Development and Application of Wirelessly-Powered Sensor...

203
Development and Application of Wirelessly-Powered Sensor Nodes Daniel James Yeager A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science University of Washington 2009 Program Authorized to Offer Degree: Electrical Engineering

Transcript of Development and Application of Wirelessly-Powered Sensor...

Page 1: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

Development and Applicationof Wirelessly-Powered Sensor Nodes

Daniel James Yeager

A thesis submitted in partial fulfillmentof the requirements for the degree of

Master of Science

University of Washington

2009

Program Authorized to Offer Degree: Electrical Engineering

Page 2: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered
Page 3: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

University of WashingtonGraduate School

This is to certify that I have examined this copy of a master’s thesis by

Daniel James Yeager

and have found that it is complete and satisfactory in all respects,and that any and all revisions required by the final

examining committee have been made.

Committee Members:

Brian Otis

Joshua Smith

Date:

Page 4: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered
Page 5: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

In presenting this thesis in partial fulfillment of the requirements for a master’s degree atthe University of Washington, I agree that the Library shall make its copies freely availablefor inspection. I further agree that extensive copying of this thesis is allowable only forscholarly purposes, consistent with “fair use” as prescribed in the U.S. Copyright Law. Anyother reproduction for any purpose or by any means shall not be allowed without my writtenpermission.

Signature

Date

Page 6: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered
Page 7: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

University of Washington

Abstract

Development and Applicationof Wirelessly-Powered Sensor Nodes

Daniel James Yeager

Chair of the Supervisory Committee:Professor Brian Otis

Electrical Engineering

This thesis presents work on three wirelessly-powered and wirelessly-interrogated compu-

tational devices (“tags”). First, the architecture and design of a software-defined passive

tag called WISP (Wireless Identification and Sensing Platform) is discussed in detail. Sev-

eral implemented applications including neural sensing and strain measurement are also

reviewed. Second, a passive data logger (“PDL”) is presented, which employs a novel

wirelessly-charged power model. Third, design and results of a fully-integrated system-on-

chip (“SoC”) sensor tag are presented. All of the work is compatible with EPC Class 1

Generation 2 protocol, allowing use with industry-standard UHF RFID readers.

Page 8: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered
Page 9: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

TABLE OF CONTENTS

Page

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi

Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii

Chapter 1: Radio Frequency Identification (RFID) Technology . . . . . . . . . . . 1

1.1 A Brief History of Wireless Power . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Trends for Wirelessly-Powered Devices . . . . . . . . . . . . . . . . . . . . . . 2

1.3 An Overview of RFID Technology . . . . . . . . . . . . . . . . . . . . . . . . 4

1.4 A Review of Tag Classifications . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.4.1 Active Tags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.4.2 Semi-Passive Tags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.4.3 Passive Tags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.5 Thesis Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

Chapter 2: RFID Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.2 RFID Basics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2.1 The link is asymmetric . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2.2 Tags cannot hear eachother . . . . . . . . . . . . . . . . . . . . . . . . 12

2.2.3 Readers can hear eachother . . . . . . . . . . . . . . . . . . . . . . . . 13

2.2.4 Readers can talk on different channels . . . . . . . . . . . . . . . . . . 13

2.2.5 Tag backscatter occurs on the reader’s channel . . . . . . . . . . . . . 13

2.2.6 Tags cannot distinguish between readers . . . . . . . . . . . . . . . . . 13

2.2.7 All communication is reader-driven . . . . . . . . . . . . . . . . . . . . 13

2.2.8 Tag discovery is stochastic . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.3 Gen2 Link Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.4 Physical Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

i

Page 10: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

2.5 Power Available To A Tag . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.6 Sending Sensor Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2.6.1 Shoehorn Data into the ID . . . . . . . . . . . . . . . . . . . . . . . . 17

2.6.2 Access Commands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.6.3 The Select Command . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Chapter 3: A Passive, Software-Defined Sensor Tag . . . . . . . . . . . . . . . . . 20

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3.2 Theory of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.3 WISP Hardware Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.3.1 RF Front End . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.3.2 Power Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.3.3 Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.4 WISP Firmware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.4.1 Power Management Firmware . . . . . . . . . . . . . . . . . . . . . . . 30

3.4.2 Communication Firmware . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.4.3 Application Firmware . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

3.5 WISP Troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.7 Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3.8 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.8.1 Light and Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . 36

3.8.2 Acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3.8.3 LED . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

3.8.4 Near Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.8.5 Other WISP Applications . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.9 Conclusions and Impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

Chapter 4: Maximalist Cryptography and Computation for UHF RFID Tags . . . 42

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4.2 Background on WISP Architecture . . . . . . . . . . . . . . . . . . . . . . . . 43

4.3 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4.4 Measurements and Estimates of Computation and Power Consumption . . . . 45

4.4.1 Measurements of Computation and Symmetric Cryptography . . . . . 45

ii

Page 11: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

4.4.2 Estimates of Maximal Computation and Measurements of the Effectof Flash Writes on Computation . . . . . . . . . . . . . . . . . . . . . 49

4.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

Chapter 5: A Wireless Strain Sensor for Embedded Material Health Monitoring . 525.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535.3 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.3.1 Power Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 555.3.2 Amplifier Gain and Bandwidth-Induced Attenuation . . . . . . . . . . 56

5.4 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585.5 Results/Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.5.1 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 595.5.2 Specimen Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

5.6 Conclusions and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

Chapter 6: NeuralWISP: A Wirelessly-Powered Brain Interface . . . . . . . . . . . 636.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 636.2 System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

6.2.1 Power and Communication . . . . . . . . . . . . . . . . . . . . . . . . 656.2.2 Analog Signal Path . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 676.2.3 Digital Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 706.2.4 Application Layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

6.3 Test Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 726.3.1 Analog Front End . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 726.3.2 Spike Detector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 756.3.3 System Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 776.3.4 In-vivo Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

6.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 826.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

Chapter 7: Wirelessly-Charged UHF Tags for Sensor Data Collection . . . . . . . 857.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 857.2 Background and Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . 877.3 Design Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

7.3.1 Run Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

iii

Page 12: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

7.3.2 Charge Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 917.3.3 RFID Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

7.4 Hardware Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 937.4.1 High Level Description . . . . . . . . . . . . . . . . . . . . . . . . . . . 947.4.2 Demodulation and Power Management . . . . . . . . . . . . . . . . . . 957.4.3 Sensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

7.5 Software Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 977.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 997.7 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1027.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

Chapter 8: A Graphical User Interface for RFID Applications . . . . . . . . . . . 1048.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1048.2 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1058.3 WISP Demo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1068.4 Attenuation Tester . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1078.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

Chapter 9: Design of an Integrated Gen2 Sensor Tag . . . . . . . . . . . . . . . . 1109.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1109.2 Prior Work on Full Tags . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1109.3 Prior Work on Tag Components . . . . . . . . . . . . . . . . . . . . . . . . . . 112

9.3.1 Rectifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1129.3.2 Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1139.3.3 Digital Core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

9.4 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1159.5 Analog Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

9.5.1 RF Rectifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1209.5.2 Bias Current Generator . . . . . . . . . . . . . . . . . . . . . . . . . . 1219.5.3 Bandgap Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1239.5.4 Regulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1249.5.5 Demodulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1259.5.6 Oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

9.6 Digital Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1319.6.1 Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1329.6.2 Receiver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

iv

Page 13: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

9.6.3 Packet and Command Parsers . . . . . . . . . . . . . . . . . . . . . . . 134

9.6.4 Random Number Generation . . . . . . . . . . . . . . . . . . . . . . . 135

9.6.5 Transmitter Architecture . . . . . . . . . . . . . . . . . . . . . . . . . 136

9.6.6 Sequencer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

9.6.7 Encoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

9.6.8 Prescaler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

9.7 Chip Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

9.8 Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

9.9 Design Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

Chapter 10: socWISP Verification and Measurement Results . . . . . . . . . . . . . 147

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

10.2 Pre-tapeout Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

10.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

10.3.1 Analog core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

10.3.2 Digital core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

10.4 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

10.4.1 Rectifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

10.4.2 Oscillator Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

10.4.3 Bandgap and Regulators . . . . . . . . . . . . . . . . . . . . . . . . . . 154

10.4.4 Bias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

10.4.5 Digital Core Performance . . . . . . . . . . . . . . . . . . . . . . . . . 155

10.4.6 UID Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

10.4.7 ADC Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

10.4.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

Chapter 11: In-flight Insect Temperature Recording . . . . . . . . . . . . . . . . . . 160

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

11.2 Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

11.3 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

11.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

11.4.1 Tethered Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

11.4.2 Untethered Test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

11.5 Thermocouple Characterization Using a Low Power On-Chip Amplifier . . . . 165

11.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

v

Page 14: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

Chapter 12: Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

vi

Page 15: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

LIST OF FIGURES

Figure Number Page

1.1 Microprocessor efficiency timeline. . . . . . . . . . . . . . . . . . . . . . . . 3

1.2 Microprocessor efficiency and predicted range versus year . . . . . . . . . . 4

1.3 Voltage scaling versus year. . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.1 RFID communication diagram . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2.2 Gen2 downlink data encoding . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3 Gen2 downlink encoding preamble . . . . . . . . . . . . . . . . . . . . . . . 16

2.4 Uplink data encoding format . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.1 Photograph of the WISP 4.1 hardware. . . . . . . . . . . . . . . . . . . . . . 21

3.2 WISP hardware block diagram. . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.3 WISP rectifier and modulator . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.4 Power management circuitry . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.5 Analog voltage sensor schematic . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.6 Demodulator input, reference and output voltage trace. . . . . . . . . . . . 28

3.7 ASK demodulator schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.8 Firmware block diagram including relevant hardware connections. . . . . . . 30

3.9 Basic power management state diagram for standard tasks. . . . . . . . . . 30

3.10 Advanced power management state diagram for power-hungry tasks. . . . . 31

3.11 WISP packet structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3.12 Photograph of the WISP Monitor . . . . . . . . . . . . . . . . . . . . . . . . 34

3.13 Rectifier performance versus input power. . . . . . . . . . . . . . . . . . . . 35

3.14 WISP communication and error rates . . . . . . . . . . . . . . . . . . . . . . 36

3.15 Temperature measurements in response to artificial stimuli . . . . . . . . . . 37

3.16 Light sensor data plotted over 24 hours . . . . . . . . . . . . . . . . . . . . . 37

3.17 Acceleration vs. filter settling time . . . . . . . . . . . . . . . . . . . . . . . 38

3.18 Acceleration measurements as the WISP is rotated in orthogonal planes . . 39

3.19 WISP wirelessly powers an LED . . . . . . . . . . . . . . . . . . . . . . . . 40

3.20 Near-field WISP embedded in playing dice . . . . . . . . . . . . . . . . . . . 41

vii

Page 16: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

4.1 WISP power cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474.2 Annotated scope trace of WISP execution and power consumption . . . . . 484.3 Computation available in one active cycle . . . . . . . . . . . . . . . . . . . 494.4 Number of cycles available after writing to flash . . . . . . . . . . . . . . . . 51

5.1 Strain gage measurement system diagram . . . . . . . . . . . . . . . . . . . 545.2 Circuit schematic of the Wheatstone bridge and strain gauge amplifier . . . 545.3 Effect of filter settling time on effective amplifier gain . . . . . . . . . . . . . 575.4 Experimental setup of the WISP(S/G) comparison with an extensometer. . 585.5 Strain recorded during the step loading of the specimen. . . . . . . . . . . 605.6 Correlation between the Wired and Wireless strain gauge measurements . . 61

6.1 Block diagram of NeuralWISP. . . . . . . . . . . . . . . . . . . . . . . . . . 646.2 Measured rectifier output power and efficiency versus input power. . . . . . 656.3 Schematic of the RF front end . . . . . . . . . . . . . . . . . . . . . . . . . . 666.4 Neural amplifier schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 686.5 Analog front end circuitry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 716.6 Software state diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 716.7 Graphical user interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 736.8 NeuralWISP board and IC photograph . . . . . . . . . . . . . . . . . . . . . 756.9 Measured neural amplifier gain versus frequency . . . . . . . . . . . . . . . . 766.10 Measured noise spectra at the output of the LNA and the post-amp. . . . 766.11 Operation of the spike detector . . . . . . . . . . . . . . . . . . . . . . . . . 776.12 Accuracy of the spike detector . . . . . . . . . . . . . . . . . . . . . . . . . . 786.13 Time domain voltage trace of spike detector . . . . . . . . . . . . . . . . . . 786.14 Single spike digitized by the on-board ADC . . . . . . . . . . . . . . . . . . 796.15 Photograph of the in vivo experiment setup . . . . . . . . . . . . . . . . . . 806.16 Wirelessly-powered data from wing muscle tissue . . . . . . . . . . . . . . . 816.17 Primate neural spike data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

7.1 Photograph of the WISP-PDL implemented using WISP version 4.1. . . . . 877.2 Maximum Iave for various run times and capacitor sizes. . . . . . . . . . . . 907.3 Predicted charge time versus range. . . . . . . . . . . . . . . . . . . . . . . . 937.4 Block diagram of the WISP-PDL. . . . . . . . . . . . . . . . . . . . . . . . . 947.5 Demodulation and power management circuitry. . . . . . . . . . . . . . . . . 957.6 WISP-PDL software state diagram. . . . . . . . . . . . . . . . . . . . . . . . 987.7 Reader antenna and control laptop outside of refrigerator. . . . . . . . . . . 100

viii

Page 17: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

7.8 Control data set logged from milk carton. . . . . . . . . . . . . . . . . . . . 101

7.9 Real world data logged from milk carton over the course of a day. . . . . . . 101

8.1 LLRP-compatible abstraction layers for RFID application development. . . 106

8.2 Screen capture of the WISP GUI application . . . . . . . . . . . . . . . . . 107

8.3 Screen capture of the Attenuator Test program . . . . . . . . . . . . . . . . 109

9.1 socWISP system-level architecture . . . . . . . . . . . . . . . . . . . . . . . 116

9.2 Analog block diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

9.3 Analog circuit schematic. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120

9.4 Rectifier cell schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

9.5 Bias current generator schematic. . . . . . . . . . . . . . . . . . . . . . . . . 122

9.6 Simulated bias current versus supply voltage. . . . . . . . . . . . . . . . . . 122

9.7 Simulated Monte Carlo bias current distribution (process and mismatch). . 122

9.8 Simulated bias current versus temperature. . . . . . . . . . . . . . . . . . . 123

9.9 Bandgap circuit schematic to generate 1.2 V reference. . . . . . . . . . . . . 124

9.10 Regulator schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

9.11 Analog demodulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

9.12 3 MHz ring oscillator schematic. . . . . . . . . . . . . . . . . . . . . . . . . . 127

9.13 Simulated oscillator performance versus process and temperature. . . . . . . 128

9.14 Simulated oscillator performance versus supply voltage and temperature. . . 130

9.15 EPC Gen2 digital core block diagram . . . . . . . . . . . . . . . . . . . . . . 131

9.16 Pulse-Interval-Encoding (PIE) demodulation logic schematic. . . . . . . . . 133

9.17 Edge detection logic state diagram. . . . . . . . . . . . . . . . . . . . . . . . 133

9.18 EPC physical layer packet digital decoder schematic. . . . . . . . . . . . . . 134

9.19 Schematic for command parsing logic. . . . . . . . . . . . . . . . . . . . . . 134

9.20 Schematic for packet parsing logic. . . . . . . . . . . . . . . . . . . . . . . . 135

9.21 EPC Gen2 uplink packet construction and modulation schematic. . . . . . . 136

9.22 Sequencer state diagram for constructing EPC Gen2 packets. . . . . . . . . 137

9.23 FM0 or Miller modulation encoder schematic. . . . . . . . . . . . . . . . . . 138

9.24 Divider residual for 15% allowable tolerance cases. . . . . . . . . . . . . . . 140

9.25 Divider residual for 10% allowable tolerance cases. . . . . . . . . . . . . . . 140

9.26 Divider residual for 7% allowable tolerance cases. . . . . . . . . . . . . . . . 141

9.27 Full chip layout with dimensions and high-level blocks annotated. . . . . . . 142

9.28 IC layout area breakdowns. . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

9.29 IC die micrograph. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

ix

Page 18: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

10.1 Photograph of test setup for socWISP verification. . . . . . . . . . . . . . . 14910.2 Measured rectifier input impedance versus frequency . . . . . . . . . . . . . 15010.3 Measured rectifier return loss versus input frequency . . . . . . . . . . . . . 15110.4 Measured rectifier return loss versus input power . . . . . . . . . . . . . . . 15110.5 Measured rectifier output voltage with analog core (approx. 1.2 µA) load. . 15210.6 Measured rectifier power efficiency at various output voltages. . . . . . . . . 15310.7 Measured oscillator frequency distribution at 0.7 V supply voltage. . . . . . 15310.8 Measured ring oscillator frequency versus temperature. . . . . . . . . . . . . 15410.9 FM0-modulated communication voltage trace . . . . . . . . . . . . . . . . . 15610.10 Miller 4-modulated communication voltage trace . . . . . . . . . . . . . . . 15610.11 Digital core power versus supply voltage and clock frequency . . . . . . . . 15710.12 Sample IDs read from different tags. . . . . . . . . . . . . . . . . . . . . . . 15710.13 ADC samples reported through a COTS RFID reader . . . . . . . . . . . . 158

11.1 Off-chip thermocouple amplifier circuit schematic. . . . . . . . . . . . . . . . 16111.2 Photograph of the socWISP attached to a moth . . . . . . . . . . . . . . . . 16411.3 Temperature time series data from the socWISP on a tethered moth. . . . . 16411.4 Photograph of an untethered, instrumented moth hovering on a flower. . . . 16511.5 Temperature time series data from the socWISP on an untethered moth. . . 166

x

Page 19: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

LIST OF TABLES

Table Number Page

1.1 Distinctions between RFID tag power models and uplink methods . . . . . . 8

3.1 Predicted available tag power versus range . . . . . . . . . . . . . . . . . . . . 22

4.1 Comparison of WISP with other RFID devices. . . . . . . . . . . . . . . . . 444.2 Execution time for operations in RC5 . . . . . . . . . . . . . . . . . . . . . . 48

6.1 Comparison of Neural Amplifiers . . . . . . . . . . . . . . . . . . . . . . . . . 746.2 Extended Comparison of Neural Amplifiers . . . . . . . . . . . . . . . . . . . 74

9.1 Comparison of published tag features. . . . . . . . . . . . . . . . . . . . . . . 1129.2 Review of oscillators presented in the literature. . . . . . . . . . . . . . . . . . 1149.3 Summary of Features vs. Design Version. . . . . . . . . . . . . . . . . . . . . 1189.4 IC layout area with and without pads. . . . . . . . . . . . . . . . . . . . . . . 1439.5 Power consumption breakdown for analog, digital, and peripheral cores. . . . 1449.6 Breakdown of simulated current consumption of analog blocks. . . . . . . . . 1449.7 Comparison of published (measured) tag performance. . . . . . . . . . . . . . 146

xi

Page 20: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

GLOSSARY

RFID: Radio Frequency Identification.

TAG: the device to be interrogated remotely.

READER: the wireless interrogator.

IC: integrated circuit.

ASIC: application-specific integrated circuit.

PCB: printed circuit board.

COB: chip-on-board.

COTS: commercial off-the-shelf.

ADC: analog-to-digital converter.

SOC: system-on-chip.

xii

Page 21: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

ACKNOWLEDGMENTS

I would like to thank Josh for his guidance and support, Brian for his patience and

constant optimism, and Alanson for mentoring my early research efforts.

xiii

Page 22: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

DEDICATION

to my father, Tom

xiv

Page 23: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

1

Chapter 1

RADIO FREQUENCY IDENTIFICATION (RFID) TECHNOLOGY

1.1 A Brief History of Wireless Power

Wireless delivery of power was first proposed by Nikola Tesla in the early 20th century to re-

motely power electric lights [1]. However, the electrical devices available during Tesla’s time,

primarily motors and electric lights, had high power consumption requirements (>10 W).

This led to large, and consequently dangerous, power densities. Similarly high-power ap-

plications were investigated in 1961 for operating aircraft [2] and in 1968 for delivering

power to earth from orbiting photovoltaic satellites [3]. Again, however, practical safety

considerations prevented adoption.

Modern implementations of wireless power systems generally fall under two categories:

inductive and radiative. Inductively-coupled systems use a magnetic field to transfer en-

ergy, and have been successfully used for powering and/or charging implantable medical

devices [4], charging consumer electronics such as the Sonicare electric toothbrush [5], and

clean room fabrication [6].

Radiative systems instead use RF propagation to transfer energy. This method is used

to send data (not power) in nearly every wireless communication system. The motivation

for using radiative energy transfer is simple: although inductive systems are generally much

more efficient than radiative systems, inductive systems are confined to centimeter ranges.

Radiative systems can extend to tens of meters with FCC-compliant transmitters, and these

systems will be discussed in great detail in this thesis.

Systems utilizing radiative energy transfer have only recently become feasible, and Sec-

tion 1.2 explains the enabling technological advances. Section 1.3 describes radio frequency

identification (RFID) technology, which is the most commercially-successful implementation

Page 24: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

2

of such systems. Section 1.4 provides context for RFID tag design discussions by reviewing

commercial tag categorizations. Finally, Section 1.5 describes the organization of this thesis.

1.2 Trends for Wirelessly-Powered Devices

For the past 30 years, both consumer and military applications have fueled demand for

semiconductor electronics performance. In 1965, Gordon Moore foresaw this growth and

predicted that the number of components per integrated circuit will double every 2 years [7].

In order to achieve this growth, Moore’s Law has driven a reduction of transistor size by

roughly 13% per year [8]. Lateral scaling has reduced the area of the transistor, which

in turn reduces the device capacitance. Vertical scaling has reduced the oxide thickness,

which in turn reduces the device threshold voltage and consequently the supply voltage.

Equation (1.1), fundamental to digital integrated circuits, states that the power of each

transistor is proportional to its capacitance (transistor size), frequency of operation, and

square of its operating voltage.

P =12CfV 2 (1.1)

In sum, after increasing the frequency of operation and number of transistors (better perfor-

mance) but decreasing the operating voltage and device capacitance through device scaling

(better technology), power consumption has not changed dramatically through the years

for high performance integrated circuits.

An opportunity has arisen, however, for applications where the required integrated cir-

cuit performance has already been met through transistor count and clock speed. As technol-

ogy improvements continue to shrink devices, the power consumption of fixed performance

integrated circuits drops significantly [9]. Figure 1.1 shows a logarithmic plot of pico-joules

per instruction of current and historical sub-10 MHz microprocessors. For a given workload

(fixed number of instructions), modern microprocessors use orders of magnitude less energy

than those of previous years. Note that while the EM 6682 has remarkably low energy per

instruction, it is not yet available at speeds necessary for wireless communication.

So why have wirelessly-powered electronics just recently become feasible? Friis’ Trans-

mission Equation, shown in Equation (1.2), models the power available to a receiver at

Page 25: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

3

Motorola HC08

Intel 4004 4bit

General

Instruments

PIC1650

Intel 8051 8bit

classic

Motorola HC05HC

TI MSP430F14x

TI MSP430F1232

EM6682

1

10

100

1000

10000

100000

1000000

10000000

1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

Year

pJ per Instruction

Figure 1.1: Microprocessor efficiency timeline.

distance R. For modest antenna sizes in the UHF band, less than 1 mW is available at 1 m

range using transmit power at the FCC limit. Section 2 will discuss this topic in greater

detail, but let it suffice that not much power is available for performing computation.

Pr = PtGtGr

4πR

)2

(1.2)

Combining Moore’s Law with Friis’ Transmission Equation for historical microprocessor

data yields Figure 1.2. Microprocessor efficiency and predicted wireless range are plotted

together, with the square-loss of received power from Friis’ Equation reducing the scaling

by 12 on the log scale. The recent emergence of RFID technology is apparent: wirelessly

powering useful computational tasks at significant wireless distances only became feasible

recently!

Figure 1.2 predicts significantly longer wireless range as instruction efficiency improves.

However, as the cost per instruction becomes increasingly small, we must consider other con-

straints in these systems. One such constraint is voltage. As the available power decreases,

so does the voltage at which that power is delivered. Technology scaling has also driven a

Page 26: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

4

0.01

0.1

1

10

100

1000

10000

100000

1000000

10000000

1970 1980 1990 2000 2010 2020

Year

Instructions per uJoule

0.01

0.1

1

10

100

1000

10000

100000

1000000

10000000

Meters

M icroprocessor Efficiency

Friis Distance (Power Limited)

Exponential Fit

Exponential Fit

Figure 1.2: Microprocessor efficiency and predicted range versus year. Note: inverse unitsof Instructions/pJ are used to show the trend with wireless range.

reduction in supply voltages in order to reduce power consumption. For example, the Inter-

national Technology Roadmap for Semiconductors (ITRS) predicts aggressive, near-linear

supply voltage scaling for cutting-edge processes (Figure 1.3(b)). However, historical mi-

croprocessor trends show less aggressive scaling with an exponential decay (Figure 1.3(a)).

At any rate, these trends will clearly benefit wirelessly powered devices.

In addition to supply scaling, advances in rectifier technology and power conversion will

also help overcome voltage-limitation challenges. For example, analog methods for tuning

transistor Vt enable the power conversion efficiency to be optimized for low input power

levels [10, 11]. Simultaneously, DC to DC power conversion technologies such as [12] are

able to convert tens of millivolts to volts. These advances will help to increase wireless

ranges even if voltage scaling becomes less aggressive in future years.

1.3 An Overview of RFID Technology

RFID systems typically consist of small, low-cost devices, called “tags”, which use the radio

signal from a specialized RFID reader for power and communication. When queried, each

Page 27: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

5

y = 6E+33e-0.0384x

1

2

3

4

5

6

1975 1990 2005 2020

Year

Voltage

(a) Historical microcontroller data.

0.5

0.7

0.9

1.1

1.3

1.5

1995 2000 2005 2010 2015

Year

Voltage

(b) ITRS roadmap.

Figure 1.3: Voltage scaling versus year.

tag responds with a unique identification number by reflecting energy back to the reader

through backscatter modulation. Tags are affixed to objects of interest, and the location

of the object is inferred from the ability of readers to communicate with the tag. Typical

communication ranges are 10 cm to 50 cm for inductively-coupled devices and 3 m to 10 m

for UHF tags. These characteristics motivate use of RFID tags as a replacement for bar-

codes in such applications as supply chain monitoring and asset management. Additionally,

RFID tags are used as identification in building security and passports as well as electronic

payment in credit cards [13].

Page 28: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

6

Industrial applications have spurred the development of RFID technology, producing a

robust physical layer capable of wirelessly powering and querying a tag. This solid foun-

dation, coupled with the widespread deployment of RFID readers, has inspired two general

categories of RFID research. The first is novel application spaces for conventional RFID

technology. The second is fundamental advances to RFID technology such as tag designs,

protocol efficiency, and reader architectures.

Various novel applications for conventional RFID technology have been proposed in

the literature. For example, the authors in [14] present network architectures for an RFID-

enhanced environment where objects are seamlessly tracked and monitored. A more specific

example discusses the implementation of an environment augmented with RFID to enhance

the quality of life and independence of elderly citizens [15]. Participants wear small RFID

reader bracelets that report interaction with tagged objects. Activities can be inferred from

this data and reported to caregivers.

Rudimentary sensor tags have been also constructed from commercial tags. To date,

there are several approaches to enhancing RFID tags with sensing capabilities. The α-WISP

multiplexes two commercial tag ICs onto one antenna, creating a 1-bit accelerometer [16].

The authors in [17] describe a humidity sensor for detecting moisture in walls of buildings

and houses by placing a sponge in front of a tag. Moisture in the sponge detunes the tag’s

antenna, allowing the approximation of humidity levels from the read range of the tag.

Another approach uses a custom tag with a built in fuse for sensing high temperatures in

food products. The fuse melts above a particular threshold, which permanently disables

the tag [18].

Fundamental advances in RFID technology have also been proposed in the literature,

and this will be an integral topic of this thesis. On the tag side, tradeoffs exist between

configurability, performance, size and cost - optimizations in this space are the subject of

ongoing research. Furthermore, enhancements such as sensing, computation, encryption and

data logging promise to enhance and unlock new applications for RFID. Examples include

infrastructure and object monitoring, automatic product tamper detection, identification

of harmful agents, and biomedical devices for noninvasive monitoring [19]. On the reader

side, opportunities exist for improvement on several fronts. Examples include physical and

Page 29: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

7

link layer protocol improvement, cost reduction, security and privacy issues, and system

integration. Recent advances in software radio technology may aid in this research; a

software-defined reader can allow direct control over the wireless protocol [20]. One specific

issue that will arise in this thesis is that commercial readers are optimized for reading many

tags quickly. However, performance when streaming sensor data from one tag is suboptimal.

Flexibility is required on both the tag and reader sides in order to tackle these challenges.

The evolution of “augmented” RFID tags to include sensory, encryption and data pro-

cessing features has led researchers to re-examine the definition of a tag. For example, Kevin

Fu at the University of Massachusetts Amherst coined the term Computational RFIDs to

describe “battery-less computers that harvest all of their operating power from an interrogat-

ing reader.” Essentially, passive RFID technology has enabled micro-computing platforms

with wireless power and communication capabilities. This thesis will explore implications of

the additional flexibility, performance, and features afforded to wirelessly-powered devices

as enabled by Moore’s Law.

