipless RFID system.

Chapter 4

ML Detection Techniques for SISOChipless RFID Tags

4.1 Introduction

Research on chipless RFID systems were mainly emphasizing on improving

the RFID reader architecture and the chipless tag design. As a result, they

were using primitive signal processing techniques at the RFID reader for tag

detection. The hypothesis focus is on improving the signal processing tech-

niques so that the success rate in tag detection and tag reading range can be

further improved using the same reader architecture and tag design. There-

fore, the proposed tag detection techniques are compatible with the existing

RFID systems.

The existing chipless signal processing techniques for tag detection was as

primitive as threshold based detection. Maximum Likelihood (ML) based de-

tection techniques have shown improved performances in communication sys-

tems over primitive techniques such as threshold based detection techniques.

The motivation for this work is to apply the ML detection techniques for chip-

less RFID tag detection so that the existing RFID system would produce better

results in terms of the detection error rate and the reading range.

The rest of the chapter is organized as follows. First the theory behind de-

riving four ML expressions for a SISO based chipless RFID system is presented.

43

44 ML Detection Techniques for SISO Chipless RFID Tags

Chipless RFID System Models

Frequency Domain Models

Time Domain Models

Figure 4.1: RFID System Models

The different expressions are derived based on the availability of the channel

information (known or unknown channel) and real or complex signal process-

ing. Then a computationally feasible tag detection technique is presented so

that the detection technique can be implemented on a portable RFID reader.

Next, the detection techniques were implemented in MATLAB and its results

are compared. Finally the original contribution and a discussion on each de-

tection technique is presented in the conclusion section.

The system models presented in the chapter are for frequency domain based

chipless RFID tags. However, the models are based on either signals using time

domain samples or frequency domain samples. Hence the system models pre-

sented can be categorised into two sections as shown in Figure 4.1.

4.2 System models - Time domain

A multi-resonator based chipless RFID system consists of three main compo-

nents, namely a reader, a tag and the middleware as shown in Figure 4.2. The

RFID reader generates an interrogating signal and transmits towards the tag

using a transmitting antenna (Txr). The interrogating signal will be received

by the tag using its dedicated receiving antenna (Rxt), which has the same

polarization as the transmitting antenna of the reader. Then the received sig-

4.2 System models - Time domain 45

Forward Channel

RFID Tag Reverse Channel

Txr

Rxr RFID

Reader

Txt

Rxt

Figure 4.2: Overview of Chipless RFID System

nal propagates via a frequency modulation circuit that comprises a cascade of

spiral resonators. Depending on the resonator combinations (presence and ab-

sence of resonators), a unique tag response is available at the end of the micro-

strip line. This process is called as tag modulation from here onwards. Then

the tag response will be transmitted using the dedicated transmitting antenna

of the tag (Txt). The polarization of the transmitting antennas is orthogonal

to that of the receiving antenna of the tag. The transmitted tag response is

received at the reader using an antennas having matching polarity. This care-

ful selection of the antenna configuration limits any unwanted cross coupling

between antennas.

Next the system is modeled using a number of system models in the next

sections.

4.2.1 System model I - Real signals

The system described in Figure 4.2 is modeled firstly using the following sim-

ple signal model. Later, more assumptions are relieved and a comprehensive

analysis is performed so that the new models closely describes the real system.


The signals considered in System Model I are assumed to be real signals,

meaning the received signal is directly sampled at a very high rate. Firstly, the

interrogating signal transmitted from the RFID reader reaches the tag through

the forward channel as shown in Figure 4.2. Then the signal received by the tag

is modulated by the resonator combination presents in the tag. Sm is defined

as this resultant signal available for transmitting back towards the reader. Sm

is a unique tag response for a given tag m and it is a vector having a length

of K. Sm is then transmitted from the tag to the RFID reader via the reverse

channel as shown in Figure 4.2.

Both the forward and reverse channels are in short range with a strong line

of sight. The channels are assumed to be real and known constants. Mixing

with the forward channel, tag modulation and mixing with the reverse channel

happen in a cascaded manner. As a result, the product of both channels can

be represented using a real constant h. The received signal at the reader is

added with noise (ω) produced by the receiver circuit at the RFID reader. The

resultant signal is called as y and can be represented using (4.1)

y = hSm + ω (4.1)

When the RFID reader transmits the interrogating signal it is first received

by the receiving antenna of the tag. Then the signal is modulated by the tag

and transmitted back towards the RFID reader. Sm includes both the resonator

response as well as the noise introduced by the tag antennas. Therefore, Sm

is actually dependent on the tag combination. However, the amplitude of the

noise added by the receiving antenna of the RFID reader is much higher than

the noise added by the tag antennas. This happens due to the close proximity

of the transmitter and receiver electronics of the reader to the reader antenna.

Also the reader antenna is expected to have higher gain compared to the tag

antennas. Therefore it may also pick more surrounding noises. Hence, noise


presented in Sm can be neglected and Sm can be treated as the pure tag re-

sponse. The noise presented at the received signal y is only due to the noise

added at the reader. Therefore, noise ω added at the reader can be assumed to

be independent of the tag response Sm. In addition, individual time samples of

ω vector is assumed to follow an independent and identical Gaussian distribu-

tion (i.i.d.) with zero mean and a variance of σ2ω. As a result, the distributions

of ω and y can be derived as follow in (4.2).

ω ∼ N (0, σ2ω IK)

y ∼ N (hSm, σ2ω IK)

(4.2)

where IK is the identity matrix with K× K dimensions.

Since y and Sm are independent from each other and Sm follows an i.i.d., the

probability of receiving y given that Sm has been transmitted can be calculated

as follows.

Pr(y|Sm) =K

∏i=1

Pr(yi|Sm,i) (4.3)

Pr(yi|Sm,i) in (4.3) is the conditional probability of receiving the ith time sample

of y, given ith time sample of tag response Sm. Pr(yi|Sm,i) can be calculated us-

ing the well known probability density function (pdf) of Gaussian distribution

as shown in (4.4).

Pr(yi|Sm,i) =1√

2πσ2ω

exp(− 1

2(yi − hSm,i)

1σ2

ω(yi − hSm,i)

)(4.4)

Using (4.3) and (4.4), Pr(y|Sm) can be calculated as follows.

Pr(y|Sm) =1√

2πσ2ω

exp( K

∑i=1

(− 1

2(yi − hSm,i)

1σ2

ω(yi − hSm,i)

))(4.5)


(4.5) can be represented in vectorized form as below. [.]T is the hermitian trans-

pose.

Pr(y|Sm) =1√

2πσ2ω

exp(− 1

2(y− hSm)

1σ2

ω(y− hSm)

T)

(4.6)

The received signal, y and all the tag combinations, Sm are used to evaluate

probabilities given by (4.6). The tag combination Sm producing highest prob-

ability is selected to be the detected tag, m. Therefore, Pr(y|Sm) is maximised

over all possible Sm combinations for tag detection as follows.

maxSm

Pr(y|Sm) (4.7)

In (4.6), only the exp(.) component is varying with Sm. Hence, the detector

proposed for this model simplifies to (4.8).

maxSm

Pr(y|Sm) = minSm

((y− hSm) (y− hSm)

T)

(4.8)

The optimization given by (4.8) performs the tag detection. Under these

assumptions, the detector for the proposed signal model is the same as the

minimum distance detector. Tag detector used in this section assumes the per-

fect channel knowledge and both the channel and received signals are consid-

ered to be real, even though in reality a typical RFID reader may perform I/Q

demodulation hence dealing with complex signals. We relieve that assump-

tion on next sections by allowing signals to be having both real and imaginary

components.

The proposed signal models under different scenarios is listed in Figure 4.3.

Signal model II still assumes perfect channel knowledge, however it utilises the

information available in both the amplitude and phase for decision making.

Signal model III needs to know only the statistical properties of the channel


Signal Models

Model IReal signalsKnown channel

Model IIComplex signalsKnown channel

Model IIIComplex signals

Known ch. distribution

Model IVComplex signals

Unknown channel

Model VPower magnitudes

Known channel

Figure 4.3: Proposed Signal Models

while the actual channel realization is not required for decision making. In

signal model IV, channel is assumed to be unknown and a joint optimization on

both the channel and tag type detection is performed. Signal V is derived for

an existing chipless RFID system which utilizes only the power magnitudes of

the backscattered tag response.

4.2.2 System model II - Complex signals

The signal model considered in this section is very similar to the System Model I

discussed in the previous section. The same assumptions in System Model I ap-

plies however, both the channel and all signals are treated as complex signals.

Therefore, I/Q demodulation is implemented at the RFID reader which is the

case for some of the existing RFID readers. As a result, the readers no longer

needs very high sampling rates and only samples the baseband signals for I &

Q. For such a reader, the received signal is called as y and can be represented

similar to (4.1).

The product of the forward and reverse channels, h is assumed to be known

and the signals considered in this model are all complex numbers. They can

be represented using real and imaginary quantities for computation simplicity


as follows.

h = hr + jhi

ω = ωr + jωi

Therefore, the received signal can be written as,

y = yr + jyi

yr = hrSm,r − hiSm,i + ωr

yi = hrSm,i + hiSm,r + ωi

(4.9)

The noise ω added at the reader can be assumed to be the independent

of the filter response Sm. In addition, real and imaginary components of in-

dividual time samples of ω vector is assumed to follow an independent and

identical Gaussian distribution (i.i.d.) with zero mean and a variance of σ2ω.

ωr ∼ N (0, σ2ω IK)

ωi ∼ N (0, σ2ω IK)

(4.10)

A new vector, y0 having only real values is created using yr and yi as fol-

lows.

y0 =(yr , yi

)(4.11)

Mean and covariance of y0 can be calculated as follows.

E[y0] = µ =[hrSm,r − hiSm,i , hrSm,i + hiSm,r

]Cov[y0] = E[(y0 − µ)T(y0 − µ)]

= σ2ω I2K

(4.12)

I2K in (4.12) is the identity matrix with a dimension of 2K × 2K. Using the

statistical properties calculated in (4.12) the distribution of the real vector, y0


can be represented as follows.

y0 ∼ K (µ, σ2ω I2K)

Similar to previous sections, the probability of receiving y0 given that Sm

has been transmitted can be calculated as follows.

Pr(y0|Sm) =1

2π√|Cov[y0]|

exp(− 1

2(y0 − µ) Cov(y0)

−1 (y0 − µ)T)

=1

2πσωexp

(− 1

2σ2ω(y0 − µ) (y0 − µ)T

)(4.13)

Similar to previous detectors, (4.13) is evaluated for all the possible tag

combinations and the one with the highest probability is taken as the detec-

tor output. However, it can be seen that the detector can be further simplified

by minimizing the exp(.) component. Therefore, the objective function of the

detector can be represented as follows.

maxSm

Pr(y0|Sm) = minSm

((y0 − µ) (y0 − µ)T

)(4.14)

Under the assumptions followed for the proposed signal model, the opti-

mum detector is the same as the minimum distance detector. However, y0 and

µ can be calculated using (4.11) and (4.12) respectively. Tag detector used in

this section assumes the perfect channel knowledge and both the channel and

received signals are considered to be complex, which means in reality the RFID

reader performs I/Q modulation/demodulation. However, the expression in

(4.14) needs only real number calculations, hence lowers the computation com-

plexity. In the next section, we assume perfect channel knowledge is no longer

available. However, statistical properties of the channel is assumed to be avail-

able while I/Q modulation/demodulation is assumed to be performed at the


RFID reader.

4.2.3 System model III - Channel with a known distribution

The signal model discussed here assumed that the channel is no longer known.

However, the statistical properties of the product of forward and reverse chan-

nel is approximated by a Gaussian distribution with a known mean and a

variance. Like in previous models, the channel is assumed to be a constant

throughout each tag reading. In addition, both the channel and the signals

considered in this model are assumed to be complex, meaning I/Q modula-

tion is performed at the RFID reader. Then the received signal can be modeled

similar to (4.1).

For computation simplicity, the complex channel (h) and the noise (ω) can

be represented using two real components as follows.

h = hr + jhi

ω = ωr + jωi

Similar to previous signal models, Sm is the signal transmitted by the tag

and ω is the noise added at the receiver which is independent of Sm and each

noise sample follows an independent and identical Gaussian distribution. The

statistical properties of the real and imaginary components of the noise (ωr and

ωi) is given by (4.15).

ωr ∼ N (0, σ2ω IK)

ωi ∼ N (0, σ2ω IK)

(4.15)

It was assumed that both the real and imaginary components of noise is

having the same statistical properties. Next, the product of forward and re-

verse channels, h is assumed to have Gaussian distributions for the real and


imaginary components as shown in (4.16).

hr ∼ N (µr, σ2r )

hi ∼ N (µi, σ2i )

(4.16)

Then the real and imaginary components of the received signal y can be

represented using the following relationship.

y = yr + jyi



(4.17)

A new real vector, y0 is created using yr and yi as follows.

y0 =[yr , yi

](4.18)

The statistical properties of yr and yi are examined next. It can easily be

seen that they too follow a Gaussian distribution. The mean of yr and yi are

given by (4.19).

E[yr] = µrSm,r − µiSm,i

E[yi] = µrSm,i + µiSm,r

(4.19)

Covariances of yr and yi can be calculated using the following formula.

Cov[X] = E[(X− E[X])T (X− E[X])

]After some calculations, it can be shown that the covariances of yr and yi

are as follows.

Cov(yr) = σ2r ST

m,rSm,r + σ2i ST

m,iSm,i + σ2ω IK

Cov(yi) = σ2r ST

m,iSm,i + σ2i ST

m,rSm,r + σ2ω IK

(4.20)


Therefore, the distribution of yr and yi can be listed as follows.

yr ∼ N(µrSm,r − µiSm,i , σ2

r STm,rSm,r + σ2

i STm,iSm,i + σ2

ω IK)

yi ∼ N(µrSm,i + µiSm,r , σ2

r STm,iSm,i + σ2

i STm,rSm,r + σ2

ω IK) (4.21)

Using (4.18) and (4.21) it can be concluded that y0 has a multivariate Gaus-

sian distribution with a dimension of 2. The mean of y0 can be written as,

E[y0] =[µrSm,r − µiSm,i , µrSm,i + µiSm,r

](4.22)

The covariance of y0 can be calculated as follows.

Cov[y0] = E[(y0 − E[y0])

T (y0 − E[y0])]

After some calculations, it can be seen that cov[y0] simplifies to,

cov[y0] =

σ2r ST

m,rSm,r + σ2i ST

m,iSm,i σ2r ST

m,rSm,i − σ2i ST

m,iSm,r

σ2r ST

m,iSm,r − σ2i ST

m,rSm,i σ2r ST

m,iSm,i + σ2i ST

m,rSm,r

+ σ2ω I2K

(4.23)

Then the conditional probability on receiving y0 given Sm has been trans-

mitted was given by (4.24).

Pr(y0|Sm) =1

2π√|Cov[y0]|

exp(− 1

2(y0 − E[y0]) Cov(y0)

−1 (y0 − E[y0])T)

(4.24)

Similar to previous models, now the probability calculated in (4.24) is max-

imized over all possible tag combinations, Sm.


maxSm

Pr(y0|Sm) (4.25)

In this model, the product of the forward and reverse channels are modeled

using a Gaussian distribution. In the next model, we assume no channel infor-

mation is available. As a result, it is a joint optimization problem of deciding

both the channel and the tag combination.

4.2.4 System model IV - Unknown channel

In this model, we assume no channel knowledge is available to the RFID reader.

In addition, signals considered here are complex meaning I/Q modulation/

demodulation is utilized at the reader. It is assumed that the product of the

forward and reverse channels, h is an unknown complex number and is a con-

stant during the interrogation time. Then the received signal y can be written

as given in (4.1).

For computation simplicity, the complex channel and the noise can be rep-

resented using two real components as follows.

h = hr + jhi

ω = ωr + jωi

Then the received signal y can be represented using the real and imaginary

components similar to previous models.

y = yr + jyi



(4.26)

From (4.26), it can be clearly seen that conditional probability of receiving yr

and yi given Sm and h has a Gaussian distribution. Their statistical properties


can be calculated as follows.

E[yr|h, Sm] = hrSm,r − hiSm,i

E[yi|h, Sm] = hrSm,i + hiSm,r

Cov[yr|h, Sm] = σ2ω IK

Cov[yi|h, Sm] = σ2ω IK

(4.27)

A vector (y0) containing real values is created by stacking yr and yi on a

row vector as follows.

y0 =[yr , yi

](4.28)

Similar to yr and yi, when both the channel h and the tag combination Sm

are given, the conditional probability of receiving y0 is having a Gaussian dis-

tribution. The mean is given by,

E[yr|h, Sm] = µ =[hrSm,r − hiSm,i , hrSm,i + hiSm,r

](4.29)

In order to derive an expression for the conditional probability, covariance

has to be calculated. The following formulas show how to calculate the covari-

ance.

Cov[y0|h, Sm] = E[[y0 − µ]T[y0 − µ]

]= E

[[ωr , ωi]

T[ωr , ωi]]

= σ2ω I2K

Then the conditional probability of receiving y0 given h and Sm can be cal-

culated as in (4.30)


y0|h, Sm ∼ N(µ , σ2

ω I2K)

Pr(y0|h, Sm) =1

2πσωexp

(− 1

2σ2ω(y0 − µ)T(y0 − µ)

) (4.30)

The probability given in (4.30) is maximized over all possible combinations

of h and Sm. Therefore, this is a joint optimization problem.

maxhr,hi,Sm

Pr(y0|hr, hi, Sm) = minhr,hi,Sm

((y0 − µ) (y0 − µ)T)

= minhr,hi,Sm

(y0yT

0 − 2y0µT + µµT) (4.31)

4.2.5 Joint optimization of h and tag type

µ is calculated using (4.29). However, there are infinitely large number of com-

binations for hr and hi, hence it is not computationally feasible. A feasible

solution would be to first find the optimum channel for a given tag combina-

tion. Then the given tag combination response and the optimum channel are

used for calculating the conditional probability given in (4.30). Then the same

process is repeated for all possible tag combinations like previous detectors to

calculate the highest probability. Next, calculating the optimum channel for a

given tag combination is discussed. L(hr, hi) is defined as follows.

L(hr, hi) = y0yT0 − 2y0µT + µµT (4.32)

For optimum hr and hi following conditions have to be satisfied.

∂

∂hrL(hr, hi) = 0

∂

∂hiL(hr, hi) = 0

(4.33)

Using (4.29) it can be shown that


∂µ

∂hr= [Sm,r , Sm,i]

∂µ

∂hi= [−Sm,i , Sm,r].

(4.34)

Using the relationships in (4.32), (4.33) and (4.34), the optimum channel hr

and hi for a given Sm can be derived as follows.

hr =y0[Sm,r , Sm,i]

T

[Sm,rSTm,r + Sm,iST

m,i]

hi =y0[−Sm,i , Sm,r]T

[Sm,rSTm,r + Sm,iST

m,i]

(4.35)

The optimum channel estimates obtained from (4.35) are used to calculate

the optimum µ (µ0). (4.29) and (4.35) yields,

µ0 =[hrSm,r − hiSm,i , hrSm,i + hiSm,r

](4.36)

Then the new optimization problem reduces to,

maxhr,hi,Sm

Pr(y0|hr, hi, Sm) = minhr,hi,Sm

((y0 − µ0) (y0 − µ0)

T) (4.37)

In this model, no channel information is available to the reader, and the

only assumption is that the channel is static during the short interrogation time

period. The tag detector derived in the model, uses the received signal at the

RFID reader and all the possible tag responses to determine both the channel

and the tag combination that provides the highest probability.