1.4 A Review of Tag Classifications

There are a number of different RFID technologies operating in frequency bands ranging

from 125 kHz to 5.8 GHz. Some systems use inductive coupling while others use radiative

coupling - some even use both forms of energy transfer! A subset of tags employ active trans-

mitters like conventional wireless sensor nodes, but the majority of commercial deployments

use passive (battery-free) tags. These technologies cause wide disparities in wireless range

from several centimeters to tens or hundreds of meters. This section aims to differentiate

between these various tag classes.

Two main distinctions between tags are the power source and the communication method.

RFID officially includes provisions for active (battery-powered) and semi-passive (battery-

assisted) tags in addition to passive (battery-free) tags. With respect to the uplink com-

munication method, active tags contain a low power radio like conventional wireless nodes,

while passive and semi-passive tags transmit data by modulating power reflected back to the

reader (backscatter). However, the main advantage of RFID tags versus traditional battery-

powered sensor nodes is freedom from battery-life constraints. Therefore the remainder of

Page 30: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

8

Wirelessly Powered Battery Powered

Backscatter Passive Semi-Passive

Active TX Active

Table 1.1: Distinctions between RFID tag power models and uplink methods. The rowsdelineate the uplink method, and the columns delineate the power model. Wirelessly-powered active tags are not defined in the RFID specification; however, they are simplyactive tags plus energy harvesting such as solar, thermal, mechanical or ambient RF.

this thesis will focus on wirelessly-powered and wirelessly-charged tags, which do not have

fundamental lifespan limitations. Table 1.1 summarizes the distinctions between these tag

types.

1.4.1 Active Tags

Typical active tags include a microcontroller, battery, and low power radio. Commercial,

off-the-shelf (COTS) 2.4 GHz solutions such as Zigbee and Wifi (802.11) are popular. These

radios have wireless ranges greater than 100 m, which actually poses a problem for item

tracking. Short range tags only communicate in physical proximity to a reader (which

implies its location), but active tags are not necessarily in close proximity to the reader

when they communicate. To address this problem, time of flight, signal strength, and

triangulation are used to estimate the tag location. In theory, the tag could wirelessly

report its own location using an on-tag GPS receiver, although, the energy requirements

are currently prohibitive. In practice, the main drawback of active tags is not locating the

tags - it is replacing the batteries. The use of an active radio consumes significant amounts

of power and this limits the battery lifespan on the order of weeks to months.

1.4.2 Semi-Passive Tags

Semi-passive tags employ a battery for extending wireless range and enabling sensor mea-

surement/logging without reader proximity. Typically, passive tags are downlink limited

due to power constraints. A battery eliminates power constraints, and these systems then

Page 31: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

9

become uplink limited due to finite reader sensitivity. Semi-passive tags have very low power

budgets because they communicate using backscatter modulation like passive tags. This

means that their lifespan can be several years despite reliance on a battery. Furthermore,

they can easily conform to passive tag protocol specifications, allowing use in existing RFID

infrastructure.

1.4.3 Passive Tags

Passive tags, unlike semi-passive and active tags, have no battery-life constraints. They are

powered by converting RF to DC power. Low frequency tags (125 kHz and 13.56 MHz) rely

on inductive coupling. The advantage of inductive coupling is that it operates in water, and

the disadvantage is that it limits wireless range to less than 1 m. Low frequency tags are

limited to inductive energy transfer because antenna sizes are too large at these frequencies.

High frequency tags (433 MHz, 915 MHz, 2.4 GHz, 5.8 GHz) typically employ radiative

energy transfer. Radiative tags have the opposite properties of inductively coupled tags;

they work poorly near water and have long wireless ranges of up to 20 m. High frequency

tags are not limited to radiative energy transfer, however. Companies such as Impinj offer

inductively-coupled UHF tags which trade the benefits of long wireless range for usability

near metal and water. Advantages of inductively-coupled UHF tags include protocol and

frequency compatibility with existing UHF infrastructure, unlike low frequency inductively-

coupled tags.

The industry has converged on UHF (915 MHz) tags for commercial deployment due to a

balance between cost (which increases with RF frequency), tag size (antenna scales inversely

with RF frequency) and operating distance (low frequency tags have insufficient operating

range). A link and physical layer standard called EPC (Electronic Product Code) Class

1, Generation 2 - or “Gen2” - has emerged as an industry standard. This specification

is described in detail in Chapter 2. Much of the work in this thesis will focus on tags

that conform to this specification, which allows operation with low-cost COTS readers and

existing infrastructure.

Page 32: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

10

1.5 Thesis Overview

This thesis presents work on three wirelessly-powered and wirelessly-interrogated compu-

tational devices (“tags”). All 3 tags are compatible with the EPC Class 1 Generation 2

protocol, allowing use with industry-standard UHF RFID readers. Chapter 2 discusses

the implications of building on this specification. Chapter 3 presents the architecture and

design of a software-defined passive tag called WISP (Wireless Identification and Sensing

Platform) [21]. Implementation details of several WISP applications are presented in Chap-

ter 4 (encryption [22]), Chapter 5 (strain measurement [23]), and Chapter 6 (neural spike

sensing [24]). Chapter 7 presents a novel wirelessly-charged RFID tag power model and

a prototype implementation called the Passive Data Logger [25]. Chapters 9, 10, and 11

present the design, performance, and in vivo deployment of a fully-integrated system-on-chip

sensor tag called “socWISP” [26].

Page 33: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

11

Chapter 2

RFID PROTOCOL

2.1 Introduction

Today, commercially-available, long-range, radiative tags and readers primarily use the Elec-

tronic Product Code (EPC) Class 1 Generation 2 UHF Air Interface Protocol Standard [27].

Formally known as ISO/IEC 18000-6 Type C, and informally referred to as “Gen2,” this

specification defines the link and protocol layers for RFID communication. Other specifica-

tions exist for the tag data format [28], and the interface for communicating with the reader

[29]. These specifications are meant to provide interoperability between manufactures and

uniformity for the user. For a overview of other RFID specifications, see [30].

Gen2 improves on its predecessor, Gen1, in several ways. First, it improves inventory

speed for large tag populations. Second, provisions are added for reading and writing of

arbitrary “user” memory. This is especially useful for sensor applications. Lastly, the spec-

ification is more flexible. In the author’s experience, Gen1 readers often operated outside of

specifications. For example, the physical layer timing parameters of the Alien Technologies

9RE-0001 violate the Gen1 specification. This was likely done to compensate for tag design

errors which are costly to redesign. Gen2 has more “wiggle room”, which makes it much

easier for tags and readers to fit into the specification. It also allows manufactures room to

optimize performance and gain a competitive edge. For example, academic findings indicate

significant performance tradeoffs in selection of protocol parameters [20, 31].

However, the purpose of this work is not to optimize tag throughput. Instead, we leverage

the core technologies of RFID in order to enable novel applications beyond item tracking.

To do this, we first review some of the basic principles of RFID systems in Section 2.2.

Next, the link and physical layers are described in Section 2.3 and Section 2.4, respectively.

Estimates of the power available to the tag are given in Section 2.5. Finally, methods for

sensor node communication are discussed in Sections 2.6.

Page 34: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

12

2.2 RFID Basics

Passive, radiative RFID systems are dissimilar from traditional wireless networks. Several

distinctions are enumerated below:

2.2.1 The link is asymmetric

State of the art tags require greater than -20 dBm RF input power in order to generate

enough DC power to operate. In order to power a tag at several meters range, readers

typically transmit at the FCC limit of 1 W (30 dBm). This allows for 50 dB of path loss

given -20 dBm tag sensitivity. Assuming the tag perfectly backscatters the received signal

with no loss, the reader received signal is -70 dBm after the same path loss. The reader

thus requires sensitivity on the order of -80 dBm in order to guarantee that performance

is not limited by the reader. This link asymmetry is acceptable because the reader is

often stationary (ex, in a warehouse) and has a wired power supply. The tag performance

requirements must be low in order for the tag to operate with no battery.

2.2.2 Tags cannot hear eachother

Because tags require a relatively large input power to operate (-20 dBm), the reader signal

is trivial to demodulate. This is important because the tag cannot employ an active re-

ceiver due to its limited power budget. Tags typically employ a low power, low sensitivity

diode detector receiver. This low receive sensitivity, coupled with the low backscatter signal

strength from the tag (-20 dBm) and any tag-to-tag path loss, prevents communication

between tags. A workaround involves retransmission by the reader, but this involves signif-

icant overhead because the RFID protocol has no native support for this operation. Other

implications of limited tag receive sensitivity are that tags cannot employ listen-before-talk

channel slotting or perform multi-hop/peer-to-peer communication as in traditional sensor

networks.

Page 35: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

13

2.2.3 Readers can hear eachother

Readers have powerful transmitters ( 30 dBm) and very sensitive receivers ( 80dBm). The

signal from another reader is easily stronger than the backscatter from a tag. This means

that a reader talking on the same channel can block another reader from hearing tags. The

interfering reader may or may not affect whether the tag can hear the intended reader.

2.2.4 Readers can talk on different channels

Readers are narrow-band communication systems. Allocated channel bandwidth is given

in [27], and readers are nominally insensitive to activity on other channels. Of course, this

requires sufficient linearity in the reader’s RF frontend.

2.2.5 Tag backscatter occurs on the reader’s channel

Tags reflect received signals back to the reader, and thus tag backscatter occurs on the

same channel (at the same frequency) as the reader. Consequently, readers (which are

narrowband systems) can only hear tags on their channel. The occupied bandwidth of the

tag’s backscatter depends on the backscatter link frequency, which is specified to the tag

by the reader.

2.2.6 Tags cannot distinguish between readers

Tags have a wideband RF front-end. Because rectifiers mix all received signals to DC, there

is no way for the tag to distinguish between received signals on different frequency channels.

Some provisions for tags to recognize different readers in close temporal proximity do exist.

For example, readers can use different “sessions,” which allows tags to distinguish between

one of four time-multiplexed channels.

2.2.7 All communication is reader-driven

The reader always talks before the tag, and compliant tags only talk when requested. This

is inefficient for sensor tags because it necessitates constant polling by the reader in order

to obtain information from tags.

Page 36: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

14

Select

(Optional)Query Ack

Req_RN

(Handle)Read / Write

RN16 EPC IDRN16

(Handle)Data

Reader

Tag

Req_RN

(Cover code)

RN16

(Cover Code)

Figure 2.1: RFID communication diagram. The required packets are shown solid, whileoptional packets are dashed.

2.2.8 Tag discovery is stochastic

A reader can directly talk to a specific tag if the tag’s ID is already known (using the Select

command) or if the reader has an open communication session with that tag. However, the

discovery of new tags is a stochastic process. Much like the Slotted-Aloha algorithm, the

reader allocates a number of time slots in which the tags may respond. The tags then choose

a random slot in which to respond. Collisions must be detected and resolved by the reader

by increasing the number of slots or excluding already-detected tags from the next query

round. This process of tag discovery is often referred to as “singulation”. To singulate a

tag refers to the discovery of a tag’s ID through the above process.

2.3 Gen2 Link Layer

The RFID protocol, as it pertains to reading sensor data, is shown in Figure 2.1. First, one

or more Select commands may optionally be issued. This command is intended to allow

inclusion or exclusion of a subset of the tag population using a mask and a logic primitive

AND/OR/XOR as well as a pointer into tag memory.

Second, a inventory round occurs. The reader initializes tags with a Query command,

which specifies number of time slots (2Q) as well as communication parameters as the tag

backscatter link frequency and modulation type. Tags respond with a 16-bit random handle

in a randomly chosen time slot (the chosen slot is based on Q LSBs of the handle). Time

slots are delineated with a short reader command called Query Repeat (QueryRep). A

Query Adjust (QueryAdj) command may be used to change the number of available slots

if the reader detects that it has allocated too many or too few slots.

Page 37: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

15

Figure 2.2: Gen2 downlink data encoding. Adapted from [27].

Third, the reader may ask for EPC tag IDs via the handle given by the tag using the

Acknowledge (Ack) command. Tags compare the handle in the Ack command to their own

to determine if they should reply.

Fourth, the reader may optionally perform an “Access” operation. This allows reading

(“Read”) and writing (“Write” and “Block Write”) of arbitrary memory, as well as opera-

tions such as locking the tag into read-only mode (“Lock”) and permanently disabling the

tag (“Kill”). The Read and Write commands are of particular interest for querying sensor

data and sending information to the tag.

2.4 Physical Layer

The Gen2 physical layer employs pulse interval encoding (PIE) for downlink communication

and phase modulation for uplink encoding. Examples of a downlink data-zero and data-one

are shown in Figure 2.2. The tag measures the pulse interval with a digital counter and

compares the measurement against the “pivot” to determine if the bit is a zero or one.

The pivot is specified by a parameter “RTcal”. RTcal is a pulse interval of duration

equal to the sum of a data-zero and data-one. Half of RTcal is the arithmetic average of a

data-zero and data-one, and this is used as the pivot. RTcal is specified at the beginning

of every reader packet in either a Preamble or Framesync. This is depicted in Figure 2.3.

The Query command is preceded by a Preamble, while all other commands have a

Framesync. The Preamble contains a pulse interval called TRcal, which specifies the tag

Page 38: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

16

Figure 2.3: Gen2 downlink encoding preamble, which specifies how the tag should parse thebaseband waveform. Adapted from [27].

Figure 2.4: Uplink data encoding format. Adapted from [27].

Page 39: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

17

backscatter link frequency (BLF). BLF is a square wave that the tag backscatters to com-

municate which has phase inversions added to encode data. Tags use one of 4 modulation

types: FM 0, Miller 2, Miller 4, or Miller 8. The modulation type is also specified in the

Query command by the reader. Examples of uplink modulation (two consecutive data-

zeros) are shown in Figure 2.4. Notice that each successive modulation has the same BLF

but half the data rate. The Gen2 spec has a number of example waveforms that offer a

good reference for the designer. Additionally, Section 9.6.8 and Section 9.6.7 present design

details for generation of these encodings.

2.5 Power Available To A Tag

Modulation parameters affect how much power is available to the tag. In the downlink

communication, reader amplitude modulation reduces average transmitted power. In the

uplink communication, tag backscatter reflects energy back to the reader instead of con-

verting it to DC power. Unfortunately, for all backscatter settings, the modulator is on

approximately half of the time. This results in 3 dB of power loss while the tag is commu-

nicating. For downlink communications, the settings are more flexible. The low-amplitude

duration, PW, becomes a smaller portion of the pulse interval as Tari increases (Tari is

the length of a data-0, shown in Figure 2.2). In the worst case, PW is 52.5% of Tari, and

the packet consists mostly of data-zeros. This results in roughly 3 dB power loss while the

reader is communicating. In the best case, PW is 26.5% of Tari and the packet consists

mostly of data-ones with PIE = 2. This results in 13.5% power loss, or 1.23 dB while the

reader is communicating.

2.6 Sending Sensor Data

There are a variety of methods to communicate data other than the EPC ID using Gen2

protocol. These are discussed below:

2.6.1 Shoehorn Data into the ID

The simplest method, which we also employed using the Gen1 protocol, is to embed data

into the ID. This causes minimal overhead aside from reducing the available ID space.

Page 40: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

18

However, Gen2 allows for greater than 450-bit IDs - this easily accommodates a long ID in

addition to significant sensor data. Even standard 96-bit IDs provide adequate space for

many applications. The obvious drawback of this method is that it violates the protocol

and likely voids compatibility with middle-ware used in commercial deployments.

2.6.2 Access Commands

At the penalty of slightly increased tag complexity, dedicated Read and Write commands

are available. This enables bidirectional communication, although as previously mentioned,

it is reader-driven. Tags only respond to reader commands, which means that for the tag

to send data to the reader, the reader must issue a Read command. The reader would send

data to the tag using the Write command. Note that packet delivery is not guaranteed.

The user would need to build this functionality on top of the Gen2 protocol.

There is significant overhead for performing a single Read or Write operation because

the reader must first singulate the tag. At minimum, this includes the Query, Ack, and Req-

RN commands or approximately 50% increased overhead. If more than one tag is present,

a Select command can be added to reduce singulation time. Thus, performing one Read

command is much slower than sending data in the ID. However, if many Read commands

are performed in succession after the tag is singulated, the effective throughput is greatly

improved. Note that some commercial readers (ex, the Impinj Speedway reader) are not

capable of multiple successive access commands at this time.

2.6.3 The Select Command

The Select command presents an opportunity to communicate with tags with very little

overhead (no Query/Ack/ReqRN/Write). The reader may send a long data mask, a log-

ical operation, and a memory pointer. However, the tag has no opportunity to respond.

This could be used for selecting only tags with particular sensor values (ex, temperature too

warm), sending broadcast information to all tags, or instructing the tags to have some infor-

mation ready (encrypted data, sensor measurement, etc.) when a Read or Write command

is issued later.

Page 41: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

19

2.7 Conclusions

The Gen2 specification is very flexible and feature-rich. The biggest hurdle is often not the

protocol itself but the way commercial readers implement it. For example, many readers

will only perform one access command and then reset. This forces significant overhead in

singulating tags repeatedly. However, as reader technology improves, so will the flexibility

they offer the user. Nevertheless, the remainder of this thesis presents work built around

commercial readers. This demonstrates the usability of the protocol for a wide variety of

sensor and security applications in spite of any limitations imposed by COTS readers.

Page 42: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

20

Chapter 3

A PASSIVE, SOFTWARE-DEFINED SENSOR TAG

3.1 Introduction

The Wireless Identification and Sensing Platform (WISP) was born out of the question

”What novel applications can be enabled through RFID technology?” At the time, RFID

tags were mainly considered to be wireless identifiers for tracking objects. Indeed, many

initial experiments employed commercial tags to enable smarter living spaces, help provide

elder care through behavioral monitoring, and even localize robots. However, it quickly

became evident that visibility into the state of the tagged objects could be improved by

collecting richer, sensor-enhanced data.

Early prototypes encoded sensor information by multiplexing several commercial tag

ICs onto a single antenna. For example, mercury switch accelerometers connected one of

two tag ICs to an antenna. This provided 3 states of information: up, down or not present

[16]. Later, a RF-powered microcontroller multiplexed the commercial tag ICs using a GaAs

switch [32]. However, the use of commercial tag ICs severely constrained the communication

bandwidth because there was little control over what information the IC transmitted. This

lack of control motivated development of a tag design. In order to explore a variety of new

applications, a tag with software-defined behavior and easily-configurable hardware was

needed. This first WISP consisted of a low-power microcontroller, energy storage capacitor,

Schottky diode RF rectifier, modulator/demodulator, and low-quiescent-current regulator

on a custom PCB. Using this relatively simple design, light and temperature data were

wirelessly reported at a distance of approximately 1 m [33].

The WISP hardware and firmware have evolved to enable greater performance and a va-

riety of new features. Examples include extended wireless range of up to 3 m, compatibility

with the latest EPC RFID protocol, and integration of new sensors measuring everything

from mechanical strain to biomedical neural signals [34][21][24]. The latest hardware is

Page 43: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

21

Figure 3.1: Photograph of the WISP 4.1 hardware.

shown in Figure 3.1, and extensive documentation is available online as part of the WISP

Challenge Giveaway.

The remainder of this chapter will discuss the WISP platform in detail. Section 3.2

presents the theory of operation including the wireless power model used by WISP. Sec-

tion 3.3 describes the hardware design, and Section 3.4 describes the firmware design. Fi-

nally, Sections 3.8 and 3.9 discuss the applications and impacts of this work including

educational outreach and research collaboration.

3.2 Theory of Operation

The WISP, like conventional RFID tags, is powered and interrogated wirelessly by commer-

cial RFID readers. Due to RF path loss, the power available to the tag can be very small

- on the order of microwatts - despite reader transmit power of 1 W. Consequently, WISP

shares some design constraints and operational principles with commercial tags. However,

due to the need for software programmability and hardware configurability, WISP requires

a more complex power model than conventional tags.

A first-order approximation of the received power, Pr, based on Friis’ Transmission

Equation is given in Equation (3.1).

Pr = PtGtGr

4πR

)2

ηLmLp (3.1)

Pr is the power received given a transmit power Pt = 1 W = 30 dBm, antenna gains

Gt = 6 dB and Gr = 2 dB, distance R, wavelength λ = cf = 3×108

915×106 = 0.33 m, RF to

DC conversion efficiency η = 25% = -6 dB, amplitude modulation losses Lm and antenna

Page 44: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

22

Distance (m) Power (dBm) Power (µW)

1 -5.67 271

2 -11.7 67.6

3 -15.2 30.1

4 -17.7 17.0

5 -19.7 10.8

Table 3.1: Predicted available tag power versus range

polarization losses Lp. Section 2.5 discussed how amplitude modulation in the Gen2 protocol

physical layer affects the power available to the tag. For 50% duty cycle, the modulation

loss Lm = 12 = -3 dB. Finally, for a circularly-polarized reader patch antenna and linearly-

polarized tag dipole antenna, polarization loss Lp = 12 = -3 dB. Table 3.1 lists a few

examples of available power at increasing distances. For a fixed-function ASIC tag with

several microwatts power requirement, greater than 5 m range is clearly achievable even

with a modest rectifier efficiency of 25%.

A key feature of WISP is its programmability, which is enabled through a low-power

Texas Instruments (TI) MSP430 microcontroller. This microcontroller consumes approxi-

mately 2 mW while performing communication, which would limit operation to sub-meter

distances based on available power shown in Table 3.1. However, while the microcontroller

is in low power sleep mode it consumes less than 1 µW, and the WISP rectifier can gen-

erate sufficient operating voltage of 1.8 V at input power levels of approximately -10 dBm

(distance = ?). To enable operation when received power is less than active power, a new

tag power model is required where energy is accumulated until sufficient to complete a task.

This can be represented by the following equations, where energy is stored in a capacitor:

Ptask = VddItask (3.2)

if(Ptask > Pr), Estored ≥ Etask,net (3.3)

Etask,net = (Ptask − Pr)Ttask (3.4)

Page 45: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

23

Estored =12C(V 2

s − V 2dd) (3.5)

Vs ≥√

2CEtask + V 2

dd (3.6)

Equation (3.2) defines the active power required to perform a task such as communica-

tion, computation or sensing. Equation (3.3) states that if the power required to perform a

task is greater than the power being received from the RFID reader, some power must be

supplied by the storage capacitor. Specifically, the energy required to perform the task must

be less than the usable energy stored in the capacitor plus the energy to be received while

performing the task (Equation (3.4)). The usable energy stored in the capacitor is given by

Equation (3.5), which counts voltage above the microcontroller operating voltage. Finally,

the required voltage on the storage capacitor can be derived from the previous equations,

and the result is shown in Equation (3.6).

The result is that the amount of energy storage required is a function of the largest

required contiguous task, and the amount of stored energy is determined by Vs and the

capacitor size. An example of a contiguous task is communication; an ID inventory requires,

at minimum, two consecutive reader-tag interactions while the Read command requires 4

consecutive reader-tag interactions. Sensing tasks can also demand significant Ttask as

analog filters settle out. Section 3.8 demonstrates that energy can be saved by sampling

before the filters settle and reducing error by modeling the filter time constant to predict the

final value. For computational tasks, checkpoint-based methods can allow long tasks such

as encryption to be broken into subtasks [35]. Because the tag may move out of range from

the reader at any time, computations stored in volatile memory may be lost. However, the

cost of writing to non-volatile memory can be significant. So, a balance is required between

aggressively performing computation and keeping stored progress updated.

Alternating between accumulating energy and performing tasks effectively reduces the

duty cycle of operation. At short distances no duty cycling is needed because Ptask < Pr.

For long distances where Ptask ≥ Pr, equations (3.7) and (3.8) compute the duty cycle, DC,

and expected task rate per second, ftask, in terms of Pr, the power received at distance R.

DC = min(Pr

Ptask, 1) (3.7)

Page 46: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

24

ftask =DC

Ttask= min(

Pr

Etask,

1Ttask

) (3.8)

These equations assume the reader is continually sending modulated power. This is often not

the case, as the reader periodically stops to frequency hop, communicate with application

software, and check for interference with other nearby readers. Generally, Pr is now reduced

by the duty cycle of the reader. For communication tasks, Ttask ≈ Tquery where Tquery is

the amount of time the reader is on for a communication round. The duty cycle given

in Equation (3.7) now represents the fraction of queries that can be responded to. The

communication rate, fcomm, is now a function of the reader query rate, freader and is given

by Equation (3.9).

fcomm = DC × freader (3.9)

The reader rate depends significantly on the choice of Gen2 physical layer parameters, tag

population size, and amount of data retrieved from each tag as discussed in Section 2.5. On

the tag side, the duty cycle, DC, is maximized through energy-efficient firmware, minimal-

complexity protocol, signal processing to reduce sensor energy requirements, etc.

3.3 WISP Hardware Design

The hardware is tasked with converting RF energy into regulated DC power for the system,

facilitating RFID communication, and incorporating useful sensors onto a minimally-sized

PCB. The power model of the WISP is described in detail in Section 3.2, and the design

goals can be concisely stated here to motivate the design choices. In order to maximize

wireless range, the quiescent current consumption must be minimized. In order to maximize

operational duty cycle, the active current consumption must be minimized. Fortunately,

digital wristwatches share these design goals in order to maximize battery life, so COTS

parts are available to meet our specifications.

A high level diagram of the WISP design is shown in Figure 3.2. Power transmitted

by the reader is received by the antenna on the left. The RF envelope is recovered by a

rectifier and used to supply power to the tag. The communication block extracts amplitude-

modulated downlink communication and facilitates uplink backscatter communication. The

power management block enables the firmware to operate despite an extremely limited power

Page 47: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

25

Matching

Network and

Power

Harvester

MSP430

µC

Ant

Communication

Sensors

Power

Management

Figure 3.2: WISP hardware block diagram.

Modulator

Input

915 MHz

Dipole Antenna

Rectifier

Output

Figure 3.3: WISP rectifier and modulator. Low-capacitance RF Schottky diodes are filledblack, while low-leakage DC Schottky diodes are filled white.

budget. Finally, sensors are powered and read by the MSP430 µC.

3.3.1 RF Front End

The RF front end is shown in detail in Figure 3.3. An L-match network transforms the tag

input impedance to match that of the antenna (72 Ω). The matching network capacitor Q is

not a concern; however, high-Q (Q = 30 to 40) inductors in the 1-20 nH range from Coilcraft

are used to minimize loss. The series and shunt capacitors in the voltage-multiplying rectifier

are 7 pF and 10 pF respectively, but their values are not critical. Conversely, the number

of voltage-doubling stages and the RF Schottky diode selection can affect performance

dramatically. This design uses Avago HSMS-285x diodes in a 5-stage configuration to

Page 48: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

26

Regulated

Supply

Supervisor

Interrupt

Cstorage

SV

1.8 V

Reg.

Level

Shifter

Clamp

VS

Rectifier

Output

Figure 3.4: Power management circuitry. A regulator creates a stable supply voltage for themicrocontroller and sensors, and a voltage supervisor detects sufficient voltage for operation.The unregulated voltage is clamped to 5.5 V by a zener diode.

maximize input power sensitivity. Note that the rectifier may instead be optimized for

maximum power conversion efficiency. This would improve operational duty cycle at the

cost of maximum-achievable range.

3.3.2 Power Management

One particular challenge is that the WISP hardware and firmware design are closely in-

terrelated. Unlike battery-powered systems, the firmware may only remain in active mode

for milliseconds at a time before brownout occurs and the microcontroller resets due to

under-voltage. Several hardware mechanisms facilitate efficient firmware operation, and

energy-conscious firmware design helps maximize wireless range and operational duty cycle.

As mentioned, the rectifier recovers the RF envelope, which is amplitude-modulated to

encode downlink data. This envelope, which is essentially the baseband signal, is again

rectified and stored in Cstorage to provide an energy source for the tag (see Figure 3.4). The

stored energy is regulated to 1.8 V by a 1 µA quiescent current linear regulator (On-Semi

NCP583). Here a tradeoff is made between active duty cycle and wireless range: switching

regulators provide high efficiency at the cost of additional quiescent current. Finally, an

overvoltage protection diode (On-Semi ESD5Z3.3) clamps the unregulated voltage to 5.5 V.

Section 3.2 describes the WISP theory of operation: when received power is less than the

Page 49: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

27

Figure 3.5: Analog voltage sensor schematic. The V oltageSense pin connects to a µC ADCinput, and the V oltageSenseEnable pin is driven by a µC digital output. Note that Cstorage

is repeated from 3.4 for clarity.

power to perform a task such as communication or sensing, the difference must be supplied

from the storage capacitor. The amount of energy required, represented as voltage on the

storage capacitor, is given by Equation 3.6. For many tasks, a standard storage voltage of

2.0 V (200 mV) is detected by a 300 nA quiescent current voltage supervisor (Seiko S-1000,

see Figure 3.4). The advantage of this mode of operation is that the microcontroller does

not need to periodically poll for sufficient voltage; the supervisor asyncronously provides a

hardware interrupt to the microcontroller when sufficient voltage is reached. The hardware

voltage supervisor is used for routine communication and sensing tasks.

Energy-intensive tasks such as encryption, measuring high-power sensors, and memory-

access communication require precise power management. For these tasks, an analog voltage

sensor enables the microcontroller to poll for a specific voltage. The schematic is shown in

Figure 3.5. Because the unregulated voltage can be much larger than the regulator supply

voltage, it must be scaled down by 13 using a voltage divider before reaching the ADC input.

There are conflicting requirements for the voltage divider: large resistors reduce quiescent

current draw, but small resistors allow for quick sampling due to their ability to drive the

ADC input capacitance. An enable circuit bypasses this tradeoff by allowing small resistors

to be disabled while not in use. An analog switch (TI TS5A3166) is needed to prevent

overvoltage to the µC when the resistors are disabled.

A low power level shifter (On-Semi NLSV1T244), is used to translate the supervisor and

demodulator signals (to be discussed in Section 3.3.3) to the logic level of the microcontroller.

Page 50: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

28

0

0.5

1

1.5

2

2.5

0 20 40 60 80 100

Time (microseconds)

Voltage

Demodulator Output Reference Rectifier Output

Figure 3.6: Demodulator input, reference and output voltage trace.

Without the level shifters, the regulated supply is pulled up to the unregulated voltage

through the over-voltage protection diodes in the microcontroller. Under low input power

conditions (unregulated voltage < 2.8 V), the total supply current of each level shifter is

less than 500 nA. At high unregulated voltages up to 6 V, total supply current remains less

than 2 µA.

3.3.3 Communication

The RF envelope is the baseband modulated data, but signal conditioning is required to

provide a logic-level signal to the microcontroller. Figure 3.6 shows an oscilloscope trace of

the demodulator in action. The raw rectifier output has insufficient swing to function as a

logic level signal. The demodulator, shown in Figure 3.7 compares the instantaneous voltage

output of the rectifier, which represents the RF envelope, to the peak value (minus a diode

drop), which acts as a reference for comparison. This is effective for two reasons. First, the

internal capacitance and back-leakage in the rectifier cause the rectifier output voltage to

decay exponentially when the input RF envelope is low. Second, the low-amplitude envelope

durations are short in duration, which means that the rectifier output has little time to decay.

This conspires to create short dips below the peak rectifier output, which are conveniently

detected by placing the reference one Schottky diode drop below the peak output. The

Page 51: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

29

Demodulator

Output

+

_

Demodulator

Enable

Level

Shifter

Regulated

SupplyRectifier

Output

1 µF

Figure 3.7: ASK demodulator schematic. A comparator uses the peak envelope minus a0.3 V diode drop as a reference. A footer transistor disables 10 µA of quiescent current.

comparator supply current is used to create the diode voltage drop without wasting power

through a resistor, and this also functions to provide power to the comparator only while

the reader is actually communicating to the WISP. A smaller, 1 µF cap simultaneously

stabilizes the reference voltage and comparator supply voltage.

3.4 WISP Firmware

The WISP firmware has several tasks. At the lowest level, it must provide power manage-

ment to keep the system due to undervoltage. In the middle is the communication code,

which implements the EPC Class 1 Generation 2 (Gen2) protocol and provides a means for

transmitting and receiving data. At the highest level is the application layer, which governs

the behavior of the WISP. The use of a programmable microcontroller allows WISP to be

easily configured for a wide variety of applications. Figure 3.8 shows these firmware blocks

along with their respective hardware connections. The latest WISP hardware (WISP 4.1)

employs a TI MSP430F2132 microcontroller. As TI has made improvements to the MSP430

product line, WISP has evolved to take advantage of new features, reduced power consump-

tion, faster clock speeds, and additional flash/RAM. As discussed in Section 1.2, Moore’s

Law is expected to drive down the cost per instruction as technology improves. This will

enable a richer feature set in wirelessly-powered devices. For now, however, fine-tuned

Page 52: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

30

Power Management

Voltage

Detector

Voltage

Sense

Communication

Application

Mod &

DemodSensors Hardware

Firmware

Figure 3.8: Firmware block diagram including relevant hardware connections.

Charging

(LPM4)

Power On Contiguous

Task

SV High

SV High

SV Low

Figure 3.9: Basic power management state diagram for standard tasks.

assembly code is required to simply meet timing constraints while operating the micro-

controller at its lowest voltage - a seemingly small detail which has a significant effect on

wireless range and operational duty cycle.

3.4.1 Power Management Firmware

The amount of RF power available to WISP varies greatly with distance, and this necessi-

tates robust power management circuitry for reliable operation. The basic power manage-

ment algorithm of WISP is shown in Figure 3.9. The µC powers on into Active Mode at

approximately 1.5 V. This mode draws several hundred µA of current, and would quickly

brown out the system if left active. The power management scheme immediately puts the

µC into low power sleep mode with an interrupt enabled for the hardware supervisor. When

Page 53: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

31

VS > VReq.Charging

(LPM4)

Power OnPoll VS

SV High

SV High

SV Low

Do Task

Task Complete

Figure 3.10: Advanced power management state diagram for power-hungry tasks.

the supervisor detects 2.0 V, an external input interrupt is triggered and a task may be

performed. The supervisor output can also be sampled directly, and more tasks may be

performed as long as the supervisor is high.

Not all tasks can be completed using the hardware supervisor. When the energy required

to perform the task requires Vs > 2.0V , a more advanced power management scheme is

employed. Figure 3.10 illustrates the state transitions. The bootup scheme is similar to

the basic power management scheme. When power is very low, the hardware supervisor is

used to wake the µC. However, when the voltage supervisor is high, the next state polls

the analog voltage sensor to determine the exact stored voltage. Polling requires a slightly

higher quiescent current because a clock source must be enabled in the µC. A fraction of

the very-low-frequency oscillator (”VLO”) or 32 kHz crystal oscillator can be used to set

the polling interval. These modes require approximately 1 µA instead of 0.1 µA quiescent

current from the µC. Voltage sampling is optimized for speed in order to reduce energy

expenditure (power times time). Sampling takes approximately 20 µs, and thus polling at

1 ms intervals or greater causes a very minimal impact on system power consumption. When

the voltage polling determines that sufficient voltage has been accumulated to perform the

task, the task is executed.

3.4.2 Communication Firmware

WISP performs EPC communication with minimum clock cycles via fine-tuned assembly

code. Both downlink and uplink communication rely heavily on the hardware timer module.

Downlink communication measures pulse intervals using the timer capture module, and

Page 54: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

32

uplink communication creates the periodic output waveforms using the timer output and

compare modules. Uplink communication code is synchronized in instruction count to the

output timer. The firmware updates the compare-to value real-time in synchrony with the

timer to affect the output frequency, which in turn encodes ones and zeros.

The most significant protocol limitation of WISP is that it currently only supports

Miller-4 uplink encoding with Tari values between roughly 10 µs and 25 µs. WISP previ-

ously supported Miller-2 encoding; however, lack of reader support motivated conversion

to the Miller-4 protocol. Furthermore, previous microcontrollers lacked flash memory to

support multiple uplink encodings. The latest WISP has additional flash memory, so a

motivated researcher could expand WISP’s protocol capabilities. Unfortunately the fastest

communication rates are unachievable with 1.8 V supply voltage clock rates (up to 6 MHz).

We experimented with using a Coolrunner-II CPLD for protocol encoding/decoding and

MSP430 for application and power management. We found that two XC2C256 (256 macro-

cell) CPLDs (one for uplink, one for downlink), or perhaps one XC2C512 (512 macrocell),

could implement the full protocol. However, the macrocell to package size ratio is unattrac-

tive for WISP-form-factor applications.

3.4.3 Application Firmware

The RFID protocol is described in detail in Chapter 2. The minimum amount of commu-

nication necessary to gather WISP data is Query/Ack, although formal Read and Write

memory operations are also supported. WISP requires a high clock rate of approximately

3 MHz while communicating, which consumes substantial power. For comparison, the cost

of a programmable µC versus synthesized Verilog code is roughly 100 to 1 in power con-

sumption. However, by re-purposing part or all of the 96-bit EPC ID for sensor data,

significant energy savings are realized versus a complete memory read operation. Other

EPC bit lengths are available, but 96-bits is common for commercial tags. The data format

is shown in Figure 3.11.

WISP can also process Select, Request RN, Read and Write commands. The WISP

allows use of standard EPC Read commands, enabling sensor data retrieval from user mem-

Page 55: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

33

Tag Type (1) Data (8) Version (1) Serial Number (2)

Figure 3.11: WISP packet structure. The information and number of bytes are shown ineach box. The tag type is often used to specify the sensor type. The data is typically sensordata. The WISP version allows marking of hardware or firmware revisions to be used forperformance evaluation or debugging. Finally, a tag-specific ID is given.

ory instead of re-purposing the EPC ID to encode the data. However, WISP populations

are typically less than 100 tags, and so ID space limitations have not been a major issue

in our experience. One reason that re-purposing the ID has not been a problem is due to

tag singulation changes between Gen1 and Gen2 RFID. Specifically, the use of a random

handle instead of ID tree-walking removes the effect of the ID on tag singulation. Various

sensor applications are described in detail in Section 3.8 which utilize this packet format.

3.5 WISP Troubleshooting

WISP requires non-conventional debugging methods due to its precarious power source. In

fact, many of the development challenges arise from power management issues. The WISP

Monitor [36] is a board consisting of a microcontroller and a UART-to-USB chip. The

Monitor allows a developer to monitor all pins (except the Voltage Regulator pin) on the

WISP 4.1 and stream debugging data to a PC via an FTDI FT232R UART-to-USB chip.

For simplicity, the Monitor uses the same µC model as the WISP. The WISP Monitor is

also powered by the USB cable, which allows it to not affect the WISP’s power budget. A

photo of the system is shown in Figure 3.12.

3.6 Discussion

Many applications require robust, high-bandwidth uplink (tag-to-reader) and/or downlink

(reader-to-tag) communication. Several realized examples will be presented in Section 3.8

including logged-data transfer, real-time sensor reading, actuation (ex, blink LED) in addi-

tion to standard Gen2 inventory activities. In many of these applications, achieving desired

Page 56: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

34

Figure 3.12: Photograph of the WISP Monitor connected to a WISP [36]. Data is sent backover the USB port.

performance can be difficult or infeasible because the Gen2 protocol, and RFID in general,

were designed primarily for inventory applications. That is, the protocol was optimized to

robustly and completely identify large tag populations. Operations such as tag memory

access are supported but inefficient due to protocol overhead, computer-to-reader latencies,

security issues and even the uncertainty of tag power availability. Despite these drawbacks,

the commercial availability of Gen2 readers and widespread industrial deployments of Gen2

RFID systems (see Walmart, DoD) allow Gen2 applications to be leveraged and expanded

by others in academia and industry.

3.7 Performance

The received voltage and power versus input power are shown in Figure 3.13(a), and the

rectifier efficiency versus input power is shown in Figure 3.13(b). The sensitivity suffers

approximately 10 dB versus integrated designs due to parasitics in packaged components.

Page 57: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

35

0

1

2

3

4

5

6

-20.0-15.0-10.0-5.00.05.0

Input Power (dBm)

Output Voltage

-270

-170

-70

30

130

230

330

430

530

630

Available Power (uw)

Low-to-High (V)

Turn On

Power (uW)

(a) Output voltage and power.

0.0%

5.0%

10.0%

15.0%

20.0%

25.0%

30.0%

35.0%

-10.0 -8.0 -6.0 -4.0 -2.0 0.0 2.0 4.0

Input Power (dBm)

Efficiency

(b) Power conversion efficiency.

Figure 3.13: Rectifier performance versus input power.

Page 58: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

36

0

10

20

30

40

50

60

70

80

90

100

1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0

Calculated Distance (m)

Percent (%)

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

5.0

5.5

0 2 4 6 8 10 12 14

Attenuation (dB)

Voltage (V)

Successful Responses Per Query (%)

Uplink Packet Errors (%)

Output (V)

Figure 3.14: WISP communication and error rates as a function of RF attenuation andcalculated distance.

As shown in Figure 3.14, the read rate also drops with input power unlike IC tags due to

the need for duty cycling (see Section 3.2).

Figure 3.14 shows communication and error rates as a function of RF attenuation and

calculated distance. The WISP was placed at 1 m from the reader antenna and an adjustable

attenuator was connected in series between the reader and antenna. Equivalent distance

is calculated using Friis’ Transmission Equation. The error rate is measured by counting

dropped packets. Because the reader silently rejects packets with errors, the WISP is

configured to number each ID sequentially. This allows missing packets to be detected on

the application side.

3.8 Applications

3.8.1 Light and Temperature

The WISP has been used for measuring a variety of sensors. Temperature and light data

are shown in Figure 3.15 and Figure 3.16, respectively.

Page 59: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

37

Figure 3.15: Temperature measurements in response to artificial stimuli [33].

Figure 3.16: Light sensor data plotted over a 24 hour period [33].

Page 60: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

38

0

50

100

150

200

250

300

350

400

450

500

0 5 10 15 20

Time (ms)

ADC Sample Value

0

5

10

15

20

25

30

35

Peak to Peak Noise

Sample

Absolute Noise

Scaled Noise

Figure 3.17: Acceleration measurements in time as the sensor output filter settles out. Thepeak to peak “absolute” noise for 100 samples is also plotted for each sampling time. Thenoise, scaled to correct for the fact that the sensor has not settled, is plotted as ScaledNoise.

3.8.2 Acceleration

Acceleration data is useful for user input devices as well as determining the orientation of

an object relative to gravity. The accelerometer (Analog Devices ADXL330) is a relatively

high-power sensor, drawing 200 µA. Also, a long filter time constant is used to prevent

vibration artifacts (aliasing) due to the WISP’s slow sample rate of 1 Hz to 50 Hz. This

long RC settling, combined with the high sensor current consumption, causes a significant

performance penalty due to the high amount of energy required per sample. Samples plotted

as the sensor value settles are shown in Figure 3.17. Because the filter is a simple linear RC

low-pass filter, the attenuation caused by the filter at any point is time can be calculated.

For a given settling time, the filter attenuates by a constant fraction regardless of the final

value. This can be compensated by multiplying the measured value by a scaling factor. The

scaling factor is calculated based on the RC time constant and the allowed settling time.

There is a tradeoff between noise and settling time (energy expenditure) shown in Fig-

ure 3.17. The more time is given to settle, the lower the sampling noise (Absolute Noise)

becomes. However, the noise is also scaled by the same constant used to compute the final

Page 61: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

39

Figure 3.18: Acceleration measurements as the WISP is rotated in orthogonal planes [34].

acceleration value. This is plotted as Scaled Noise. By approximately 5 ms, the noise has

reached its minimum and there is no benefit by waiting longer. Aggressive designs can use

2 ms or 3 ms settling time at the cost of higher noise.

Figure 3.18 shows acceleration data as the WISP is rotated in orthogonal planes. This

data can be streamed to a log file or used real time by the reader application. For example,

the WISP GUI (see Section 8) includes a demo where the user can orient a 3D object

by manipulating the orientation of WISP. The WISP accelerometer has also been used to

prototype a passive RFID security keycard or credit card activation mechanism [37]. The

user makes a motion gesture to activate the card, which prevents ghost-and-leech attacks.

3.8.3 LED

WISP can wirelessly power an LED. At short distances, the LED can be illuminated con-

tinuously. At long distances, the LED blinks at a duty cycle that is limited by the available

Page 62: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

40

Figure 3.19: WISP wirelessly powers an LED when commanded by the reader. The lightedLED indicates the location of an item in a bin.

power. An example application is shown in Figure 3.19. Here, the user selects an item on

the computer. The computer instructs a particular tag to blink to indicate the location of

the item.

3.8.4 Near Field

UHF RFID tags can also work with in the inductively-coupled regime. Figure 3.20 shows an

example of WISPs embedded in playing dice. WISP’s accelerometer allows the dice states

to be wirelessly reported, where they are displayed to the user.

3.8.5 Other WISP Applications

Implementation details of several WISP sensor applications are presented in Chapter 4 [22]

(encryption), Chapter 5 [23] (strain measurement), and Chapter 6 [24] (neural spike sensing).

3.9 Conclusions and Impacts

WISP’s software programmability and hardware flexibility enable its application towards

a wide variety of applications. Collaborative efforts with researchers changing from aero-

Page 63: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

41

Figure 3.20: Near-field WISP embedded in playing dice. WISP’s accelerometer allows thedice states to be wirelessly reported.

nautics engineers to neuroscientists has led to exciting technological advances. Perceiving

a continued interest in RFID research, the WISP developers launched a WISP Challenge.

After accepting proposals for WISP hardware grants, Intel donated over 150 WISPs to 23

research groups. Extensive documentation, tutorials and informational pages are available

online via a public wiki at http://wisp.wikispaces.com. As of July 2009, there were

over 70 unique visitors per day, over 50 members, and 51 wiki pages. It is our hope that

this donation will help spur exciting RFID research, especially in other disciplines such as

computer science, biology, and medicine.

Page 64: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

42

Chapter 4

MAXIMALIST CRYPTOGRAPHY AND COMPUTATION FOR UHFRFID TAGS

4.1 Introduction

Because of computational constraints on many RFID tags, classical cryptographic primi-

tives such as block ciphers and asymmetric cryptography were thought to be unrealistic on

a low-resource tag [38]. To this end, many lightweight cryptographic protocols have been

proposed [39, 40, 41, 42, 43, 44, 45]. However, many such protocols have serious vulner-

abilities [46, 47, 48]. Moreover, the lack of a development platform makes it difficult to

determine the feasibility of proposed cryptographic schemes. Thus, a popular approach is

to minimize cryptographic operations to ensure feasibility on an RFID tag. This minimalist

approach [38] can leave spare computational resources unused. An open question is then

how to best maximize the security on an RFID tag to fully utilize available computational

resources.

Our approach to maximizing security relies on low-power microcontrollers. Continuous

improvements in the efficiency of microelectronics (i.e., required energy per instruction) now

enables wirelessly-powered, general-purpose microcontrollers — infeasible at any reasonable

range a few years ago. Such microcontrollers make traditional cryptographic methods more

feasible on RFID tags.

Using a maximalist approach to cryptography, our results show that an RF-powered UHF

tag can perform strong encryption. We provide preliminary experimental results of imple-

menting RC5-32/12/16 [49] on WISP (Wireless Identification and Sensing Platform) [50]

— a battery-free platform powered and read by a standards compliant UHF RFID reader

running the EPC Class 1 Gen 1 protocol. Our contributions include:

1. We provide preliminary experimental data on how much computation is available on

a microcontroller-based RFID tag.

Page 65: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

43

2. We show that symmetric cryptography is feasible on an RF-powered, general-purpose

RFID tag. To the best of our knowledge, this is the first implementation of conven-

tional cryptography on an RF-powered UHF RFID tag.

4.2 Background on WISP Architecture

The WISP’s general-purpose computation capabilities are provided by an ultra low power

microcontroller, the TI MSP430F1232. This 16-bit device has just over 8 kBytes of flash

memory, 256 bytes of RAM and a 10-bit, 200 kilo-samples-per-second Analog to Digital

Converter (ADC). It can run at 8 MHz with a 3.3 V supply voltage, 6 MHz with a 3.0 V

supply, and 4 MHz with a 1.8 V supply. To erase and write to flash memory, a minimum

2.7 V supply is required. In this application, we used a 3.0 V supply because a suitable

ultra-low quiescent current 2.7 V regulator was not available. In active operation at 3.0 V,

the microcontroller consumes approximately 690 uA to 1.4 mA. Erasing and writing flash

both require approximately 3 mA. Reading from flash requires no additional power (above

ordinary active-mode operation). The flash in the MSP430’s information memory must

be erased in 128 byte blocks; its main memory is erased in 512 byte blocks. Bytes are

individually writable. Erasing the flash memory takes more time and therefore more energy

than writing or reading flash. Erasing a block requires 1.51 ms; writing a byte takes 32.7 us;

reading a byte from flash requires no more time than accessing RAM.

The MSP430 has various low power modes that are very useful for wirelessly powered

operation. Its lowest-power mode, “RAM-retention mode,” which stops all computation

but maintains state, requires only 0.5 uA at 1.5 V. The low power consumption of this

relatively new device is a critical factor in enabling use of a general-purpose microcontroller

in RF-powered RFID systems.

4.3 Related Work

Because of the resource-constrained nature of nodes in sensor networks, many of the design

criteria for security coincide with that for security of RFID tags. SPINS [51] and TinySec [52]

both present experimental data on implementation of security protocols in sensor networks.

Both works implemented RC5 for their block ciphers because of its small code size and high

Page 66: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

44

Platform Power

Com

puting

Stor

age

Com

mun

icat

ion

Dista

nce

EPC Gen1 [53] UHF RF State

machine

64 bits UHF

backscatter

3-7.5 m [54]

EPC Gen2 [53] UHF RF State

machine

96/128 bits UHF

backscatter

3-7.5 m [54]

WISP [55] UHF RF 16-bit

8MHz

8KB+

256 bytes flash

256 bytes RAM

UHF

backscatter

< 4.5 m

DemoTag [56] Battery 8-bit

16MHz

4KB EEPROM

4KB SRAM

128KB flash

HF

backscatter

N/A

Mica2 Battery 8-bit

8MHz

128KB ROM

4KB RAM

512KB flash

UHF/

433MHz

40-50 m [57]

Microchip

MCRF202

Inductive 12-bit

400kHz

96/128 bits LF

backscatter

1.3-10.2 cm [58]

Proxmark3 [59] USB 32-bit

16MHz

256KB flash

64KB SRAM

LF and HF

backscatter

N/A

RFIDGuardian [60] Battery 32-bit

520MHz

16MB flash

64MB SRAM

HF

backscatter

.5 m

Table 4.1: Comparison of WISP with other RFID devices.

efficiency, and they show that RC5 provides a balance between security and performance

for sensor networks. Although sensor nodes have limited resources, general-purpose RFID

tags are even more limited in memory, power, and computing capabilities. Sensor nodes

like the Mica2 platform have their own power source onboard, and they have orders of

magnitude more storage. Table 4.3 provides side-by-side comparison of different devices

with RF interfaces. WISP is most comparable to general-purpose RFID tags such as EPC

Gen1 and Gen2.

Page 67: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

45

There are many HF tags capable of cryptography because of their common usage in

building access control and authentication applications [61]. HF tags can afford to be more

expensive in terms of manufacture cost and therefore provide stronger cryptography because

the demand for tags with built-in security and privacy mechanisms, which is usually achieved

with a separate cryptographic engine. Israsena proposes the Tiny Encryption Algorithm

(TEA) as a suitable encryption engine for low-cost RFID applications [62]. Three different

architectures for the TEA encryption algorithm are presented that fit within the budget of a

5 cent per tag cost. Aigner and Feldhofer discuss their results on an implementation of Tiny

AES (TINA) as an ASIC [63]. Although both systems comply with stringent requirements

for low-cost RFID systems (average current consumption below 10 uA), they depend on

separate special-purpose circuitry. While ASICs are efficient in terms of power consumption

and cost, they are inflexible and limited to a narrow set of applications. Because of this

inflexibility, the cost for prototyping and development is relatively high — making it difficult

and time consuming to realize in actual applications. To the best of our knowledge, our

system provides the first UHF RFID tag using general-purpose microcontroller to provide

cryptographic capabilities under non-trivial computational constraints.

4.4 Measurements and Estimates of Computation and Power Consumption

In this section, we demonstrate that symmetric cryptography is feasible on microcontroller-

based RFID tags through an empirical study of RC5-32/12/16 on the WISP. Experimental

data shows how much computation is available on WISP with varying workloads. Although

WISP is relatively power-inefficient compared to ASIC designs, the logic gate count of WISP

is comparable to an EPC UHF tag — enabling a reasonable measurement of computational

capabilities and limitations of an RFID tag.

4.4.1 Measurements of Computation and Symmetric Cryptography

To show the feasibility of classical cryptography on a general-purpose RFID tag, we present

experimental results of implementing RC5 on WISP. We have chosen RC5 because of its

simplicity and relatively small memory requirements. RC5 is implemented with 32-bit word,

12 rounds, and 16 bytes of secret key. The 16-byte secret key is stored in flash. Even with

Page 68: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

46

careful choices of block cipher parameters, there exist practical challenges in implementing

RC5 on such a resource-constrained platform. Because of the extremely limited RAM

memory (256 bytes), minimizing stack size is crucial. For instance, our first implementation

resulted in an immediate stack overflow upon running the key scheduler. Since RC5 requires

expanded key table of size 2(r+1) words, where r is the number of rounds, careful memory

planning is required to reduce any unnecessary memory writes. The current implementation

fully utilizes all of its RAM memory so that expanded key table (2(r+1)*4 bytes = 104

bytes) is kept in RAM along with just enough room for the stack. One of the approaches

to remedy possible stack overflow would be to use a precomputed expanded key table. It

can be precomputed and stored in ROM or flash as long as the secret key remains the same

without hurting the performance since reading from flash or ROM takes the same amount of

power and cycles as reading from RAM. In our implementation, we compute the extended

key table once in every hardware reset. In other words, the key table is computed during the

first active cycle and kept in RAM unless WISP reaches a brownout voltage. This decision

is based on the assumption that it is unlikely for the secret key to be changed while an

RFID reader is polling.

Although WISP can be running at 3MHz at 1.8 V, our data comes from running WISP

at 3MHz at 3.0 V at the cost of reduced operating range to ensure reliable flash writes. Since

many applications with security requirements assume a relatively short read range during

the reader-tag communication, we believe that this is a reasonable demonstration of present

capabilities, which may be enhanced for greater computation and range with newer, more

efficient microcontrollers. There is already a newly released MSP430 that allows internal

flash writes with only 2.2 V [64]. A more recent version of WISP not used in this paper

has an external serial flash that operates at 1.8 V, allowing flash writes at much longer

ranges. We expect the range to continue to increase as the energy-efficient performance of

microelectronics continues to improve. Here, we aim to show any patterns or characteristics

involved rather than providing exact figures.

All three major functions of RC5 — setupKey(), encrypt(), and decrypt() — have

been implemented and measured on WISP. setupKey() is executed once after hardware

reset, and the 64-bit ID value is encrypted or decrypted once every duty cycle (e.g., during

Page 69: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

47

Power Save Mode (LP4)Power Save Mode (LP4)

Generate Packet(RC5, Calculate CRC) Generate Packet(RC5, Calculate CRC)

Receive and TransmitReceive and Transmit

Wait for Query (LP4)Wait for Query (LP4)

HardwareReset

Sufficient Voltage

SufficientVoltage

QueryInsufficient Voltage

Figure 4.1: WISP power cycle

“Generate Packet” stage in Fig. 4.1).

The duty cycling is enabled by a hardware voltage supervisor. This supervisor creates

software interrupts by toggling a microcontroller input high to indicate sufficient voltage for

operation. When the supervisor detects a voltage of 3.3V or greater, the microcontroller

is enabled and a computation is initiated. At the end of the computation, if the voltage

is between 2.7V and 3.3V the microcontroller enters very low power (RAM retention only)

mode. While in this low power mode, the harvester provides power to re-charge the supply

capacitor back to 3.3V for the next computation. Below 2.7V, a hardware reset is generated

by an insufficient voltage indicator output on the voltage regulator.

Fig. 4.2 depicts a typical lifecycle of WISP with RC5 encryption enabled and voltage

level patterns throughout the WISP duty cycle after a hardware reset, captured by an oscil-

loscope. The timeline shows that WISP spends most of its time in Power Save Mode(LP4),

approximately 90-95% of its duty cycle. WISP is in active mode only during “Generate

Packet” and “Transmit and Receive” stages.

Table 4.2 presents the execution time for three operations of RC5 on WISP tags operating

at a distance of 1 foot (0.3 m). The time measurements are the mean of ten separate trials of

execution after a hardware reset. The RFID reader constantly queries the WISP tag (about

15 polls per second) until the first response is received. Beyond the reader-tag distance of

Page 70: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

48

Power Save Mode (LP4)

Generate Packet(RC5, Compute CRC)

Transmit & Receive

VOUT

LP4RESET

Wait for Query(LP4)Reader ON

Reader OFF

Figure 4.2: An annotated scope trace of WISP execution and power consumption. Thedotted line measures the supply voltage VOUT in 1 V increments on the y-axis. While theReader is ON (i.e., sending queries), the WISP’s voltage level stairs up. The ramp up from0 V is omitted on the left. The voltage supervisor wakes up the WISP from LP4 when thevoltage level exceeds 3.3 V, and WISP begins its computation (“Generate Packet”). Whenthe reader receives the WISP’s response, the reader stops sending queries, therefore cuttingoff RF-power to the WISP. This is observed as a gradual decline of voltage at the right sideof the figure. The WISP first enters LP4 and then resets as the voltage level falls belowthe minimum operating voltage. The total latency from 0 V until the end of RF responsetransmission in this case is approximately two seconds.

2 feet (0.6 m), WISP generally does not function except for a few sporadic reads. However,

within the range of 2 feet, we observed no major difference in execution time.

RC5-32/12/16 function Execution time (msec)

setupKey() 7.93

encrypt() 1.43

decrypt() 1.39

Table 4.2: Execution time for operations in RC5 with 32-bit words, 12 rounds, and a 16-bytesecret key on a WISP tag. The WISP received uninterrupted power. Encryption happenson 64-bit messages.

Page 71: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

49

Figure 4.3: This estimate shows the computation available in one active cycle, where thepower harvester charges a 10 µF capacitor to its peak value and then the microcontrollerenters active mode, discharging the capacitor down to the minimum operating voltage. TheInstructions Available line ends abruptly at the distance where the harvester is no longerable to supply the microcontroller’s minimum operating voltage.