All the four signal models discussed in this section are based on time do-

main signal samples [43]. Some of the existing RFID readers work based on

frequency domain samples. The next section discusses about the tag detectors

that can work based on frequency domain signal samples.

4.3 System models - Frequency domain 59

4.3 System models - Frequency domain

Chipless RFID readers are based on either time domain tags [58–60] or fre-

quency domain tags [68, 69, 72, 74–77] Time domain tags encode the informa-

tion in time samples of the signal leaving a unique time signature whereas, the

frequency domain tags encode information in the frequency samples of the

signal leaving a unique frequency domain signature. Therefore, it is impor-

tant to examine the tag detection techniques for the frequency domain based

chipless RFID tags. There is another very important benefit of using frequency

domain tags. Tag detectors derived for time domain based tags has a high com-

putational complexity. However, frequency domain based tag detection can

be achieved with relatively a lower computational complexity as explained in

Chapter 4 in detail. The rest of this section describes 5 tag detection techniques

for frequency based chipless RFID tags.

4.3.1 System models I - IV

The tag detectors derived for time domain chipless RFID tags are first revisited

briefly. The signal model used in all the four detectors is as follows.

y(t) = hSm(t) + ω(t) (4.38)

Model I

In model 1, channel h is a known real constant. Noise is having a zero mean

normal distribution with a covariance σ2ω IK. Then the probability distribution

function of receiving y given that Sm has been transmitted is given by (4.6). If

fourier transformation is performed on (4.38), the result is shown below.


Y( f ) = hSm( f ) + ω( f ) (4.39)

Y( f ) is the fourier transform of the received signal y. Sm( f ) and ω( f ) are

the fourier transforms of the mth tag response (Sm) and the noise ω respec-

tively. Fourier transform is a unitary transformation. Therefore the statistical

properties of the signals should remain the same. As a result, the probability

distribution function of receiving y( f ) given that Sm( f ) has been transmitted

is the same as (4.6). Therefore the frequency domain based chipless RFID tag

detector for system model 1 can be derived using the frequency samples of the

signal as follows.

maxSm( f )

Pr(Y( f )|Sm( f )) = minSm( f )

((Y( f )− hSm( f ) (Y( f )− hSm( f ))T

)(4.40)

Model II

Similarly, frequency domain chipless RFID tag detector for system model 2 can

be derived as below.

maxSm( f )

Pr(Y0( f )|Sm( f ) = minSm( f )

((Y0( f )− µ( f )) (Y0( f )− µ( f )T

)(4.41)

Similar to the previous model, Sm( f ) is the fourier transformation of Sm.

Y0( f ) and µ( f ) are the fourier transformations of y0 in (4.11) and µ in (4.12)

respectively.


Model III

The statistical properties of the signal model 3 too remains the same hence, the

tag detector in frequency domain is given by,

maxSm( f )

Pr(Y0( f )|Sm( f )) =1

2π√|Cov[Y0( f )]|

exp(− 1

2(Y0( f )− E[Y0( f )]) Cov(Y0( f ))−1

(Y0( f )− E[Y0( f )])T)

(4.42)

E[Y0( f )] and Cov[Y0( f )] are the fourier transformations of E[y0] in (4.22)

and Cov[y0] in (4.23) respectively.

Model IV

Like in previous 3 models, statistical properties of the detector in signal model

does not change with the fourier transform. Therefore the tag detector for

signal model IV can be written as,

maxhr,hi,Sm( f )

P(Y0( f )|hr, hi, Sm( f )

)= min

hr,hi,Sm( f )

((Y0( f )− µ0( f )

) (Y0( f )− µ0( f )

)T)

(4.43)

Y0( f ) and µ0( f ) are the fourier transformations of y0(t) in (4.28) and µ0(t)

in (4.36) respectively.

It is clear that the frequency samples of the signal can be used to detect tags

using the same detectors derived for the time based models. It is true that the

frequency signature of the frequency domain tags can be seen in fourier trans-

formation of the time domain signal. However, there are chipless RFID read-

ers that work on the power spectral density of the signal, rather than fourier


transformation. In the next section, a tag detector is derived for a power based

chipless RFID reading method.

4.3.2 System model V - Power magnitudes

There are chipless RFID readers that operates based on the power measure-

ments rather than on voltage samples. In these readers, a narrow banded sinu-

soidal waveform is transmitted as the interrogating signal and the magnitude

of the tag response compared to the transmitted signal is measured using a

gain detector. Then the frequency of the narrow-banded interrogating signal

is swept across the frequency of interest and the frequency signature of the tag

is obtained.

The frequency signature obtained in this method, requires a lesser sam-

pling rate at the reader compared to the frequency signature obtained using

the fourier transformation performed on time domain based measurements. In

addition, the narrow-banded signals are subjected to lesser noise which could

lead to better tag reading reliability.

System Model V describes an existing chipless RFID reader developed at

Monash Microwave, Antenna, RFID and Sensor Laboratory (MMARS) under

Australian Research Councils Linkage Project Grant: LP0991435: Back-scatter

based RFID system capable of reading multiple chipless tags for regional and suburban

libraries. The reader works on power magnitude of the received tag response.

The application assumes a fixed distance between the reader and the tags, and

at short distances such as 15 cm, the line of sight component dominates over

any multi paths. Therefore, the channel undergoes a very slow variation with

respect to time. During the calibration phase, all possible tag responses were

measured and recorded for the given distance between the reader and the tags.

The recorded tag responses were used with the likelihood based detector de-


rived below.

If the fourier transformation of the tag response is Sm( f ) and the fourier

transformation of the noise added at the reader is ω( f ) then the fourier trans-

formation of the received signal at the reader is given by Y( f ). Then Y( f ) is

separated into real and imaginary components as shown in 4.44.

Y( f ) = Sm( f ) + ω( f )

Y( f ) =(

Sm,r( f ) + ωr( f ))+ j(

Sm,i( f ) + ωi( f )) (4.44)

As shown in Section 4.3.1, the statistical properties of ω( f ) is the same as its

time domain samples. Then the power magnitude Z of the frequency domain

samples are given by,

Z = |Y( f )|2 =

(Sm,r( f ) + ωr( f )

)2

+

(Sm,i( f ) + ωi( f )

)2

Statistical properties of ωr( f ) and ωi( f ) can be written as below.

ωr( f ) ∼ N (0, σ2r IK)

ωi( f ) ∼ N (0, σ2i IK)

Then the real (R) and imaginary (I) components of Y( f ) are defined as

shown below.

R = Sm,r( f ) + ωr( f )

I = Sm,i( f ) + ωi( f )

The statistical properties of R and I can be derived as shown in (4.45).

R ∼ N (Sm,r( f ), σ2r IK)

I ∼ N (Sm,i( f ), σ2i IK)

(4.45)

Assuming independence between individual samples of each R and I, the


probability of receiving Z given that Sm,r( f ) and Sm,i( f ) had been transmitted

is given by 4.46. (.)(k) is the kth sample of the corresponding vector.

Pr(Z|Sm,r( f ), Sm,i( f )) =K

∏k=1

Pr(Z(k)|S(k)m,r( f ), S(k)

m,r( f )) (4.46)

It is clear that R and I are independent. As a result, Z has a non-central chi-

square distribution with λ being the non-centrality parameter and d f being the

degree of freedom, which is 2. λ for kth sample can be calculated as follows.

λ(k) =

[S(k)

m,r( f )σr

]2

+

[S(k)m,i( f )

σi

]2

(4.47)

Then Pr(Z(k)|S(k)m,r( f ), S(k)

m,r( f )) can be expressed as follows.

Pr(

Z(k)|S(k)m,r( f ), S(k)

m,r( f ))=

12

exp(− Z(k) + λ(k)

2

)(Z(k)

λ(k)

) d f4 −

12

Id f /2−1

(√λ(k).Z(k)

) (4.48)

Iv(.) is the modified Bessel function of the first kind. The above expression

can be simplified using d f = 2 and assuming the variance for both real and

imaginary noise components is the same (σr = σi = σω). Then,

Pr(

Z(k)|S(k)m,r( f ), S(k)

m,r( f ))=

12

exp(− Z(k) + λ(k)

2

)I0

(√λ(k).Z(k)

)(4.49)

Using (4.46) and (4.49) Pr(Z|Sm,r( f ), Sm,i( f )) can be calculated as follows.

4.4 Simulations 65

Pr(

Z|Sm,r( f ), Sm,i( f ))=

12K exp

(− 1

2(K E[Z] + Em/σ2

ω

))K

∏k=1

I0

(√λ(k)Z(k)

) (4.50)

Em in (4.50) is the energy of the selected tag response Sm( f ) defined as,

Em =K

∑k=1

([S(k)

m,r( f )]2

+[S(k)

m,i( f )]2) (4.51)

Then the probability calculated in (4.50) is maximized over Sm,r( f ) and

Sm,i( f ) to detect the most likelihood tag combination.

maxSm,r( f ),Sm,i( f )

Pr(

Z|Sm,r( f ), Sm,i( f ))

(4.52)

The tag detection technique developed in this section was used with an

existing chipless RFID reader. During the calibration phase, the tag responses

for all possible combinations are recorded. Then the above detection technique

is used to detect a random tag.

So far in the chapter, several tag detection techniques have been derived

based on both time domain samples as well as frequency domain samples. In

order to test these detection techniques comprehensively, a MATLAB simula-

tion was performed, which is explained in the next section.

4.4 Simulations

The validity of the above tag detection techniques were verified using MAT-

LAB and Computer Simulation Technology (CST) simulations. The steps car-

ried out in the simulation are given in Figure 4.4. Firstly, an interrogating

signal was generated to provide a flat frequency response in 2.2-2.6 GHz fre-


Figure 4.4: Flowchart of the MATLAB simulation in conjunction with CST full-wave EM solver simulation

quency range. Four resonators were designed using CST with resonating fre-

quencies as shown in Table 5.1. The resonance frequencies are selected as 100

MHz apart following the specifications provided in [68]. Then the combina-

tions of resonators were placed besides a micro-strip line to cover all possible

tag IDs. One end of the bandstop filters loaded micro-strip line was fed with

the interrogating signal and the tag responses were collected at the other end.

These collected tag responses were saved in a look up table for the algorithms

to be used later. More details about the tag design is available in Chapter 6.

Then the tag responses were fed through a channel which is given by the

product of the forward and reverse channels of the RFID system. Depending

on the detection technique the channel values are selected to be a known con-

stant or a variable with a known or unknown statistical distribution. Finally

noise is added to the resultant signal according to the specified SNR.

SNR is calculated compared to the average power of all the tag combina-

4.4 Simulations 67

tions. It can be summarized as follows.

I(t) – interrogating signal

h f – forward channel

hr – reverse channel

h – product of forward and reverse channels

Fm(t) – impulse response of the mth filter

Sm(t) – mth tag response

y(t) – received signal

ω(t) – noise added at the reader

Then the received signal can be represented using (4.53).

y(t) =[[h f I(t)] ∗ Fm(t)

]hr + ω(t)

= h f hr[I(t) ∗ Fm(t)] + ω(t)

= hSm(t) + w(t)

(4.53)

The power of each tag response is calculated and averaged to obtain the

average power of a given tag response. For example, 2-bit tags have four dif-

ferent tag responses (Sm(t)) and power of each tag response is calculated and

averaged to obtain the average power of 2-bit tag responses. Then the average

tag response power is multiplied using the channel to calculate the average

signal power available at the reader. For a given SNR, noise power is calcu-

lated using this available signal power.

MATLAB simulation parameters are outlined in 4.1. Four band-stop filters

were used, and most-significant-bit (MSB) corresponds to the lowest resonance

frequency and least-significant-bit (LSB) to the highest.

I/Q demodulation is performed with the received signal at the RFID reader.

Then the two output time domain signal vectors were used to evaluate likeli-



Parameter Value

Center Frequency 2.4 GHzTotal bits encoded in a tag 4 bitsFlat frequency response 400 MHzBand-stop filter attenuation 10 dBBand-stop filter 3dB bandwidth 40 MHzGuard band 60 MHzResonance frequency set 1 (MSB to LSB) [2.2, 2.3, 2.4, 2.5] GHzResonance frequency set 2 (MSB to LSB) [2.34, 2.38, 2.42, 2.46] GHzResonance frequency set 3 (MSB to LSB) [22.5, 23.5, 24.5, 25.5] GHzChannel mean 0.4Channel standard deviation 0.1No. of iterations up to 10,000,000

hood expressions for each detection technique. In order to verify their fre-

quency domain performances the time domain vectors were converted using

fast fourier transform (FFT). Then these frequency domain samples were used

for tag detection.

Finally the detection error rate (DER) is calculated for each tag detection

technique at different SNR levels. DER is defined as the probability of having

at least one erroneous bit out of the all data bits. It can be represented using the

throughput as in (4.54). Throughput is defined as the ratio between the num-

ber of successful tag readings (NS) and total number of tag readings performed

(NT) as given in (4.55).

DER = 1− throughput (4.54)

throughput =NS

NT(4.55)

Existing chipless RFID systems use a threshold based detection technique

based on frequency domain based samples. The presence and absence of each

resonator in this method is detected based on a magnitude threshold at cor-

responding resonator frequency bands. This threshold based detection tech-

4.5 Experimental setup 69

Figure 4.5: A chipless tag coded with bits [1111]

nique is implemented and the DER is calculated to compare the performances

of the proposed detection techniques.

4.5 Experimental setup

The likelihood detector derived under this System model V uses only the mag-

nitude of the tag responses in frequency domain. This tag detection technique

is derived for an existing chipless RFID system that operates between 21 - 27

GHz. A circular resonator based chipless RFID tags were designed and fab-

ricated as described in Chapter 3. A printed tag encoded with bits [1111] is

shown in Figure 4.5.

Two tag prototypes were experimented to validate the theory. Tag type I

was a retransmission tag operating over the frequency band of 2.2 - 2.6 GHz.

The tag type II is a backscattering tag operating at 21 - 27 GHz. The experiment

set up for validating tag type II is shown in Figure 4.6. In both experiments, the

reader transmits a narrow-banded sinusoid signal using one horn antenna and

the reader receives the tag response using a second antenna. The magnitude

of the received signal is recorded along with the frequency of the signal sent.


Tag Reader RF Electronics

Horn Antenna Under

Perspex Table

LCD Display Perspex Table

Digital Control & Power Supply Unit

Figure 4.6: Experimental setup

The frequency is swept across the band 21 - 27 GHz and the complete band

response is recorded.

This process is repeated with all the possible 16 combinations for a 4-bit

tag and the recorded tag responses are used with the tag detection technique

derived for System model V to detect the encoded tag data bits.

In the next section both the simulation and experimental results are ana-

lyzed for the 5 tag detection techniques presented.

4.6 Results 71

0 1 2 3 4 5 6 7 8 9−1

−0.5

0

0.5

1

Time (ns)

Nor

mal

ized

am

plitu

de

1 1.5 2 2.5 3 3.5−150

−140

−130

−120

−110

−100

−90

Frequency (GHz)

Pow

er s

pect

ral d

ensi

ty (

dB/H

z)

400 MHz

Figure 4.7: Interrogating signal in time and frequency domain

4.6 Results

An interrogating signal was designed to provide a flat frequency response for

at least 400 MHz around 2.4 GHz and it was used as the port excitation signal

in CST simulations. Figure 4.7 shows the time and frequency domain interro-

gating signals. It can clearly be seen that the above requirement is achieved

quite easily.

Then the multi-resonator tags were designed in CST according to the spec-

ifications given in Table 4.1. Initially, resonators were designed according to

the frequency set 1 given in Table 5.1 which includes a guard band between

resonance frequencies. Then the resonators were redesigned in CST using the

frequency set 2 without any guard band. Then both the time and frequency

domain tag responses were obtained from CST simulations and stored in a

lookup table for MATLAB simulations.

Figure 4.8 shows the frequency domain response of a tag encoded with


4 6 8 10 12 14 16 18 20 22−0.4

−0.2

0

0.2

0.4

Time (ns)

Nor

mal

ized

am

plitu

de

2.1 2.2 2.3 2.4 2.5 2.60

0.2

0.4

0.6

0.8

1

Nor

mal

ized

mag

nitu

de

Frequency (GHz)

Figure 4.8: Tag responses for [1111] with a guard band

bits [1111]. It can clearly be observed that the resonances occur at the designed

frequencies given by set 1 in Table 4.1. Figure 4.8 also shows the corresponding

time domain response of a tag encoded with bits [1111].

Then a new set of tag resonators were designed according the frequency set

2 given in Table 4.1 without any guard band. The frequency and time domain

tag responses are illustrated in Figure 4.9. Similar to the previous simulation

results, it can be concluded that the resonators were performing as expected

by observing the resonance frequencies.

Then these four resonators are arranged to implement the 16 tag combina-

tions for a 4-bit tag. Simulations were repeated for 16 times and both the time

domain and frequency domain tag responses were obtained and recorded.

Next the system models discussed in the previous section were implemented

in MATLAB and the performances were analyzed. Detection error rates (DER)

under different SNR levels were calculated for each system model and the

4.6 Results 73

5 10 15 20 25−0.4

−0.2

0

0.2

0.4

Time (ns)

Nor

mal

ized

am

plitu

de

2.2 2.25 2.3 2.35 2.4 2.45 2.5 2.55 2.60

0.2

0.4

0.6

0.8

1

Nor

mal

ized

mag

nitu

de

Frequency (GHz)

Figure 4.9: Tag responses for [1111] without a guard band

compared with a threshold based tag detection system used in existing RFID

readers. Simulation results for each system model is presented in the following

sub sections.

4.6.1 System model I

Firstly, the detection error rate is calculated at different noise power levels,

effectively changing SIR and the result for System Model I is shown in Figure

4.10.

It can be seen that the DER is the same for both the time and frequency do-

main based samples. Therefore, it verifies the argument that the time domain

based detection expression remains valid for the frequency domain based sam-

ples. When the frequency domain based samples are used, tag detection can

be performed using the presence and absence of the power dips at resonating

frequencies. This is achieved in existing chipless RFID readers using a thresh-


0 2 4 6 8 10 12 14 16 18 2010

−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e (D

ER

)

ML detector I − TimeML detector I − FrequencyThreshold based detector

Figure 4.10: DER vs SNR for 4-bit tag with 60 MHz guard band

old based detection method that detects the power dips. This threshold based

detection method is used as the baseline comparison for the proposed tag de-

tection methods.

As can be seen in 4.10, the proposed detection methods provide a signifi-

cant improvement on detection error rate over the threshold based detection

method. It can be seen that in order to achieve 99.99% reading accuracy, thresh-

old based detector need a SNR of 19 dB. However the proposed detector re-

quires a SNR of only 11 dB which is a SNR gain of 8 dB over the threshold

based detector. SNR can be related to the reading distance and SNR gain re-

sults in an increment in the tag reading range. The signal travels twice the

distance between the reader and the tag. As a result, if the indoor propagation

constant is assumed to be 2 [100], this SNR improvement can be related to im-

prove the reader distance by a factor of 2.5. Therefore, the tag reading range

can be improved by 2.5 times with the proposed tag detection technique while

4.6 Results 75

0 2 4 6 8 10 12 14 16 18 2010

−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e (D

ER

)

ML detector I − TimeML detector I − FrequencyThreshold based detector

Figure 4.11: DER vs SNR without a guard band between resonator frequencies

achieving a target reading accuracy of 99.99%.

On the other hand, this improved performance can be viewed as an incre-

ment in the tag reading accuracy at a given SNR. For example, the detection

error rate of the threshold based detector at SNR = 10 dB is about 90%. With

the proposed detector the accuracy can be improved up to 99.95%. This avoids

the requirement to perform multiple tag readings to detect one tag, specially

under low SNR scenarios. Therefore, depending on the application, improve-

ment can be viewed on either the tag reading accuracy or the tag reading range.

Then the System Model I was used to detect the tag responses obtained using

frequency set 2 given in Table 4.1. The removal of the guard band causes the

number of resonators allowed per unit bandwidth to be more than doubled in

this example which in turn double the tag bit capacity. As can be seen in Figure

4.11, traditional threshold based decoder produces very poor performances

when the guard-band is absent. For example, at a SNR of 10 dB the detection


0 2 4 6 8 10 12 14 16 18 2010

−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e (D

ER

)

ML detector I − with GBML detector I − w/o GBThreshold detector − with GBThreshold detector − w/o GB

Figure 4.12: DER vs SNR for ML decoder 1

error rate of the threshold based method is about 60%. However, the proposed

Signal Model 1 detector still achieves a high accuracy of 99% at the same SNR.

Similar to the increment in the tag reading accuracy, the SNR gain of the

proposed method compared to the baseline is also significant. For example, in

order to achieve an accuracy of 90% threshold based detector requires at least

a SNR of 19 dB while the proposed method requires only 3 dB providing a

SNR gain of 16 dB which is equivalent to a tag reading range improvement

by 6 times. However, in reality the maximum reading range is limited by the

mismatches in the tag orientation, antenna gain, transmitted power and re-

ceiver’s dynamic range. At large distances all these factors contribute to cause

performance deterioration.

For further clarity, figure 4.12 shows a comparison of the DER for both

the threshold method and ML decoder under the presence and absence of a

guard-band (GB). At lower SNR levels ( <5 dB) such as noisy industrial en-

4.6 Results 77

vironment, the proposed detection method on both resonator sets performs

similarly. However, at higher SNR levels resonators with a guard-band per-

forms better for obvious reasons. The key observation is that, the threshold

based detector has very poor performances at compact tag resonators (with-

out guard-band) however the proposed detector still has an accepted level of

tag reading accuracy. Therefore, it can be argued that the proposed detection

method allows the tag data bit capacity to be doubled without compromising

the tag reading performance. In addition, the likelihood expression is valid for

both time and frequency domain based sampling.

4.6.2 System model II

After obtaining satisfactory performances from Model I, the System Model II is

tested with both the real (I) and imaginary (Q) components of the received

signal. Figure 4.13 shows the real and imaginary components of the baseband

signal obtained after I/Q demodulation for tag response of [1111] when a 60

MHz guard-band is used between resonator frequencies.

FFT was performed on these I/Q samples and the tag response in frequency

domain was obtained. Figure 4.14 illustrates the magnitude and phase of the

frequency domain tag response. It can clearly be seen that the information is

encoded in both the magnitude and phase. As a result, it can be concluded

that unlike in Model I, Model II uses the information used in both the I/Q sam-

ples. Therefore it is expected to outperform the detector derived for Model I.

These time as well as the frequency domain samples obtained by applying fast

fourier transformation to the time samples were used to calculate the DER at

different SNR levels. Figure 4.15 shows a comparison of the DER for both the

threshold method and ML decoder 2 when a guard band is presented. It can

be seen that, the threshold detector achieves a tag reading accuracy of 99.99%


-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

3 5 7 9 11 13 15 17 19 21

No

rmal

ize

d in

-ph

ase

am

plit

ud

e (

Rea

l)

Time /ns

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

3 5 7 9 11 13 15 17 19 21

No

rmal

ize

d q

uad

-ph

ase

am

plit

ud

e (

Imag

.)

Time /ns

Figure 4.13: Real and imaginary samples of the tag response [1111]

at a SNR of 17 dB. The same level of reading accuracy can be achieved at 6 dB

using ML decoder 2. The ML detector provides a SNR gain of 11 dB at 99.99%

accuracy level which can be related to the improvement of the reading range

by a factor of 3.5.

On the other hand, at a SNR of 8 dB, the ML detector achieves and reading

accuracy of 99.999% while the threshold detector manages only 70%. In addi-

tion, the proposed ML detector 2 has a tag reading accuracy of 95% at as low as

0 dB SNR. Therefore, it can also be concluded that the proposed ML detector

2 is performing well under low SNR levels. It is interesting to notice that, ML

detector 2 has a SNR gain of 3 dB compared to ML detector 1. This is due to

the fact that, ML detector 2 uses in information available in both the real and

4.6 Results 79

2.15 2.2 2.25 2.3 2.35 2.4 2.45 2.5 2.550

0.5

1

Frequency /GHz

Nor

mal

ized

Am

plitu

de

2.15 2.2 2.25 2.3 2.35 2.4 2.45 2.5 2.55−2

−1

0

1

2

3

Frequency /GHz

Pha

se

Figure 4.14: Frequency signature of tag type [1111]

0 2 4 6 8 10 12 14 16 18 2010

−6

10−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector II − TimeML detector II − FrequencyThreshold based detector

Figure 4.15: DER vs SNR for ML decoder 2 with the presence of a guard-band


0 2 4 6 8 10 12 14 16 18 2010

−6

10−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector II − TimeML detector II − FrequencyThreshold based detector

Figure 4.16: DER vs SNR for ML decoder 2 without a guard-band

imaginary components of the received signal whereas ML detector uses only

the real component of the signal. A detailed comparison of the results obtained

in different models are presented later in this section.

Figure 4.16 shows a comparison of the DER for both the threshold method

and ML decoder 2 when a guard band is removed. Similar to System Model 1,

the threshold based detector perform very poorly when the guard-band is re-

moved. At a SNR of 10 dB, threshold based detector has a reading accuracy of

50% while ML detector 2 achieves well over 99.99% tag reading accuracy. On

the other hand, it is obvious that ML detector 2 provides an enormous SNR im-

provement over the threshold detector. This significant improvement is mainly

due to the fact that unlike threshold detector the likelihood expression derived

in 4.14 is the optimum decoder as it uses the information available in both the

real and imaginary components of the signal. A comprehensive comparison of

the performances of the proposed detection techniques is presented at the end

4.6 Results 81

of the 4.6.


and ML decoder 2 under the presence and absence of a guard-band. It is clear

0 2 4 6 8 10 12 14 16 18 2010

−6

10−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector II − with GBML detector II − w/o GBThreshold detector − with GBThreshold detector − w/o GB


that for obvious reasons, both detectors perform well with the presence of a

guard-band. However, once the guard-band is removed, threshold detector get

affected the most while ML decoder 2 still provides acceptable performances.

ML detector performs well at lower noise levels regardless of the guard-band.

In addition, it can be seen that the even after removing the guard-band,

Model II detector performs better than the existing threshold based method.

It can be interpreted as doubling the data capacity per unit bandwidth. In

order to verify this claim, the simulations were repeated with 8 resonators in

the same bandwidth compared to 4 in the previous case. Figure 4.18 shows

the calculated DER under 8 bits and compared against 4 bits tags with and

without a guard-band. It can be seen that 4 bit tag with a guard-band has


the best performance. Even though, the 8 bit tag without a guard-band has

the worst performance out of the 3 tags, it is comparable to that of the 4-bit

tag without a guard-band. Hence, it is possible to conclude that the proposed

detection algorithm doubles the tag data capacity.

0 2 4 6 8 10 1210

−6

10−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

4 bit with GB

4 bit w/o GB

8 bit w/o GB

Figure 4.18: DER comparison for 8-bit tags

4.6.3 System model III

System Model III assumes an unknown channel with a known channel distribu-

tion. It was assumed that both the real and imaginary part of the channel has

the same Gaussian distribution due to symmetry. The mean and the standard

deviation of the distribution was taken as 0.4 and 0.1 respectively. At different

noise levels, DER was calculated based on both time and frequency samples

4.6 Results 83

0 2 4 6 8 10 12 14 16 18 2010

−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector III − TimeML detector III − FrequencyThreshold based detector


and compared that with the baseline detection applied on frequency domain

samples.


and ML decoder 3 when a guard-band is presented. Unlike in previous two

models, channel information is not available at the RFID reader. ML detector

derived in this model, uses the statistical properties of the channel for tag de-

tection. Therefore, as expected the performances are not very good as previous

models, specially under low SNR scenarios. ML detector 3 achieves about 90%

reading accuracy at SNR = 5 dB which is still better than the 50% accuracy level

of threshold based detection. However, as SNR improves the reading accuracy

improves exponentially. For example, at a SNR of 10 dB, ML detector achieves

an accuracy level of 99.95% whereas threshold detector achieves only 90%.

Moreover, ML detector achieves an accuracy level of 99.9% at SNR = 9 dB

while threshold detector needs 15 dB in order to achieve the same accuracy


0 2 4 6 8 10 12 14 16 18 2010

−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector III − TimeML detector III − FrequencyThreshold based detector


level. This is a SNR gain of 6 dB which is equivalent to the improvement of the

reading range by a factor of 2. In addition, similar to previous models, time

and frequency based ML detectors provides the same results.


and ML decoder 3 when there is no guard-band presented. As expected, thresh-

old based detector has high detection error rates. For example, the reading ac-

curacy of threshold detector at SNR = 10 dB is about 50% while the proposed

ML detector achieves an accuracy level of 99% at the same SNR. Even though,

the tag reading accuracy at lower SNR is poor, it improves exponentially as

SNR increases. On the other hand, the proposed ML detector has an SNR gain

of 10 dB over threshold detector when both detectors are expected to achieve

an accuracy level of 95%. This SNR gain provides an improvement in reading

range by a factor of 3. It can be concluded that, the proposed ML based de-

tector performs better than the threshold based detector used in existing RFID

4.6 Results 85

0 2 4 6 8 10 12 14 16 18 2010

−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector III − with GBML detector III − w/o GBThreshold detector − with GBThreshold detector − w/o GB


readers.


and ML decoder 2 under the presence and absence of a guard-band.

It can clearly be seen that, all of the detection methods failed to operate

at lower SNR levels (<5 dB). So it is safe to conclude that the assumptions to

represent both the real and imaginary parts of the channel in a Gaussian dis-

tribution System Model 3 is valid only for higher SNR levels (>5 dB). However,

ML based detection technique performs better as the SNR improves regardless

of the guard-band. The guard-band provides on average a SNR improvement

of 2 dB for obvious reasons. Finally it can be concluded that, the proposed ML

detection method allows to double the bit capacity of the chipless RFID tags

without compromising the reading accuracy.


0 2 4 6 8 10 12 14 16 18 2010

−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector IV − TimeML detector IV − FrequencyThreshold based detector


4.6.4 System model IV

As described earlier, no channel information is available at the reader in System

Model IV. This ML based decoder detect both the tag type and the channel

simultaneously. Like in previous cases, DER was calculated based on both

time and frequency samples and compared that with the baseline detection

applied on frequency domain samples.


and ML decoder 3 when a guard-band is presented. Unlike the System Model

III, this model performs better even at lower SNR levels. In this model, no

assumptions are made to represent the channel. Instead, channel values are

estimated along with the tag detection. It can be seen that, with the presence

of a guard-band, ML detector achieves a reading accuracy level of 99.9% at

SNR = 5 dB. However, threshold detector provides only 50% accuracy at the

same SNR.

4.6 Results 87

0 2 4 6 8 10 12 14 16 18 2010

−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector IV − TimeML detector IV − FrequencyThreshold based detector


On the other hand, a reading accuracy of 99.9% is achieved by the proposed

ML detector with a SNR gain of 10 dB over the threshold method. This results

in improving the tag reading range by a factor of 3. In addition, both the time

and frequency domain samples based data provides the same performances as

expected.


and ML decoder 3 when there is no guard-band presented. Following the trend

in previous models, System Model IV performs well, even without a guard-

band between resonance frequencies while the threshold detector has very

poor performances. For example, at SNR = 5 dB, ML detector provides a read-

ing accuracy of 99.5% while threshold detector provides on 20%. Apart from

that, the proposed ML decoder provides a SNR gain of 17 dB over the thresh-

old method when both are achieving a reading accuracy of 97%. This can be

related to tag reading range being improved by a factor of 7. However, as ex-


0 2 4 6 8 10 12 14 16 18 2010

−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector IV − with GBML detector IV − w/o GBThreshold detector − with GBThreshold detector − w/o GB


plained earlier, this is possible only if the perfect tag orientation is achieved.

The miss-orientation of the tag and the reader antennas may deteriorate the

performance. After observing the both aspects, it can be concluded that the

proposed detection method allows to double the tag data bit capacity without

compromising the tag reading performance.


and ML decoder 2 under the presence and absence of a guard-band. A com-

mon feature of the results for System Model IV is that, it performs well even

under the lower SNR levels, regardless of the guard-band. As expected, at

higher SNR levels, the guard-band provides a SNR gain of 2 dB. As mentioned

at the beginning, ML detector 4 not only detects the tag, but also estimates the

channel.

4.6 Results 89

Figure 4.25: Channel Estimation Samples when a guard-band is presented

Channel Estimation:

The channel estimation accuracy is analysed next. In MATLAB, a random

channel is generated and used for system simulation. Then the ML decoder

4 is used to calculate an estimate of the channel and compared with the actual

channel. Figure 4.25 compares the estimated channel values obtained under

number of iterations with the actual channel realization when a guard-band is

presented between resonators in the chipless tag. It can be seen that, the com-

plex channel estimations are centered around the actual channel realization

and the accuracy level of the estimations are very high at SNR = 14 dB. How-

ever, it does not demonstrate a clear picture of the distribution of the channel

estimate.

Figure 4.26 shows the probability distribution function (pdf) of the channel

estimation for both real and imaginary components and compares them with


0.3 0.35 0.40

1000

2000

3000

4000

5000

6000

Real part of channel

Pro

babi

lity

Den

sity

Fun

ctio

n (P

DF

)

0.22 0.24 0.26 0.28 0.3 0.320

1000

2000

3000

4000

5000

6000

Imaginary part of channel

Pro

babi

lity

Den

sity

Fun

ctio

n (P

DF

)

Figure 4.26: PDF of Channel Estimation when a guard-band is presented

the actual channel realisation which is given by white coloured asterisks (*).

It can be clearly seen that, the mean of each distribution is the same as

the actual channel realisation. In addition, both the distributions have very

similar but low variances. Therefore, it can be concluded that the proposed

ML detection method estimates the channel accurately when a guard-band is

presented.

Figure 4.27 compares the estimated channel values with the actual channel

realisation when more tag bits are presented leaving no guard-band between

resonator frequencies in the chipless tag. The results shows channel estimates

obtained after a number of iterations at SNR = 10 dB. It is clear that even when

the guard-band is not presented, the proposed ML detector manages to esti-

mate the channel accurately. The pdf of the estimates were calculated to closely

observe the performance of channel estimation.

Figure 4.28 compares pdf of the channel estimation at SNR = 10 dB. Simi-

lar to the case with the guard-band, the mean of the distributions for both the

real and imaginary components of channel are approximately same as the ac-

4.6 Results 91

Figure 4.27: Channel Estimation Samples without a guard-band

0 0.1 0.20

1000

2000

3000

4000

5000

6000

7000

Real part of channel

Pro

babi

lity

Den

sity

Fun

ctio

n (P

DF

)

0.35 0.4 0.45 0.5 0.550

1000

2000

3000

4000

5000

6000

7000

Imaginary part of channel

Pro

babi

lity

Den

sity

Fun

ctio

n (P

DF

)

Figure 4.28: PDF of Channel Estimation without a guard-band


0 5 10 1510

−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector I − with GBML detector II − with GBML detector III − with GBML detector IV − with GBSSR method

Figure 4.29: DER comparison with a guard-band

tual channel realisation. Moreover, the variance of both the real and imaginary

components of the channel are very similar and low. The main difference be-

tween the cases having a guard-band and not, is the latter has a high variance

compared to the former. It can be justified as the guard-band prevents inter

resonator interference causing less errors in the outcome of the ML detector

used with tag responses having a guard-band.

Comparative Study:

Finally, detection error rate (DER) performances of all the detection techniques

described above is compared and discussed. Figure 4.29 compares the DER

performances of all the detection when a guard-band is presented between the

tag resonance frequencies. It can be seen that, ML detector 2 has the best per-

formance out of all the detectors presented. ML detector 2 assumes, perfect

channel knowledge is available at the reader and uses information available

4.6 Results 93

in both the real and imaginary components of the signal. This is the optimum

detector for the chipless RFID system and hence can be treated as the upper

margin for tag reading accuracy performances. The next best detector is ML

detector 4 where the channel is completely unknown to the reader. ML detec-

tor 4 is a powerful detector as it not only achieves a higher tag reading accuracy

but also estimates the channel with a high accuracy. ML detector 1 is the next

best performer where it assumes perfect channel knowledge and uses only the

information available in the real part. This decoder does not exist in reality as

this is derives first as the fundamental decoder and every other decoder is an

extension of it.

Signal space representation (SSR) method used in [91] has similar perfor-

mances to ML detector 1. This can be explained as both the detectors uses the

information encoded in only on the magnitude. However, only 5 most domi-

nant basis functions are used for SSR and leaving others causes performances

to be slightly decay. It can be concluded that SSR method almost achieves its

upper limit performances which is the performances of ML detector 1. Even

though, both SSR and the proposed ML based detection techniques require

the same number of computations, SSR has slightly lower computation com-

plexity in each iteration. For example, ML detector 1 calculates the minimum

distance using the total number of samples of the received signal while SSR

does it using only 5 samples which are the coordinates in 5-dimensional space.

Finally, ML detector 3 shows similar performances to ML detector 1 at

higher SNR levels. Its performance at lower SNR though is poor compared

to other detection techniques. This pin points that the Gaussian distribution

model used for each real and imaginary part of the channel is not very accu-

rate at lower SNR levels.

Figure 4.30 compares the DER performances of all the detection techniques

when there is no guard-band presented between the tag resonator frequencies.


Similar to the case with a guard-band, ML detector 2 has the best performance.

It is interesting to notice that ML detector 4 achieves almost the upper margin

even though no channel information is available. SSR method for this scenario

uses only the most significant 7 base functions. It can be seen with 7 base

functions SSR method achieves its upper margin performances, which is the

performances of ML detector 1. This confirms that the proposed detection

methods are performing better than the SSR method.