It is not apparent from the displayed data, but WISP with RC5 enabled does not greatly

affect the average tag response time. At a distance of 1 foot, the difference in the average

tag response time (e.g., time between when the first reader command is sent and the cor-

responding response is received) of WISP with and without RC5 is negligible (average of

462.6ms).

4.4.2 Estimates of Maximal Computation and Measurements of the Effect of Flash Writes

on Computation

Fig. 4.3 provides an estimate of available microcontroller instructions versus wireless dis-

tance. The model used to generate this plot is based on experimental data of WISP per-

formance, on the Friis transmission equation, and on published microcontroller power con-

sumption specifications.

In [55], experimental data of output voltage as a function of input power for the WISP

power harvester is given. Thus this data, together with the Friis equation, allows us to

predict the WISP’s output voltage Vrec as a function of distance d. This data captures

Page 72: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

50

the finite efficiency of the WISP harvester. An ideal, lossless harvester could produce any

desired voltage if it were able to accumulate energy long enough. Actual WISP devices

reach a steady state voltage, at which point power lost and power harvested balance.

The TI-MSP430 data sheet specifies its current consumption for given voltage and op-

erating frequencies. This allows us to calculate its energy per instruction: 281 pJ per

instruction at Vdd = 1.8 V or 900 pJ per instruction at Vdd = 3.0 V.

Of the energy stored in the WISP’s storage capacitor, only the voltage above the mi-

crocontroller’s minimum operating voltage is usable. If the capacitor is initially charged to

Vrec and its minimum operating voltage is Vdd, then 12CVrec

2 − 12CVdd

2 Joules are available

to run the computation. Using the previously calculated dependence of Vrec on distance, we

find energy available as a function of distance. Dividing by energy per instruction provides

number of instructions as a function of distance.

Our actual measurements in Fig. 4.4 show the relationship between the reader-to-tag

distance and number of cycles available in one duty cycle with varying workloads of flash

writes to the MSP430 information memory. While Fig. 4.3 provides the estimated upper

bound on how much computation we can afford on WISP, Fig. 4.4 presents how many cycles

are available after flash writes of different sizes. For consistency, the experiment involves

two steps, initialization and measurement. During the initialization, WISP’s flash memory

is checked if its state is erased or not. The flash is wiped out (writing ’1’ in all bits) before

each measurement is taken. During the measurement phase, WISP consumes power writing

to flash then raises a bit in a loop until WISP completely runs out of power (brownout

voltage of 1.5 V). These bit raises appear as a pulse on the oscilloscope. Since raising a

bit requires 4 cycles, the number of bits are counted to compute available number of cycles

after writing different number of bytes to the flash. Note that a minimal set of operations

are done during the “Generate Packet” stage in this experiment (e.g., preloading the 64-bit

ID into an array), and there is no RC5 computation or CRC computation.

We observed that the number of cycles available does not deviate significantly from each

other at short read ranges. At the medium range of 0.5 m, the power consumption of a

flash write causes a significant decrease in available computation. Beyond the distance of

0.75 m, WISP is not able to reach the minimum operating voltage of 2.7 V for flash writes,

Page 73: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

51

Figure 4.4: Preliminary measurements of the number of cycles available after writing toflash — computed after a minimal set of operations during the “Generate Packet” stageand without tag response over RF. Each data point represents a mean of ten samples.

therefore the plot lines for flash writes end abruptly. These measurements are preliminary,

and further research is necessary to determine a more precise relationship between flash

writes, erase segment sizes, and computation. However, one conclusion is that unnecessary

writes to flash at mid-range distances will significantly reduce available computation.

4.5 Conclusions

Our preliminary experimental data shows that UHF RFID tags with cryptographic capa-

bilities are no longer infeasible. Although our experimental platform exceeds the current

EPC UHF Class 1 tags in terms of computing power and storage, we believe that the de-

vice is comparable and a good representation of the future of UHF RFID tags. We believe

that the trend in microelectronics will continue to bring power-efficient and cost-effective

microcontrollers capable of more sophisticated computation. We hope that our work will

encourage further research to determine the feasibility of maximal-strength cryptography

with actual power measurements of RF-powered UHF RFID tags — an beyond algorithmic

estimates of space and running times.

Page 74: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

52

Chapter 5

A WIRELESS STRAIN SENSOR FOR EMBEDDED MATERIALHEALTH MONITORING

5.1 Introduction

With the adoption of composite materials in the next generation of cutting edge mechanical

structures, the pursuit of technology that will allow the health of these structures to be

easily monitored is increasingly important for maximum safety and minimum maintenance

cost. In next-generation commercial aircraft composites make up 16% of the airframe weight

of the Airbus A380 [65] and 50% of the airframe weight of the Boeing 787 [66]. Composites

are also widely used in military aircraft and astronautics due to their high strength and low

weight, resulting in improved fuel economy and reduced maintenance [67]. Using a composite

monitoring system that is embedded in the structure could help reduce the amount of time

required to perform the inspection and even allow continuous, real-time monitoring of future

aircraft, vehicles and other composite structures.

The use of composites is also accompanied by the potential for structural failure due to

weakening from moisture absorption and/or heat, stress fractures, and delamination [68].

One method of monitoring the health of composite structures is to affix numerous strain

gauges to the structure and detect anomalous conditions by excessive strain [69]. However,

the weight, complexity and maintenance of wires to each strain gauge are often prohibitive,

especially during use of the structure (ex, in flight). These wires are also a weak point when

cycled in the dynamic load conditions many composite structures experience.

To address the disadvantages in wired strain gauges, the Wireless Identification and

Sensing Platform (WISP) can be leveraged to power and measure each strain gauge. Fur-

thermore, WISP communicates bi-directionally with a UHF RFID reader, allowing data

to be collected and analyzed wirelessly. Because each of these devices has a unique bi-

nary thumbprint, they can be queried in serial, reducing the number of data channels and

Page 75: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

53

removing the need for heavy and weak wiring throughout the composite structure. The

following paper will discuss the WISP technology, its integration with a strain gauge and

the evaluation of its correlation with a traditional wired strain gauge.

5.2 Overview

The WISP is a wireless, battery-free sensing and computation device that communicates

with and is powered by a commercial UHF RFID reader. Much like conventional passive

RFID tags, WISP rectifies RF energy to power its onboard circuitry, and communicates

by reflecting power back to the reader through a technique called backscatter radiation.

However, unlike most RFID tags which backscatter a fixed identification number, WISP

encodes data into the tag ID using an ultra-low-power programmable microcontroller, the

Texas Instruments MSP430. Each ID sent from the WISP is reported by the reader to a

computer, which extracts the encoded data from the ID for plotting and/or logging. Using

the microcontrollers onboard ADC, WISP can power and measure attached sensors such as

temperature and ambient light. This sensor data is then reported to the computer through

each ID sent by the WISP. To measure a resistive strain gauge, an instrumentation amplifier

and Wheatstone bridge were designed to maximize strain sensitivity while keeping power

consumption to a minimum. The microcontroller takes a strain measurement by briefly

powering the bridge and amplifier and sampling the output with the ADC.

5.3 Design

Figure 5.1 shows a diagram of the WISP strain gauge measurement system. A computer

drives the RFID reader, which in turn queries the WISP. The WISP rectifier converts

incoming RF power to data and DC voltage. The harvested DC voltage is regulated to

1.8 V and used to power the microcontroller. When the voltage supervisor detects 1.9 V,

the microcontroller wakes from sleep mode via an external interrupt. The microcontroller

then makes a strain gauge measurement by powering the amplifier and Wheatstone bridge,

waiting for the amplifier output to stabilize, and converting the analog voltage to a digital

value. Next, the microcontroller embeds the sensor data in an EPC tag ID and computes

the required CRC for the ID. Finally, the microcontroller waits for a query from the RFID

Page 76: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

54

Wheatstone

Bridge

Instrumentation

Amplifier

MSP430

Microcontroller

Voltage

Regulator

RF

Rectifier

Sensor Power

RFID

Reader

UHF

Wireless

Link

ComputerRS-232

Voltage

Supervisor

Figure 5.1: High-level system diagram showing the communication path between the straingauge (Wheatstone bridge), WISP, RFID reader and computer.

2k

R4Resistor

3k

R3Resistor

3k

R1Resistor

2.99k

R2Resistor 1 3

2

R6POT

8

1

4

3

21

U1A

MAX4242EUA

84

75

6

2

U1B

MAX4242EUA

84

75

62

U2BMAX4242EUA

8

1

4

3

2

1

U2AMAX4242EUA

300k

R5Resistor

300k

R5Resistor

GND

VCC

VCC

VCC

VCC

GND

GND

GND

GND

300k

R5

Resistor

300k

R5

Resistor300k

R5

Resistor

VCC

300k

R5

Resistor

6k

R7Resistor

6k

R7Resistor

VCC

GND

200 ohms

R6Resistor

13

2

R2,POT

Vin,1VOUT

12

RSG

Strain Gage

0.1uF

C4

0.1uFC3

GND

GND

20 ohm pot

100 ohm pot

Vin,2

Figure 5.2: Circuit schematic of the Wheatstone bridge and strain gauge amplifier. Thedifferential signal of the Wheatstone bridge is converted to the single-ended output, VOUT,which is measured by the WISP ADC.

reader and responds using the generated ID. This process repeats as long as the reader is

polled by the computer, and update rates of approximately 10 to 20 sensor measurements

per second are achieved over a range of 1 m. Measurement speed decreases with range

because available RF power decreases with the square of the range from the reader antenna.

Page 77: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

55

The strain gauge instrumentation amplifier (IA) was built using discrete components

because an integrated IA chip that met both the low voltage (1.8 V) and low power (1 mW)

requirements of WISP is not available. A schematic of the design is shown in Figure 5.2.

The Wheatstone bridge, shown on the left of the schematic, has additional series resistance,

R4, added to the strain gauge in order to reduce current consumption of the bridge to

600 µA at 1.8 V. The resistance of the strain gauge, RSG, is

RSG = GF ∗Rg ∗ µStrain (5.1)

where GF, the gauge factor for the strain gauge, is 1.99 and Rg, the intrinsic strain gauge

resistance, is 1 kΩ. The differential output of the bridge, Vin,2 − Vin,1, is

Vin,2 − Vin,1 = Vdd

[R2 +R2,POT

R1 +R2 +R2,POT− R4 +RSG

R3 +R4 +RSG

](5.2)

where the resistor values are shown in Fig. 5.2, Vdd is 1.8 V and RSG is given by (1). The

output of the instrumentation amplifier, provided that the values of the resistances for R5

are equal, is

Vout = (Vin,2 − Vin,1)(1 +2R5

R6 +R6,POT) (5.3)

where Vin,2−Vin,1 is specified in (5.2) and resistor values are shown in Fig. 5.2. Finally, the

10-bit analog to digital converter output is

ADC =1024Vdd

Vdd(5.4)

where Vout is given by (5.3) and Vdd is the supply voltage, 1.8 V. R2,POT adjusts the zero-

strain output, and R6,POT adjusts the amplifier gain between 2000 and 3000. This allows

the full range of the ADC to be utilized by amplifying the strain gauge signal to the full

ADC input range of 0 V to 1.8 V. At a gain of 3000, a range of 2000 microstrain can be

measured, and at a gain of 2000, a range of 3000 microstrain can be measured.

5.3.1 Power Constraints

The power consumption of the strain gauge measurement circuit exceeds the power received

by WISP both due to the fact that the RFID reader transmits power less at less than 50%

Page 78: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

56

duty cycle and also due to the RF transmission loss over wireless distance. Therefore power

to the strain gauge is partially supplied from a storage capacitor which must be charged

to the supervisor detection voltage before a strain measurement or RFID communication

takes place. A larger capacitor requires more time to charge and consequently affects the

query rate of the WISP. The allowable sensor settling time, given by Equation (5.6), is a

direct function of the storage capacitor size, C, the power consumption of the sensor and

microcontroller, I, and the difference, 4V , between the supervisor detection voltage, VSV ,

and the minimum operating voltage, Vdd.

I = C4V4t

(5.5)

For the WISP strain gauge measurement system, C = 47uF , VSV = 1.9V , Vdd = 1.8V ,

and I is approximately 600 µA. This allows 7.8 ms if no RF power is received. An actual

settling time of 8.6 ms is used to allow as much settling as possible. This is feasible because

some power is also received by the reader.

5.3.2 Amplifier Gain and Bandwidth-Induced Attenuation

The differential gain of the amplifier is 2700 when R6 + R6,POT = 222 ohms. Using a power

supply, this gain is achieved in steady state; however a lower gain of approximately 2100

was observed experimentally during wireless operation. Due to power constraints, the strain

gauge and amplifier were powered for only 8.6 ms before a sample was taken by the ADC.

The low gain-bandwidth product of the MAX4242 op-amps used in the instrumentation

amplifier causes a low-pass filter response by the amplifier on startup. This causes atten-

uation when an ADC sample is taken before the output has reached its final, steady-state

value.

Figure 5.3 shows the relationship between WISP ADC output and the resistance of the

strain gauge in the wired and wireless power cases. In this test, a potentiometer (pot)

was used instead of a strain gauge and an ohmmeter was used to measure the true value

of the pot for each sample. The amplifier was adjusted to center the ADC output at 500

for a pot value of approximately 1004 ohms, which results in the centering of the two

series in Figure 5.3. Neglecting this expected offset, the gain of the pulse-powered amplifier

Page 79: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

57

0

200

400

600

800

1000

1200

1001 1002 1003 1004 1005 1006 1007

Strain Gauge Resistance (ohms)

ADC Value

Wireless Power (Duty Cycled)

Wired Power (Steady State)

Wireless Power Linear Fit

Wired Power Linear Fit

Figure 5.3: Comparison of instrumentation amplifier gain using an 8.6 ms pulse or contin-uous power to the amplifier.

configuration is 81% of the continuously-powered amplifier gain. This gain attenuation

is well modeled by a first-order low pass filter, which is created by the finite-bandwidth

op-amps of the instrumentation amplifier. The attenuation is given by

VADC

Vfinal= 1− exp(−2πtf−3dB) (5.6)

The gain-bandwidth-product(GBP) of the MAX4242 op-amps used is 90,000, and by

conservation of GBP, the bandwidth (f−3dB) is 90,000 / 2700, or 33.3 Hz. The strain gauge

was powered for t = 8.6 ms after which the ADC samples the amplifier output (VADC).

The attenuation of VADC with respect to the steady-state voltage, Vfinal, is predicted to

be 83.5% using (5), and this agrees with the experimental attenuation to within a factor of

1%. Offline compensation easily removes the linear filtering caused by the amplifier time

constant.

Page 80: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

58

(a) WISP, Wheatstone Bridge, and Strain

gauge.

(b) WISP attached to sample under strain.

Figure 5.4: Experimental setup of the WISP(S/G) comparison with an extensometer.

5.4 Experimental Setup

Tests were conducted with a 1/4” and 1” wide specimen of S-glass composite. A 1 kohm

Omega strain gauge with a gauge factor of 1.99 was attached to the test specimen and then

evaluated with both a standard wired setup and an integrated setup with the Intel WISP.

Along with the wired strain gauge and WISP strain gauge combination [WISP(S/G)] an

extensometer was attached to the specimen and run in parallel with both the wired and

wireless strain tests of the specimen (Figure 5.4).

The specimen was loaded in steps from 0 to 6400 psi. First the wired strain gauge and

extensometer measured the strain of the specimen in parallel. Following the traditional ten-

sion test of the specimen the wireless WISP(S/G) and extensometer measured the specimen

in parallel. Data was logged through Labview for all the wired applications and through a

lab top integrated with the RFID reader for the WISP(S/G). Along with this discrete load-

ing test a positive ramp from 0 to 6800 psi at 25 lb/sec was used to test the time response

of the WISP(S/G) combination and a sine wave loading between -1000 and 1000 lbs. at

0.25 Hz. The following will detail a comparison of the WISP(S/G) data with a traditional

Page 81: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

59

wired strain gauge to evaluate the proposed use of the WISP(S/G) for the construction of

a wireless intelligent composite for use in advanced composite structures.

5.5 Results/Discussion

Present the response of the WISP compared to other systems of measurement. Big conclu-

sion to draw seems to be that it is promising. Do you have any noise or integration results

you want to add? If we haven’t fixed the gain problem we may want to show something

that makes us think it matches up.

5.5.1 Data Analysis

To measure the strain from the 1 kohm strain gauge the WISP(S/G) has both a modified

quarter Wheatstone bridge and a 2070x gain amplifier before being converted to digital

in a 10-bit ADC. Data logged was the raw numbers form the ADC from 0 to 1024 which

represented a range from 1.8 to 0 V. To convert the rawADC data to a strain for comparison

with the extensometer the following formulas were used (Equation (5.7) and (5.8)). Equation

(5.7) converts the rawADC value logged from the WISP(S/G) to the voltage measured across

the Wheatstone bridge.

VADC =1024− rawADC

1024Vw

gain(5.7)

rawADC is the binary output of the ADC, 1024 is the full-scale value for the 10-bit

ADC, Vw = 1.8 V is the voltage across the Wheatstone bridge, and gain = 2070x is the

amount of amplification between the Wheatstone bridge and the ADC. Equation (5.8) then

converts the voltage measured across the Wheatstone bridge to the strain applied to the

strain gauge.

Vo ≈1124RR

Vw =112KsεVw (5.8)

The simplification in Equation (5.8) is warranted as long as the change in resistance is

much smaller than the resistance of the strain gauge. The resistors used in this configuration

are much greater than the change in resistance of the strain gauge.

Page 82: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

60

Figure 5.5: Strain recorded during the step loading of the specimen.

5.5.2 Specimen Tests

Figure 5.5 shows the step loading of the specimen and the strain measured by the traditional

wired strain gauge, WISP(S/G), and the extensometer for both cases. The results of this

test show the low noise of the WISP(S/G) measurement and the corroboration of its results

by both the extensometer and the traditional strain measurement. A ramped force was

applied to the specimen at 25 lb/sec to evaluate the stability of the WISP(S/G) signal

under a changing force (Figure 5.5).

Using a linear ramp of applied pressure, measurements of strain were taken from both

the wired strain gauge, the WISP strain gauge and an extensometer for reference. Figure 5.6

demonstrates the highly linear correlation between the wired ADC measurements and the

WISP wireless measurements. The low noise and high correlation to traditional strain

measurements of the WISP(S/G) strain measurement support the use of the WISP(S/G)

as a technology that can be developed into a wireless health monitoring system embedded

into composite structures.

Page 83: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

61

y = 1.0046x - 21.427

R2 = 0.9982

0

500

1000

1500

2000

2500

0 500 1000 1500 2000 2500

Wired Strain Measurement

WISP Strain Measurement

Figure 5.6: Correlation between the Wired and Wireless strain gauge measurements. WISPdata has been corrected to account for bandwidth-induced attenuation.

5.6 Conclusions and Future Work

The conclusion of these experiments is that the Intel WISP can be successfully integrated

with a strain gauge and used to monitor a specimen to create a useful stress v. strain plot.

The capability of wirelessly monitoring the strain in a composite structure can lead to the

creation a large, high resolution, network of WISP(S/G) that can monitor strain through

out a structure wirelessly. With discrete nodes of strain information, interpolation between

strain gauges will allow for the loading condition and health of a structure to be diagnosed

in pseudo real time. This not only makes the structures safer but promises to reduce the

cost of maintenance future commercial and military vehicles. The ability to embed these

components with a minimum of interference with the composite structure itself will allow

for the health monitoring system to last as long as the vehicle with a minimum of added

weight and complexity.

Page 84: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

62

The WISP(S/G) is a PCB design, and could not currently be embedded in a composite.

However, it is feasible to make WISP into a small IC that is roughly the size of commercial

RFID chips (less than 1mm2). We have successfully embedded a functional, commercial

Alien Class 1, Generation 1 RFID tag in an epoxy laminate composite. Similarly, a WISP

IC and strain gauge could be embedded in a laminate.

One challenge that arises when embedding WISP(S/G) in a carbon-fiber composite

airplane wing is metal shielding from lightening-hardening. This would severely attenuate

the interrogator signal, and the system would not work. There are several other methods

of applying the system, however. One important example is flight testing. It takes several

weeks to instrument an airplane with the required equipment to perform these tests. Instead,

WISP(S/G) could be attached like a label to the outside of the wing, or inside the wing

box, and measurements could be wirelessly obtained in this manner. This setup would take

hours instead of weeks to install, and provide tremendous cost savings.

A second parameter of interest for flight testing is pressure data along the airplane

wing. Ethan Shih at the University of Washington recently integrated a BMP085 pressure

sensor with WISP. This would allow pressure measurements to be performed in much the

same manner. Furthermore, there are commercially available MEMS pressure sensors which

affirm the possibility of a fully-integrated solution.

Finally, there have been numerous other power scavenging sources proposed in the lit-

erature which could power a low power sensor solution like WISP. Examples include piezo-

electric, solar and thermoelectric power harvesting. Application-tailored solutions may help

overcome specific challenges in the deployment of these sensors.

Page 85: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

63

Chapter 6

NEURALWISP: A WIRELESSLY-POWERED BRAIN INTERFACE

6.1 Introduction

Neural interfaces have made tremendous technology-driven advances in the recent past.

Cochlear implants are an early example of clinically relevant implantable devices [70], mod-

ern electronics are enabling previously impossible brain research experiments [71], and major

progress has been made toward neurally controlled prosthetics [72]. Because transcutaneous

wiring poses a significant infection risk, it is desirable that a neural interface communicate

and receive power wirelessly. Previous systems [73, 74] have achieved wireless operation by

using a near-field inductive link to transmit power and data. However, these systems require

that the external coil be located within a few centimeters of the internal coil. A wireless

neural interface with a range of 1 m or more will enable the removal of the interrogator

from the head and would allow wireless interfaces to be placed on small animals incapable

of carrying the interrogator hardware, such as mice.

We present a wireless neural interface which harvests power from the radio-frequency

(RF) energy provided by a standard commercial UHF RFID reader. The system operates

at a distance of up to 1 m from the reader. It records the spike count in a programmable

window (typically 1-10 s) and subsequently transmits the spike count to the reader as part

of the tag identification number that the reader is designed to acquire. This allows the

neuroscientist a wireless, battery-free method of recording spike density ( spikessecond) as various

tasks are performed or stimuli are presented.

6.2 System Design

The NeuralWISP builds upon our prior work on the Wireless Identification and Sensing Plat-

form (WISP) [34][21]. The WISP is a fully-passive UHF RFID tag that uses an ultra-low

power, general-purpose microcontroller (µC) for sensing, computation and RFID commu-

Page 86: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

64

RF Match and

Power Harvester

MSP430

μC

Preamp

Modulation

Storage Cap

Spike Detection

Voltage Sense

Power Management

Demodulation

Neural LNA

Electrode

Figure 6.1: Block diagram of NeuralWISP.

nication. The use of a programmable µC allows WISP to be easily configured for different

applications including measurement of temperature, light level, strain, and acceleration.

These environmental sensor signals change slowly and thus permit periodic, low-frequency

(1 to 50 Hz) measurement. However, a much faster sampling rate (at least 8 kHz) is neces-

sary to detect neural spikes due to spectral content between 500 Hz and 2 kHz.

Achieving an 8 kHz sampling rate under the constraints of the limited power budget of

an RFID tag is not possible with general purpose microcontrollers available today. Instead,

a continuous-time analog spike detector was designed to wake the system when spikes occur

via a µC interrupt pin. This helps minimize average power consumption because the µC

can remain in a low-power sleep mode during periods of inactivity and wake only to process

spikes or communicate with the RFID reader. The µC counts spikes during a programmable

window and is reset after the spike count is transmitted to the reader.

The architecture of the NeuralWISP is shown in Fig. 6.1. Like a typical RFID tag,

power is received at the antenna, voltage-multiplied, rectified, and regulated to provide a

stable system power supply. The neural input signal is amplified and applied to an analog

spike detector in addition to an analog-to-digital converter (ADC) integrated in the µC.

The µC performs the control and timing tasks, and implements the RFID communication

protocol.

Page 87: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

65

0%

5%

10%

15%

20%

25%

30%

35%

40%

-10.0-5.00.05.0

Input Power (dBm)

Eff

icie

ncy

(%

)

0

100

200

300

400

500

600

Po

wer

(u

w)

Rectif ier Eff iciencyAvailable Pow er (uW)

Figure 6.2: Measured rectifier output power and efficiency versus input power.

6.2.1 Power and Communication

NeuralWISP receives all of its power from the RFID reader. Commercial off-the-shelf

(COTS) UHF RFID readers operating in the U.S. Industrial, Scientific and Medical (ISM)

band (902 MHz to 928 MHz) are limited in transmit power to the FCC limit of 1 W

(+30 dBm). Readers often use an 8 dBi circularly-polarized patch antenna, which has an

approximately 60-degree beam width. Tags typically employ a 2 dBi dipole antenna. How-

ever, a number of factors decrease the power available to the tag. The most significant

effect is path loss; Friis’ Transmission Equation predicts that the amount of power received

decreases with the square of wireless range. The circularly-polarized reader antenna in

conjunction with a (linearly-polarized) dipole tag antenna also incurs 3 dB of polarization

loss. Additionally, amplitude modulation in the downlink (reader-to-tag) and backscatter

modulation in the uplink (tag-to-reader) communication causes up to 3 dB of loss. Finally,

tag rectifier efficiency, shown in Fig. 6.2, of about 25% causes 6 dB of loss. All included,

427 µW (-3.7 dBm) is available after the rectifier at 1 m. Implantation causes additional

losses due to dielectric constant mismatch at the air-tissue interface; however, experiments

have demonstrated the feasibility of radiative power transfer in this regime [75].

Page 88: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

66

Modulator

Demodulator

Regulated

Supply

Supervisor

Interrupt

+

_

Demodulator

Enable

915 MHz

Dipole Antenna

Cstorage

Level

Shifter

SV

1.8 V

Reg.

Figure 6.3: Schematic of the RF front end including voltage-multiplying rectifier, modulator,demodulator with level shifter (LS), and voltage supervisor (SV). RF (low capacitance)Schottky diodes are filled black, and DC (low leakage) Schottky diodes are filled white.Also shown is the 1.8 V regulator, providing a stable supply for the rest of the system.

The power and communication circuitry, shown in Fig. 6.3, are similar to that of con-

ventional RFID tags, and [21] presents the design in detail. A 5-stage voltage-multiplying

rectifier converts the received RF signal from the reader into an unregulated voltage source

for the tag, which is stored on Cstorage. An L-match network transforms the tag impedance

to match that of the antenna. The harvested voltage is converted to 1.8 V by a low-drop-

out linear regulator with 1 µA quiescent current, which provides a stable supply voltage for

the system. The voltage supervisor provides a digital interrupt to wake the microcontroller

while waiting for sufficient energy to operate.

NeuralWISP communicates using the EPC Class 1 Generation 2 RFID protocol [27],

allowing compatibility with industry-standard COTS readers. Downlink (reader to tag)

communication is accomplished through reader amplitude modulation. The demodulator

employs a comparator to threshold the instantaneous received voltage against the average

(minus a diode drop). A footer transistor disables the comparator to save 10 µA of qui-

Page 89: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

67

escent current while communication is not needed. Uplink (tag to reader) communication

is accomplished by modulation of the tag reflection coefficient. The modulator uses an RF

transistor to short the antenna terminals together, thereby producing a strong reflected

signal. In the worst case, approximately half of the nominal power is available to the tag

due to reader amplitude modulation and tag reflection modulation. Note that tag ID and

memory communication errors are minimized through use of a 16-bit CRC. Further details

on the protocol can be found in [27].

6.2.2 Analog Signal Path

The extremely low signal levels recorded from neural probes place severe constraints on the

analog front-end. Input-referred noise levels must be < 10µVRMS while providing good lin-

earity and high gain. These requirements frequently result in the low-noise neural amplifier

consuming a majority of the system power. In the NeuralWISP, the power dissipation limits

the wireless range, so power must be minimized. We designed a custom low-noise amplifier

(LNA) in a 0.5 µ m SOI BiCMOS process to meet these requirements. The amplifier is

designed to provide a gain of 40 dB. A schematic is shown in Fig. 6.4.

The amplifier is built using a two-stage op-amp with capacitive feedback, shown in

Fig. 6.4(b). The mid-band gain is set to 100 by the ratio of the C1 and C2, which are 20 pF

and 200 fF, respectively. Sizing the input capacitor C1 presents a tradeoff between noise

performance and chip area. Because large input transistors are used to minimize 1/f noise,

the op-amp input capacitance Copamp is on the order of 1 pF. The input signal is attenuated

by the capacitive divider formed by the C1 and the op-amp. Since any attenuation before the

op-amp will directly increase the input-referred noise, C1 should be much larger than Copamp

to minimize attenuation. However, overly large values for C1 would lead to unnecessarily

large chip area.

A closed-loop configuration was chosen for this system because open-loop amplifiers,

while demonstrating superior noise efficiency factors (NEF), typically suffer from inferior

power-supply rejection [76]. The LNA is AC-coupled to reject DC offsets resulting from

tissue-electrode interactions [77].

Page 90: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

68

-

+

IN20pF

20pF

200fF

PR1

PR1

Out

Ref

800kΩ

.5μA

0.5μA

2μA

RST

RST

(a)

VOutVIn- VIn+IBias

(b)

Figure 6.4: (a) Schematic of custom 8 µA low noise neural amplifier fabricated in a 0.5µmSOI CMOS process (b) Schematic of the op-amp used in the low-noise amplifier.

MOS-bipolar pseudo-resistors [77] (PR) were used to set the DC bias point. The high-

pass pole frequency is set by C2 and the resistance of the PR. For small signals, the PRs

have an incremental resistance of about 1012 Ω, resulting in a low frequency pole below

1 Hz. The high-pass corner frequency set by the PRs is much lower than is necessary for

the extra-cellular recording task being demonstrated here. However, all other options have

significant drawbacks. A poly-silicon resistor would be prohibitively large, while transcon-

ductor implementations of high-valued resistors consume additional power and contribute

Page 91: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

69

noise. Previous amplifiers have used a subthrehold MOS transistor as a resistive element,

but typically require a variable gate voltage to tune the resistance [78, 79]. Because the

amplifier described here was designed to operate without any programming or configuration

and to be suitable for a variety of neural recording tasks targeting both action potentials

and field potentials, the PR was deemed to be a simple, flexible, and reliable method for

setting the amplifier’s DC bias point.

To avoid long settling times on power up due to the PRs, a power-on-reset circuit is

included. When the reset signal is asserted, MOS switches temporarily short the PRs,

allowing the amplifier to reach its equilibrium bias point quickly.

A source-follower output stage was chosen for its flexibility with respect to load con-

ditions. A resistive load to ground will increase the current in the NMOS source follower

transistor, allowing the amplifier to automatically adapt to resistive loads without consum-

ing extra static bias current under lightly loaded conditions or using a complicated class AB

output stage. A PMOS source follower is used to shift the output level up into the input

range of the NMOS source follower, which drives the off-chip load. The LNA chip is com-

pletely self-contained, and includes a supply-independent bias current generator allowing

consistent operation over a range of 1-5 V.

An additional gain of 20 dB is provided by a second amplifier built from two OPA349

op-amps, shown in Fig. 6.5. The first op-amp is used to establish a 0.6 V reference for AC

coupling the amplifier stages. While the reference is derived from the supply, local regulation

(see Fig. 6.3) ensures low-frequency stability and the 100 pF capacitor shunts high-frequency

noise to ground. The second opamp is used in a non-inverting gain configuration. The gain

of the first stage allows relatively noisy micro-power op-amps to be used for the second gain

stage. Consequently, the second stage consumes only 1.9 µA from a 1.8 V supply, including

the reference.

The amplified signal is applied to an analog spike detector. Numerous algorithms for

spike detection have been developed, ranging from simple thresholding detectors to sophis-

ticated supervised learning classifiers such as support vector machines [80]. Even with many

other options available, the simple thresholding detector remains popular [73, 74]. An anal-

ysis of spike detection methods in the context of power constraints found thresholding on

Page 92: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

70

the absolute value to provide the best tradeoff between accuracy and computational re-

quirements [81]. For this work, a single-ended thresholding detector was found to have the

appropriate combination of simplicity and efficacy.

The signal is low-pass filtered with a time constant of 0.1 s to generate the detection

threshold. Deriving the threshold from the amplified signal prevents any offsets in the

op-amps from corrupting the detection results. The signal is also shifted towards 0 V

and attenuated by up to 15% via a variable-ratio resistive divider. A digitally-controlled

resistor, variable from 0 Ω - 50 kΩ, determines the attenuation of the divider and thus the

sensitivity of the spike detector. The shifted signal is compared to the low-pass filtered

signal to generate the detection signal, which triggers an interrupt in the µC. The spike

detector’s programmable threshold is set by the µC to a level stored in flash memory. The

threshold level can be chosen by the user prior to deployment. Firmware control of the

threshold also allows for future implementation of real-time adjustments to the threshold.

Such adjustments could be made based on the observed input or by including adjustment

commands in the data sent from the reader to the NeuralWISP.

The output of the second amplifier is also connected to the ADC input of the MSP430

microcontroller to allow for direct digitization of the neural signal. The ADC has 10-bit

resolution and uses the supply as the reference. This provides a 1.76 mV LSB, or 1.76 µ V

input-referred.

6.2.3 Digital Control

An MSP430F2274 microcontroller (µC) is used to implement control, timing, and commu-

nication tasks. Fig. 6.6 shows the software architecture. On bootup, the µC configures the

adjustable resistor in the spike detector. During the primary mode of operation, the µC

will count spikes during a user-specified time interval (typically 1-10 s) and transmit the

number of spikes detected at the end of the interval. During the counting interval, the µC

is in a low-power sleep state for the majority of the time. The spike detector triggers an

interrupt, which causes the µC to wake up, increment the spike count, and return to sleep.

A timer drives another interrupt, which signals the end of the counting interval, causing

Page 93: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

71

MCP4012

50k

40dB LNA

2M

OPA349

300k

TLV3491

4M

100pFOPA349

1M

0.1uF900k

100k

1M

0.1uF

Electrode

Digital Pot Control Spike

Interrupt

Figure 6.5: Analog front end circuitry, including custom 40 dB LNA, 20 dB post-amp,and spike detector with programmable threshold. The 600 mV reference is generated bya voltage divider from the regulated 1.8 V supply and an op-amp buffer. A non-invertingop-amp configuration provides the additional 20 dB gain. The spike detector compares theaverage (low-pass) signal to a programmable fraction of the instantaneous signal.

Comm State

(Sleep Mode)

Increment Counter

Construct Packet

Reset Counter

Send Packet

Spike State

(Sleep Mode)

Spike Interrupt

Timer Interrupt

Holdoff

RFID Query

Timer Interrupt

Figure 6.6: Software state diagram. The µC is in the low-power Spike State for the ma-jority of the time, awakening only to increment the spike counter after a detection or tocommunicate with the reader.

the µC to exit the spike-counting mode and await a communication session with the reader.

After communicating with the reader, the µC pauses for 3 s to allow the analog circuits

to recover from RF interference that occurred during the read, then returns to the spike

Page 94: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

72

counting phase and repeats the cycle.

6.2.4 Application Layer

A graphical user interface was developed to read, graph and log spike density measurements

reported by the NeuralWISP to the RFID reader. The application communicates via Eth-

ernet using LLRP (Low Level Reader Protocol) [29]. A screen capture of the application in

action is shown in Fig. 6.7.

The state cycle for the application begins by polling the reader until the NeuralWISP

powers up and reports a spike count. Then the application turns off the reader for a

predetermined spike sampling period, which is set in the code of the NeuralWISP and in the

application. During this time, the NeuralWISP counts spikes. After the sampling period has

ended, the application again polls the reader. The NeuralWISP recharges and the reports

its spike count. The cycle then repeats. Note that the application stays synchronized with

the NeuralWISP’s charge-record-upload state cycle by waiting for the WISP to respond in

the upload state. This ensures that timing errors do not accumulate over several iterations

of the state cycle.

6.3 Test Results

The fabricated board is shown in Fig. 6.8. The populated board alone weighs 1.0 g, and

a 900 MHz wire dipole antenna (not shown) weighs approximately 0.6 g. During spike

counting, the system draws an average of about 20 µ A of current from its unregulated

supply, of which 8 µ A is consumed by the neural LNA. A commercial RFID reader with

+30 dBm transmitted power was used to wirelessly supply power and communicate with

the NeuralWISP.

6.3.1 Analog Front End

The amplifier’s noise was measured using an HP 35670A dynamic signal analyzer with the

input grounded. Fig. 6.10 shows the input-referred noise spectra at the output of both the

LNA and the post-amp, which was integrated from 0.5 Hz to 25 kHz and divided by the

Page 95: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

73

Figure 6.7: A graphical user interface processes data from an Impinj Speedway RFID reader,displaying a graph of spike detections versus time.

measured gain to find the input-referred RMS noise voltage, which is 4.4 µVRMS for the low-

noise amplifier. Even with the use of the micro-power commercial op-amps, it can be seen

that the gain of the LNA suppresses the noise contribution of the post-amp. Additionally,

very low-frequency noise is filtered by the AC coupling between the LNA and post-amp.

The LNA can operate from a supply between 1 V and 5 V, and provides a measured

gain of 39 dB with a bandwidth spanning from 0.5 Hz to 5.9 kHz. Operating from the

NeuralWISP’s regulated 1.8 V supply, the amplifier consumes 8 µ A, including the bias

generator and output buffer. Fig. 6.9 shows the measured frequency response of the first

stage and the combined response of both gain stages. The LNA combined with the second

amplifier provides a mid-band gain of 56 dB with a bandwidth from 2 Hz to 4.9 kHz.

Table 6.1 and Table 6.2 show a comparison of the LNA used here to other published neural

amplifiers. The core amplifier has a current consumption of 3.3 µ A, corresponding to an

NEF of 4.0, which is comparable to other published amplifiers. Including buffering and bias

Page 96: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

74

Table 6.1: Comparison of Neural Amplifiers

Gain IAmp NEF vni,RMS

This Work1 39dB 3.3µ A 4.0 4.4µ V

This Work2 39dB 8µ A 6.2 4.4µ V

Holleman [82] 36.1dB 805nA 1.8 3.6µ V

Rai [83] 38.3 12.5µ A 2.5 1.95µ V

Harrison [77] 39.5dB 16µ A 4.0 2.2µ V

Denison [84] 45.5dB 1.2µ A 4.9 .93µ V

Wu [85] 40.2dB 330nA 3.8 .94µ V

Wattanapanitch [86] 40.9dB 2.7µ A 2.7 3.1µ V

1Core amplifier only

2Complete amplifier chip, including buffers reference and biasgeneration

Table 6.2: Extended Comparison of Neural Amplifiers

THD (@ Input) PSRR Bandwidth Tech.

This Work1 0.2% @ 5mVpp 45dB .5Hz-5.9kHz .5µ m

This Work2 0.2% @ 5mVpp 45dB .5Hz-5.9kHz .5µ m

Holleman [82] 7.1% @ 1mVpp 5.5dB .3Hz-4.7kHz .5µ m

Rai [83] 1% @ 1mVpp 63dB .023Hz-11.5kHz .13µ m

Harrison [77] 1% @ 16.7mVpp ≥ 85dB .025Hz-7.2kHz 1.5µ m

Denison [84] — — .5Hz-250Hz 0.8µ m

Wu [85] .053% @ 5mVpp 62dB 3mHz-245Hz .35µ m

Wattanapanitch [86] 1% @ 7.3mVpp 75dB 45Hz-5.3kHz .5µ m

1Core amplifier only

2Complete amplifier chip, including buffers reference and bias generation

generation, the total current is 8 µ A, leading to an NEF of 6.2. It should also be noted

that some of the amplifiers in the table [84, 85] are designed with much lower bandwidth for

monitoring EEG signals or field potentials, and as a result have lower current consumption.

Page 97: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

75

Figure 6.8: System photograph. Inset shows chip-on-board mounting of the custom low-noise amplifier IC.

The NEF accounts for different bandwidths and noise levels, allowing comparison between

amplifiers designed to different specifications.

6.3.2 Spike Detector

To characterize the spike detector, we applied a synthesized neural recording [80] to the

NeuralWISP input. This technique allowed us to vary the SNR and spike rate in the

recording and provided a reference against which to compare our measured spike detection

results in order to characterize the detector accuracy. Fig. 6.11 shows the operation of the

detector on a single spike, with an 85 µVP−P input signal. Software debouncing in the

interrupt handler prevents any glitches in the spike detection signal from causing errors in

the spike count.

Fig. 6.12 shows the spike detector accuracy. Spikes were detected using the hardware

analog spike detector (circle tick) and also using a PC-based threshold-crossing detector

Page 98: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

76

10−1

100

101

102

103

104

20

30

40

50

60

Frequency (Hz)

Gain

(dB

)

LNA

LNA+2nd Amp

Figure 6.9: Measured gain versus frequency for both low-noise amplifier (LNA, bottom),and the combined gain of the LNA and 2nd amplifier.

100

101

102

103

104

10−9

10−8

10−7

10−6

Frequency (Hz)

Vn

i,rm

s (

V/√

Hz)

LNA

LNA+Post−Amp

Figure 6.10: Measured noise spectra at the output of the LNA and the post-amp.

(square tick) for comparison. Both detectors were run on synthetic recordings with an am-

plitude of approximately 400 µVP−P and SNR of 10 dB (left) and 6 dB (right). The results

were compared with the known spike times provided by the signal synthesis software. The

analog detector demonstrates comparable discriminative abilities to the software detector,

indicating that noise contributions from the analog front end do not limit spike detection

performance.

Page 99: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

77

−40

−20

0

20

40

Post−

am

p O

utp

ut (m

v)

0 20 40 60 80 1000

1

2

Time (ms)

Dete

ct (V

)

Figure 6.11: Operation of the spike detector. The output of the post-amp is shown at thetop with an input signal containing a single spike with SNR of approximately 10 dB. Theamplitude at the input to the NeuralWISP is approximately 85 µ Vpp.

6.3.3 System Operation

Fig. 6.13 demonstrates the operation of the NeuralWISP. The middle trace is the unregu-

lated voltage stored on a 100 µ F capacitor Cstorage, which begins at 0 V, since the WISP

starts out with no stored energy. Initially, the reader is configured to transmit power in

continuous-wave (CW) mode, which charges the storage capacitor to 5.5 V where it is

clamped by a zener diode. As the stored voltage rises, the µC boots up (A). At point (B)

continuous-wave transmission stops and the RFID reader reads data from the WISP. The

first read following bootup will contain empty data. Following the read, the µC enters a 3 s

waiting state (C) in order to allow the analog circuits to recover from RF interference which

occurred during the read. After 3 s, the WISP begins counting spikes (D) for 5 s. After

the spike-counting phase, the reader again transmits CW power (E) to recharge the storage

capacitor, followed by another read, which retrieves data from the previous spike-counting

phase (D). This cycle is repeated indefinitely.

Page 100: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

78

0 0.2 0.4 0.6 0.8 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

FNR

FP

R

Thresholding in SW

WISP HW Detector

0 0.2 0.4 0.6 0.8 10

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

FNRF

PR

Thresholding in SW

WISP HW Detector

Figure 6.12: Accuracy of the spike detector compared to a software spike detector for SNR= 10 dB (left) and SNR = 6 dB (right). The x-axis is the false negative rate (FNR =Number of missed spikes / Number of total true spikes). The y-axis is the false positiverate (FPR = Number of false detections / Number of total detections).

Figure 6.13: Two read cycles of wireless operation, showing the spike detector output (top),the unregulated stored voltage (middle), and a microcontroller output (bottom) pulsed toshow operation. The data was taken at a distance of approximately 1 m from the reader.

The NeuralWISP’s flexible hardware platform allows for many potential extensions to

the present functionality. For example, it could be configured to sample spike waveforms

after a spike is detected, and transmit the digitized data. An appropriate duty cycle would

Page 101: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

79

0 0.5 1 1.5

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1

Am

plif

ied S

ignal (V

)

time (ms)

Original Data

Digitized Data

Figure 6.14: A single spike digitized by the on-board ADC. The µC began sampling andconverting in response to an interrupt from the spike detector.

need to be chosen in order to meet the constraints imposed by the data rate allowed by the

tag/reader interface. Fig. 6.14 shows a spike captured and digitized by the NeuralWISP,

where the digitization was begun in response to a detected spike. The digitized spike

waveform is superimposed on the original waveform. This experiment was performed with

wireless power, but the data was retrieved through a wired interface. It demonstrates that

accurate reconstruction of a spike can be accomplished by waking the µC and ADC from low-

power sleep after spike detection, dramatically reducing average system power. Digitization

of detected events could also be used to implement secondary screening of spikes in software,

improving effective detection accuracy.

6.3.4 In-vivo Results

To validate the NeuralWISP’s ability to detect spikes in vivo, measurements of wing muscle

activity from a Manduca Sexta moth were taken. Note that while the prototype NeuralWISP

is too heavy to be carried by a moth, integration of NeuralWISP onto an IC could allow

in-flight measurements to be performed. Because the recording device is wirelessly powered,

no batteries or wires are required. Since the battery consumes a large fraction of the weight

budget of flying-insect-mounted electronics [87], a wirelessly-powered interface would permit

significant weight reduction compared to traditional sensing schemes. The setup is shown

Page 102: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

80

Figure 6.15: In vivo experiment setup showing Manduca Sexta moth with tungsten wireelectrodes in wing muscle tissue. The electrodes are connected to the NeuralWISP via aresistive attenuator. Spike density measurements are wirelessly recorded and communicatedto the RFID reader.

in Fig. 6.15, and a wirelessly-powered recording captured by an oscilloscope is shown in

Fig. 6.16.

NeuralWISP relies on extremely low-power custom analog front end circuitry to allow

operation from a wireless power source. In order to test the compatibility of the analog

front end with extra-cellular neural recording, we performed in vivo measurements on a

macaque monkey (Macaca nemestrina). Fig. 6.17(a) shows spikes recorded with the Neu-

ralWISP LNA and post-amp, digitized with a standard rack-mounted acquisition system,

and high-pass filtered off-line. The clean grouping of the spikes demonstrates that the Neu-

ralWISP’s analog front end has compatible input impedance and sufficient linearity to allow

accurate recording of small neural signals in the presence of 60 Hz interference and local

field potentials.

Fig. 6.17(b) shows the same spikes without the high-pass filtering. A combination of

60 Hz interference and low-frequency field potentials cause the spikes to be superimposed

on a varying baseline level, complicating detection. For use in situations with significant

Page 103: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

81

Figure 6.16: Wirelessly-powered data from wing muscle tissue captured over a 2 s timespanby an oscilloscope. The top trace shows the post-amplifier output, which corresponds toapproximately 2.4 mVpp at the input. The bottom trace shows the NeuralWISP spikedetector output.

0 0.2 0.4 0.6−40

−20

0

20

40

time (ms)

Input (µ

V)

(a)

0 0.2 0.4 0.6−200

−100

0

100

200

300

400

time (ms)

Input (µ

V)

(b)

Figure 6.17: Spikes recorded through the NeuralWISP’s amplifiers and digitized with stan-dard rack-mounted recording equipment. (a) High-pass filtered spikes line up well. (b)Unfiltered data indicates significant variation in the baseline.

low-frequency interference similar to that shown here, future systems should incorporate

additional high-pass filtering. Fortunately, such filtering (and additional gain, if desired)

can easily be implemented at a negligible cost in power using micro-power op-amps such as

the one used for the post-amp described above.

Page 104: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

82

6.4 Discussion

There are a number of trade-offs in the design of wirelessly powered neural recording systems.

Some involve the use of the RFID protocol. RFID is a relatively mature technology, which

provides a robust communication layer and low cost (approx. $1000) interrogators. This

enables wide-scale deployment as well as collaboration with other researchers. Conversely,

RFID technology was designed for reading many nodes once. In sensing applications, the

goal is often to read one node repeatedly. This repurposing of the protocol constrains data

throughput due to increased overhead. For example, the query rate for 96-bit tag IDs is

limited to about 1 kHz. Reading other tag memory increases throughput but decreases the

tag access rate to about 360 Hz for multi-byte memory access. These estimates are based

on 160 kbps downlink and 640 kbps uplink with preamble overheads, even bit distributions,

and a single tag present (no singulation) [27].

Radiative energy transfer also has a number of trade-offs. Advantages include the abil-

ity to power many nodes with one reader and transfer power over many wavelengths of

distance (wavelength λ = cf = 0.33 m for UHF RFID). Limitations on the number of nodes

include shadowing (where one node blocks the reader signal from reaching a second node)

and detuning (nodes in close proximity may detune the frequency band they are sensitive

to). However, compared to other approaches such as inductive coupling, radiative energy

transfer enables operation at much greater range. Disadvantages include low power transfer

efficiency and the necessity of an antenna on the scale of the RF wavelength. Low power

transfer efficiency, in turn, limits the features of the nodes due to power constraints. How-

ever, reductions in power consumption due to technology scaling are enabling increased

functionality for the same power budget.

Lastly, while one reader may power many tags, multiple readers in close proximity must

time-multiplex. This is due to the tag demodulator design: simple peak detection lacks the

ability to perform channel selection. Therefore, tags cannot selectively listen to one reader

while other readers are transmitting simultaneously. This effectively means that tags in close

proximity must share the communication bandwidth of one reader. Spike sorting, binning,

and other forms of signal processing will allow maximum information to be gathered from

Page 105: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

83

a collection of nodes under the limited bandwidth constrains of one reader.

Wireless neural sensing inherently places an RF transmitter in close proximity to the

signal to be detected. While pickup of 60 Hz noise can be greatly reduced due to the sensor

being wirelessly interrogated, the radio can introduce interference in the measurement. In

this work, we separated the sensing and communication temporally, which places time gaps

in the recording. In order to sense and communicate simultaneously, several techniques may

help eliminate RF interference. First, integration of the electrodes, amplifiers, and spike

detection circuitry to a size well below the RF wavelength helps prevent the electrodes from

acting as an antenna. Second, integration of RF-rejection filters in the signal path helps

eliminate noise picked up by the electrode. Finally, shielding of the low-noise amplifier and

filters helps reduce direct interference from the amplifier.

As the NeuralWISP is currently configured, it measures spike density in a time window

of a few seconds. While this is not the typical protocol for neuroscience experiments, it

provides an indication of the activity level of the neuron being observed. This can provide

useful information about sleep-wake state [88]. Simple measures of neural activity levels

can also be used to gauge recovery from brain insult, such as due to cardiac arrest. In [89]

EEG was used to measure activity, but spike density could serve as another indicator. With

additions to the firmware, even the modest data rate supported by the RFID protocol

would allow other information to be transmitted, such as histograms of inter-spike intervals

or spike amplitudes.

NeuralWISP enables a variety of interesting applications including monitoring of small

animals and insects in laboratory environments as well as implantable neural sensors pro-

vided the device size is reduced through ASIC integration. Similar applications for monitor-

ing heart rate, blood sugar, blood pressure, etc. could leverage this technology by applying

it to other biomedical sensors.

6.5 Conclusions

We have demonstrated a wirelessly powered neural interface with a range of 1 m. Using

harvested RF power, the NeuralWISP transmits spike counts to a commercial RFID reader

at user-programmable intervals. In addition to testing with simulation data, in vivo mea-

Page 106: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

84

surements with Manduca Sexta moth and macaque monkey validated the feasibility of this

system in real-world conditions.

By operating from a wireless power source, the NeuralWISP allows indefinite operation

without the need to change batteries, a critical need for implanted neural interfaces. The

platform is also flexible and can be programmed to operate in different modes, such as spike

time-stamp recording, or continuous recording on a duty-cycled basis. Future work reducing

the size and weight of NeuralWISP will help lead to the practical deployment of wireless,

battery-free neural recording systems.

Page 107: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

85

Chapter 7

WIRELESSLY-CHARGED UHF TAGS FOR SENSOR DATACOLLECTION

7.1 Introduction

RFID tags equipped with sensors are being developed because of their potential to add value

over identifier-only tags in a variety of settings. For example, perishable goods could be

monitored while they are in transit to check that they are not subject to unsafe temperatures,

and fragile goods could be monitored to detect unsafe accelerations that may cause breakage.

Other applications include the detection of harmful agents and non-invasive biomedical

monitoring.

To date, sensor-enhanced tags fall into two broad categories based on conventional mod-

els for powering RFID tags. There are several examples of application-specific, long-range

(UHF) passive tags with integrated temperature and light sensors, as well as an Analog to

Digital Converter (ADC) [1,2,3]. The benefits of passive tags are that they are inexpensive,

disposable and can potentially last for tens of years if not longer. Consequently they can

be embedded in objects for structural, medical, or product monitoring. The key limitation

of these sensor-enhanced passive tags is that their operation requires proximity to an RFID

reader. This greatly limits their use for capturing data through supply chains or other

applications in which there is only partial reader coverage.

In contrast, active, battery-powered tags with data logging capabilities have the advan-

tage that they can be used when tags are not in proximity to an RFID reader because

they carry their own power source. From a hardware standpoint, active tags range in com-

plexity from battery-assisted passive tags that communicate with RFID readers at the low

end to wireless sensor nodes at the high end. High end wireless sensor nodes can em-

ploy a reprogrammable microcontroller, large flash memory, numerous sensors, and wireless

communication over Wifi, Bluetooth, Zigbee or custom protocols [4,5]. Unfortunately, the

Page 108: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

86

inclusion of batteries leads to several limitations. The first is finite lifespan, which ranges

from 1 to 5 years in the aforementioned examples. Rechargeable batteries (which also have

lifespan limits) require a charging infrastructure, usually a wired connection to a power

source, and also incur the overhead costs of connecting the battery to the charger. Second,

many batteries deteriorate or become hazardous at extreme temperatures. Third, batteries

add cost, both directly through purchase cost and indirectly through the cost of regula-

tory disposal procedures. Lastly, the inclusion of battery chemicals such as lithium may be

prohibitive for safety reasons.

This chapter presents a novel wirelessly-charged power model for sensor-enabled RFID

tags in the form of a passive data logger (PDL) tag. PDLs aim to combine the best features

of passive (battery-free) and active (battery-powered) tags. Like passive tags, PDLs use

only the energy that they acquire from wireless signals to perform tasks. Like active tags,

PDLs continue to operate when away from an RFID reader. To do so, PDLs store energy

harvested from RFID reader signals in a capacitor, and later operate using this energy. For

a similar physical size, the amount of energy PDLs can store is likely to be less than a semi-

active tag. However, many settings include occasional reader contacts, e.g., supply chains

feature relatively short intervals of data logging between manufacturers, distributors, and

retail stores. Thus, there are typically many opportunities for tags to wirelessly recharge

at the various locations where passive tags are normally interrogated. This allows PDLs to

be used repeatedly and to sustain their operation across occasional reader contacts despite

their lack of an internal, long-lived power source.

To demonstrate the feasibility of wireless-charging, we have built a prototype PDL,

called the WISP-PDL, and used it for a small data collection study that we report later

in this paper. A picture of the latest WISP-PDL is shown in Figure 7.1. Our prototype

builds upon and extends our prior work on WISP (Wireless Identification and Sensing

Platform) [1]. WISP is a fully-passive UHF RFID tag that uses an ultra-low power, 16-bit,

general-purpose microcontroller for sensing, computation and RFID communication. The

WISP-PDL inherits all the WISPs design requirements, but adds an additional one: it needs

to operate without proximity to an RFID reader. The WISP-PDL described in this paper

is a new generation of WISP, presented here for the first time, that implements the EPC

Page 109: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

87

Figure 7.1: Photograph of the WISP-PDL implemented using WISP version 4.1.

Class 1 Generation 2 communication protocol.

Our prototype effort caused us to consider the problem of how to retrieve logged sensor

from PDL tags. Earlier sensor-enabled WISPs shoehorned sensor data into the low-order

bits of Gen 1 EPC IDs; the tag emulated different IDs as it was queried for data. However,

this strategy did not extend well to retrieving stored sensor data. Instead, our solution

for the PDL presented here is to upload sensor data using the EPC Gen 2 protocols tag

memory functionality. This has the advantages of compatibility with standard RFID tags

via EPC-compliant tag IDs, and it allows us to perform data collection from PDLs as

an application running on top of commercial RFID readers. The rest of this paper is

organized as follows. After describing related work (Section II) we discuss the main design

considerations in adding Passive Data Logging capabilities to WISP (Section III). We then

present the hardware and software architecture of the WISP-PDL (Sections IV and V). We

report on a short use study in which the WISP-PDL monitors the temperature of a milk

carton over the course of a day (Section VI). Finally, we discuss other applications (Section

VII) and conclude (Section VIII).

7.2 Background and Related Work

Typical UHF RFID systems are designed to increase the visibility of goods in supply chains.

A uniquely-numbered tag is attached to an item of interest, which is then wirelessly scanned

at checkpoints in the supply chain. RFID readers wirelessly power and communicate with

the tags at distances up to 10 m.

Standard commercial tags have also been used for detecting parameters other than loca-

Page 110: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

88

tion. In our prior work, mercury tilt switches were used to toggle between two commercial

tag ICs, allowing one bit measurement of object orientation [6]. Others have made a hu-

midity sensor for detecting moisture in walls of buildings and houses by placing a sponge

in front of a tag [7]. Moisture in the sponge detunes the tags antenna, allowing approxi-

mation of humidity level from the read range of the tag. Finally, a custom tag for sensing

high temperatures in food products used a fuse that melts above a particular threshold to

enable or disable the tag [8]. These passive tags based on physical properties (e.g., a fuse

that melts above a particular temperature) can sense while away from a reader but most

are extremely limited in what they can report and are not reusable. PDLs aim to provide

general and reusable sensing when away from a reader.

Efforts have been made to retrieve richer, multi-bit sensor data from RFID tags for a

wide variety of applications. Possible examples include infrastructure and object monitor-

ing, automatic product tamper detection, identification of harmful agents, and biomedical

devices for noninvasive monitoring [9]. To enable these applications, two regimes appeared:

active battery-powered and passive battery-free tags.

Most active tags are essentially wireless sensor nodes [4,5], as noted previously, though

from a software standpoint they are mainly designed to communicate directly with a bases-

tation, rather than peer-to-peer like many wireless sensor nodes. An active tag with adaptive

analog sensor thresholds for triggering sensor measurements was proposed in [10]. By using

a quasi-active state where analog circuitry wakes a microcontroller when sensors exceed

defined thresholds, it can achieve better accuracy in capturing events than devices that rou-

tinely poll sensors on a time interval. Another class of battery-powered data-loggers used in

supply chains is semi-active tags. They are battery-assisted rather than battery-operated,

using backscatter modulation for communication rather than a radio. [15] illustrates the

use of 2.4 GHz semi-active tags in cold chain food monitoring for the army. However, semi-

active tags generally suffer the same disadvantages as active tags due to their dependence

on a battery.