However, at lower SNR levels (<10 dB), ML detector 3 performs worse

than ML detector 1. As the SNR increases (>10 dB), ML detector 3 performs

similar to ML detector 1. Similar to the case with a guard-band, even though

System Model 3 models the system quite poorly at lower SNR levels, it is a valid

detector for higher SNR levels. In addition, all ML based detectors performs

significantly better than the threshold based detectors used in existing chipless

RFID readers. Table ?? summarises the results of all investigated methods so

far.

Table 4.2: DER comparison for different detection methods

SNR 2 4 6 8 10 12

Threshold methodwith GB 7.0E-1 6.0E-1 4.0E-1 2.0E-1 8.0E-2 2.0E-2w/o GB 9.0E-1 8.0E-1 7.0E-1 6.0E-1 5.0E-1 3.0E-1

SSR methodwith GB 2.5E-1 1.0E-1 4.0E-2 6.0E-3 8.0E-4 N/Aw/o GB 2.2E-1 1.2E-1 4.6E-2 1.3E-2 2.7E-3 3.3E-4

Model Iwith GB 2.1E-1 1.0E-1 3.1E-2 5.5E-3 4.5E-4 3.0E-5w/o GB 2.2E-1 1.1E-1 4.7E-2 1.4E-2 2.6E-3 2.9E-4

Model IIwith GB 1.0E-2 1.0E-3 1.0E-4 3.0E-6 N/A N/Aw/o GB 2.0E-2 6.0E-3 2.0E-3 3.0E-4 2.0E-5 N/A

Model IIIwith GB 7.0E-1 2.8E-1 6.6E-2 7.7E-3 5.0E-4 N/Aw/o GB 8.0E-1 6.1E-1 2.6E-1 6.6E-2 1.0E-2 8.0E-4

Model IVwith GB 2.3E-2 4.9E-3 6.2E-4 4.6E-5 N/A N/Aw/o GB 2.6E-2 8.4E-3 2.1E-3 3.4E-4 3.0E-5 N/A

4.6 Results 95

0 2 4 6 8 10 12 14 16 18 2010

−6

10−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector I − w/o GBML detector II − w/o GBML detector III − w/o GBML detector IV − w/o GBSSR method

Figure 4.30: DER comparison without a guard-band

4.6.5 System model V

The likelihood detector derived under this model, uses only the magnitude of

the tag responses in frequency domain. This tag detection technique is derived

for an existing chipless RFID system that operates between 21 - 27 GHz. Figure

4.31 shows the tag response recorded by the chipless RFID reader.

The tag response received in Figure 4.31 together with all possible 16 com-

binations were used for the detection algorithm derived for System model V.

Table 4.3 shows the likelihood values obtained for each tag type. In order to

avoid underflow, probabilities were calculated in log to the base 10.

It can be clearly seen from the Table 4.3 that tag type [1111] has the high-

est likelihood out of the 16 combinations. Therefore, the proposed detection

technique has managed to detect the tag bits successfully.

A comprehensive analysis was performed using MATLAB simulations and

Figure 4.32 shows both the simulated and experimented DER variation at dif-


0

0.2

0.4

0.6

0.8

1

21 22 23 24 25 26 27

No

rmal

ized

po

we

r m

agin

itu

de

Frequency ( /GHz)

f1 f2f3 f4

Figure 4.31: Magnitude of the tag response for tag [1111]

Table 4.3: Likelihood for each tag type

Tag Type Likelihood in log scale Tag Type Likelihood in log scale

[0000] -35.41 [1000] -28.54

[0001] -28.50 [1001] -19.77

[0010] -27.76 [1010] -20.70

[0011] -22.87 [1011] -15.59

[0100] -29.76 [1100] -24.22

[0101] -22.47 [1101] -16.87

[0110] -22.28 [1110] -15.58

[0111] -17.31 [1111] -11.06

ferent SNR levels. In addition the results were compared with the threshold

based detection technique. It can be seen that the proposed tag detection tech-

nique is having superior reading accuracy compared with the existing thresh-

old based detection technique. It is also important to notice that the experi-

mental results agree with the simulated results. The experimental data was

gathered only for two SNR levels as achieving low DERs at higher SNR levels

are practically not feasible.

It can be concluded that the proposed tag detection technique for the mag-

4.7 Conclusion 97

0 2 4 6 8 10 12 14 16 18 2010

−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

Likelihood detector V − SimulationLikelihood detector V − ExperimentalThreshold based detector

Figure 4.32: DER vs SNR for likelihood based detector 5 for 21-27 GHzbackscattering tag

nitude based chipless RFID reader achieves higher reading accuracy over the

existing tag detection technique.

4.7 Conclusion

Detection error rate of a number of likelihood based detectors were presented

and compared against the threshold based detector used in existing chipless

RFID systems and the SSR method proposed in [91]. It is evident that all the

likelihood based detectors performs better than the threshold based detector.

ML detector 4 that jointly detects both the channel and the tag type achieves al-

most the optimum performance (ML detector 2). The improved performance

of the proposed tag detection techniques directly relates to an increased tag

reading accuracy at a given SNR level. On the other hand, it can also be repre-


sented as an increment in the reading range while achieving a particular goal

of reading accuracy. Therefore the improved performance can be represented

either as increased reading accuracy or the reading range depending on the

application.

However, there is a common drawback of all the likelihood based detec-

tion methods discussed so far. All these methods require higher computation

complexity compared to the primitive detection techniques such as threshold

based detection. For example a detecting a tag having N bits involves eval-

uating the likelihood expressions for 2N number of occasions. Two compu-

tationally feasible tag detection techniques have been introduced in Chapter 4

that can reduce the computation complexity from exponential to linear order

without compromising on the chipless RFID tag reading accuracy.

Chapter 5

Computationally Feasible TagDetection Techniques

5.1 Introduction

Maximum likelihood (ML) based detectors generally produce better tag detec-

tion performances as they utilize all information available, prior to making a

decision. However, one of the main drawbacks of ML detectors is its higher

computation complexity. In the proposed tag detection techniques, they com-

pare the received signal with all possible tag combinations and select the one

with the highest probability as the detected tag. If an RFID tag has N bits, they

compare the received signal with all 2N tag combinations to calculate individ-

ual probabilities. In the case of tags being used to identify the object category

not the object itself, the number of bits required in a tag can be small. In such

cases, direct application of the tag detection techniques presented in Chapter 4

may be feasible. However, when the number of bits in the tag is large, com-

putation complexity is increasing exponentially hence utilising the tag detec-

tion techniques presented in Chapter 4 may not be feasible. Higher computa-

tion complexity brings up two main challenges. Firstly, the RFID reader needs

higher computation capability to evaluate the likelihood expressions derived

in Chapter 4. The second challenge is the increased computation time, which

directly affect the tag reading time.

99

100 Computationally Feasible Tag Detection Techniques

Regarding the former challenge, it can afford to deploy higher computa-

tion complexity at the RFID reader, with ever evolving low cost single board

computers available in few tens of dollars. However, some of the repeated cal-

culations can be pre-processed and stored in a look up table for real-time use.

In addition memory required to store 2N tag responses is manageable with

few Mega Bytes. For example, if each tag response has 100 samples and each

sample is denoted by a signed short number format (16 bits), then the memory

required for each tag response can be calculated as 200 bytes. For 10-bit tags

there are 1024 tag combinations. Therefore the memory required for storing

all possible combinations for 10-bit tags is only 200 kilo bytes. Hence, it can

be argued that memory and the processing power can be overcome with the

readily available hardware.

However, in practical applications, 10-bit tags are not sufficient to tag indi-

vidual items, e.g. grocery items in a superstore. In such applications large data

bits in the order of 30-60 bits are needed to tag each item with its serial num-

ber. Therefore, efficient tag detection algorithm is to be developed for practical

applications.

Computation time is mainly dominated by 2N number of calculations per-

formed for each tag combination. This chapter describes two techniques that

is capable of reducing the total number of computations. The first techniques

reads bit by bit in the tag rather than reading all bits at the same time. The

second technique uses a trellis tree based Viterbi decoding method to reduce

the number of computations from 2N to 8× N.

The rest of the chapter is organized as follows. Firstly, the proposed bit

by bit tag detection technique is presented. Next, the trellis based Viterbi de-

coding method is explained. Then the simulation setup is described and the

results for each method are presented in the following section. Finally, the

chapter is concluded and recommendations for SISO chipless RFID tag read-

5.2 Bit by bit detection method 101

ing is presented.

5.2 Bit by bit detection method

The tag detection methods presented in Chapter 3 first evaluates the likelihood

expression for all the possible tag combinations and the one with the high-

est likelihood is selected as the detected tag. It has been proved in Chapter 3

that the proposed exhaustive detection methods work based on both time and

frequency domain samples. Frequency spectrum of the multi-resonator tags

has a unique advantage when it comes to observing the individual resonator’s

contribution to the overall tag response. Unlike in time domain tag responses,

frequency domain tag responses have dedicated blocks of samples represent-

ing contribution from each resonator. It is more evident when there is a guard

band to minimize any inter resonator interference (IRI). The guard band helps

to make a decision only based on the frequency samples in the given block.

Therefore, samples in each frequency block are observed separately and the

presence or the absence of the resonators in that block are detected using the

highest likelihood. The process is described in detail next.

The first step is to separate the frequency domain samples in to N blocks

representing the N number of tag data bits. Each block ( yk ) has only two

possibilities namely the response when a resonator is present (bit ‘1’) and a

resonator is absent (bit ‘0’). The kth resonator response in the frequency band

around the resonance frequency is given by Sk,1. The absence of resonance is

denoted by a signal (Sk,0) having a constant amplitude of 1 and a linear phase

variation. For example, a 4-bit tag would have at most 4 resonator dips. In each

block the tag response is compared against the response when a resonator is

present and absent. Figure 5.1 shows the comparison of overall tag response

with each resonator response in individual blocks. The comparison is shown


2.15 2.2 2.25 2.3 2.35 2.4 2.45 2.5 2.55 2.6 2.650

0.2

0.4

0.6

0.8

1

Frequency ( /GHz)

Nor

mal

ized

Mag

nitu

de

Bit 1Bit 2Bit 3Bit 4Tag [1111]

Bit 1 Bit 2 Bit 4Bit 3

Figure 5.1: Bit by bit detection for a tag having [1111]

only in the magnitude, however it can be extended to the phase as well. Phase

comparison has to perform with care. As there could be a phase offset in the

overall tag response compared to individual resonator response it should be

removed before performing the real and imaginary comparison.

Then the likelihood of receiving a bit ‘1’ and a bit ‘0’ is calculated using the

likelihood function derived in Chapter 4 and the decision is made based on the

highest probability. It can be represent as follows.

maxSi={Sk,1,Sk,0}

Pr(yk|Sk,i) (5.1)

This process is repeated for all N blocks until the all the tag bits are de-

tected. A flowchart of the bit by bit tag detection technique is presented in

Figure 5.2. The likelihood functions of all five frequency domain based detec-

tion techniques derived in Chapter 4 is compatible with the proposed bit by

bit detection method assuming there is a guard-band between the resonator

frequencies.

Since the tag bits are detected sequentially, this method can be termed as

5.2 Bit by bit detection method 103

Start

Separate frequency domain samples into N blocks representing N bits

Do Loop k = 0, N, 1

Divide kth block in to 2

Perform MLF

Si = Max Pr (Yk| Ski) {Sk1, Sk0} i=0,1

i = 0,1

Tag Detected

Yes

Stop

No

k = k+1

Figure 5.2: Flowchart of bit by bit detection technique


serial bit reading whereas the methods presented in Chapter 4 that reads all tag

bits simultaneously hence termed as parallel bit reading. The advantage of serial

bit reading is the less computation complexity compared to parallel bit reading.

In this method, detection of each tag data bit requires evaluating the likelihood

function two times; with the presence and the absence of the resonator. If N is

the total number of tag bits then the number of likelihood function evaluations

required in the serial reading is 2× N whereas parallel reading requires 2N it-

erations. It can be seen that the order of complexity in serial reading is linear

and which is a significant improvement over the exponential complexity pre-

sented in the exhaustive ML based detection techniques presented in Chapter

4. Also only the samples in a given frequency block is evaluated, the length

of the vectors used in the calculations are smaller hence less computation ca-

pabilities are required. In addition, the memory requirement to store the tag

responses have reduced from exponential (2N) to linear (2×N). As a result the

serial bit reading method also benefit from having less memory management in

the RFID reader.

However, one disadvantage of the serial bit reading method proposed, is

that this method assumes a guard band to minimise IRI. That limits the bit ca-

pacity available for the tags. Moreover, in this method only the information

embedded in magnitude was used for decision making, hence it is a subopti-

mal detector. With the presence of the guard band most of the data is confined

in each frequency block itself. However, there could be more data available

in the neighbouring frequency blocks in terms of interference. This method

does not utilise those extra information for decision making hence not an op-

timal detector and as a result this method can not be treated as a maximum

likelihood method. These shortcomings are addressed in the Trellis tree based

Viterbi decoding detection presented in the next section.

5.3 Trellis tree based Viterbi decoding 105

5.3 Trellis tree based Viterbi decoding

5.3.1 Introduction

In the bit by bit tag detection method presented in Section 5.2, assumes there

is minimal or no IRI. In other words, the presence of a guard band minimises

the interference from neighbouring resonators. Therefore, the above technique

is valid only when there are guard bands presented in multi-resonator tag de-

sign. As discussed earlier, this will limit the bit capacity for a given bandwidth.

Moreover, the detector presented in Section 5.2 is suboptimal as some use-

ful information is thrown out by not considering the neighbouring frequency

blocks. Trellis tree based Viterbi decoding technique discussed in this section,

can be treated as an optimal decoder because of its utilisation of the most use-

ful information available in the tag. Trellis tree creates a specific orthogonal

codes structure that eliminates redundant calculations and achieves exhaus-

tive likelihood performances using limited calculations.

Viterbi decoding [101] is a forward error correction decoding method. A

transmitter sends extra information along with data for better error perfor-

mance. It assumes the encoding scheme involves convolutional coding. In

traditional Viterbi decoding, a convolutional encoding system is selected and

that limits the number of allowed state transitions and their outputs. The ba-

sic idea here is to reduce the computational complexity by disregarding the

invalid state transitions. The number of output bits generated per input bit is

determined by the code rate used for encoding. For example, 12 rate involves

doubling the number of output bits compared to the input bits. However, in

chipless RFID systems, data bit capacity is relatively small and cannot afford to

have lower code error rates. Therefore the traditional Viterbi decoding cannot

be applied directly. A novel approach needs to be taken to smartly utilize the

Viterbi decoding algorithm. In the following section the proposed approach is


explained.

5.3.2 Signal model

As discussed earlier, multi-resonator based chipless RFID tags consist of a

number of resonators which are designed to resonate at unique frequencies

in the given bandwidth. The presence or absence of a resonator is used to

encode data bit ‘1’ or ‘0’ respectively. For minimum interference with neigh-

boring resonators, a guard-band is used between two resonance frequencies.

This will limit the number of resonances placed in the given bandwidth hence

limiting the total bit capacity in the tag. However, if the resonator frequencies

in the chipless RFID tag are placed close to each other without a guard-band

the individual resonances interfere with each other. This is very similar to in-

ter symbol interference (ISI) in communication channels [102]. The proposed

Viterbi algorithm is designed to perform under IRI which means no guard

band is required when designing the tags. This will allow improving tag data

bit capacity.

In frequency domain, the final tag response can be treated as the product

of individual resonator responses. If the final tag response has N number of

samples and L number of resonators, each resonator band has on average NL

number of samples.

It can be observed that when the guard band is removed, ith resonator re-

sponse (Xi) is mainly interfered by the neighbouring two resonators. The in-

terference from the resonators further away can be neglected. In physical de-

sign approach, resonators with neighbouring resonance frequencies are placed

apart to reduce inter resonator mutual coupling. Therefore, the actual response

at ith resonator response in frequency domain can be interpreted as a product

of 3 resonators as shown in 5.2. The assumption here is that individual reso-


nator responses are treated as orthogonal functions and the multiple resonator

responses can be calculated using the product of orthogonal individual reso-

nator responses. As a result, this method uses almost all the useful informa-

tion available unlike in the bit by bit detection method discussed in Section 5.2.

Therefore ith resonator response (S(i)) can be represented in 5.2.

S(i) = Xi−1 × Xi × Xi+1 (5.2)

For simplicity, Signal Model I with frequency domain samples is used to

explain the operation principle of Trellis tree based Viterbi decoding method.

Therefore, the received signal, y is given by,

y = hSm + ω.

Notations are as the same as in Chapter 4. Likelihood function for Signal

Model I is similar to the one derived in (4.8) with the exception of using only

a portion of the overall tag response for evaluation. As mentioned earlier, in

frequency domain, each resonator has a bandwidth given by NL number of

samples. Then likelihood function is evaluated using 3× NL number of sam-

ples. The modified received signal is given by yb. Sb is one of the 8 (23) possi-

ble combinations for the three resonators considered. The modified likelihood

function can be written as in (5.3) and it is optimized over the 8 possible com-

binations of Sb.

maxSb

Pr(yb|Sb) = minSb

((yb − hSb) (yb − hSb)

T)

(5.3)

Now that the likelihood for a given condition can be performed, the opera-

tion of the Trellis tree based Viterbi decoding algorithm is described next.

Figure 5.3 illustrates both the states and the transitions allowed in a Trellis


0 0 0

0 1

0 0

0 1

1 0

1 1

0 0

0 1

1 0

1 1

0 0

0 1

1 0

1 1

B0 B0 B1 B1 B2 B2 B3 BN-1 BN BN BN+1 BN+1

Next bit 0

Next bit 1

0 0

1 0

0

Figure 5.3: Operation of Trellis tree based Viterbi detection technique

tree. Most significant bit (MSB) of the tag data bits is considered as the first

bit and the least significant bit (LSB) as the last bit. A state is constructed from

two neighboring bits in the tag. The first bit in the state is always closer to the

MSB than the second. Each bit is repeated in the neighboring states as shown

in Figure 5.3. In each bit position, there is a maximum of four possible states

hence four rows in the diagram. Each column denotes the bit position in a tag.

Therefore, third column represents all possible states considering first (MSB)

and the second bits in the tag. It can be seen that at the two ends of the tree

only a limited number of states are valid. It is explained using the initial and

final conditions. As explained earlier, each resonator bit is interfered by the

neighboring bits. The first resonator does not have any neighboring resonators

to the left hence the initial condition (B0) for the analysis is assumed to be

having a bit ‘0’. Similarly, final condition (BN+1) is also assumed to be having

a bit ‘0’.

Although, there are many states in the Trellis tree, only a limited number

of transitions are possible. Figure 5.3 shows state transition from left to right

only, even though both directions are possible. Transition between two states

involves three bits. For example, transition between two states in [B1B2] and


[B2B3] associates first three bits of the tag (B1, B2 and B3). Transition between

each state depends on the next bit value. For example transition from state [0 0]

to state [0 1] requires the next bit to be bit ‘1’. No transition is possible from [0 0]

state to [1 0] state as the first bit of the next state must have current bit, which

is 0. It can be easily seen that a maximum of 8 transitions are allowed between

any two columns. The state transitions are further restricted at the two ends of

the tree with ‘0’ initial and final conditions.