Many passive sensor-enhanced RFID tags tend to be application specific. A commercially-

available, low-frequency RFID tag for detecting dangerous temperatures in food products

during transit is reported in [9], although its range is too small for use in many applications.

Page 111: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

89

Other examples of application-specific, long-range (UHF) passive tags with integrated tem-

perature and light sensors, as well as an Analog to Digital Converter (ADC) are [2,3]. We

previously developed a general-purpose tag called WISP [1], which is a fully programmable

passively-powered UHF RFID sensor tag. While these devices provide the benefits of pas-

sive, battery-free operation, none can collect sensor data without proximity to an RFID

reader.

7.3 Design Considerations

The design constraints and challenges of passive sensor-enhanced UHF tags have been thor-

oughly discussed in our prior work [1]. Additionally, there are a number of considerations

we encountered while developing a PDL from our work on WISP. In general, these fall under

three categories: run time, charge time, and RFID compatibility.

7.3.1 Run Time

The primary design consideration for a PDL is the sensor measurement and logging run

time from a single capacitor charge. This run time is determined by the storage capacitor

size and the time-averaged current consumption, Iave, during both sleep and active periods.

As the maximum capacitor size is often fixed by size or cost, the maximum run time is then

determined by Iave. We designed for a 1 day run time using a 0.1 F capacitor. A significant

amount of effort was made reducing Iave to achieve the desired run time.

Fig. 7.2 plots the approximate maximum Iave for various run times and capacitor sizes

using the relationship I = C∆v∆t . This relationship allows accurate computation of the

runtime if the current consumption is constant over voltage, and this is a reasonable ap-

proximation when the quiescent current is larger than the average active current. When

the active current dominates and is drawn from a regulator, the amount of current drawn

from the capacitor by the regulator is a function of the voltage on the capacitor.

For applications with short-duration (100 µs to 10 ms) sensing / data logging and long

sampling interval (1 second or more between samples), the quiescent current can potentially

determine the minimum achievable Iave. A brief example illustrates this point. Assume a

sensor measurement and flash memory write occur every 10 seconds and each event consumes

Page 112: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

90

1.E-08

1.E-07

1.E-06

1.E-05

1.E-04

1.E-03

1.E-02

1.E-01

1.E+00

1.E+01

1.E+02

1 10 100 1000 10000 100000

Capacitance (uF)

Average Current Budget (uA)

1 Hour1 Day1 Week1 Month1 Year

Figure 7.2: Maximum Iave for various run times and capacitor sizes.

500 µA for 1 ms. The average current consumption of this workload (sensing and logging)

is only 56 nA. However, the quiescent current of discrete component designs (not IC), can

require several microamps (as shown in our design).

The current consumption of the WISP-PDL, built with commercial off-the-shelf (COTS)

components, is roughly equal to that of the previous example. Excluding parasitic leakages,

the quiescent current budget includes a 1 µA ultra-low-power linear regulator, a 0.5 µA

interval wakeup timer and a 0.3 µA voltage supervisor. This corresponds to a sleep current

consumption of 1.8 µA. The active current is approximately 500 µA for 1 ms, but the sample

time can be adjusted between hundreds of milliseconds to tens of seconds.

As the quiescent current is the dominant consumer, a considerable improvement could

be made using an external wakeup circuit to power on the remaining circuitry at regular

intervals. For example, real-time clocks from Seiko [11] and Swatch [12] allow operation from

an unregulated power source at quiescent currents of approximately 350 nA. This RTC could

be used to enable the remaining circuitry. Although the reduced current consumption of a

RTC represents some improvement over the regulated design, it was not enough to justify

Page 113: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

91

the added design complexity since we had met our run time design goal.

There is a point of diminishing returns in optimizing both the active or sleep currents.

When either current consumption is less than 10% of the other, little run-time is gained

from further reduction of the smaller current consumption. Also, some capacitors have

internal leakage resistances in the megaohm range, which also causes an upper bound on

maximum runtime. For example, ceramic capacitors have low-nanoamp to sub-nanoamp

leakages. Electrolytic and tantalum capacitors can have hundreds of nanoamps of leakage.

Double layer capacitors, which we employed for our design, have widely varying leakages

from low-nanoamps to low-microamps.

7.3.2 Charge Time

A second consideration is the wireless charge time, with acceptable charge times depending

on the intended PDL application. In general, it is important to characterize how fast the

PDL can charge at any particular distance. Conservation of energy requires that the energy

consumed during runtime first be charged into the storage capacitor, which is illustrated in

Equation (7.1).

Estored =12C(V 2

charged − V 2dd) = PchargeTcharge (7.1)

The usable energy, Estored, is stored in voltage above the minimum operating voltage,

Vdd. Energy stored in the capacitor at voltages below the PDLs minimum operating voltage,

Vdd, cannot be used. Thus the capacitor voltage must first charge up to Vdd after which

surplus (i.e. usable) energy begins to accumulate. Due to component ratings, a maximum

voltage, Vcharged, is enforced via overvoltage protection. The charge time is somewhat

longer than the recharge time, because the charge time includes voltage from 0 V to Vdd as

well as Vdd to Vcharged, while recharges only charge from Vdd to Vcharged.

Equation (7.1) allows prediction of the charge time if the power supplied by the harvester,

Pcharge, is known. Friis transmission equation allows prediction of the RF power available

to the PDL over distance, and empirical RF to DC rectifier efficiencies can then be used

to estimate Pcharge as a function of distance. Equation (7.2) is the linear form of Friis’

Page 114: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

92

equation for path loss, with terms for polarization loss and rectifier efficiency included.

Pr = PtGtGr

4πR

)2

ηLmLp (7.2)

The transmit power of the reader Pt = 1 W = 30 dBm. Its center frequency is 915 MHz,

corresponding to wavelength λ = 0.33 m. The transmit antenna gain GT = 6 dBi (this

yields an effective isotropic radiated power of 4 WEIRP , the United States regulatory limit

for this ISM band). The receive antenna gain Gr = 2 dBi (the standard gain figure for

a dipole antenna), and the polarization loss Lp = 3 dB. Loss Lp occurs because only half

of the power transmitted from the circularly-polarized transmit antenna is received by the

linearly-polarized receive dipole antenna. Lastly, η = 30% is the rectifier efficiency, and this

is approximated from empirical data using the WISP. (Since η is a fraction between 0 and

1, its logarithm is negative, which is why it is added on the right hand side.)

Fig. 7.3 shows the predicted charge time versus range for the previously discussed effi-

ciency, antenna gains, transmit power and stored energy based on Friis equation, assuming

rectified power can be converted to Vcharged at a particular distance. WISP becomes

voltage limited at 4.5 m; however, integrated circuit designs are able to achieve high con-

version efficiencies up to 10 m [9]. Thus even at long distances from the reader, a PDL

can charge overnight or over the course of a day, which is not unreasonable for many ap-

plications. Finally, an integrated circuit design can potentially achieve much lower power

consumption and consequently much faster charge time for the same runtime as our discrete

implementation.

7.3.3 RFID Compatibility

Many supply chains have existing RFID infrastructure for use with standard commercial

tags. Creating secondary wireless networks for sensing in supply chains is generally imprac-

tical, so it is desirable for PDLs to integrate into the existing RFID network. WISPs already

harvest energy from standard reader signals. In addition, we needed to communicate sensor

data using only standard commands. Fortunately, the EPC Class 1, Generation 2 specifi-

cation provides a suitable interface for this task in the form of arbitrary, i.e., user-defined

Page 115: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

93

0

5

10

15

20

25

30

35

0 2 4 6 8 10

Meters

Time (hours)

Recharge Time

Charge Time

Figure 7.3: Predicted charge time versus range.

memory. This memory can be read and written with standard commands. We log sensor

data into this memory and upload it by having the RFID reader perform a series of reads.

We can also use this memory scheme to signal associated metadata such as whether there

is more data to upload. This approach allows WISP-PDLs to operate on existing RFID

networks without interfering with standard ID-only tags. Adding EPC Gen 2 support re-

quired a significant change in our software-defined protocol from previous versions of WISP

as well as some modification to WISPs hardware demodulation circuitry.

7.4 Hardware Architecture

The WISP-PDL is a new generation of WISP hardware with modifications for long run time

data logging, EPC Gen 2 communication, and fluid level sensing. Previous WISPs were

fully-passive, sensor-enhanced EPC Gen 1 RFID tags. In the WISP-PDL as well as prior

WISPs, communication, sensing and computation are software defined in a general-purpose,

reprogrammable Texas Instruments MSP430 microcontroller. WISPs include circuitry for

RF power harvesting, modulation, demodulation, and voltage regulation that enable op-

eration as a passive RFID tag. A detailed design discussion of WISP is found in [1], and

therefore the bulk of this section will focus on areas specific to WSP-PDL. First, a high

Page 116: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

94

Figure 7.4: Block diagram of the WISP-PDL.

level description of the WISP-PDL is presented. Second, WISP-PDL-specific components

are described in detail. Finally, a capacitive sensor is described.

7.4.1 High Level Description

The WISP-PDLs high-level implementation is similar to previous WISPs, and a block di-

agram of the WISP-PDL is shown in Fig. 7.4. The analog front end contains a power

harvester block that rectifies incoming RF energy into DC voltage to charge the storage

capacitor. The power management block stores energy and provides regulated voltage for

the system. The demodulator thresholds the enveloped RF carrier wave and converts the

downlink (reader to tag) logic level to the regulated logic level. The modulation circuitry

converts the uplink (tag to reader) waveform into a modulated tag reflection coefficient

to the antenna. Lastly, a MSP430 microcontroller facilitates communication and sensor

measurement.

There are a few significant hardware changes in the PDL-WISP from the previous

WISPs. First, the demodulator was revised to enable Gen 2 communication. Second,

the power management block was revised to reduce leakage and thereby lengthen run time.

Third, an external EEPROM allows low-voltage non-volatile data storage. Fourth, a capaci-

tive sensor was added to enable various sensing applications. Lastly, a newer microcontroller

(MSP430F2012) was used which allows 6 MHz operation at 1.8 V (versus 4 MHz at 1.8 V

with the MSP430F1232) for nearly the same power consumption. This was needed in order

to perform the more complex Gen 2 communication protocol.

Page 117: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

95

Figure 7.5: Demodulation and power management circuitry.

7.4.2 Demodulation and Power Management

The demodulation and power management circuitry is shown in Fig. 7.5. Gen 2 reader-to-

tag physical layer modulation (PR-ASK) causes smaller amplitude modulation deviations

than Gen 1 ASK modulation. This required a new demodulator, which is similar to those

used in silicon tag designs. Two low threshold DC schottky diodes are used to rectify the

enveloped signal from the harvester. The current consumption of the comparator is used

as a constant-current source, and the voltage across D7 detects current (high amplitude

RF modulation) supplied by the harvester as a positive voltage, and lack of current (low

amplitude RF modulation) as negative voltage. The comparator is used to generate a rail-

to-rail logic level waveform, and the level shifter converts the unregulated logic level to

the regulated logic level. It is important to optimize current consumption and speed when

choosing a comparator; the Gen 2 protocol requires roughly 1 µs propagation time, and

minimal power consumption is desirable for maximizing the wireless range of the device.

Storage capacitor C1 accumulates energy, which is rectified by D6. Over-voltage pro-

tection (OVP) clamps high rectified voltages to 5.5 V, and a low dropout, ultra-low-power

Page 118: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

96

linear regulator is used to provide 1.8 V regulated power for the system.

Fig. 7.4 shows external sensors and EEPROM attached. External sensors are described

shortly. An 8 Kbyte, 1.8 V EEPROM is used for logging data. Alternatively, the internal

flash memory of the microcontroller may be used for logging small amounts of data; however,

this necessitates a slightly higher regulated voltage of 2.2 V. Regardless of implementation,

the EEPROM or flash memory will hence be referred to generally as non-volatile memory

(NVM).

7.4.3 Sensors

A number of sensors can be integrated into the WISP-PDL, including those previously

implemented on WISP (rectified voltage, light level, temperature, and orientation). Other

interesting sensors that we have not yet integrated but have been explored for active and

semi-active tags include humidity, vibration, sound, bearing (via digital compass) and others

[10].

We implemented a capacitive sensor for fluid level sensing in our demonstration study.

Capacitive sensing is a popular method for sensing user input in devices such as the Ap-

ple iPod, human presence in automatic airbag enable/disable in automobiles, distance in

digital calipers, and displacement in MEMS accelerometers, as well as directly sensing gas

pressure, liquid level, and other parameters. It is also the case that capacitive sensing can

be accomplished with minimal components and low power consumption. In its simplest

incarnation, measurement of the discharge time of an RC circuit can be used to estimate

capacitance. More advanced designs provide enhanced noise immunity, directivity of sensing

and accuracy [13].

We achieved reasonable accuracy with a simple RC discharge measurement. A config-

urable microcontroller port I/O pin is used for sensing. The pin is first set as an output to

charge the sense capacitor to Vdd. Next, the pin is set as an input (output tri-state buffer

set to high-z) and the microcontrollers timer is used to measure the discharge time from Vdd

to the input buffer threshold of approximately Vdd / 2. In some microcontrollers, the pin

input leakage could be used as the parallel resistor, but the MSP430 has such low leakage

Page 119: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

97

that a 10 MΩ resistor is necessary to create a discharge rate in the hundreds of microseconds

range.

%fill = 100Cmeasured − Cmin

Cmax − Cmin(7.3)

We characterized the sensor to understand the relationship between fluid level and ca-

pacitance. Table I shows experimental data versus linear steps for fill percentage. Fill

percentage was calculated using Equation (7.3). Note that the measurement was taken in-

side the physical range of the sensor (the bottom of the sense wires to the top of the sense

wires in Fig. 7.4).

7.5 Software Algorithms

The microcontroller firmware implements EPC Gen 2 communication, periodic sensor sam-

pling and data logging. Generally, the microcontroller spends most of its time in a low

power sleep mode. An internal timer wakes the microcontroller periodically to take sensor

measurements and potentially log the measurement. When the WISP-PDL is queried by

an RFID reader and there is new logged data, the four communication states on the right

side of Fig. 7.6 are executed. Note that portions of the Gen 2 specification not used for

retrieving the ID or sensor data are not yet implemented.

There are two primary challenges in the design of the software. The first is maximizing

time spent in the microcontrollers low-power sleep states. This minimizes the active current

consumption. Second, general purpose microcontrollers are not designed to efficiently per-

form RFID communication because they process software instructions while custom silicon

RFID tag state machines process many instructions in parallel. Custom, optimized com-

munication algorithms are necessary to achieve RFID data rates with a 6 MHz clock speed

and also to keep power consumption at a minimum.

Fig. 7.6 shows a state diagram of the WISP-PDL data logging and communication

cycles. The center box represents the sleep state, and a low-frequency, low-power oscillator

is used to periodically wake the device from this sleep. These timer interrupts trigger

the data logging cycle, shown on the left side of Fig. 7.6. First, sensors are powered on

Page 120: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

98

Sleep Mode

(LPM 3+VLO)

Query Sensors

Compare

sensor data to

last logged

data → ∆

Write to EEPROM/Flash and

Enable RFID

Send RN

Send ID

Send RN

Send Data

VLO

WD

T

Interrupt

∆> V

t

∆ < Vt

RF

ID O

n +

Q

uery

Tim

eo

ut

Rea

d

Read

Request

RN

Ack

Figure 7.6: WISP-PDL software state diagram.

and sampled using the microcontrollers 10-bit ADC. To conserve NVM and power, not all

sensor measurements are stored. Current sensor measurements are compared against the

last stored sensor measurements, and if the difference ? exceeds a threshold Vt, the new

sample is logged to NVM. Also, each sample is time stamped using a count of the number

of timer interrupts. Finally, when a new sample is logged to NVM, the RFID interface is

enabled so that the new data may be retrieved by a nearby reader.

The implemented portion of the EPC Gen 2 communication cycle is illustrated on the

right half of Fig. 7.6. If the RFID interface is enabled, a Query command from the reader

begins the communication session. Several further commands precede the Read command,

which allows the reader to download logged data from NVM. A fully erased NVM Read data

packet is used to mark the end of valid NVM data. The reader then issues an additional read

command past the end marker, which signals the WISP-PDL to disable its RFID interface.

This last step is required because Gen 2 protocol does not provide an acknowledgment to

the tag that the read command was successful, and the tag must wait until the reader

successfully receives the last packet before disabling its RFID interface.

Depending on the usage scenario, it may be necessary to clear the NVM after a download

Page 121: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

99

so that memory can be freed up for further logging. To reset the NVM address where sensor

data is stored, a write command or a read to a non-valid address can be used to trigger the

reset. For devices using flash memory, this also triggers an erase operation on the memory.

7.6 Results

We conducted two tests with the WISP-PDL, instrumenting a milk carton to measure

temperature and fill percent over the course of a day. This test was chosen for two reasons.

First, it is a controllable, convenient, small-scale setting. Second, it has two scenarios that

resemble a supply chain: (1) the refrigerator is temperature-controlled and RF-isolated like

a truck trailer or shipping crate and (2) there are random times in which the tagged object

is removed from the metal box and scanned by an RFID reader. In either case, the tag does

not know when it will be removed from the fridge for use or transferred from a shipping

crate or trailer. Fig. 7.7 shows the reader antenna (the reader is on top of the refrigerator)

and laptop which controls the reader and logs data retrieved from the WISP-PDL.

Two tests were performed with the instrumented milk carton, which is shown in Fig. 7.6.

The first is a controlled experiment where the milk carton is filled with water and dispensed

in fixed amounts. The second is an uncontrolled experiment where the milk carton is used

by various people in an office environment for tea and coffee drinks.

The results of the first experiment are shown in Fig. 7.8. In this test, a carton was

filled with water and poured out in regular increments much like the test in Table I. The

difference is that the carton is stored in the fridge between pours, and the water is poured

out instead of added. Similar to Table I, the carton is physically separated from other

cartons to prevent the sensor from detecting the milk in other cartons.

Each time the carton is poured, an upward spike in fill percentage occurs. This is caused

by the position of the sensor; when the carton is tilted to pour out liquid, the tilting causes

the fluid to cover the entire sensor. Positioning the sensor of the opposite side of the carton

would cause dips in the level during pouring.

Fig. 7.9 shows an uncontrolled experiment where a milk carton is instrumented the

same way as Fig. 7.8 and used over the course of a day. The high milk consumption is due

primarily to an espresso machine near the office refrigerator; one carton of milk makes 10

Page 122: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

100

Figure 7.7: Reader antenna and control laptop outside of refrigerator.

(a) WISP-PDL attached to milk carton. (b) Fridge environment with other cartons nearby.

Page 123: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

101

0

10

20

30

40

50

60

70

80

90

100

0 2 4 6 8

Time (hours)

Percent Full (%) / Temperature (C)

Temperature Celcius

Percent Fill

Figure 7.8: Control data set logged from milk carton.

0

10

20

30

40

50

60

70

80

90

100

0 5 10 15 20Time (hours)

Percent Full (%) / Temperature (C)

Temperature (°C)Percent Full

Figure 7.9: Real world data logged from milk carton over the course of a day.

to 15 lattes.

The data in the uncontrolled experiment (Fig. 7.9) shows two primary differences from

the controlled experiment (Fig. 7.8). First, the carton was not physically separated from

the other cartons in the uncontrolled experiment, which caused the level to increase at times

Page 124: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

102

when the sensor was near other cartons. Second, the amount of milk removed each time was

not regulated in the uncontrolled experiment, so the level changed by varying amounts after

each use. By shielding the capacitive sensor, the interference effects of other milk cartons

could be eliminated.

7.7 Applications

There are a host of potential applications for PDLs. Monitoring sensitive parameters in sup-

ply chains is an obvious extension for current RFID networks. In particular, temperature-

controlled supply chains for food, chemicals, blood plasma, and pharmaceuticals are target

application areas. Shock, vibration, and humidity are other parameters than can be mea-

sured for sensitive goods.

In addition to logged parameters, PDLs can supplement their sensor data with infor-

mation provided by RFID readers. Examples include location, time, package contents, and

transport vehicle information. Further, this data could easily be provided using the stan-

dard Gen 2 RFID Write command. Note that storing reader-supplied information be done

with any RFID tag (passive, active or PDL) that has unused non-volatile memory, but

this information gains value when it supplements logged sensor data. This would allow for

detailed information at the destination without the complexity of requiring communication

between parties (such as manufacturing company, distributor and retailer). For example,

if the temperature exceeded a threshold during transit and the delivery truck license plate

number were supplied by the loading dock RFID reader, the problem could be discovered

and easily identified.

RFID has also been employed for non-conventional applications, which can benefit from

PDLs. For example, RFID has been explored for use in eldercare monitoring, allowing

caregivers to unobtrusively monitor daily living tasks such as cooking and cleaning [14].

However, data in these applications has typically been limited to the binary visibility model

(ie, is the tag near the reader?). Using PDLs, an RFID reader in the fridge and pantry

cabinets can recharge the tags, and the tags can sense and log data when out of range

of the reader. This allows continual data acquisition without the stringent requirement of

proximity to a reader.

Page 125: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

103

Another application for PDLs is implantable medical sensors. Parameters such as blood

pressure, blood sugar, temperature, and heart rate could be monitored and logged through-

out the day. At night, a bedside reader would then download the days data and charge the

PDL. Finally, medical analysis software would allow the user to see the results and submit

data to a doctor for professional analysis.

7.8 Conclusions

The WISP-PDL demonstrates a new power model that differs from conventional passive

and active models for powering RFID tags. Like passive tags, PDLs use only the energy

that they acquire from wireless signals (via standard RFID readers) to perform tasks. Like

active tags, PDLs can continue to operate and collect sensor data when away from an RFID

reader. We have achieved this combination by having PDLs charge themselves by harvesting

energy from RFID reader signals and storing the energy in a capacitor for later use when

they are away from a reader. This model is intended for use in settings such as supply chain

monitoring, in which tags have occasional contact with readers.

Our prototype WISP-PDL extends our prior work on WISP (a passive, sensor-enabled

UHF tag) by adding energy storage and away-from-reader operation. WISP-PDLs run on

EPC Gen 2 infrastructure to upload collected sensor data via commercial readers and in

a manner that is compatible with standard EPC Gen 2 tags. The short demonstration

study we present demonstrates the feasibility of the novel power model. The WISP-PDL

successfully monitored the temperature and fullness of a milk carton during the course of

a day. The results are sufficiently promising that we are looking at larger and different

use cases, and we hope that wirelessly-charged tags prove to be a viable model for many

applications.

Page 126: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

104

Chapter 8

A GRAPHICAL USER INTERFACE FOR RFID APPLICATIONS

8.1 Introduction

Many RFID readers include a basic tag display application that connects to the reader via

serial (RS232) or ethernet. However, these applications are primarily intended to demon-

strate the reader capabilities - not provide a useful service to the end user. For example,

there is often no method of manipulating tag data in real time or logging tag data to a

text file. For commercial reader deployment, the user typically hires a middleware company

who, in turn, writes software (middleware) to interface the reader with business manage-

ment software such as BizTalk [90]. Consequently, the default reader application is of little

use for RFID research.

Until recently, there was no standard interface for RFID readers. Each company imple-

mented its own protocol, and the middleware company wrote tailored interfaces for each

reader. For example, the author has written serial interfaces for both Alien and Thing-

Magic readers in Microsoft Visual Basic. These applications are similar in appearance and

architecture to the standards-compliant application that is presented in this chapter and

will thus be ignored. Clearly, writing unique software for each vendor (and sometimes each

reader from each vendor) is inefficient and slows commercial adoption. In 2007, the RFID

consortium EPCglobal ratified a standard reader interface called Low Level Reader Pro-

tocol (LLRP) which defines a standardized network interface for readers [29]. Just as a

standardized air protocol (Gen2) helped advance commercial RFID success through vendor

interoperability, LLRP is intended to further standardize protocols across vendors and thus

allow ease of commercial adoption.

This chapter discusses two LLRP-compatible reader applications. The first application,

described in Section 8.3, provides sensor data display, graphing and logging. The second

application, described in Section 8.4, varies the power available to the tag in order to

Page 127: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

105

test tag performance across input power levels (input power is a proxy for multipath-free

wireless range). These applications are built on a common abstraction, which is described

in Section 8.2. This abstraction optionally hides much of the LLRP library complexity from

the application developer. The LLRP library provides low-level ethernet calls and XML

encoding/decoding of the reader communications.

8.2 Architecture

EPCglobal provides an LLRP library which facilitates ethernet communication with the

reader. Libraries are provided in Java, Perl, C, C++ and C#.NET. This work uses the

C# library in order to leverage the rapid GUI development tools available in Microsoft

Visual Studio. This choice of framework comes at the cost of operating system inflexibility.

However, a virtual machine could potentially sidestep this issue.

For the application developer, however, the library includes significant complexity that

is unnecessary for controlling the reader (connect / disconnect, start / stop) and receiving

tag data. The goal of this work is to abstract away much of the complexity of the LLRP

library to the application developer. To this end, the architecture shown in Figure 8.1 was

developed. The RFID Reader class acts as an interpreter, encoding basic function calls into

LLRP-compatible commands as well as decoding XML-encoded tag data and placing it into

convenient, typed data structures. The Reader Manager class stays synchronized with the

reader state (ex, connected or disconnected) and handles errors. This class also prevents the

user from sending errant commands to the reader, which improves application robustness.

The Settings module includes graphical forms and LLRP encoding/decoding code to

enable the user to manipulate the reader air protocol settings and link layer behavior. The

Logging module provides standard menus, file naming, and function calls for data logging

to text files. Minimal effort is required to add logging functionality to applications - the

developer need only define what information the user may select to log and define how their

data should be formatted in the log file. Graphing is provided through a GPL-licensed

library called ZedGraph [91]. This library is an incredibly simple turn-key solution. The

user need only add (x,y) points to a list and refresh the graph. The GUI elements are

provided through Microsoft Visual Studio, allowing extremely rapid and simple application

Page 128: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

106

LLRP Library

RFID Reader

Reader Manager

Tags / Sensor Data Instructions

Parsed Data Control Signals

Raw Tag XML LLRP Commands

ApplicationSettings Logging

Graphing GUI

Figure 8.1: LLRP-compatible abstraction layers for RFID application development.

development. Finally, several interfaces are used to define the interactions between the

abstraction layer and the application. For example, the application must implement a

means for handling received tags and for displaying status and error messages.

8.3 WISP Demo

The first application is primarily used to display and log sensor data from WISPs. A screen

capture of the application is shown in Figure 8.2. On the right, a summary table shows

statistics about each tag in view of the reader. The read rate is shown in large text in

the upper right corner. Various tabs in the left panel show graphs of sensor data retrieved

from WISPs. Connecting and starting the reader are provided through panels along the

top of the window. Reader settings manipulation and text file logging are available through

buttons near the bottom of the window.

Starting the reader is simple due to the abstraction layers mentioned before. The GUI

/ application code simply calls the Reader Manager with the appropriate command. For

example, to connect to the reader, the application simply sends a connect command with the

reader address to the Reader Manager. This allows for simple, straightforward application

code which would not be the case if interacting directly with the LLRP library.

Page 129: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

107

Figure 8.2: Screen capture of the WISP GUI application. A summary of tag statistics isshown on the right, and sensor details are available in various tabs on the left. Text filelogging and control of reader settings are available through buttons at the bottom of theapplication.

8.4 Attenuation Tester

As tag to reader distance increases, the power available to the tag decreases rapidly. WISP

maintains operation at low input power levels by lowering its duty cycle. That is, the read

rate gracefully declines with input power. However, firmware bugs can easily cause errant

behavior, resulting in a “cliff” effect. That is, for some input power where the WISP has

sufficient voltage to operate, it is unable to communicate. The Attenuation Tester allows

for repeatable testing without physically changing the tag to reader distance. It does this

by sweeping the transmit power of the reader through the reader settings.

A screen capture of the Attenuation Tester is shown in Figure 8.3. The features are

very similar to the WISP Demo application due to the required manipulation of the Reader

Page 130: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

108

Manager in order to connect and start the reader. Read rate versus attenuation is auto-

matically plotted during the sweep, and it may also be logged to a text file. The logging

takes advantage of the Logging module, and requires only a few additional lines of code to

implement for this application.

8.5 Conclusions

A common reader interface represents an important step in the development of RFID.

Writing custom interfaces for each vendor is wasteful and inefficient. However, LLRP is

relatively new, and it disenfranchises readers that cannot talk over ethernet. In theory, the

data sent by ethernet could just as easily be sent by serial, and perhaps future versions of

LLRP will allow the user to select the physical layer in use. Regardless, there are good

reasons for using ethernet versus serial due to its ability to support many readers on one

network and the existence of LAN infrastructure in many environments. Wireless readers

are also compatible with LLRP via 802.11 (wifi) and standard network infrastructure. It

is likely that ethernet will become the standard interface of new readers in order to enable

LLRP compatibility.

Page 131: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

109

Figure 8.3: Screen capture of the Attenuator Test program. Tag read rate versus attenuationlevel is plotted using ZedGraph. Text file logging and control of reader settings are availablethrough buttons in the center of the application.

Page 132: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

110

Chapter 9

DESIGN OF AN INTEGRATED GEN2 SENSOR TAG

9.1 Introduction

The tremendous impact of WISP, both as a research platform and an educational tool,

demonstrated the usefulness of a Gen2-compatible, programmable RFID tag. WISP’s im-

plementation as a PCB design using COTS components has a number of drawbacks, how-

ever, compared to IC implementations. Chiefly, the WISP rectifier has poor sensitivity (-10

dBm vs. -20 dBm), and the microcontroller (µC) is clumsy and inefficient at performing

RFID communication (100x power consumption, limited protocol support). Work presented

in the literature - see Section 9.2 - has largely ignored Gen2 compatibility (for deployabil-

ity), programmability/configurability (for application development), and addressability (for

multi-tag environments). It is the aim of this work to address these oversights and apply

knowledge and experience from the WISP project to an IC implementation. The design of

this system-on-chip solution, or socWISP, is described in Sections 9.5 and 9.6. Verification

and measurement results are presented in Chapter 10, and Chapter 11 presents in-flight, in

vivo temperature recordings from a Manduca Sexta moth.

9.2 Prior Work on Full Tags

A plethora of RFID-related research has appeared in the literature, predominately since

Karthaus [92] influential and widely-cited paper in 2003. As described in Chapter 1, the

recent technological feasibility of RFID, coupled with numerous application spaces and

widespread commercial adoption, have galvanized academic research in this field. Literature

contributions on specific tag components (ex., rectifiers, oscillators) will be reviewed in

Section 9.3. However, a number of full tag implementations have also been presented, and

their features are summarized in Table 9.1.

Table 9.1 groups tags into four categories. The first category, including this work and

Page 133: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

111

the WISP, represents fully-passive tags with sensors, addressability, and Gen2-compliant

protocol. The WISP, which is discussed in detail in Chapter 3, is incredibly feature-rich.

However, as stated in the introduction, performance is cramped by use of a µC. WISP has

proven most useful as a prototyping tool; its configurability and programmability facilitate

rapid application development. This work preserves some of the WISP’s flexibility while

dramatically improving performance, and design details will be presented in this chapter.

The second category, tags with sensors but no protocol/addressability, includes Shen,

Cho, Shenghua, and Paek. It is unclear how one would deploy this type of tag because

without an ID, the sensor data they report cannot be associated with the object to which

the tag is attached. The reported power consumption is remarkable: Paek claims a tag

power budget of only 95nW. Unfortunately the back-leakage of UHF rectifiers or the self-

leakage of most batteries will likely limit performance.

The third category, tags with addressability but no sensors, includes Nakamoto, Pillai,

Karthaus, and Barnett. These tags target conventional RFID applications and provide a

useful reference upon which a more feature-rich tag can be designed.

The fourth category, including Kim and Kocer, includes tags primarily designed for

battery-assisted operation. Kim [94] presents a Gen2 compatible tag with integrated tem-

perature sensor and addressability via FeRAM-stored ID. The tag has the capacity to op-

erate in passive mode without a battery; however, it appears that the tag functions as a

standard ID tag in this regime. Furthermore, their publication shows temperature data sent

in lieu of the ID. The authors are ambiguous about the exact functionality, but it appears

the tag can function as one of the two above types: a sensor tag or an addressable tag.

Finally, there is no ADC, so external sensors cannot be measured. The on-chip temper-

ature is measured with a ring oscillator, similar to temperature tags such as Shen, Cho,

etc. Kocer [95] presents a tag with 5-bit ADC and 3-bit programmable ID. The lack of a

commercially-compatible protocol and limited ID space makes deployment of this tag infea-

sible. Furthermore, the tag’s active current of 2 mW necessitates a battery or a WISP-like

power management scheme.

Page 134: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

112

Table 9.1: Comparison of published tag features.

Protocol NVM Addressable Sensors Battery-Free

This Work G2 - Y1 8-Bit ADC Y

WISP G22 EEPROM+Flash Y 10-Bit ADC Y

Kim [94] G23 FeRAM Y Temp N

Kocer [95] N - 3-Bits 5-Bit ADC N

Nakamoto [96] ISO FeRAM Y - Y

Pillai [97] ISO EEPROM Y - Y

Karthaus [92] ISO4 EEPROM Y - Y

Barnett [98] G2 EEPROM Y - Y

Shen [99] N - N 8-Bit ADC + Temp Y

Cho [100] N - N Temp Y

Shenghua [101] N - N Temp Y

1Unique ID Generator provides addressability[93]

2Partial protocol support

3Sensor data not sent by Gen2 protocol

4Specific protocol not stated in publication

9.3 Prior Work on Tag Components

Remarkably few of the full tag implementations present implementation details or perfor-

mance metrics on the individual tag components. However, various subsections of the tag

have been published individually and in greater detail. [102] includes a literature review for

the regulator, bandgap reference and demodulator. Prior work on the rectifier, oscillator

and digital core are reviewed here.

9.3.1 Rectifier

There have been a number of in-depth analyses of UHF and microwave rectifier design.

This research has been spurred not only by RFID applications (where the power source is

intentionally “planted”), but also by the prospect of ambient RF harvesting. For example,

Page 135: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

113

[103] wirelessly powers a consumer clock from television signals. This could enable a low

power, low duty cycle sensor node such as a Mote [104] to operate in environments where

solar power is infeasible and without battery-life constraints.

Much of the literature focuses on optimization of conventional topologies. For exam-

ple, Umeda [105], Vita [106], and Barnett [107] present design strategies for conventional

Dickson charge pumps (voltage-doubling ladders), and Yi [108, 109] includes an in-depth

mathematical analysis. Mandal [110, 111] discusses fundamental physical relationships that

link the operating bandwidth and range are related to technology dependent quantities like

threshold voltage and parasitic capacitances.

Several papers have investigated non-conventional topologies. For example, Nakamoto [10]

tunes transistor Vt to maximize rectifier power conversion, achieving 36.6% efficiency. Le [112,

11] replicates this work using floating gate PMOS transistors and optimizes for sensitivity

instead of efficiency. They achieve the best results to date: 1 V output voltage is reported

at -22.5 dBm input power. Unfortunately, however, Impinj patented “adaptive silicon”

technology using floating gate Vt tuning several years ago [113].

9.3.2 Oscillators

Gen2 RFID tags require a local oscillator to clock the digital core. In particular, the tag

backscatter specifications necessitate either a precise 1.28 MHz clock or a higher clock

rate with a programmable divider. Unfortunately many full-tag publications rarely provide

details on the oscillator; however, there have been several dedicated papers which are listed

in Table 9.2.

Most designs employ a current-starved ring oscillator, and efforts focus on reducing vari-

ation and power consumption. Ring oscillators suffer from high sensitivity to almost every

parameter including supply voltage, bias current, temperature, mismatch, parasitics, and

process variation. An alternative low-power oscillator topology is the relaxation oscilla-

tor. Achieving nanowatt power levels can be difficult with these designs; a 1 MΩ resistor

with 1 V supply draws 1 µW. Precision resistors consume significant area (especially for

megaohm values) and are not available in many processes. Additionally, a comparator with

Page 136: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

114

Author Frequency Current (nA) Supply (V) Variation Tech.

This Work 3 380 0.5-0.9 20% max 130 nm

Cilek [114] 1.28 240 0.8-1.0 14% 130 nm

Song [115] 5.65 1200 0.6 - 130 nm

Klapf [116] 1.9 800 0.7-1.6 PLL 120 nm

Barnett [117] 1.52 400 0.8-1.5 - 130 nm

Lincoln [118] 3.5 14000 0.5-1 Inj. Lock 180 nm

Table 9.2: Review of oscillators presented in the literature.

sufficient bandwidth for oscillation must also be factored into the power budget. Due to

these tradeoffs, ring oscillators are more widely used than relaxation oscillators.

Cilek [114] presents a 1.28 MHz 3-stage current-starved ring oscillator in 0.14 µm CMOS.

Reported power is 191 nW minimum, and the supply voltage can range from 0.8 V to 1.0 V.

The authors claim 14% maximum and minimum frequency deviation, however, this includes

an impractical, manual adjustment of the bias current. In contrast to some older RFID

specifications, the Gen2 protocol provides no opportunity to phase-lock the oscillator to a

downlink preamble square wave. This means that the oscillator needs to meet specifications

without tuning. Lincoln [118] instead injection locks to the RF carrier. However, the

tradeoff in power consumption is unsuitable for a passive tag. The table cites the full clock

generator current of 14 µA. The relaxation oscillator core consumes 3 µA, but no results

are presented on the standalone operation of the oscillator.

Barnett [117] presents a 1.52 MHz relaxation oscillator. The advantage of this topology is

supply-independence; the frequency is stable from 0.8 V to 1.5 V. Low current consumption

of 400 nA is achieved by using a 14 MΩ resistor. However, they rely on wafer-trim or on-

chip calibration in order to correct for process variation. This is unrealistic for ultra-low-

cost RFID tag manufacture and as mentioned above, cannot be corrected using phase-lock

techniques.

There are various problems with the remaining designs. Klapf [116] presents an oscil-

lator with poor performance given the technology node - current consumption is an order

Page 137: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

115

to magnitude higher than other results in the table. Finally, the 5.7 MHz oscillator from

Song [115] is more than twice the required frequency needed to meet Gen2 specifications.

They do not explain why such a frequency was chosen, and this high frequency will cause

unnecessarily high digital power consumption. Their implementation as a relaxation oscil-

lator provides low frequency variation over supply voltage. However, design sensitivities to

transistor Vt will undoubtedly affect both the bias current generator and oscillator core over

process and temperature variation, and no analysis or results are presented.

9.3.3 Digital Core

Relatively few publications discuss the design and optimization of the digital core. However,

the digital core can consume significant power if implemented poorly, and this would in

turn degrade performance of the tag as a whole. Two exceptions exist. First, Zalbide [119]

presents a Gen2 digital core architecture. Unfortunately they target just one allowable link

frequency and consequently may not work with commercial readers (which also only support

certain subsets of the protocol). They do, however, offer simulation results comparing

performance gains from various power reduction strategies. Second, Ricci [120] presents

simulation results for a 2 µW Gen2 digital core with cryptography, and some analysis

on clock frequency selection. A more detailed analysis is presented in Section 9.6.8. For

reference, Cilek [121] discusses the impact of clock frequency variation on the baseband

processing in the tag. Oddly, they focus on the downlink communication, which is not the

limiting constraint (see Section 9.6.2).

9.4 Architecture

The socWISP has an RF front end, nano-power analog circuitry, and digital architecture

similar to conventional ID-only RFID tags. In this manner, the socWISP is a standalone,

Gen2-compatible RFID tag. However, a number of features provide novel functionality for

an integrated tag solution as compared to the prior work shown in Table 9.1. Most notably,

this work functions simultaneously as an addressable, passive Gen2 tag and also as a sensor

tag. Figure 9.1 shows a block diagram of the system.

Page 138: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

116

MSP430 µC

ADC

RegulatorsBandgap Ref.

RF Rectifier

Modulator & Demodulator

Packet Parse

Controller

Data Mux

Receiver

Transmitter

UID

On-Chip

Analog Core Digital Core Peripherals

Oscillator

Voltage Clamp, Supply DecouplingAntenna

Matching

Network

Figure 9.1: System-level architecture. On-chip components are shown in the tan area,including the analog (blue) and digital (green) cores as well as peripheral (red) circuitry.

There are 4 supply voltages used in the system. The analog circuitry can operate from

an unregulated rectifier output of 0.5 V to 3.6 V due to the use of thick-oxide (52A) devices.

The remainder of the on-chip circuitry uses standard (22A) devices and has a nominal supply

voltage of 1.2 V. Finally, a 1.8 V supply is provided for external devices such as an MSP430

µC and COTS opamps.

The digital core is functional with supply voltages ranging from 0.55 V to 1.2 V. To

reduce power consumption, a 0.7 V nominal supply is generated using two series diode-

connected transistors as a reference. This supply is intentionally CTAT (varies complemen-

tary to absolute temperature) in order to track the device Vt, and it is used to power both

the oscillator and digital core.

Many tags presented in the literature rely on precise clock frequencies to meet baseband

timing constraints. This work uses a programmable divider to output the correct baseband

waveform regardless of the local oscillator frequency. The divider is programmed on-the-fly

based on timing information provided by the Gen2 protocol. For clock frequencies ranging

from 1.7 MHz to 10 MHz, the digital core can output modulation that conforms to EPC

Gen2 specifications for frequency tolerance. This is discussed in detail in Section 9.6.8.

A nano-power current-starved 3 MHz ring oscillator clocks the digital core, and is de-

signed for a 0.6 V to 0.8 V supply. The 50% margin in clock frequency above the minimum of

1.7 MHz allows for supply, temperature, process and mismatch variation and also eliminates

need for costly post-manufacture trimming.

Page 139: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

117

The peripheral circuitry, which includes an 8-bit ADC and a unique ID generator (UID),

operates from a 1.2 V supply. There are numerous examples in the literature of sub-1 V

(even down to 500 mV) ADCs in the literature; however, it was not the aim of this project to

optimize the ADC. This general-purpose ADC can be used in conjunction with an external

sensor to transmit sensor data in a fully-Gen2-compliant manner. Chapter 11 presents an

example of such.

The UID [93] exploits process variation in the start-up configuration of an SRAM to

achieve a randomly-distributed, but 96% stable, tag ID. This alleviates need for non-volatile

memory, which adds complexity through high-voltage flash memory charge pumps or spe-

cial FeRAM processes. This circuit is also optimized for 1.2 V operation. Again, however,

numerous publications in the literature have demonstrated low-voltage subthreshold opera-

tion of SRAM cells. Note that tag circuitry is naturally compatible with subthreshold clock

speeds; byte-wise ID readout at the maximum uplink rate requires less than 100 kHz clock

rate.

The breadth of applications to which WISP was applied reinforced the need for config-

urability. In addition to allowing an arbitrary analog signal to be measured by the ADC,

the socWISP facilitates bidirectional RFID communication with an off-chip device. This

is intended to provide an interface for a MSP430 µC, which can perform data logging,

advanced power management, duty cycling and measuring of high-power sensors, and actu-

ation/control of other devices such as a powered sensor node. These devices can be powered

by a 1.8 V on-chip linear regulator, and µC power management is aided by a 2.2 V voltage

detector as in the WISP design.

Two versions of this design were fabricated, and Table 9.3 summarizes the differences.

Version 1 includes all of the circuitry described above. This is meant to be a full-featured

extensible sensor tag. Version 2 does not include a microcontroller interface, 1.8 V regulator,

or voltage detector. Instead, a FIFO asynchronously buffers ADC sampling from irregular

RFID communication. This allows periodic sampling for time-sensitive applications, and

this version functions as a stand-alone tag.

Page 140: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

118

Feature Version 1 Version 2

8-b ADC Yes Yes

ADC Access Un-Buffered FIFO-Buffered

ADC Sampling On Demand Periodic

1.2V, 0.7V Regulators Yes Yes

1.8V Regulator Yes No

2.2V Supervisor Yes No

Gen2 µC Interface Yes No

Table 9.3: Summary of Features vs. Design Version.

On-ChipRF RectifierMatching

Network

Analog Core

1.2 V Bandgap Reference

1.2 V Regulator 1.8 V Regulator

Bias Generator

0.7 V Regulator

Demodulator Ring Oscillator

Modulator

From Digital Core Digital

Supply

Analog

Supply

µC

Supply

Unregulated

Voltage

To Digital CoreAntenna

Figure 9.2: Analog block diagram. On-chip components are shown in the tan area.

9.5 Analog Design

Figure 9.2 shows a block diagram of the analog core components. [102] presents design

equations, layout diagrams, and simulation results for the bandgap, demodulator, regula-

tors and voltage detector. However, the schematics and underlying design methodology

will be repeated here for completeness. Additionally, measurement results are included in

Chapter 10. Lastly, design of the oscillator and bias generator will be described in detail.

Figure 9.3 shows a more detailed schematic of the analog core components. An off-chip

matching network, which could easily be absorbed into the antenna impedance, follows the

Page 141: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

119

antenna. On chip, six voltage-doubling rectifier stages recover the baseband reader signal.

An off-chip Schottky diode and storage capacitor form the unregulated supply voltage for

the tag by rectifying DC from the recovered baseband waveform. The Schottky diode could

easily be replaced with a low- or zero-Vt diode-connected transistor. However, this design

simplified testing and added minimal cost/area to the final design. A zener diode clamps

the unregulated voltage to approximately 3 V. An off-chip clamp was used due to its low

cost, small size and simplicity. If necessary, one of several on-chip clamps presented in the

literature could be implemented.

A large 10 µF storage capacitor is used for two reasons. First, the sensor signal con-

ditioning hardware needs to stay powered even while the reader is off. Because commer-

cial readers often operate at less than 50% duty cycle, this necessitates an energy storage

mechanism. Using a reader with 100% duty cycle and/or a sensor which can be quickly en-

abled/disabled, much smaller supply capacitors could be used. Second, an optional off-chip

µC can be powered by the 1.8 V regulator. COTS µCs require high current (∼ 500µA) for

short periods of time (∼ 100µs). A large supply cap is necessary source this current at long

wireless distances.

All of the analog circuits shown on Figure 9.3 are powered by the unregulated voltage.

The use of thick-oxide devices allows operation up to 3.6 V. A unit 50 nA reference current

is used to bias the analog blocks, which comes from a VGSR current reference. Two reference

voltages are used to create the supply voltages. First, the bias generator creates a 0.7 V

reference using two diode-connected series transistors. Second, a bandgap creates a 1.2 V

reference voltage for the regulators. Two unity-gain-feedback linear regulators create the

0.7 V and 1.2 V supply directly from the reference voltages. Lastly, a 1.8 V supply is created

with resistive feedback using the 1.2 V reference.

The demodulator compares the instantaneous rectifier output to the same signal passed

through a low-pass filter. This output, the recovered baseband reader-to-tag communica-

tion, is level-shifted to the digital core voltage. Finally, uplink communication is achieved

by a single transistor which shorts the antenna terminals together. Each component of

Figure 9.3 is described in detail in the sections below.

Page 142: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

120

Modulator

50

0.12

RF

IN

DC

OUT

6-Stage Rectifier

Clamp

Unreg

Rect

Out

Demod

Out

Un

reg

Dvd

d

Bia

s

Demodulator

Demodulator

µC Vdd

Avdd

Dvdd

Regulator

Unreg

FB

OUT

Bias

Ref

Regulator

Unreg

FB

OUT

Bias

Ref

Regulator

Unreg

FB

OUT

Bias

Ref

22nH

2pF

10µF

Antenna

2.5 MΩ

5.0 MΩUnreg

1.2VBandgap

Bias

50 nA

Bias Generator

Unreg 0.7V

Figure 9.3: Analog circuit schematic.

9.5.1 RF Rectifier

Figure 9.4 shows the voltage-doubling rectifier cell. A plethora of publications on rectifier

design have appeared in the literature as described in Section 9.3. Most focus on design for

sensitivity or efficiency. In reality, the most crucial feature is transistor Vt. This work was

fabricated in IBM 8RF 130 nm CMOS, which has standard-, low-, and zero-Vt transistors

in various oxide thicknesses. Back-leakage was tolerable with zero-Vt devices, and their low

forward drop provided good sensitivity and efficiency. As sensitivity and efficiency were

not dramatically affected by transistor size, the design was optimized for convenient input

impedance. After resonating the input capacitance with an inductor (which is often built

Page 143: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

121

Figure 9.4: Rectifier cell schematic. Six cells are stacked to increase output voltage at lowinput power.

into the antenna as an inductive short), the tag presents a 50-ohm load. This eases testing

and allows easy integration with standard antennas.

It should be noted that the rectifier requires careful simulation for accurate modeling

due to its highly-nonlinear nature. Periodic steady state (PSS) and periodic s-parameter

analyses were performed in Cadence (Spectre) to measure the input impedance versus input

power and frequency as well as rectification efficiency.

9.5.2 Bias Current Generator

A conservative textbook design is used for the bias current generation, shown in Figure 9.5

[122]. Despite use of thick-oxide devices, the bias generator starts reliably at 0.6 V. The

high output impedance of the cascode devices, even in subthreshold operation, maintains

a remarkably constant output current of 50 nA from 0.6 V to 3.6 V as shown in Fig-

ure 9.6. Simulated Monte Carlo process and mismatch bias current distribution is shown

in Figure 9.7. Variation is minimized through use of relatively-large device sizes as well as

precision resistors. This achieves a standard deviation of just 2.8 nA without trimming.

Layout area could be reduced significantly with the use of less precise resistors, but at the

cost of post-fabrication calibration.

Temperature variation is shown in Figure 9.8. The temperature coefficient is PTAT with

0.21nAC slope. This results in a ±12% variation between 0C and 50C from nominal 25C.

Page 144: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

122

vbn

vbp

vbnc

vbpc

1.6 MΩ

…..

50 nA Bias

Currents

vbpvbpc

Unreg

200 fFvbp

Startup Circuit Cascode Bias Voltage Bias Current Generator

vbnc

M1

M2

M27

M3

M4

M5

M6

M7

M8

M9

M10

M12

M14

M13

M15

M16

M17

M18

M19M11

M20

M22

M24M26

M21

M23

M25

Figure 9.5: Bias current generator schematic.

0

10

20

30

40

50

60

0 1 2 3 4

Input Voltage

Bias Current (nA)

Figure 9.6: Simulated bias current versus supply voltage.

0

10

20

30

40

50

60

70

80

90

100

39 40 42 43 44 45 47 48 49 50 52 53 54 55

Bias Current (nA)

Frequency

N=500,

Mean=46nA,

StDev=2.8nA

Figure 9.7: Simulated Monte Carlo bias current distribution (process and mismatch).

Page 145: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

123

30

35

40

45

50

55

60

-50 0 50 100

Temperature (Celcius)

Bias Current (nA)

Figure 9.8: Simulated bias current versus temperature.

Most startup circuits consume substantial quiescent current. This design uses MOS

pseudoresistors to minimize static power to several nanoamps. A 200 fF capacitor ensures

that the startup circuitry time constant is much greater than the bias generator core, and

this in turn prevents any oscillations from developing.

A fountain design is used to distribute bias currents. This is necessary because of the

very small currents used: transients take a long time to settle out leaving bias voltages

vulnerable to capacitively-coupled interference. Additionally, hard-switching circuits such

as the demodulator and oscillator can create noise on the NMOS bias voltage. This in

turn can upset sensitive circuits such as the bandgap references. To combat this effect, the

demodulator and oscillator generate their own NMOS bias voltages at the cost of increased

bias current.

9.5.3 Bandgap Reference

Figure 9.9 shows the bandgap reference voltage generator. The nominal output voltage is

1.2 V, which allows easy generation of the 1.2 V supply using unity-gain feedback. Design

equations and simulation results are presented in [102].

Like the bias current generator, this circuit employs a textbook topology. The bandgap

reference does not suffer the same tradeoff between precision and layout area like the bias

generator despite use of significantly-higher-valued resistors. Instead, output voltage ac-

Page 146: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

124

vbp

Unreg

8x1x

+

_

5 MΩ

1.2 V

5.5 MΩ

1x

M1 M2 M3

Figure 9.9: Bandgap circuit schematic to generate 1.2 V reference.

curacy is ensured through accurate matching between resistors. This is achieved using an

interleaved resistor layout. Parasitic junction diodes, which are characterized in IBM 8RF

130 nm, are used to form the bandgap.

9.5.4 Regulators

This design uses a linear regulator core to create various regulated voltages. By changing

the feedback and input reference voltage, various output voltages can be created. The core is

shown in Figure 9.10, and three incarnations that generate 0.7 V, 1.2 V, and 1.8 V supplies

are shown in Figure 9.3. The regulator core is biased with 100 nA quiescent current.

There are several design considerations. First, the PMOS pass transistor is 200 µm wide

to minimize VDS (I ·RON ) and thus provide low drop-out voltage. Second, a cascode stage

is used to increase opamp gain, which is low due to subthreshold biasing of transistors.

Addition of the cascode stage increases gain to approximately 67 dB. Third, a common-

centroid input transistor layout is employed to minimize input offset voltage which directly

impacts output voltage accuracy. Finally, feedback resistors for the 1.8 V regulator must be

quite large to keep quiescent current minimized. A resistive network of 5 MΩ and 2.5 MΩ

is used, which consumes about 250 µA. The 1.2 V and 0.7 V regulators use unity-gain

feedback, which does not consume quiescent current.

Many of the sensor tags reported in the literature employ shunt regulators. This is

Page 147: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

125

Bias

10pF

Unreg

Ref

Out

FB

M1

M2

M3

M4

M5

M6

M9

M1 M4

vbp

Figure 9.10: Regulator schematic. The feedback and reference voltage determine the regu-lated output voltage. The PMOS bias voltage is reused from the bias current generator.

feasible because they have no ADC - the on-chip temperature sensor does not require dig-

itization of an analog signal. Instead, they employ an open-loop ring oscillator that has

a known temperature vs. frequency characteristic. However, these sensors suffer from low

absolute accuracy accuracy and errors from fluctuations in the supply voltage. For preci-

sion measurement of most sensor outputs, analog to digital conversion is required. For a

wireless system, a known reference voltage is required to interpret the code returned by

the ADC. This is because the ADC code represents the input voltage as a fraction of the

reference voltage. This design uses a precision reference to generate the supply voltage, and

the supply is also used as the ADC reference voltage.

9.5.5 Demodulator

Gen2 protocol is discussed in detail in Chapter 2, but downlink communication will be briefly

reviewed here for context. Reader to tag communication employs pulse-interval encoding

(PIE). Specifically, the duration of positive pulse width determines whether each bit is a zero

or one. The PIE ratio, or ratio between a data-zero length and a data-one length, is 1.5:1 at

Page 148: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

126

minimum and 2:1 at maximum. The data-zero is called Tari, and can range from 6.25 µs to

25 µs. Obviously the positive pulse width cannot be determined without a negative pulse

to separate each bit. The negative pulse is denoted PW, and is the greater of either 26.5%

of Tari or 2 µs. The takeaway is that PW can be as small as 1.66 µs, which is greater than

600 kHz in bandwidth. This sets the lower limit on the demodulator bandwidth.

The digital core measures and converts the positive pulse durations into data and clock.

However, the received signal (the output of the rectifier) is not a suitable digital signal (ex,

see Figure 3.6). The purpose of the demodulator is to recover a logic-level (rail to rail)

signal from the rectifier output. The demodulator inputs are the rectifier output and a

low-pass-filtered version of the rectifier output. The RC filter has approximately 16 kHz

bandwidth and works to filter out the negative pulses. This is used as a reference voltage

to compare against.

The demodulator schematic is shown in Figure 9.11. It is essentially a 2-stage opamp;

the first stage provides differential to single-ended conversion with at least unity-gain am-

plification at the signal bandwidth. The second stage significantly boosts the signal swing

in order to achieve a logic level swing. Thick-oxide buffers with a low supply voltage (0.7 V)

prevent crowbar current due to the limited slew rate of the comparator. Phase margin is

not a concern; the comparator is an open-loop amplifier. The differential pair is biased with

75 nA, and the common-source stage is biased with 300 nA.

The second stage intentionally limits on the supply rails. The demodulator was designed

such that static current of cascaded second stage does not draw bias current while the

reader is off (which is more than 50% of the time). In order to maximize the differential

pair bandwidth, transistor sizes were minimized at the expense of mismatch. However, the

input drive is greater than 50 mV, so significant input offset voltage is tolerable.

9.5.6 Oscillator

A 3-stage current-starved ring oscillator is used to clock the digital core. Figure 9.12 shows

a schematic of the design. The oscillator is designed for a minimum frequency of 3 MHz,

which consumes a nominal 267 nA including the NMOS bias voltage generation. Large 1 µm

Page 149: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

127

Bias

10pF

Unreg

DvddRect Out

21 pF

550

Demod

Out

M1 M2

M3 M4

M5M6

M7

M8

M9

M10

M11

M12

Figure 9.11: Analog demodulator that is used to recover a logic-level baseband signal fromthe rectifier output.

Figure 9.12: 3 MHz ring oscillator schematic.

by 1 µm devices are used to reduce mismatch. It should be noted that parasitic extraction

is crucial for achieving useful simulation results. Without parasitics, simulation results vary

from 3.5 MHz to 6.8 MHz. With parasitics, simulation results vary from 3 MHz to 4.7 MHz.

Frequency control by use of current-starving naturally produces a poor slew rate. This

tends to cause large crowbar currents in the digital buffer following the oscillator. An

open-loop common-source amplifier significantly improves slew rate for the following digital

buffers, while consuming a fixed bias current itself.

Ring oscillators are essentially relaxation oscillators with an effective resistance defined

Page 150: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

128

2.0

2.5

3.0

3.5

4.0

4.5

5.0

5.5

6.0

6.5

7.0

-50 0 50 100

Temperature (Celcius)

Oscillator Frequency (MHz)

FFF

FS

SF

SSF

(a) Simulated oscillator frequency versus process and temperature without compensation circuit.

2.5

3.0

3.5

4.0

4.5

5.0

-50 0 50 100

Temperature (Celcius)

Oscillator Frequency (MHz)

FFF

FS

SF

SSF

(b) Simulated oscillator frequency versus process and temperature with compensation circuit.