Each state transition is associated with a likelihood given by (5.3). For ex-

ample transition from state [0 0] state in the third column to [0 1] state in the

fourth column assumes B1 = 0, B2 = 0 and B3 = 1. The likelihood for that

transition can be calculated using 3× NL long vectors of yb and Sb, which is the

frequency response obtained with only the first three resonators are presented

in the tag.

The algorithm starts from state [B0] and likelihood for each state transition

is calculated first. Until state [B1B2], each state has only one allowed transition.

Therefore, there is only one path from the initial state to each of the states in

[B1B2]. The total likelihood of the path can be calculated by taking the prod-

uct of each transition likelihood along the path. Likelihood of transiting via

this path is stored at each state and named as the state probability from now on-

wards. In addition to the state probability, states also store the previous state

of the path. However, any state in [B2B3] and onwards can be reached using

two different states of the previous column. Then the path from the initial state

to a state in [B2B3] has two options. Likelihood for each path is calculated and

the one with the highest likelihood is stored as the state probability. Since previ-

ous states already have the best possible path to the initial state, previous state

probabilities and the transition likelihood can be used for current state probabil-

ity calculation. Each of the path probabilities for current state is given by the

product between the previous state probability and the matching transition like-


0 0 0

0 1

0 0

0 1

1 0

1 1

0 0

0 1

1 0

1 1

0 0

0 1

1 0

1 1

B0 B0 B1 B1 B2 B2 B3 BN-1 BN BN BN+1 BN+1

Next bit 0

Next bit 1

0 0

1 0

0

Decoding path

0

1

01

0

Figure 5.4: Viterbi decoding in a Trellis tree

lihood. Then the highest state probability among the two paths is stored in the

state together with the respective previous state.

Once all the state probabilities are calculated and the final state is reached, the

decoding algorithm works from final state towards the initial state as shown

in Figure 5.4. Final state has stored its state probability and the previous state

(BNBN+1 = [1 0]) with the highest path likelihood. The common bit between

previous state (BNBN+1 = [1 0]) and the final state (BN+1 = ‘0′) is the decoded

bit in the optimum path. Therefore BN+1, which is the final condition is de-

coded correctly as ‘0’. Similarly, if the previous state stored in (BNBN+1 = [1 0])

is (BN−1BN = [1 1]) then the common bit between the two states is (BN = ‘1′)

and the LSB of the tag can be decoded as ‘1’. The process of traversing the

most likelihood path continues until the initial state is reached as shown in

Figure 5.4. Along the way tag bits are decoded from LSB to MSB. The Trellis

tree based Viterbi decoding process is depicted in Figure 5.5

At the beginning we assumed that once the guard band is removed, a res-

onators is interfered only by the 2 neighboring resonators. Therefore all infor-

mation required to decode a certain bit is only available in the current and its

two neighboring resonators. Since, Trellis tree based Viterbi decoding is us-

ing these three resonators to make the decision, it can be treated as an optimal


Start

Set Bo = 0

Calculate LF for each transition

Calculate/Store state probability

Store highest state probability

Is k ≥ N+1

Start decoding algorithm

Store data bit

Stop

Do loop k=1, N+1, 1

Set Bk

Bk+ . Bk

k = k +1

Yes

No

Figure 5.5: Flow chart of Trellis tree based Viterbi decoding


(maximum likelihood) detection technique with a less computation complex-

ity. Reduction in the computation complexity is achieved by disregarding the

transitions that are not valid. The total number of likelihood function evalua-

tions (valid transitions) is 8N − 4. It can be seen that the proposed detection

method is able to reduce the computation complexity from exponential order

to a linear order without compromising tag bit capacity.

Then a simulation setup was designed using CST and MATLAB to verify

the proposed tag detection techniques. The simulation setup is described next.

5.4 Simulation setup

The validity of the above tag detection techniques were verified using CST and

MATLAB simulations.

CST Simulation:

The steps carried out in the simulation are given in Figure 5.6. Firstly, an inter-

rogating signal was generated to provide a flat frequency response in 2.2-2.6

GHz frequency range. Four resonators were designed using CST with resonat-

ing frequencies as shown in Table 5.1. Then the combinations of resonators

were placed besides a micro-strip line to cover all possible tag IDs. One end of

the micro-strip line was fed with the interrogating signal and the tag responses

were collected at the other end. These collected tag responses were saved in

a look up table for the algorithms to be used later. More details about the tag

design is available in Chapter 3.

5.4 Simulation setup 113

Figure 5.6: Flowchart of the MATLAB simulation

MATLAB Simulation:

Then the tag responses were fed through a channel which is given by the prod-

uct of the forward and reverse channels of the RFID system. In order to prove

the concept, only the detection technique derived for System Model II is used

which utilise information embedded in both the amplitude and phase. The

same principle can be extended to other models as well. The channel realisa-

tion for System Model II is taken from a normalised distribution with a certain

mean and a variance. Finally Gaussian noise is added to the resultant signal

according to the specified SNR. Simulation parameters are given in Table 5.1.

SNR is calculated compared to the average power of all the tag combina-

tions. It can be summarised as follows.


I(t) – interrogating signal

h f – forward channel

hr – reverse channel

h – product of forward and reverse channels

Fm(t) – impulse response of the mth filter

Sm(t) – mth filter response

y(t) – received signal

ω(t) – noise added at the reader

Then the received signal can be represented using (5.4).

y(t) =[[h f I(t)] ∗ Fm(t)

]× hr + ω(t)

= h f hr × [I(t) ∗ Fm(t)] + ω(t)

= hSm(t) + w(t)

(5.4)

The power of each tag response is calculated and averaged to obtain the

average power of a given tag response. For example, 2-bit tags have four dif-

ferent tag responses (Sm(t)) and power of each tag response was calculated

and averaged to obtain the average power of 2-bit tag responses. Then the

average tag response power was multiplied using the channel to calculate the

average signal power available at the reader. For a given SNR, noise power is

calculated using this available signal power.

MATLAB simulation parameters are outlined in Table 5.1. Four band-stop

filters were used, and MSB corresponds to the lowest resonance frequency and

LSB to the highest.

I & Q demodulation is performed with the received signal at the RFID

reader. Then the two output time domain signal vectors were used to evaluate

likelihood expressions for each detection technique. In order to verify their

frequency domain performances the time domain vectors were converted us-

5.5 Results 115


Parameter Value

Centre Frequency 2.4 GHzTotal bits encoded in a tag 4 bitsFlat frequency response 400 MHzBand-stop filter attenuation 10 dBBand-stop filter 3dB bandwidth 50 MHzGuard band 50 MHzResonance frequency set 1 (MSB to LSB) [2.2, 2.3, 2.4, 2.5] GHzResonance frequency set 2 (MSB to LSB) [2.34, 2.38, 2.42, 2.46] GHzChannel mean 0.4Channel standard deviation 0.1No. of iterations up to 10,000,000

ing FFT. Then these frequency domain samples were used for tag detection.

Finally the DER is calculated for each tag detection technique at different SNR

levels. Existing chipless RFID systems use a threshold based detection tech-

nique based on frequency domain based samples. The presence and absence

of each resonator in this method is detected based on a magnitude threshold

at corresponding resonator frequency bands. In the MATLAB simulation, this

threshold based detection technique is implemented and the DER is calculated

to compare the performances of the proposed detection techniques. In addi-

tion, the average computation time for one reading using the above MATLAB

simulation setup is also calculated.

5.5 Results

The simulations were carried out as above and the results are presented in

the next two subsections. The first subsection presents the detection error rate

performances under different noise levels. The second subsection compares

the computation time for each detection method.


0 2 4 6 8 10 1210

−5

10−4

10−3

10−2

10−1

100

SNR /dB

Det

ectio

n E

rror

Rat

e

ML detector II − with GBML detector II − w/o GBTrellis detector II − with GBTrellis detector II − w/o GBBit by Bit detection − with GB

Figure 5.7: DER comparison for 10-bit tags

5.5.1 Detection error rate (DER)

System Model II in Section 4.2.2 utilises all the information available when mak-

ing a decision. Therefore, System Model 2 is used for verification in this section.

The expression derived in (4.14) was used to evaluate the likelihood of a tran-

sition from one state to the other in Trellis tree diagram.

Figure 5.7 compares the DER results for System model 2 under 5 cases.

Firstly, it presents the DER using exhaustive ML detection method derived in

Chapter 4 with the presence and absence of a guard-band. Then the Trellis tree

diagram having less computation complexity is used to calculate DER with

and without a guard-band. Finally above four cases were compared with bit

by bit detection method when a guard-band is presented.

It can be seen that Trellis decoding method has very similar results to the

fully optimal detection method, i.e. ML detector 2. As a result, it can be con-

cluded that the Trellis decoder is a fully optimal detector. As expected bit by

5.5 Results 117

1 2 3 4 5 6 7 80

50

100

150

200

250

300

No. of tag bits

No.

of e

valu

atio

ns r

equi

red

Bit by Bit detectionTrellis decodingOriginal ML

Figure 5.8: Computation complexity comparison

bit detector has the poorest performance.

5.5.2 Computation time

The main drawback of exhaustive likelihood based detection techniques is its

exponential computation complexity as the number of tag bits increase. Fig-

ure 5.8 presents the number of computations required for each tag detection

type against the number of tag data bits. It can be noticed that both the Trel-

lis decoding and bit by bit detection technique has linear computation com-

plexity against the exponential computation complexity of original likelihood

detection methods. As expected, bit by bit detection technique has the low-

est computation complexity whereas Trellis decoding provides a manageable

complexity at higher tag bits.


Based on the DER and computation complexity results presented above, it

can be concluded that, bit by bit detection method is preferred under very high

SNR environments, with larger tag bits having a guard-band. In low tag data

capacity applications, original ML detection techniques are preferred. Trellis

decoding technique is the favorable choice when the tag data capacity high.

5.6 Conclusion

Computationally feasible two tag detection techniques have been proposed to

reduce the computation complexity from exponential to linear in Chapter 4. It

was found that the bit by bit detection method, which is a suboptimal detec-

tion method performs successfully when a resonator guard-band is used in

tag design. It was shown that the computation time has significantly dropped

compared to exhaustive maximum likelihood detection methods without com-

promising the reading accuracy. In addition, a fully optimal Trellis tree based

Viterbi decoding technique has been introduced to reduce the computation

complexity from exponential to linear order while achieving similar reading

accuracy to original likelihood detection techniques.

Chapter 6

Signal Processing for MIMO basedChipless RFID Systems

6.1 Introduction

Under literature review, several chipless tags have been studied and it was

concluded that, multi-resonator based chipless tags will be further investi-

gated in order to incorporate them with the proposed MIMO based chipless

RFID system. The main reason is that the multi-resonator tags are reported

as one of the chipless tag types with the largest data capacity. Also, Monash

Microwave, Antennas, RFID and Sensors Laboratory (MMARS) has all the re-

quired equipment and facilities for tag fabrication and testing.

An overview of the proposed MIMO based chipless RFID system is illus-

trated in Figure 6.1. The chipless tag used in the proposed system is multi-

resonator based, where each resonator acts as a band-stop filter introducing a

frequency signature to the tag.

The tag considered above has one receiving antenna (Rx) and two trans-

mitting antennas (Tx1 and Tx2) cross polarized with Rx as shown in Figure

6.1. The RFID reader has one transmitting antenna (Tx) that is cross polarized

with its two receiving antennas (Rx1 and Rx2), hence minimizing the coupling

between transmitting and receiving antennas. The transmitting antenna of the

reader (Tx) and the receiving antenna of the tag (Rx) are co-polarized which

119

120 Signal Processing for MIMO based Chipless RFID Systems

Rx

Tx2

Forward Channel

Tx1

MIMO RFID Tag Reverse Channel

Tx

Rx2

Rx1 RFID

Reader

Figure 6.1: MIMO based chipless RFID system

are already cross polarized with the transmitting antennas of the tags (Tx1 and

Tx2) so that, the undesired coupling throughout the system is minimized. As

a result, when the reader transmits, it is safe to assume that only Rx of the tag

receives the signal, hence forming a Single Input Single Output (SISO) channel

from the reader to the tag named as the forward channel hereafter.

Then the received signal at the tag will be divided into two parts using an

equal power divider. Each RF component will then be traveled towards its

transmitting antenna surrounding the frequency resonators as shown in 6.2.

When a RF signal travels surrounding the resonators, the resonators start to

resonate at their resonating frequency, hence losing the power in the corre-

sponding frequency of the signal. Once the RF signal reaches the end of the

transmission line, it contains the frequency signature of the tag and this pro-

cess is called tag modulation hereafter. The tag modulated RF signals will

then be transmitted back to the reader using each transmitting antenna of the

tag. There will be two different signals transmitting from the tag towards the

reader and the reader will receive them using two receiving antennas (Rx1 and

6.1 Introduction 121

Figure 6.2: MIMO tag

Rx2) having the same polarization to that of the transmitting antennas of the

tag. This forms a 2x2 MIMO channel and is called reverse channel hereafter.

Tag detection involves two stages. First stage is about MIMO decomposing.

When the RFID reader receives two streams of signals from its two receiving

antennas, those signals are already mixed with the 2x2 MIMO channel. First

MIMO decoding techniques are used to decompose these already mixed two

signal streams. Once they are separated, in second stage, ML based tag detec-

tion techniques are used to identify encoded tag data.

The resonator combination used in each branch of the MIMO tag can per-

form tag modulation individually. As a result, the bit capacity can be improved

by several factors if the tag contains multiple branches. For example, a MIMO

tag with two branches can double the tag bit capacity using the same frequency

band of the SISO tag. This becomes achievable thanks to the MIMO decompos-

ing algorithms, which will be discussed in next section.


6.2 MIMO decomposing techniques

The signal model used in the proposed MIMO based chipless RFID system is

shown in Figure 6.3 and explained in this section.

Figure 6.3: MIMO tag operation overview

If the interrogating signal is denoted by x(t) and the forward channel by h f

then, the received signal at the tag rt(t) can be represented using (6.1) where

as nt(t) is the noise added by the receiving antenna at the tag.

rt(t) = h f x(t) + nt(t) (6.1)

Then the received noisy signal is divided into two power-equal compo-

nents tx′1 and tx′2.

tx′1 = tx′2 =1√2

rt(t)

These two identical signals are tag modulated using independently selected

resonator combination. Assume that the equivalent bandstop filters of the

resonator combinations in each branch are given by f1(t) and f2(t). Then the

final tag response in each branch (tx1(t) and tx2(t)) can be calculated as (6.2).

6.2 MIMO decomposing techniques 123

tx1(t) = f1(t) ∗ tx′1

= f1(t) ∗(

1√2

(h f x(t) + nt(t)

))tx2(t) = f2(t) ∗ tx′2

= f2(t) ∗(

1√2

(h f x(t) + nt(t)

))(6.2)

When the tags are located closer to the reader the signal power is consider-

ably higher than the noise added by the receiving antenna of the tag. There-

fore, noise can be neglected and the above expressions further simplify to (6.3).

S(1)m (t) and S(2)

m (t) are the resultant signals after tag modulation in branches 1

and 2 respectively which are independent of the forward channel.

tx1(t) ≈1√2

h f

(f1(t) ∗ x(t)

)≈ 1√

2h f (t)S

(1)m (t)

tx2(t) ≈1√2

h f

(f2(t) ∗ x(t)

)≈ 1√

2h f (t)S

(2)m (t)

(6.3)

If the tag responses in (6.3) has L number of samples then these two re-

sponses can be stacked to form a matrix tx(t)2×L as shown (6.4).

tx(t) =

tx1(t)

tx2(t)

2×L

=1√2

h f

S(1)m (t)

S(2)m (t)

2×L

(6.4)

The received signals (rx1(t) and rx2(t)) at the reader antenna array can be

represented using a matrix as follows.

rr(t) =

rx1(t)

rx2(t)

2×L


If reverse channel is given by hr2×2 the received signal array at the RFID

reader can be calculated using (6.5). H2×2 is the product of the forward and

the reverse channel weighted by a factor of 1√2

and the noise matrix added by

the both receiving antennas of the reader is given by nr(t)2×L.

rr(t)2×L = hr2×2 × tx(t)2×L + nr(t)2×L

=1√2

h f hr2×2 ×

S(1)m (t)

S(2)m (t)

2×L

+ nr(t)2×L

= H2×2 ×

S(1)m (t)

S(2)m (t)

2×L

+ nr(t)2×L

(6.5)

Assuming the channel, H2×2 is known to the RFID reader, the estimated

tag responses can be calculated using standard MIMO decomposing methods.

In this section two methods are presented namely, zero forcing (ZF) equalizer

and the minimum mean square error equalizer (MMSE) as presented in (6.6).

WZF =

(H#H

)−1

H#

WMMSE =

(H#H + N0 I2×2

)−1

H#

(6.6)

H# is the Hermitian transpose of the channel matrix H and N0 is the noise

power available at the reader which is calculated using SNR.

The ML based tag detection technique derived in Section IV assumes a

Gaussian distribution for noise available at the reader. We have used MMSE

equalizer as the decomposing technique in prior work [81]. The results are

shown in Section 6.6.1 under method 1. Even though it leads to better signal to

interference plus noise ratio (SINR) performance, in the process it makes noise

distribution to be bimodal hence no longer can be treated as Gaussian. On

the other hand, the noise produced after ZF method still follows a Gaussian

distribution subjected to an amplified noise. Modeling the bimodal noise dis-

6.3 Tag detection in MIMO 125

tribution to derive a likelihood based detector makes the signal processing ex-

tremely complex. Therefore, only the ZF equalizer is used with the likelihood

detector and the results show that even under this scenario the ML detection

technique performs better than the performance reported in [81]. This method

is named as method 2 and the results are presented in Section 6.6.2.

After applying the ZF equalizer, an estimated tag response for each branch

can be obtained as shown in (6.7). Y2×L(t) represents the estimated tag re-

sponses in each branch.

Y2×L(t) = WZF × rr(t)2×L (6.7)

The estimated tag response in (6.7) is derived for time domain based signal

samples. However, the above relationship will still be applicable if a unitary

transformation such as Fourier transform is performed. As a result, the esti-

mated tag responses in frequency domain can be calculated as follows.

Y2×L( f ) = WZF × Rr( f )2×L (6.8)

An ML based tag detection technique is derived which is later applied on

these estimated tag responses. Derivation of the tag detection technique is

discussed in next section.

6.3 Tag detection in MIMO

In this section we derive an expression for the maximum likelihood (ML) func-

tion, to detect which resonator combination (notch filters) the signal has gone

through. ZF equalizer derived in previous section produces an estimated tag

response for each branch in the MIMO tag. However, it could also amplifies

the noise during the decomposing. As a result the output of the ZF equaliser


can be expected to be noisy. The task is to find which resonator combination

has the highest maximum likelihood, out of the all the possibilities. For exam-

ple, if each branch had N resonators then there are 2N unique tag responses in

each branch. Therefore, each estimated tag response should be compared with

2N tag responses and select the one with the highest likelihood.

The signal model used is explained below. If Sm is the mth tag response

vector out of all the 2N number of combinations and ω is the noise vector

available after ZF equalizer, then the estimated tag response vector, y is given

by,

y = Sm + ω.

Due to I & Q demodulation, these signals are complex and they can be repre-

sented using real and imaginary components ([.]r and [.]i) as follows.

y = yr + jyi

Sm = Sm,r + jSm,i

ω = ωr + jωi

Therefore, the received signal can be written as,

yr = Sm,r + ωr

yi = Sm,i + ωi

(6.9)

Each noise sample in ω is assumed to have an independent and identical dis-

tribution (i.i.d.) with zero mean and a variance of σ2. Original noise added

at the reader is assumed to be Gaussian due to the reader architecture. As

pointed out in the previous section, ZF equaliser may only amplify the noise.

As a result, the noise after the equaliser can still be treated as Gaussian. There-

fore, it was assumed both the real and imaginary parts of each noise sample

(ωi) has a Gaussian distribution given by, ωi ∼ N (0, σ2). Then for calculation

6.3 Tag detection in MIMO 127

purposes, the results can be vectorised as (6.10).

ωr ∼ N (0, σ2 IN)

ωi ∼ N (0, σ2 IN)(6.10)

IN is the identity matrix with dimensions of N × N. A new real valued vector,

y0 is created using yr and yi as follows.

y0 =(yr , yi

)(6.11)

Mean and covariance of y0 can be calculated as follows.

E[y0] = µ =[Sm,r , Sm,i

]Cov[y0] = E[(y0 − µ)T(y0 − µ)]

= σ2 I2N

(6.12)

I2N in (6.12) is the identity matrix with a dimension of 2N × 2N. Using the

statistical properties calculated in (6.12) the distribution of the real vector, y0

can be represented as follows.

y0 ∼ N (µ , σ2 I2N)

Using probability theory, the conditional probability of receiving y given that

Sm has been transmitted, can be derived as (6.13)

Pr(y0|Sm) =1

2π√|Cov[y0]|

×

exp(− 1

2× (y0 − µ) Cov(y0)

−1 (y0 − µ)T)

=1

2πσexp

(− 1

2σ2 (y0 − µ) (y0 − µ)T) (6.13)


(6.13) is evaluated for all the possible tag combinations and the one with

the highest probability is taken as the detector output. However, it can be seen

that the detector can be further simplified to minimising the exp(.) component.

Therefore, the objective function of the detector can be represented as follows.

maxSm

Pr(y0|Sm) = minSm

((y0 − µ) (y0 − µ)T

)(6.14)

Under the assumptions followed for the proposed signal model, it can be

proved that the optimum detector is the same as the minimum distance detec-

tor. However, y0 and µ can be calculated using (6.11) and (6.12) respectively. In

addition, this expression is valid for frequency based samples as unitary trans-

formations like FFT does not change the statistical properties of the signals.

6.4 Experimental setup

An experiment was conducted to measure the MIMO tag response using an ar-

bitrary waveform generator (AWG) and an oscilloscope with a high sampling

rate as shown in Figure 6.4. However, antennas were replaced using cables as

the sole purpose of this work is to verify the validity of the ML based detection

method.

Figure 6.5 shows the CST generated tag response and the measured tag

response for tag bits [1010]. It can be seen that they are closely matched.

An 8-bit MIMO tag is fabricated to have tag bits [1010] in the first branch

and [0000] in the other. An AWG was used to generate the interrogating signal

at 2.4 GHz and oscilloscope with 20 GSamples/sec was used to capture the

tag responses available at each branch. ML based tag detection was performed

only on the 1st branch for demonstrative purposes and the same technique can

be applied to the 2nd branch too. Table 6.1 shows the distance value obtained

for each tag type after evaluating the expression in (4.14). It can be clearly

6.4 Experimental setup 129

Figure 6.4: MIMO tag experiment

0

0.2

0.4

0.6

0.8

1

2.1 2.2 2.3 2.4 2.5 2.6 2.7

No

rmal

ize

d A

mp

litu

de

Frequency ( GHz)

CST generated tagresponse

Experimental tagresponse

Figure 6.5: Tag response for [1010]

seen that minimum distance occurs at the tag type [1010]. A comprehensive

analysis was carried out using MATLAB simulations, as it is not feasible to

take large number of experimental data.


Table 6.1: An Example of a Table

Tag Type Distance in (6.14) Tag Type Distance in (6.14)

[0000] 10.84 [1000] 4.73

[0001] 7.57 [1001] 4.12

[0010] 5.03 [1010] 0.56

[0011] 7.89 [1011] 2.80

[0100] 4.49 [1100] 1.62

[0101] 3.58 [1101] 2.75

[0110] 3.51 [1110] 2.61

[0111] 5.74 [1111] 5.29

6.5 Simulations

Simulations were carried out in two methods. In the first method, tag re-

sponses were generated in MATLAB using bandstop filters. The integrating

signal is generated using orthogonal frequency division multiplexing (OFDM)

techniques. The MIMO decomposing technique explained earlier was imple-

mented in MATLAB and the tag responses in each branch of the MIMO tag

were estimated. In this method, tag detection was performed using a thresh-

old based valley detection technique applied on the power spectral density of

the estimated tag responses. This information was used to calculated the DER

at different SNR levels.

In the second method, both the interrogating signals and the tag responses

were generated using CST simulations. The MIMO decomposing technique

and the tag detection technique described in the previous section was imple-

mented in MATLAB. Finally the DER at different SNR levels were calculated.

Firstly, the details about the first method is described next.

6.5 Simulations 131


6.5.1 Method 1

The signal path from the RFID reader through the MIMO tag back to the reader

is modeled in MATLAB using the baseband signal representation. As briefly

explained earlier, in this method firstly, the interrogating signal is generated

using OFDM techniques. This was achieved using binary phase shift keying

(BPSK) modulated test bits. All 200 test bits were selected as ’1’. Other OFDM

parameters used in MATLAB simulations are displayed in Table (6.2).

Then the resonators were emulated using bandstop filters in MATLAB and

the filter attenuation is selected as 10 dB. Center frequencies of the bandstop

filters were selected as shown in Table 6.2. Multiple resonators were emulated

using cascaded bandstop filters. Then the tag responses were fed through a


channel which is given by the product of the forward and reverse channels

of the RFID system. In the MATLAB simulations, the channel is assumed to

be known at the RFID reader. Finally noise is added to the resultant signal

according to the specified SNR.


Parameter Value

No. of BPSK modulated of test symbols 200OFDM block size 25Length of Cyclic Prefix 2No. of FFT / IFFT points 25Sampling frequency 600 MHzOFDM signal bandwidth 300 MHzTotal bits encoded in the tag 6 bitsBand-stop filter attenuation 10 dBNumber of branches in the MIMO tag 2Resonance frequency set (MSB to LSB) [50, 150, 250, 350, 450, 550] MHz

Baseband signals received at the RFID reader is used for MIMO decompos-

ing. Then ZF equalizer is used to decompose the received signal and obtain

an estimate of the tag response in each branch. Then a threshold based valley

detector is used to identify the presence and absence of resonators which will

be used to read the encoded tag data bits. Finally the DER is calculated for

each tag detection technique at different SNR levels.

However, the calculation of tag responses using bandstop filters does not

take into account any coupling between the two branches in the tag. There-

fore a more realistic simulation is performed using CST. In addition, valley

detection technique used to identify the tag data bits is very primitive and

can be further improved using advanced tag detection techniques. The second

method explained in next section rectify these limitations.

6.5 Simulations 133

6.5.2 Method 2

As explained in previous section, method 2 uses CST to simulate tag responses

which takes in to account any coupling between the two branches in the MIMO

tag. The tag responses generated using CST simulations are more realistic than

bandstop filter emulation performed in previous method. Then the validity of

the tag detection technique derived in Section 6.3 was verified using MATLAB.


Parameter Value

Center Frequency 2.4 GHzTotal bits encoded in a tag 4 bitsFlat frequency response 400 MHzBand-stop filter attenuation 10 dBGuard band 50 MHzResonance frequency set (MSB to LSB) [2.2, 2.3, 2.4, 2.5] GHzNo. of iterations 1,000,000

The steps carried out in the simulation are given in Figure 6.7. Firstly, an in-

terrogating signal was generated to provide a flat frequency response in 2.2-2.6

GHz frequency range. Four resonators were designed using CST with resonat-

ing frequencies as shown in Table 6.3. Then the combinations of resonators

were placed besides a microstrip line to cover all possible tag IDs. One end of

the micro-strip line was fed with the interrogating signal and the tag responses

were collected at the other end. These collected tag responses were saved in a

look up table for the algorithms to be used later.

Then the signal flow from the MIMO tag to the reader was modeled in

MATLAB and the channel information is assumed to be available at the reader.

ZF equalizer was used to decompose the tag responses in each branch of the

MIMO tag. Then the tag detection technique derived in Section 6.3 was used to

identify the encoded tag data bits. Finally the detection error rate was calcu-

lated at different SNR levels. The results obtained in each simulation method



is discussed next.

6.6 Results

In this section, results obtained using the two simulation methods are pre-

sented. Firstly, the results of the method using OFDM technique and bandstop

filters is discussed.

6.6.1 Method 1

In the simulation the test bits were selected as all ones and they were BPSK

modulated. Then the BPSK modulated signals were OFDM modulated to gen-

6.6 Results 135

Figure 6.8: Interrogating signal in time domain

erate the interrogating signal in time domain as shown in Figure 6.8. When the

test bits were taken as all ones, the resulting interrogating signal will be similar

to having a train of impulses.

The two sided power spectral density (PSD) of the interrogating signal is

shown in Figure 6.9. It can be considered that the signal has a flat frequency

response throughout the signal bandwidth of 300 MHz.

This interrogating signal was then transmitted through a SISO channel to

the tag and the time domain representation of the received signal is shown in

Figure 6.10.

Then received signal at the tag will be divided into two equal RF com-

ponents and each component will travel surrounding spiral resonators. The

spiral resonators were implemented as band-stop filters in Matlab and Figure

6.11 shows the filter response of such a spiral resonator. All together three res-

onators were emulated at 47, 150 and 253 MHz. The presence of a resonator


Figure 6.9: Two sided PSD of the Interrogating Signal

Figure 6.10: Received Signal at the Tag

6.6 Results 137

Figure 6.11: Filter response of a spiral resonator

was represented as bit 1 while the absence as bit 0. As there are 2 components

which are tag modulated independently, the considered prototype has a ca-

pacity of 6 bits.

Once each component reaches its transmitting antenna, it contains the fre-

quency signature of all the spirals presented along the way. Figure 6.12 shows

the two-sided PSDs of each signal (Tx1 and Tx2) after traveling via spiral res-

onators. Top graph in Figure 6.12 represents the case of having only 2 res-

onators at 47 and 253 MHz (corresponds to [101]) while the bottom represents

having 3 resonators at 47, 150 and 253 MHz (corresponds to bits [111]). It can

clearly be observed the presence and the absence of the spiral resonators in

each branch.

The time domain representation of the above two signals, is shown in Fig-

ure 6.13. The noise introduced by the receiving tag antennas are visible at the

two signals already.


Figure 6.12: Two-sided PSD of the tag modulated signals (Tx1 and Tx2)

Figure 6.13: Tag modulated signals (Tx1 and Tx2) in time domain

6.6 Results 139

Figure 6.14: Channel realizations

Then the two transmitted signals were propagated via a 2x2 MIMO chan-

nel. Figure 6.14 shows the channel realizations for both forward and reverse

channels. As a result of Rician distribution the four MIMO channel realizations

looked very similar. After being mixed with the channel, the received signals

at the receiver antenna array are shown in Figure 6.15.

The received signal array will then be decoded using MMSE equalizing

method and the estimated transmitted signals (Tx1H and Tx2H) are obtained.

Figures 6.16 and 6.17 compare the actual and the estimated transmitted sig-

nals in time domain for Tx1 and Tx2 respectively. It can be observed that the

estimates were very similar to the actual signals.

The two-sided PSDs of each of the estimated transmitted signals (Tx1 and

Tx2) are shown in Figure 6.18. A separate algorithm was implemented to de-

tect the presence and absence of the frequency dips at the resonating frequen-

cies. Using the algorithm, estimated data bits were obtained and were com-


Figure 6.15: Received Signals at the two Rx antennas of the Reader

Figure 6.16: Actual and the Estimated Tx1

6.6 Results 141

Figure 6.17: Actual and the Estimated Tx2

pared with the data bits encoded in the tag.

The simulation was repeated for 100 times and Figure 6.19 shows the com-

bined tag response obtained for 100 iterations. It can be concluded that for

different channel realisations, the performances are consistent.

Simulations were carried out to investigate the BER performance under dif-

ferent signal to noise ratios (SNR). SNR is defined as the signal to noise ratio

at each of the receiving antenna array at the reader. Figure 6.20 shows the bit

error rate (BER) performance of the proposed system versus SNR and also a

comparison to the theoretical BER of a traditional BPSK modulation scheme.

Even though, the definition of the BERs in two schemes are different, it is in-

teresting to learn that the simulated system performances closely follows the

theoretical BER performance for a BPSK modulated 2x2 MIMO system. In tra-

ditional BPSK modulation schemes, the bits are modulated into either raised

cosine symbols or no signal at all based on the bit value. In the power spectral


Figure 6.18: Combined Tag Response

Figure 6.19: Combined Tag Response for 100 iterations

6.6 Results 143

Figure 6.20: BER of the Proposed System vs SNR

density it is similar to be represented using the presence and absence of a val-

ley. As a result, the two schemes can be compared and should display similar

performance which they do.

Figure 6.21 shows a comparison between the proposed 2x2 MIMO system

and the traditional SISO multi-resonator based chipless RFID system. In this

simulation, the same data bits were encoded in both the branches introduc-

ing diversity. Compared with the traditional system, it is evident that the two

branches in the MIMO multi-resonator tag causes less errors due to the ex-

tra reliability in the proposed system. This extra reliability can also be seen

differently as encoding more data bits in the tag with an acceptable reading

accuracy. Apart from that, the diversity gain of the MIMO system is clearly

visible compared to linear variation of the SISO system.


Figure 6.21: Noise Performance of the Proposed System vs SISO counterpart

6.6.2 Method 2

The results obtained using both CST and MATLAB simulations are presented

and discussed in this section. The resonator response simulation is similar to

that of the SISO tag simulation. Figure 6.22 shows the resonator response of

a branch when all four resonators are presented. It can clearly see the four

resonances at the designed frequencies 2.2, 2.3, 2.4 and 2.5 GHz.

The received signal at each receiving antenna of the reader is calculated

using (6.9) under different noise power levels. Then the received signal array

was decomposed using ZF decoder as shown in (6.7) to calculate the estimated

tag responses in each branch. Finally the likelihood based detector derived

in (6.14) is used to identify the tag data bits. Figure 6.23 compares the DER

against SNR for the two methods.

It can clearly see that the likelihood based detection technique (method 2)

is performing better than the valley detection based technique in method 1.

6.6 Results 145

2.1 2.2 2.3 2.4 2.5 2.60

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Frequency /GHz

Nor

mal

ized

Am

plitu

de

Figure 6.22: CST generated tag response for a branch having [1111] tag bits

5 7 9 11 13 1510

−7

10−6

10−5

10−4

10−3

10−2

SNR /dB

DE

R

Method 1Method 2

Figure 6.23: Comparison of DER performances for 6 bit tags


For example, at SNR = 10 dB, method 1 has a DER of 99.8 % while the method

2 produces an accuracy of 99.99 %. At higher SNR levels both methods per-

forms better for obvious reasons. It can be concluded that the likelihood based

detection method performs better than the valley detection method at all SNR

levels. However, method 2 assumes perfect channel state information. In case

if there is errors in channel estimation, that could magnify the noise amplitude

with ZF decoder and performances could be degraded.

6.7 Conclusion

This chapter presented signal processing techniques for successfully detect-

ing the tag bits of a MIMO based chipless RFID system. Firstly, zero forcing

decomposing techniques were used on the received data array at the reader

to estimate the tag response. Then two methods were used to detect the tag

data bits encoded in the MIMO tag. The first method uses a threshold based

valley detection method. The other method uses the proposed ML detection

technique to detect the tag bits in each branch of the MIMO chipless tag. An

experiment was setup to test the ML detection technique and it was shown

that the encoded tag data bits were identified successfully. In order to per-

form a comprehensive analysis, a CST and MATLAB based simulation was

performed. The results show that the proposed detection technique provides

better detection error rate performance at different SNR values over traditional

threshold based detection. This benefit can be interpreted in two different met-

rics. Firstly, it can be seen as an SNR gain over the existing threshold based de-

tection technique which effectively increase the reading range. Secondly, the

high accuracy in tag reading avoids multiple reading cycles, which yields an

energy efficient reading method.

Theoretically, higher number of branches can encode higher tag data bits

6.7 Conclusion 147

within the same frequency band. However, the number of receiving antennas

used in the reader should be always equal or higher than the total branches in

the MIMO tag to successfully decompose the signals. In addition, due to short

distance the proposed setup operates under line-of-sight (LOS) MIMO. In LOS

MIMO, the physical distances between antennas have to be maintained such

that they will not form similar channel gains between transmitter and receiver

antennas. As a result, the maximum number of branches allowed in the tag is

limited. Main drawback of ML detection technique is the computation com-

plexity. It was demonstrated in Chapter 5 that the proposed computationally

feasible detection techniques reduce the complexity from exponential to linear

order without compromising the tag reading accuracy.

Chapter 7

Conclusion

The conclusion chapter first summarises key finding and how it relates to the

research objectives of the thesis. The second half of the chapter discusses about

the future directions of the tag detection techniques for chipless RFID systems

and also open issues that needs to be addressed. Finally, a set of recommenda-

tions for chipless RFID tag detections is listed.

This research work is part of a research project funded by an Australian Re-

search Council (ARC) Linkage Project Grant number DP110105606: Electroni-

cally Controlled Phased Array Antenna for Universal UHF RFID Applications.

A number of chipless RFID tag detection techniques have been produced in-

cluding two computationally feasible tag detection techniques. In addition, a

MIMO based chipless RFID system has been proposed which is the first of its

kind reported to the best of the knowledge of the author. This project com-

menced in 2011 and was successfully completed by the end of 2014. The work

presented in this thesis is one outcome of the project.

7.1 Fulfilling the goals of the thesis

The main goals of the thesis is to develop advanced yet computationally fea-

sible tag detection techniques for chipless RFID systems that is capable of im-

proving the tag reading accuracy, reading range and the data bit capacity. De-

149

150 Conclusion

tection error rate of a number of likelihood based detectors were presented and

compared against the threshold based detector used in existing chipless RFID

systems. It is evident that all of the likelihood based detectors performs bet-

ter than the popular threshold based detector. The performance improvement

of the proposed tag detection techniques can be interpreted as an increased

tag reading accuracy at a given SNR level. On the other hand, it can also be

represented as an increment in the reading range while achieving a particular

goal of reading accuracy. Therefore the improved performance can be repre-

sented either as increased reading accuracy or the reading range depending on

the application requirement. Another main objective of the thesis is improv-

ing the tag data bits. Two approaches have been taken to improve the tag data

bits. Firstly, the proposed tag detection techniques have allowed to remove the

guard-band presented in the frequency domain tags. It has been shown that it

allows to increase the data capacity by a factor up to 2. The second approach

is to design a new MIMO chipless RFID tag and the relevant signal processing

techniques. Theoretically, it can be proved that the tag data capacity can be

improved by a factor of 2 or more.

However, there is a common drawback of all the likelihood based detec-

tion methods discussed so far. All these methods require higher computation

complexity compared to the primitive detection techniques such as threshold

based detection. Computationally feasible two tag detection techniques have

been proposed to reduce the computation complexity from exponential to lin-

ear in Chapter 4. It was found that the bit by bit detection method, which is a

suboptimal detection method performs successfully when a resonator guard-

band is used in tag design. It was shown that the computation time has signif-

icantly dropped compared to exhaustive maximum likelihood detection meth-

ods without compromising the reading accuracy. In addition, a fully optimal

Trellis tree based Viterbi decoding technique has been introduced to reduce

7.1 Fulfilling the goals of the thesis 151

the computation complexity from exponential to linear order while achieving

similar reading accuracy to original likelihood detection techniques.

Chapter 6 presented signal processing techniques for successfully detect-

ing the tag bits of a MIMO based chipless RFID system. Firstly, zero forcing

decomposing techniques were used on the received data array at the reader

to estimate the tag response. Then two methods were used to detect the tag

data bits encoded in the MIMO tag. The first method uses a threshold based

valley detection method. The other method uses the proposed ML detection

technique to detect the tag bits in each branch of the MIMO chipless tag. An

experiment was setup to test the ML detection technique and it was shown

that the encoded tag data bits were identified successfully. In order to per-

form a comprehensive analysis, a CST and MATLAB based simulation was

performed. The results show that the proposed detection technique provides

better detection error rate performance at different SNR values over traditional

threshold based detection. This benefit can be interpreted in two different met-

rics. Firstly, it can be seen as an SNR gain over the existing threshold based de-

tection technique which effectively increase the reading range. Secondly, the

high accuracy in tag reading avoids multiple reading cycles, which yields an

energy efficient reading method.

Theoretically, higher number of branches can encode higher tag data bits

within the same frequency band. However, the number of receiving antennas

used in the reader should be always equal or higher than the total branches in

the MIMO tag to successfully decompose the signals. In addition, due to short

distance the proposed setup operates under line-of-sight (LOS) MIMO. In LOS

MIMO, the physical distances between antennas have to be maintained such

that they will not form similar channel gains between transmitter and receiver

antennas. As a result, the maximum number of branches allowed in the tag is

limited. Main drawback of ML detection technique is the computation com-

152 Conclusion

plexity. It was demonstrated in Chapter 5 that the proposed computationally

feasible detection techniques reduce the complexity from exponential to linear

order without compromising the tag reading accuracy.

After analysing the simulations, it is noteworthy to pinpoint that, even

though there are only two transmitting branches presented in the RFID tag

considered, it is theoretically possible to add more branches and still recover

the transmitted signals given that, the number of receiving antennas in the

reader is larger or equal to the number of transmitting branches in the tag.

Hence, without increasing the bandwidth, the bit capacity can be further in-

creased using the same frequency resonators compared with having only one

branch at the tag. However, it is required to evaluate the effect of mutual cou-

pling between antennas with higher number of transmitting branches in the

tag.

In the RFID tag proposed, there is only one receiving antenna through

which, the received signal will be divided into two equal components. The

proposed concept can be extended to having a dedicated receiving antenna

for each component, hence increasing the effective signal-to-noise ratio (SNR)

at each branch. Therefore, with multiple dedicated transmitting and receiv-

ing antennas on the tag can further improve the performances. In addition,

the concept can be further extended to multiple tag detection if each branch is

considered as a separate tag.

Furthermore, the use of I/Q modulation/demodulation allows an extra de-

gree of freedom to increase the bit capacity. Since the baseband signal consid-

ered is complex it is possible to have asymmetric frequency response in pos-

itive and negative frequencies. Therefore, the eligible frequency band in the

passband centred around the RF carrier doubles, allowing more resonators to

be placed in the tag, without increasing the sampling rate of the ADC at the

receiving end of the reader. After analysing the above results, it can be con-

7.2 Limitations of the proposed system 153

cluded that, MIMO is a competitive candidate for improving reliability or the

bit capacity of a resonator based chipless RFID system.

It was found that, the proposed tag detection techniques for SISO systems

provides significantly higher tag reading accuracy over the existing threshold

based detector. In addition, they are capable of operating without a guard-

band which makes the tag data bit capacity to be doubled without compromis-

ing the reading accuracy. Moreover, the effective SNR gain provided by the

proposed techniques can be represented as increasing the tag reading range.

All these benefits were achieved without compromising the low computation

complexity. The MIMO tag with 2 branches is capable of encoding up to 4

times the total bits stored in existing SISO tags. Due to the highly reliable tag

detection techniques, chipless RFID tag readers does not need to read the same

tag multiple times unlike the existing readers. This introduces the new ONE

TIME tag reading philosophy.

These smart tag detection techniques are expected to increase the data bit

capacity in chipless RFID tags that can be detected at a higher success rate

and that can be detected further away from the reader. These advances in

knowledge help producing commercialised chipless RFID systems in future.

7.2 Limitations of the proposed system

However, the performance of the proposed tag detection method could be lim-

ited by few factors. One of the main factor is the fabrication defects such as the

dielectric constant of the substrate and the precision of the line widths. Due

to these inaccuracies in tag design, two tags with the same tag data bits could

have slightly different tag responses. These imperfections could affect the suc-

cessful tag detection rate. In addition, when the tags are fabricated on paper

the resonance level is less compared to that on substrates. As a result, it could

154 Conclusion

be more susceptible to noise conditions which can cause the detection error

rate to be increased.

The tag detection technique presented in System Model IV, involves esti-

mating the channel and then using the estimated channel for tag detection.

Due to various conditions such as interference and object movement channel

may change suddenly. The proposed tag detection technique assumes a slowly

varying channel for the interrogation period which is in the order of few 100s

of milliseconds. The sudden changes of the channel conditions introduces an

error in channel estimation. This error can cause the detection error rate to be

increased.

The proposed tag detection techniques perform well when there is only one

tag in the vicinity of the reader interrogation zone. If there are multiple tags

inside the interrogation zone, the responses from other tags interfere with the

current tag of interest. In order to eliminate this interference the channel real-

izations from all the tags to the reader should be known. However, obtaining

these channel state information is very difficult as the positions of the channel

are unknown and there is no feasible way to estimate the channel from each

of those tags to the reader. If the tag positions are known, reader can interro-

gate by beam-forming only one tag at a time and record the tag response and

estimate the channel as in one tag situation.

The tag detection techniques proposed in Chapter 4 and 5 require extra com-

putational power compared to low spec micro-controllers used in some of the

chipless RFID readers. As a result, unless the existing hardware performance is

already enough, there is a hardware upgrade for the new detection techniques

to be worked. However, it can be seen that these hardware upgrade is feasible

with single board computers as discussed in Section 7.4.

Even under the limitations presented above, the proposed smart tag detec-

tion techniques are expected to increase the data bit capacity in chipless RFID

7.3 Potential applications 155

tags that can be detected at a higher success rate and be detected further away

from the reader. These advances in knowledge is expected to produce com-

mercialised chipless RFID systems in future.

7.3 Potential applications

The proposed tag detection techniques can be used in number of potential ap-

plications. The most favourable would be conveyer belt applications when

a tag is either printed directly on the product or the already printed tag is

stuck on the product. On a conveyer belt, the items to be identified can be

controlled to appear one after the other. This avoids multiple tags been illu-

minated by the reader at the same time, hence interference limited tag reading

can be performed. In addition, this controlled item movement is important to

avoid disorientation of items as tag reading is orientation sensitive. Some of

these applications can be found in production lines in manufacturing indus-

tries, packaging, pharmaceuticals and airport luggage tracking and handling.

Another potential application is to identify counterfeit bank notes. The tag

will be printed on the polymer note with an invisible conductive ink using an

inkjet or laser printer. A chipless RFID reader with the proposed tag detec-

tion techniques can be used to interrogate the banknotes and based on the de-

tected data bits, counterfeit notes can be identified. Reserve bank of Australia

is the world leader in printing polymer based bank notes. They are currently

working with the main author’s research group to investigate the feasibility of

implementing this technology on to the bank notes.

Smart library is a concept that has been proposed for sometime now. A

smart library automates several day to day tasks with the use of RFID systems.

The most popular task is the lending where user picks a book and can checkout

using the RFID readers available at a self checkout desk. This has already

156 Conclusion

been realized in several occasions and the proposed detection techniques can

help reliably perform several other tasks such as receiving new stocks, carrying

out inventory checks, checking for misfiled items. For example, checking for

misfiled items can be performed by scanning the book with a tag printed on it

using a hand held chipless RFID reader.

It has been discussing about applying chipless RFID in retail market for

a decade or so. The cost of fabricating a chipless RFID tag is less than a frac-

tion of a cent which makes it an ideal technology for tagging low cost items ($1

bread) in the retail market such as supermarket. Current limitations for the de-

ployment are the low number of tag data bit and being unable to read multiple

tags simultaneously. The proposed techniques helps to double the data capac-

ity by removing the guard-band and with improved tag detection techniques.

So these techniques help to move one step closer for actual deployment.

There are few other areas where the proposed chipless RFID system can be

deployed. One of them is vehicle tracking where a tag is placed on the wind-

screen of the vehicle and readers are mounted at the entrance to the carpark.

The authors collaborated with a ski center already performed a trial to check

the feasibility of tracking the incoming and outgoing cars to the car park of the

ski center. Tagging and tracking of individual components used in safety criti-

cal applications is another arena where the proposed chipless RFID system can

be utilized. Some recommendations and open issues will be presented next.

7.4 Future work and open issues

The verification of the proposed tag detection techniques were performed as

a post processing exercise using MATLAB. Implementation of the detection

techniques as a firmware is a significant step in producing commercialised

chipless RFID readers having extra benefits summarised in the previous sec-

7.4 Future work and open issues 157

Table 7.1: Technical specifications of Raspberry Pi 2 Model B

Broadcom BCM2836 Arm7 Quad Core Processor running at 900MHz

1GB RAM

40pin extended GPIO

Micro SD port for loading your operating system and storing data

Micro USB power source

4 x USB 2 ports

4 pole Stereo output and Composite video port

Full size HDMI

DSI display port for connecting the Raspberry Pi touch screen display

tion. Table 7.1 outlines the specification of the latest Raspberry Pi 2 Model B

which costs less US$45 off the shelf.

Single board computers are becoming powerful that ever and is capable of

loading advanced operating systems such as Windows or Linux. 1 GB of RAM

and the quad core processor make it possible to run powerful signal process-

ing applications such as MATLAB. However, there are open source software

tools like Octave which is highly compatible with running MATLAB codes.

In addition, extended 40 pin general purpose input/output (GPIO) allows to

capture signals for realtime data processing. These advances in technology and

the cheap price have enabled single board computers to be a potential candi-

date for preparing portable chipless RFID readers with advanced tag detection

techniques.

The second part of the work presented in the thesis is on tag detection tech-

niques for MIMO based chipless RFID systems. The proposed detection tech-

niques require accurate channel state information. More advanced signal pro-

cessing techniques are required to be developed that performs when perfect

channel state information is not available.

In addition, the detection techniques were derived based on the assump-

158 Conclusion

tion that the noise has a identically independent Gaussian distributed sam-

ples. Even though the results produced has shown improved results, it might

be worth investigating a novel model to represent noise encountered in the

proposed system.

Moreover, the interrogating signal used is constructed based on having

equal power across the frequency band interested. However, once channel

state information is available, the interrogating signal shape can be optimised

to improve the tag reading accuracy.

With the proposed tag detection techniques, it is believed that the chal-

lenges for commercialising chipless RFID systems will be successfully over-

come. As a result, chipless RFID can made to be the future of barcode like the

researchers predicted about a decade ago.

List of References

[1] K. Finkenzeller, RFID Handbook: Fundamentals and Applications in Con-

tactless Smart Cards and Identification, 2nd ed. New York, NY, USA: John

Wiley & Sons, Inc., 2003.

[2] S. Preradovic and N. Karmakar., Modern RFID readers, 2007 (accessed De-

cember 24, 2014). [Online]. Available: http://www.microwavejournal.

com/articles/5271-modern-rfid-readers

[3] S. Preradovic and N. Karmakar, “Chipless RFID: Bar code of the future,”

Microwave Magazine, IEEE, vol. 11, no. 7, pp. 87–97, Dec 2010.

[4] U. Kaiser and W. Steinhagen, “A low power transponder IC for high

performance identification systems,” in Custom Integrated Circuits Con-

ference, 1994., Proceedings of the IEEE 1994, San Diego, CA, USA, May

1994, pp. 335–338.

[5] S. Preradovic, S. Roy, and N. Karmakar, “RFID system based on fully

printable chipless tag for Paper-/Plastic-ltem tagging,” Antennas and

Propagation Magazine, IEEE, vol. 53, no. 5, pp. 15–32, Oct 2011.

[6] F. Kamoun, “RFID system management: state-of-the art and open re-

search issues,” Network and Service Management, IEEE Transactions on,

vol. 6, no. 3, pp. 190–205, September 2009.

159

160 BIBLIOGRAPHY

[7] X. Chen, W. G. Yeoh, Y. B. Choi, H. Li, and R. Singh, “A 2.45-GHz near-

field RFID system with passive On-Chip antenna tags,” Microwave The-

ory and Techniques, IEEE Transactions on, vol. 56, no. 6, pp. 1397–1404, June

2008.

[8] A. Shameli, A. Safarian, A. Rofougaran, M. Rofougaran, J. Castaneda,

and F. De Flaviis, “A UHF near-field RFID system with fully integrated

transponder,” Microwave Theory and Techniques, IEEE Transactions on,

vol. 56, no. 5, pp. 1267–1277, May 2008.

[9] A. Athalye, V. Savic, M. Bolic, and P. Djuric, “Novel semi-passive RFID

system for indoor localization,” Sensors Journal, IEEE, vol. 13, no. 2, pp.

528–537, Feb 2013.

[10] Y. Weng, S. Cheung, T. Yuk, and L. Liu, “Design of chipless UWB RFID

system using A CPW multi-resonator,” Antennas and Propagation Maga-

zine, IEEE, vol. 55, no. 1, pp. 13–31, Feb 2013.

[11] C. Floerkemeier and S. Sarma, “An overview of RFID system interfaces

and reader protocols,” in RFID, 2008 IEEE International Conference on,

April 2008, pp. 232–240.

[12] H. Song, Y. Zhao, and H. Lan, “Multi-channel active RFID system for IoT

application,” in Measurement, Information and Control (MIC), 2012 Interna-

tional Conference on, vol. 2, May 2012, pp. 568–571.

[13] G. Fritz, V. Beroulle, M. Nguyen, O. Aktouf, and I. Parissis, “Read-error-

rate evaluation for RFID system on-line testing,” in Mixed-Signals, Sen-

sors and Systems Test Workshop (IMS3TW), 2010 IEEE 16th International,

June 2010, pp. 1–6.

BIBLIOGRAPHY 161

[14] S. Sabesan, M. Crisp, R. Penty, and I. White, “An error free passive UHF

RFID system using a new form of wireless signal distribution,” in RFID

(RFID), 2012 IEEE International Conference on, April 2012, pp. 58–65.

[15] S. Ma, Y. Zhang, and D. Wang, “Distributed work flow system for RFID

integration and application,” in Information, Computing and Telecommuni-

cation, 2009. YC-ICT ’09. IEEE Youth Conference on, Sept 2009, pp. 62–65.

[16] S. Mirshahi, S. Uysal, and A. Akbari, “Integration of RFID and WSN for

supply chain intelligence system,” in Electronics, Computers and Artificial

Intelligence (ECAI), 2013 International Conference on, June 2013, pp. 1–6.

[17] R. Falk, F. Kohlmayer, A. Koepf, M. Braun, H. Seuschek, and M. Li, “Ap-

plication of passive asymmetric RFID tags in a high-assurance avionics

multi-domain RFID processing system,” in RFID Systems and Technologies

(RFID SysTech), 2008 4th European Workshop on, June 2008, pp. 1–7.

[18] S.-K. Youm, J.-H. Kim, and S.-K. Cho, “A study on the methodology for

testing of RFID system at library,” in Multimedia and Ubiquitous Engineer-

ing, 2007. MUE ’07. International Conference on, April 2007, pp. 1076–1079.

[19] I. Mun, A. Kantrowitz, P. Carmel, K. Mason, and D. Engels, “Active RFID

system augmented with 2D barcode for asset management in a hospital

setting,” in RFID, 2007. IEEE International Conference on, March 2007, pp.

205–211.

[20] A. Fawky, M. Mohammed, M. El-Hadidy, and T. Kaiser, “UWB chipless

RFID system performance based on real world 3D-deterministic channel

model and ZF equalization,” in Antennas and Propagation (EuCAP), 2014

8th European Conference on, April 2014, pp. 1765–1768.

[21] Z. jun Guo, “Research on test technology of RFID system,” in Control and

Decision Conference (CCDC), 2013 25th Chinese, May 2013, pp. 2793–2796.

162 BIBLIOGRAPHY

[22] K. Kozlowski, L. Kulas, and K. Nyka, “New RFID readers for scalable

RFID system,” in Information Technology (ICIT), 2010 2nd International

Conference on, June 2010, pp. 95–98.

[23] N. Aziz, I. Alias, A. Hashim, R. Mustafa, K. Anuar, S. Ahmad, and

W. Muhamad, “Smart RFID system for oil palm bio-laboratory,” in RF

and Microwave Conference, 2008. RFM 2008. IEEE International, Dec 2008,

pp. 247–251.

[24] S.-Y. Chan, S.-W. Luan, J.-H. Teng, and M.-C. Tsai, “Design and imple-

mentation of a RFID-based power meter and outage recording system,”

in Sustainable Energy Technologies, 2008. ICSET 2008. IEEE International

Conference on, Nov 2008, pp. 750–754.

[25] C.-T. Huang, S.-J. Wang, W.-L. Wang, and Y.-S. Wang, “Construction

of an online RFID enabled supply chain system reliability monitoring

model,” in Computer, Consumer and Control (IS3C), 2012 International Sym-

posium on, June 2012, pp. 626–629.

[26] M. Nassih, I. Cherradi, Y. Maghous, B. Ouriaghli, and Y. Salih-Alj, “Ob-

stacles recognition system for the blind people using RFID,” in Next Gen-

eration Mobile Applications, Services and Technologies (NGMAST), 2012 6th

International Conference on, Sept 2012, pp. 60–63.

[27] J.-S. Cho, S.-C. Kim, and S.-S. Yeo, “RFID system security analysis, re-

sponse strategies and research directions,” in Parallel and Distributed Pro-

cessing with Applications Workshops (ISPAW), 2011 Ninth IEEE International

Symposium on, May 2011, pp. 371–376.

[28] Y. Zuo, “Survivable RFID systems: Issues, challenges, and techniques,”

Systems, Man, and Cybernetics, Part C: Applications and Reviews, IEEE

Transactions on, vol. 40, no. 4, pp. 406–418, July 2010.

BIBLIOGRAPHY 163

[29] E. Hagras, “Interleave division multiple access; robust anti-collision pro-

tocol for UWB RFID system in non-gaussian impulsive channel,” in Ra-

dio Science Conference (NRSC), 2014 31st National, April 2014, pp. 202–209.

[30] Z. Al-Amir, F. Al-Saidi, and H. Abdulkadir, “Design and implementation

of RFID system,” in Systems, Signals and Devices, 2008. IEEE SSD 2008. 5th

International Multi-Conference on, July 2008, pp. 1–6.

[31] M. Mubarak, J. Manan, and S. Yahya, “Trusted anonymizer-based RFID

system with integrity verification,” in Information Assurance and Security

(IAS), 2011 7th International Conference on, Dec 2011, pp. 98–103.

[32] A. Bhattacharjya and R. Pal, “Distributed design of universal lightweight

RFID system for large-scale RFID operation,” in Business Innovation and

Technology Management (APBITM), 2011 IEEE International Summer Con-

ference of Asia Pacific, July 2011, pp. 40–44.

[33] S. Sabesan, M. Crisp, R. Penty, and I. White, “Wide area passive UHF

RFID system using antenna diversity combined with phase and fre-

quency hopping,” Antennas and Propagation, IEEE Transactions on, vol. 62,

no. 2, pp. 878–888, Feb 2014.

[34] D.-G. Min, J.-W. Kim, and M.-S. Jun, “The entrance authentication sys-

tem in real-time using face extraction and the RFID tag,” in Ubiquitous

Computing and Multimedia Applications (UCMA), 2011 International Con-

ference on, April 2011, pp. 20–24.

[35] H. Cheng, W. Ni, and N. Li, “A systematic scheme for designing RFID

systems with high object detection reliability,” in Information Science,

Electronics and Electrical Engineering (ISEEE), 2014 International Conference

on, vol. 3, April 2014, pp. 1521–1526.

164 BIBLIOGRAPHY

[36] J. Griffin and G. Durgin, “Gains for RF tags using multiple antennas,”

Antennas and Propagation, IEEE Transactions on, vol. 56, no. 2, pp. 563–

570, Feb 2008.

[37] D.-Y. Kim, H.-S. Jo, H. Yoon, C. Mun, B.-J. Jang, and J.-G. Yook, “Reverse-

link interrogation range of a UHF MIMO-RFID system in nakagami-

m fading channels,” Industrial Electronics, IEEE Transactions on, vol. 57,

no. 4, pp. 1468–1477, April 2010.

[38] N. Saldanha and D. Malocha, “Low loss SAW RF ID tags for space appli-

cations,” in Ultrasonics Symposium, 2008. IUS 2008. IEEE, Beijing, China,

Nov 2008, pp. 292–295.

[39] N. I. of Advanced Industrial Science and T. (AIST), Printing of organic

thin-film transistor arrays on flexible substrates, 2008 (accessed December

24, 2014). [Online]. Available: http://www.idtechex.com/research/

reports/rfid forecasts players and opportunities 2006 2016 000137.asp

[40] J. Vemagiri, A. Chamarti, M. Agarwal, and K. Varahramyan, “Transm-

ission line delay-based radio frequency identification (RFID) tag,” Mi-

crowave and Optical Technology Letters, vol. 49, no. 8, pp. 1900–1904, 2007.

[41] A. Vena, E. Moradi, K. Koski, A. Babar, L. Sydanheimo, L. Ukkonen, and

M. Tentzeris, “Design and realization of stretchable sewn chipless RFID

tags and sensors for wearable applications,” in RFID (RFID), 2013 IEEE

International Conference on, April 2013, pp. 176–183.

[42] R. Koswatta and N. Karmakar, “Moving average filtering technique for

signal processing in digital section of UWB chipless RFID reader,” in

Microwave Conference Proceedings (APMC), 2010 Asia-Pacific, Dec 2010, pp.

1304–1307.

BIBLIOGRAPHY 165

[43] S. Preradovic, I. Balbin, N. Karmakar, and G. Swiegers, “Multiresonator-

based chipless RFID system for low-cost item tracking,” Microwave The-

ory and Techniques, IEEE Transactions on, vol. 57, no. 5, pp. 1411–1419, May

2009.

[44] M. Bhuiyan and N. Karmakar, “Chipless RFID tag based on split-wheel

resonators,” in Antennas and Propagation (EuCAP), 2013 7th European Con-

ference on, April 2013, pp. 3054–3057.

[45] A. Blischak and M. Manteghi, “Pole residue techniques for chipless RFID

detection,” in Antennas and Propagation Society International Symposium,

2009. APSURSI ’09. IEEE, Charleston, SC, USA, June 2009, pp. 1–4.

[46] S. Mukherjee, “Chipless radio frequency identification by remote mea-

surement of complex impedance,” in Microwave Conference, 2007. Euro-

pean, Munich, Germany, Oct 2007, pp. 1007–1010.

[47] P. Kalansuriya, N. Karmakar, and E. Viterbo, “On the detection of

frequency-spectra-based chipless RFID using UWB impulsed interro-

gation,” Microwave Theory and Techniques, IEEE Transactions on, vol. 60,

no. 12, pp. 4187–4197, Dec 2012.

[48] M. Manteghi, “A novel approach to improve noise reduction in the

matrix pencil algorithm for chipless RFID tag detection,” in Antennas

and Propagation Society International Symposium (APSURSI), 2010 IEEE,

Toronto, ON, Canada, July 2010, pp. 1–4.

[49] A. Lazaro, A. Ramos, D. Girbau, and R. Villarino, “Chipless UWB RFID

tag detection using continuous wavelet transform,” Antennas and Wire-

less Propagation Letters, IEEE, vol. 10, pp. 520–523, 2011.

166 BIBLIOGRAPHY

[50] A. Vena, E. Perret, and S. Tedjini, “Chipless RFID tag using hybrid cod-

ing technique,” Microwave Theory and Techniques, IEEE Transactions on,

vol. 59, no. 12, pp. 3356–3364, Dec 2011.

[51] D. Chizhik, G. Foschini, and R. Valenzuela, “Capacities of multi-element

transmit and receive antennas: Correlations and keyholes,” Electronics

Letters, vol. 36, no. 13, pp. 1099–1100, Jun 2000.

[52] R. Langwieser, C. Angerer, and A. Scholtz, “A UHF frontend for MIMO

applications in RFID,” in Radio and Wireless Symposium (RWS), 2010 IEEE,

New Orleans, VA, USA, Jan 2010, pp. 124–127.

[53] M. A. Ingram, M. F. Demirkol, and D. Kim, “Transmit diversity and spa-

tial multiplexing for RF links using modulated backscatter,” in Interna-

tional Symposium on Signals, Systems and Electronics (ISSSE), 2001, Tokyo,

Japan, July 2001.

[54] J. Griffin and G. Durgin, “Gains for RF tags using multiple antennas,”

Antennas and Propagation, IEEE Transactions on, vol. 56, no. 2, pp. 563–

570, Feb 2008.

[55] J. Collins, Alien cuts tag price, 2007 (accessed December 24, 2014).

[Online]. Available: http://www.rfidjournal.com/article/articleview/

857/1/1/

[56] S. DHonts, The cutting edge of RFID technology and applications for

manufacturing and distributione, accessed December 24, 2014. [Online].

Available: http://electrocom.com.au/pdfs/whitepaper manuf dist ti.

pdf

[57] P. Harrop, The price-sensitivity curve for RFID, 2006 (accessed December

24, 2014). [Online]. Available: http://www.idtechex.com/products/

en/articles/00000488.asp

BIBLIOGRAPHY 167

[58] S. Harma, V. Plessky, C. Hartmann, and W. Steichen, “PS-1 SAW RFID

tag with reduced size,” in Ultrasonics Symposium, 2006. IEEE, Vancouver,

BC, Canada, Oct 2006, pp. 2389–2392.

[59] Y.-Y. Chen, T.-T. Wu, and K.-T. Chang, “P5H-3 A COM analysis of SAW

tags operating at harmonic frequencies,” in Ultrasonics Symposium, 2007.

IEEE, New York, NY, USA, Oct 2007, pp. 2347–2350.

[60] S. Harma, V. Plessky, X. Li, and P. Hartogh, “Feasibility of ultra-

wideband SAW RFID tags meeting FCC rules,” Ultrasonics, Ferroelectrics,

and Frequency Control, IEEE Transactions on, vol. 56, no. 4, pp. 812–820,

April 2009.

[61] T. Han, W. Wang, H. Wu, and Y. Shui, “Reflection and scattering char-

acteristics of reflectors in SAW tags,” Ultrasonics, Ferroelectrics, and Fre-

quency Control, IEEE Transactions on, vol. 55, no. 6, pp. 1387–1390, June

2008.

[62] T. Han, W. Wang, J. M. Lin, H. Wu, H. Wang, and Y. Shui, “P2H-2 phases

of carrier wave in a SAW identification tags,” in Ultrasonics Symposium,

2007. IEEE, New York, NY, USA, Oct 2007, pp. 1669–1672.

[63] N. Saldanha and D. Malocha, “P4J-1 design parameters for SAW multi-

tone frequency coded reflectors,” in Ultrasonics Symposium, 2007. IEEE,

New York, NY, USA, Oct 2007, pp. 2087–2090.

[64] J. Liu and J. Yao, “Wireless RF identification system based on SAW,”

Industrial Electronics, IEEE Transactions on, vol. 55, no. 2, pp. 958–961, Feb

2008.

[65] R. Das and P. Harrop, Chipless RFID forecasts, technolo-

gies & players 2006-2016, 2006 (accessed December 24, 2014).

168 BIBLIOGRAPHY

[Online]. Available: http://www.idtechex.com/research/reports/

rfid forecasts players and opportunities 2006 2016 000137.asp

[66] A. Chamarti and K. Varahramyan, “Transmission delay line based ID

generation circuit for RFID applications,” Microwave and Wireless Compo-

nents Letters, IEEE, vol. 16, no. 11, pp. 588–590, Nov 2006.

[67] J. Vemagiri, A. Chamarti, M. Agarwal, and K. Varahramyan, “Transm-

ission line reflection and delay-based ID generation scheme for RFID

and other applications,” International Journal of Radio Frequency Identifi-

cation Technology and Applications, vol. 1, no. 4, pp. 401–416, 2007.

[68] S. Preradovic and N. Karmakar, “Design of fully printable planar chip-

less RFID transponder with 35-bit data capacity,” in Microwave Confer-

ence, 2009. EuMC 2009. European, Rome, Italy, Sept 2009, pp. 013–016.

[69] IDTechEx, Chip-less RFID? The end game, 2006 (accessed December 24,

2014). [Online]. Available: http://www.idtechex.com/products/en/

articles/00000435.asp

[70] M. Glickstein, Firewall protection for paper documents, 2004 (accessed

December 24, 2014). [Online]. Available: http://www.rfidjournal.com/

article/articleprint/790/-1/1/

[71] J. Collins, RFID fibers for secure applications, 2004 (accessed December

24, 2014). [Online]. Available: http://www.rfidjournal.com/article/

articleprint/845/-1/1/

[72] K. C. Jones, Invisible tattoo ink for chipless RFID safe,

company says, 2007 (accessed December 24, 2014). [On-

line]. Available: http://www.electronics-eetimes.com/en/

invisible-tattoo-ink-for-chipless-rfid-safe-company-says.html?cmp id=

7&news id=196900063

BIBLIOGRAPHY 169

[73] J. Dowe, SOMARKs chipless RFID ink tattoo field demo brings the company

closer to launch, 2008 (accessed December 24, 2014). [Online]. Available:

http://www.prweb.com/releases/2008/02/prweb677113.htm

[74] S. Innovations, RFID Tattoo: Somark of the Beast, 2008 (accessed

December 24, 2014). [Online]. Available: http://www.rap-con.com/

signs/rfid-tattoo-somark-beast

[75] I. Jalaly and I. Robertson, “RF barcodes using multiple frequency

bands,” in Microwave Symposium Digest, 2005 IEEE MTT-S International,

Long Beach, CA, USA, June 2005, pp. 139–142.

[76] J. McVay, A. Hoorfar, and N. Engheta, “Space-filling curve RFID tags,” in

Radio and Wireless Symposium, 2006 IEEE, San Diego, CA, USA, Jan 2006,

pp. 199–202.

[77] S. Preradovic, I. Balbin, S. M. Roy, N. C. Karmakar, and G. Swiegers,

“Radio frequency transponder,” Australian Provisional Patent Applica-

tion P30 228AUPI, April, 2008.

[78] S. Preradovic, I. Balbin, and N. Karmakar, “The development and design

of a novel chipless RFID system for low-cost item tracking,” in Microwave

Conference, 2008. APMC 2008. Asia-Pacific, Macau, China, Dec 2008, pp.

1–4.

[79] S. Preradovic, S. Roy, and N. Karmakar, “Fully printable multi-bit chip-

less RFID transponder on flexible laminate,” in Microwave Conference,

2009. APMC 2009. Asia Pacific, Singapore, Dec 2009, pp. 2371–2374.

[80] S. Preradovic, I. Balbin, N. Karmakar, and G. Swiegers, “A novel chipless

RFID system based on planar multiresonators for barcode replacement,”

in RFID, 2008 IEEE International Conference on, Las Vegas, NV, USA, April

2008, pp. 289–296.

170 BIBLIOGRAPHY

[81] C. Divarathne and N. Karmakar, “MIMO based chipless RFID system,”

in RFID-Technologies and Applications (RFID-TA), 2012 IEEE International

Conference on, Nov 2012, pp. 423–428.

[82] M. NAKAGAMI, “The m-distribution -A general formula of intensity

distribution of rapid fading,” in Statistical Methods in Radio Wave Propa-

gation, W. HOFFMAN, Ed. Pergamon, 1960, pp. 3 – 36.

[83] M. Islam and N. Karmakar, “A novel compact printable dual-polarized

chipless RFID system,” Microwave Theory and Techniques, IEEE Transac-

tions on, vol. 60, no. 7, pp. 2142–2151, July 2012.

[84] A. Vena, E. Perret, and S. Tedjni, “A depolarizing chipless RFID tag

for robust detection and its FCC compliant UWB reading system,” Mi-

crowave Theory and Techniques, IEEE Transactions on, vol. 61, no. 8, pp.

2982–2994, Aug 2013.

[85] A. Vena, E. Perret, and S. Tedjini, “A fully printable chipless RFID tag

with detuning correction technique,” Microwave and Wireless Components

Letters, IEEE, vol. 22, no. 4, pp. 209–211, April 2012.

[86] I. Balbin and N. Karmakar, “Phase-encoded chipless RFID transponder

for large-scale low-cost applications,” Microwave and Wireless Components

Letters, IEEE, vol. 19, no. 8, pp. 509–511, Aug 2009.

[87] R. Koswatta and N. Karmakar, “A novel reader architecture based on

UWB chirp signal interrogation for multiresonator-based chipless RFID

tag reading,” Microwave Theory and Techniques, IEEE Transactions on,

vol. 60, no. 9, pp. 2925–2933, Sept 2012.

[88] R. Rezaiesarlak and M. Manteghi, “A space-frequency technique for

chipless RFID tag localization,” Antennas and Propagation, IEEE Trans-

actions on, vol. 62, no. 11, pp. 5790–5797, Nov 2014.

BIBLIOGRAPHY 171

[89] ——, “A new detection technique for identifying chipless RFID tags,” in

Radio Science Meeting (USNC-URSI NRSM), 2014 United States National

Committee of URSI National, Boulder, CO, USA, Jan 2014, pp. 1–1.

[90] ——, “Time-frequency analysis of the scattered signal from chipless

RFID tags,” in Radio Science Meeting (USNC-URSI NRSM), 2014 United

States National Committee of URSI National, Boulder, CO, USA, Jan 2014,

pp. 1–1.

[91] P. Kalansuriya, N. Karmakar, and E. Viterbo, “Signal space representa-

tion of chipless RFID tag frequency signatures,” in Global Telecommunica-

tions Conference (GLOBECOM 2011), 2011 IEEE, Houston, TX, USA, Dec

2011, pp. 1–5.

[92] R. Koswatta and N. Karmakar, “A novel method of reading multi-

resonator based chipless RFID tags using an UWB chirp signal,” in

Microwave Conference Proceedings (APMC), 2011 Asia-Pacific, Melbourne,

Australia, Dec 2011, pp. 1506–1509.

[93] A. Blischak and M. Manteghi, “Embedded singularity chipless RFID

tags,” Antennas and Propagation, IEEE Transactions on, vol. 59, no. 11, pp.

3961–3968, Nov 2011.

[94] S. Mukherjee, “Chipless radio frequency identification by remote mea-

surement of complex impedance,” in Wireless Technologies, 2007 European

Conference on, Munich, Germany, Oct 2007, pp. 249–252.

[95] W. Dullaert, L. Reichardt, and H. Rogier, “Improved detection scheme

for chipless RFIDs using prolate spheroidal wave function-based noise

filtering,” Antennas and Wireless Propagation Letters, IEEE, vol. 10, pp.

472–475, 2011.

172 BIBLIOGRAPHY

[96] P. Kalansuriya and N. Karmakar, “Time domain analysis of a backscat-

tering frequency signature based chipless RFID tag,” in Microwave Con-

ference Proceedings (APMC), 2011 Asia-Pacific, Melbourne, Australia, Dec

2011, pp. 183–186.

[97] M. Manteghi, “A space-time-frequency target identification technique

for chipless RFID applications,” in Antennas and Propagation (APSURSI),

2011 IEEE International Symposium on, Spokane, WA, USA, July 2011, pp.

3350–3351.

[98] F. Costa, S. Genovesi, and A. Monorchio, “A chipless RFID based on mul-

tiresonant high-impedance surfaces,” Microwave Theory and Techniques,

IEEE Transactions on, vol. 61, no. 1, pp. 146–153, Jan 2013.

[99] V. Montilla, E. Ramon, and J. Carrabina, “Frequency scan technique

for inkjet-printed chipless sensor tag reading,” in Electronics, Circuits,

and Systems (ICECS), 2010 17th IEEE International Conference on, Athens,

Greece, Dec 2010, pp. 1100–1103.

[100] G. Janssen and R. Prasad, “Propagation measurements in an indoor ra-

dio environment at 2.4 GHz, 4.75 GHz and 11.5 GHz,” in Vehicular Tech-

nology Conference, 1992, IEEE 42nd, May 1992, pp. 617–620 vol.2.

[101] A. Viterbi, “Orthogonal tree codes for communication in the presence of

white gaussian noise,” Communication Technology, IEEE Transactions on,

vol. 15, no. 2, pp. 238–242, April 1967.

[102] R. Scolaro, “Error probabilities for orthogonal multiplex systems in the

presence of intersymbol interference and gaussian noise,” Communica-

tions, IEEE Transactions on, vol. 25, no. 5, pp. 549–557, May 1977.

Smart Tag Detection Techniques for Chipless RFID Systems · The core theme of the thesis is the...

Documents

Transcript of Smart Tag Detection Techniques for Chipless RFID Systems · The core theme of the thesis is the...