Figure 9.13: Simulated oscillator performance versus process and temperature.

Page 151: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

129

by the current sources, capacitance set by CGS , and hysteretic up/down comparison voltages

defined by the NMOS and PMOS threshold voltages, respectively. Temperature changes

significantly affect the transistor Vt, which is a substantial fraction of the 0.6 V to 0.8 V

supply voltage. A relatively-simple, zero-power, open-loop compensation is created by M1

and the 1.3 MΩ resistor, R. The input bias current of 50 nA creates a voltage across R which

sets the VGS of M1. As the Vt of M1 decreases with increasing temperature, more current is

drawn away from M3 into dummy load M2. This reduces the bias current of current mirror

M3, which provides bias current compensation that tracks Vt. By choosing an appropriate R,

simulated variation is reduced from 40% to 10% for slow and normal PMOS corners between

-50C and 100C. Because PMOS process variation affects the Vt of M1, the compensation

also helps control the PMOS transistor variation. At room temperature, variation across

process corners is reduced from 67% to 30%. The worst case variation over process and

corners is reduced by a factor of two. Simulation results are shown in Figure 9.13(a) and

Figure 9.13(b).

Using PMOS transistors without current limiting (M4, M7, M10) causes the PMOS

transistor speed to affect the oscillation frequency. This is due to the effective “charging”

resistance being set only by the PMOS on-resistance. The disadvantage of this topology is

about 40% frequency variation over corners. However, the cost of adding PMOS current

sources is relatively high: increased supply voltage, and increased bias current. Stacking

another device necessitates a higher supply voltage, and also requires an extra current mirror

to generate the PMOS bias voltage. More importantly, however, the “free-running” PMOS

transistors greatly increase the oscillation frequency for a fixed bias current. A significantly

higher bias current would be required to achieve the same oscillation frequency with PMOS

current sources added.

Because the oscillation frequency is a function of Vt, which is referenced to the supply

rails, it is strongly affected by the supply voltage. Simulation results of frequency and supply

current versus voltage and temperature are shown in Figure 9.14(b) and Figure 9.14(a). This

is left uncompensated, relying on supply regulation for consistent performance.

Page 152: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

130

3.0

3.5

4.0

4.5

5.0

0.5 0.6 0.7 0.8

Vdd

Oscillation Frequency (MHz)

-50˚C

0˚C

50˚C

(a) Simulated oscillator frequency versus temperature and supply voltage.

0.0

100.0

200.0

300.0

400.0

500.0

600.0

700.0

800.0

900.0

1000.0

0.5 0.6 0.7 0.8

Supply Voltage

Supply Current (nA)

-50˚C

0˚C

50˚C

100˚C

(b) Simulated oscillator current consumption versus temperature and voltage.

Figure 9.14: Simulated oscillator performance versus supply voltage and temperature.

Page 153: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

131

Demodulator

Digital Core

Modulator

Oscillator Prescaler

Receiver

Encoder

Preamble

CRC 16

Sequencer

ControllerPacket Parse

RNG

UID

ADC

µC

MUX

Figure 9.15: EPC Gen2 digital core block diagram. The controller implements the Gen2RFID state machine, selecting the proper packets to send in response to received packets.

9.6 Digital Design

A high level block diagram of the digital core is shown in Figure 9.15. The Receive block

converts incoming downlink communication from pulse-interval-encoding (PIE) to clock and

data. A PacketParse block decodes EPC commands and stores relevant information for the

Controller and Prescaler. The Controller block decides what packet to send after receiving

a packet from the reader, and enables the transmitter if appropriate.

The transmit architecture is very modular. A standard 3-wire interface consisting of

clock, data and done is used for data sources, which are selected through a multiplexer

(MUX) by the Controller. The Sequencer aligns all of the pieces of a packet together: a

Preamble, data from the MUX, and CRC16 if appropriate (not all packets have a CRC).

Finally, the Encoder block converts the clocked data to FM0 or Miller encoding. The output

of the Encoder drives the uplink backscatter modulation transistor.

The Encoder requires a clock at twice the backscatter link frequency (BLF). This clock is

generated by scaling the input oscillator clock using a programmable divider. Section 9.6.8

describes how a minimum oscillator frequency is determined by residual divider error and

Section 9.6.7 presents the encoder design.

Page 154: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

132

9.6.1 Controller

The Controller, shown in Figure 9.15, manages the EPC Gen2 state transitions, determines

if the tag should respond to received commands, and selects appropriate transmit data

packets. At a lower level, the Controller manages the transmit and receive states, providing

reset and enable signals to each block. The Controller also determines what data is sent in

response to the Read command. For version 1, the user memory address selects either the

µC or on-chip ADC as a data source. For version 2, the ADC is read through a FIFO and

the address determines whether the FIFO should advance the read pointer or keep it fixed.

This allows dropped packets to be repeated.

9.6.2 Receiver

A schematic of the receiver digital logic is shown in Figure 9.16. A 10-bit counter measures

incoming PIE symbols which are then compared against the pivot. Decoded data is clocked

out to the parsers. The LSB of the counter is used for random number generation, which is

discussed in Section 9.6.4.

The counter size determines the maximum clock frequency. A 10-bit counter allows

up to 10 MHz input clock frequencies without overflow for the longest symbol (TRcal) at

the slowest data rate. On the other extreme, the minimum clock frequency must be fast

enough for a digital counter to differentiate between zeros and ones. The minimum difference

between a data-zero and data-one is 3.1 µs. A clock rate greater than 323 kHz could detect

this change. However, this is irrelevant because the uplink specifications necessitate a much

higher clock frequency in the megahertz range (see Section 9.6.8).

Most information sent by the reader is digitally encoded. However, TRcal, which is used

to set the transmit frequency, is a temporally-encoded signal. In other words, the pulse

interval duration encodes information about the required uplink data rate. To store this

information, the counter value for the TRcal symbol is output directly to the Transmit (TX)

Settings module. TX Settings stores these values until the next Query.

Two state machines make up the receiver. The first is a low-level edge detector, shown

in Figure 9.17. After a low-high transition, the bit is evaluated and the physical layer state

Page 155: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

133

clockOscillator

resetcount

Demodulator

LSBRNG

timeoutTRcal

Counter

rx clock

rx data

Logic Parser

TX Settings

PIE

data

Figure 9.16: Pulse-Interval-Encoding (PIE) demodulation logic schematic.

demod

high

always

demod

low

resetWait For

Demod

Low

Reset

Counter

Wait For

Demod

High

Evaluate

Bit

always

Figure 9.17: Edge detection logic state diagram.

transition is evaluated as shown in Figure 9.18. The states progress through the various

symbols in a packet. The TRcal state is selected if the measured count is greater than

RTcal, which is the only case in which this happens. For a review of the Gen2 packet

structure, see Chapter 2.

There are no error-handling states in the code. However, a timeout signal is generated in

the bits physical layer decoder state (see Figure 9.18) if the current counter value is greater

than RTcal. This timeout signal serves two purposes. Most obviously, if a timeout occurs

before the packet is complete, the controller resets the Receiver module. A more subtle

purpose, however, is to notify the Encoder that one RTcal period has transpired following

Page 156: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

134

Data 0

always

RTcal

Bits

TRcalDelimiter

reset

else

always

count > RTcal

reset

always

* State transitions occur in the bit evaluation state

Figure 9.18: EPC physical layer packet digital decoder schematic.

rx clock

rx data

Receiver

MUX

8-Bit

Memory

Bit

Counter packet

complete

command

complete

M

DR

TRext

Logic

packet type9

2

Figure 9.19: Schematic for command parsing logic.

the last bit of the received packet. This is one of two conditions that the Encoder must wait

on before uplink transmission may occur.

9.6.3 Packet and Command Parsers

Two logic blocks are used to decode and parse the incoming data stream from the Receiver

module. The command parser, shown in Figure 9.19, detects which type of packet is being

received, and outputs signals to indicate when the command has been received and also

when the entire packet has been received. Additionally, the Query information relevant for

configuring the transmit block (TREXT, DR, M) is decoded.

Page 157: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

135

rx clock

rx data

Receiver

Logic

+

MUX

Bit

Counter

read: words, pointer, bank

Q, up/dn

write data

Memory

Logic

handle

RN16

match

write clock

packet

type

Command

Parser

Figure 9.20: Schematic for packet parsing logic.

The packet parser, shown in Figure 9.20, is responsible for interpreting and storing

the payload of each different packet type. Depending on the packet type, as indicated by

the command parser, the incoming data is parsed differently. Signals related to the Write

command are passed directly through this module. Signals related to the Read command

are stored in memory, which remains set until the Controller resets this module. Finally,

signals pertaining to a reader-requested handle are compared against the tag’s handle, and

a match indicator flag is set appropriately.

9.6.4 Random Number Generation

A random number for tag singulation is generated by sampling the LSB of the baseband

bit-length counter, and passing these sampled bits through a linear-feedback shift-register

(LFSR). Barnett [98] takes a similar approach, sampling the RF carrier directly at the cost

of higher power consumption. Three mechanisms create randomness such that two tags

do not generate the same random number. First, the bit-length counter is clocked by the

on-chip current-starved ring oscillator, which is inherently noisy and uncorrelated to the

baseband signal. Second, the precise bit-length of the demodulator output varies based

on RF input signal strength, process variation in bias current, and mismatch of transistor

sizes. Each tag generates a slightly different baseband output, which is easily greater than

the period of the oscillator. Finally, the use of an LFSR creates history-dependency in the

random number generation. At minimum, changes in multipath due to reader frequency

Page 158: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

136

do

CRC

Controller

start TX

done

CRC

16

Preamble

Generatorclock

preamble

data

done CRC data

clock out

done

data in

clock in

clock done data

ready stopclock done data

M

generate

violation

Receivertimeout

TREXT

DR TRcal

N

clock

Oscillator

Encoder

Logic ÷ N

TX

Settings TX

Settings

Data Source

MUX

Sequencer

modulator

out

Data

Sources

Figure 9.21: EPC Gen2 uplink packet construction and modulation schematic.

hopping will affect each tag’s demodulator output. This creates further uniqueness in the

random number generation.

9.6.5 Transmitter Architecture

The full transmitter logic schematic is shown in Figure 9.21. A Prescaler (Section 9.6.8) di-

vides down the clock to the appropriate backscatter link frequency (BLF). The Encoder (Sec-

tion 9.6.7) requests data from the Sequencer by clocking it. The Sequencer (Section 9.6.6)

orchestrates the construction of a packet by routing the clock to the each data source. The

data sources (Preamble, MUX, and CRC16) each provide a done indicator, which causes

the Sequencer to either move to the next data source in the packet, or tell the Encoder to

stop. Upon receiving a stop signal, the Encoder finishes up sending the current bit and then

sets a done flag. This flag from the Encoder tells the Controller that the transmission is

complete and that the transmitter can be reset.

Page 159: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

137

done & CRC

done

done

resetPreamble

Mux

CRC16done &

!CRC

Done

Figure 9.22: Sequencer state diagram for constructing EPC Gen2 packets.

9.6.6 Sequencer

The Sequencer is responsible for creating a Gen2-compliant packet and passing it to the

Encoder for conversion to FM0 or Miller encoding. As the name suggests, it sequences data

streams from the Preamble generation module, a data source from the MUX selected by the

Controller, and finally the CRC16 if enabled by the Controller. A state diagram for these

transitions is shown in Figure 9.22.

A standard interface is used for all modules connected to the Sequencer: clock is sent

to the module, data is expected starting on the second positive clock edge, and the module

asserts a done signal when it is finished sourcing data. After each module is finished sending

data, the Sequencer clocks the next module. When all of the modules are complete and

the Encoder has finished sending the last bit, the Sequencer asserts a transmit complete

signal for the Controller. Note that the Sequencer does not generate a clock; the Encoder

generates the clock to pull in data at the correct timing.

9.6.7 Encoder

The Encoder’s task is to convert data into Gen2-compliant modulation. A schematic is

shown in Figure 9.23, with some of the details of Figure 9.21 omitted for simplicity. Both

FM0 and Miller modulations encode data through phase inversions (PI) of a square wave

Page 160: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

138

2xBLF clockBLFPrescaler ÷ 2

PI 2xBLF

modulation out

data

÷ M

FF

XOR

Sequencer

FSM

ANDEnable

violation

Figure 9.23: FM0 or Miller modulation encoder schematic.

with frequency equal to the BLF. They differ in the timing of when the phase inversions

occur. The finite state machine (FSM) positions the phase inversion correctly in the outgoing

waveform depending on the data from the sequencer and on the selected modulation type

(FM0 or Miller). The Encoder is clocked at twice the BLF so that the finite state machine

(FSM) can manipulate each phase of the outgoing waveform.

The Miller modulation M sets the data rate relative to the backscatter link frequency

(BLF) via a divider. For FM0, the BLF is the data rate, and the divide ratio is 1. For

Miller 2, 4 and 8, the divide ratio is the Miller number. This clock drives the sequencer,

which returns data to the finite state machine (FSM). The finite state machine manipulates

the output waveform using an exclusive-or operation which is discussed below.

The FM0 preamble employs a violation of the encoding rules for signaling purposes.

To create this exception, the preamble sends a violation flag to the encoder, which over-

rides the standard encoding rules. Compliant modulation waveforms are found in the EPC

specification [27].

9.6.8 Prescaler

The EPC Gen2 specification defines the tag backscatter link frequency (BLF) as shown in

(9.1). TRcal is a length of time specified to the tag as a pulse interval, and DR is 8 or 643

depending on the Query settings.

BLF =DR

TRcal(9.1)

Page 161: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

139

The tag measures TRcal with a digital counter clocked at the oscillator frequency f0. This

produces a measurement of TRcal called TRcount as given by Equation (9.2).

TRcount = TRcal · f0 (9.2)

In order to manipulate both positive and negative edges, the Prescaler provides a clock

fTX to the Encoder which is double the BLF. Combining (9.1) and (9.2), the input clock

f0 requires division by N to equal fTX as given by (9.3) where DR is 8 or 643 depending on

the Query settings.

N =TRcount

2DR(9.3)

Multiply and divide hardware consumes significant area and power. A more efficient

implementation is given by (9.4) and (9.5), where one of two divider values is chosen de-

pending on the required DR. Note that << and >> represent bit-wise shift operations

with the in-shifted bit (MSB or LSB) filled by a logic 0. Shifting to the left multiplies by

two, and shifting to the right divides by two. α = 4 and β = 13 are empirically chosen to

center the residual divide error about zero.

N8 = (α+ TRcount) >> 4 (9.4)

N 643

= (β + (TRcount << 1) + TRcount) >> 7 (9.5)

To illustrate this operation with an example, let the tag oscillator frequency f0 = 3 MHz

and the desired BLF = 256 kHz. The reader would send TRcal = 31.25 µs and DR = 8

(alternatively, it could send TRcal = 83.3 µs and DR = 643 ) using (9.1). The tag would

measure TRcount = 93 using (9.2). The tag would then compute the divide ratio N8 = 6

using (9.3). This would create a 500 kHz clock for the encoder and thus a 250 kHz fTX .

The error of 6 kHz, or 2.3%, is well within the Gen2 specified tolerance.

Residual errors versus input clock f0 are shown in figures 9.24, 9.25, and 9.26 for transmit

frequencies with the lowest divide ratios and most strict tolerances. The highest errors

occur for the fastest transmit frequencies because the divide ratio becomes coarse as it

approaches 1. A popular choice of clock frequency is 1.28 MHz [98]; however, this allows

a divide ratio of only 2. However, as shown in Figure 9.24, this requires clock frequency

Page 162: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

140

-30%

-20%

-10%

0%

10%

20%

30%

40%

0.5 1 1.5 2 2.5 3 3.5

Frequency (MHz)

Residual Error

640 kHz, DR 64/3

Upper Limit

Lower Limit

Figure 9.24: Backscatter link frequency error due to divider residual for 15% allowabletolerance transmit frequencies.

-20.0%

-15.0%

-10.0%

-5.0%

0.0%

5.0%

10.0%

15.0%

20.0%

0.5 1 1.5 2 2.5 3 3.5

Frequency (MHz)

Residual Error

320 kHz, DR 64/3

256 kHz, DR 64/3

320 kHz, DR 8

256 kHz, DR 8

Upper Limit

Lower Limit

Figure 9.25: Backscatter link frequency error due to divider residual for 10% allowabletolerance transmit frequencies.

Page 163: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

141

-15%

-10%

-5%

0%

5%

10%

15%

0.5 1 1.5 2 2.5 3 3.5

Frequency (MHz)

Residual Error

160 kHz, DR 8

Upper Limit

Lower Limit

Figure 9.26: Backscatter link frequency error due to divider residual for 7% allowable tol-erance transmit frequencies.

accuracy of ±15% for 640 kHz transmit frequencies because the next divider steps of 1

and 3 do not reduce the error to a tolerable level. This is notoriously difficult to achieve

without trimming for open-loop ring oscillators. In contrast, for clock frequencies above

approximately 1.7 MHz, the divider can reduce the error to a tolerable for any increase in

frequency.

Figure 9.25 shows residual error for transmit frequencies requiring 10% tolerance. Again,

above 1.7 MHz any frequency error can be corrected with an appropriate divider setting.

Note that it is important to center the residual error around 0, otherwise the error will start

at 0, increase to twice the absolute error shown currently, and then reset to 0.

9.7 Chip Layout

The layout for the chip is shown in Figure 9.27. Version 1 and 2 appear to be the same in

size; however, version 1 contains several pads for optional probe-test debugging. Figure 9.7

provides a comparison of the chip area used for each component. The capacitor array of the

ADC consumes a large portion of the area, and the digital cores also require significant real

Page 164: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

142

Version 1 Version 2Chopper

AmpTest

Cells

2.5 mm

1.0 mm

Figure 9.27: Full chip layout with dimensions and high-level blocks annotated.

Digital Core

ADC

UID

Demodulator

Bandgap

Bias Gen.

Regulators

Rectifier

(a) Layout area for version 1.

Digital Core

FIFO

ADC

UID

Demodulator

Bandgap

Bias Gen.

Regulators

Rectifier

(b) Layout area for version 2.

Figure 9.28: IC layout area breakdowns.

estate. The complexity of the Gen2 protocol also causes significant area to be consumed by

the digital core. The FIFO size could be dramatically reduced by use of a buffered SRAM

cell. This increases chip complexity, however, due to the need for asynchronous read and

write access. The chopper amplifier area is dominated by MIM capacitors much like the

ADC. Finally, the area for each version is summarized in Table 9.4.

As mentioned, this work was designed and fabricated in IBM 8RF 130 nm CMOS tech-

nology. A die micrograph is shown in Figure 9.29. The chip was covered with a metal

plane, which was constructed in the top metal layer. The intention was to prevent RF

Page 165: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

143

Component Without Pads With Pads

Version 1 0.29mm2 0.72mm2

Version 2 0.47mm2 0.68mm2

Table 9.4: IC layout area with and without pads.

Version 1 Version 2 Chopper

Amp

2.5 mm

1.0 mm

Figure 9.29: IC die micrograph.

interference from coupling onto MIM capacitor plates and other sensitive structures. For

example, the amplifier used in Chapter 6 is extremely sensitive to RF signals from sources

such as cellular phones and RFID readers. Copper foil helped shield the amplifier from this

interference, and so similar protections were applied here on chip. The low system noise

level while sampling a DC input, shown later in Chapter 10, demonstrates the effectiveness

of this shielding.

9.8 Performance

Idle and active system level current consumptions are shown in Table 9.5. Idle current

for the digital core is roughly 500 nA unclocked. Current consumption increases to 5 µA

at 3 MHz clock rate but with no communication activity. Active current is 6 µA total at

400 Hz tag rate. The current varies little with tag read rate. The peripheral core active

current varies with tag communication rate as well as ADC and UID utilization.

Table 9.6 further breaks down the current consumption for each block of the analog core.

An extra 50 nA is included in the bias generator current to account for a current mirror

Page 166: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

144

Component Operating Voltage (V) Quiescent (µA) Active (µA)

Analog Core 0.5 to 3.6 1.4 1.7

Digital Core 0.55 to 0.9 0.5 6

Peripheral Core 1.2 0.1 1

Table 9.5: Power consumption breakdown for analog, digital, and peripheral cores. Notethat the analog core minimum operating voltage includes only the bias generator and de-modulator. Operation of all on-chip components requires 1.3 V, and optional off-chip com-ponents requires 1.9 V.

Table 9.6: Breakdown of simulated current consumption of analog blocks.

(a) Base Configuration

Component nA

Bias Gen. 250

Bandgap 220

0.7 V Reg. 100

1.2 V Reg. 100

Demod. 375

Oscillator 260

Sub-Total 1305

(b) Version 1 Extras

Component nA

2.2 V S.V. 100

1.8 V Reg. 350

(c) Totals

Version nA

V1 Total 1755

V2 Total 1305

which creates a bias voltage for the regulators and bandgap reference. The current shown

for the demodulator is worst case; when the reader is off, the 2nd stage shuts off and current

is reduced by 300 nA.

Table 9.7 compares the performance of this work to other tags presented in the literature.

For a comparison of tag features, refer back to Table 9.1. This work compares favorably to

other sensor tags without resorting to duty cycling (WISP, Kocer), loss of Gen2 compatibil-

ity (Shen), or elimination of the ADC (Kim). Drawing comparisons to tags lacking sensors

and/or addressability (all remaining tags) is less meaningful, but Karthaus and Barnett

clearly achieve the best RF sensitivity for a fully-passive tag. In part, the use of a linear

Page 167: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

145

regulator, ADC and more complex state machine simply limit the achievable performance.

However, the power consumption of the digital core can clearly be reduced significantly.

Optimization of the digital core would add 3 to 5 dB of RF sensitivity.

A number of footnotes accompany Table 9.7. Unfortunately many authors fail to report

both the tag power consumption and RF sensitivity. When omitted, the RF sensitivity

is calculated from tag power consumption based on two assumptions: a conservative 20%

rectifier efficiency and a 50% loss due to downlink and uplink modulation. Conversely, if RF

sensitivity is presented, the inverse calculation is performed using the same assumptions. In

the case of Nakamoto [96], RF input power is calculated using Friis’ Transmission Equation

with no multipath or polarization losses (which matches results shown in Figure 17.2.4 of

[96]). Again, 50% modulation losses are assumed, but rectifier efficiency is specified as

36.6%.

9.9 Design Summary

Design methodology and simulation results were presented for two tag versions. The first is

a full-featured sensor tag with optional µC interface. This enables flexibility in application

development via µC programmability (ie, system-on-chip WISP) while leveraging ASIC

performance gains (low power Gen2 state machine, and analog components). The second

version enables periodic sampling for time-sensitive applications via a FIFO buffer. Such

an application is presented in Chapter 11.

Several features enable high performance for a feature-rich tag. Addressability is pro-

vided by a unique ID generator [93] which eliminates the need for high voltage charge pumps

(EEPROM / Flash) or special processes (FeRAM). An ultra-low-power ring oscillator with

analog and digital compensation helps minimize system power consumption. Similarly,

ultra-low-power analog components provide a stable system supply for accurate ADC con-

version. Finally, the digital core provides interfaces for reading the ADC and communicating

with an off-chip microcontroller while remaining fully-Gen2 compliant.

Page 168: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

146

Table 9.7: Comparison of published (measured) tag performance.

Sensitivity Active Sleep Vdd Area Tech. Clock

This Work -12dBm 9µA 6µA 1.5V1 1.0mm2 0.13µ 3MHz

WISP -9dBm2 800µA 2µA 1.8V 1.5cm2 PCB 3MHz

Kim [94] -5.3dBm3 15µA 0.5µA 2.0V 1.0mm2 0.25µ4 -

Kocer [95] -12.3dBm2 660µA 0.9µA 3V 1.0mm2 0.25µ -

Nakamoto [96] -6.2dBm5 87.8µW - - 1.8mm2 0.25µ -

Pillai [97] 19dBm2 12µA 0.66µA 1.5V 5mm2 1.0µ -

Karthaus [92] -17.8dBm 1.5µA - 1.5V - 0.5µ4 300kHz

Barnett [98] -14dBm 2.75µA3 - 1.45V 0.55mm2 0.13µ4 1.28MHz

Shen [99] -8.2dBm3 10.2µA6 - >1.5V 0.8mm27 0.35µ4 -

Cho [100] -12.9dBm3 3.4µA - 1.5V 0.42mm27 0.25µ 330kHz

Shenghua [101] -20.5dBm3 0.9µW8 - - 0.2mm2 0.18µ -

1Optional 1.8 V regulator requires Vdd > 1.8 V. On-chip components employ 0.7 V and 1.2 V regulators.

2Duty cycling required at this input power.

3Calculated based on 20% rectifier efficiency and 50% modulation losses.

4Process includes Schottky diodes.

5No power numbers reported, but authors claim 36.6% rectifier efficiency and 4.3m range given 4W EIRP.

6Simulation result, no measurements reported.

7Excludes pad area.

8Supply voltage not specified.

Page 169: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

147

Chapter 10

SOCWISP VERIFICATION AND MEASUREMENT RESULTS

10.1 Introduction

This chapter will discuss the socWISP performance and functionality verification. Sec-

tion 10.2 discusses efforts to ensure correct functionality pre-tapeout. Section 10.3 presents

experimental methods. Section 10.4 includes measurement results, and Section 10.4.8 con-

cludes.

10.2 Pre-tapeout Verification

Analog behavior was simulated exclusively using Cadence/Spectre. With the exception of

the peripheral circuitry (UID, ADC), the demodulator and modulator interfaces to digital

core were designed to specification but not simulated together. Regulator behavior was also

designed to specification, with load regulation behavior guaranteed through maximum Iload

and minimum Cload [102].

Verifying the digital core design using simulated test cases is relatively easy for normal,

successful operation. A Verilog RFID reader was designed to stimulate the tag by generating

and sending the downlink baseband signal and interpreting the tag uplink signal. In the tag

implementation, these signals ordinarily connect to the demodulator input and modulator

output of the tag digital core. Using this setup, Gen2 physical and link layer behavior were

verified using ModelSim. Additionally, mixed-signal simulations were performed in Cadence

to verify the digital core interaction with the ADC and UID.

Note that this test reader only tests tag behavior with logic-level digital signals; that is,

the analog and RF signals signal paths are not included in this simulation. A full reader

implementation would include signal processing to recover and threshold noisy, sampled tag

backscatter from an ADC. Examples of FPGA-based RFID readers have been presented

in the literature; however, they are beyond the scope of this discussion. The reason we

Page 170: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

148

will not consider the reader implementation is because a COTS reader was used to verify

functionality. This is discussed below.

In contrast to nominal operation which is easy to verify, a wide variety of error conditions

can cause failure with first silicon and require revision. To ensure robust operation in

realistic conditions, the digital core was implemented on an Altera Cyclone II FPGA (via

the DE2 development board). An Impinj Speedway reader was used to test the system,

and a WISP analog front end (rectifier, modulator, demodulator) enabled the FPGA to

communicate with the reader. This was valuable for ensuring robust demodulation as well

as verifying the uplink encoder, prescaler, and CRC16. Although the commercial reader

does not delineate specific errors (with the exception of a bad CRC), tag read errors indicate

need for debugging. Real-time outputs such as the packet complete, packet type, and handle

match flags were monitored on an oscilloscope to debug errors in real time. These conditions

were replicated in simulation to troubleshoot and resolve issues.

10.3 Methods

A fully-instrumented test board is shown in Figure 10.1. The reader RF signal is not

transmitted by antenna; a variable attenuator is inserted in line with a coax cable between

the reader and tag. This eliminates multipath effects and simplifies testing. Specific testing

on the analog and digital cores is described below.

10.3.1 Analog core

Various tests were performed to verify functionality and yield of the design. Transient step

and DC loading tests were performed for the regulators. Demodulator functionality was

verified with an COTS RFID reader by monitoring signals on an oscilloscope and later

through full-system tests. The bandgap, bias generator and oscillator yield were measured

across 10 boards, with two versions for each board. Rectifier performance input impedance

was measured with a VNA, and output voltage/power versus input power was measured

with an RF signal generator. Results are presented in Section 10.4.

Page 171: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

149

Figure 10.1: Photograph of test setup for socWISP verification.

10.3.2 Digital core

As with pre-fabrication verification, the post-fabrication verification was aided greatly by use

of a COTS Impinj Speedway reader. Initial tests employed a WISP front end as before. After

the analog core functionality was verified, the full-chip tests were performed. Information

from tags was parsed, displayed and logged to a text file using a C# application (see

Section 8). Several Gen2 modulation parameters are available from the Speedway reader,

and functionality is demonstrated in Section 10.4.5.

10.4 Measurement Results

10.4.1 Rectifier

The rectifier input impedance versus frequency, after an on-board matching network, is

shown in Figure 10.2. Note that UHF RFID in the United States operates between 902 MHz

and 928 MHz. Figure 10.3 shows input match with the modulator turned on and off. This

demonstrates that the uplink transistor is effective in modulating a large change in power

Page 172: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

150

Figure 10.2: Measured rectifier input impedance versus frequency at -15 dBm input power,plotted on a smith chart from 850 MHz to 950 MHz.

reflected to the reader. This in turn provides the maximum backscatter signal for the

reader. Lastly, rectifiers are highly nonlinear - this causes varying input impedance with

input power. Figure 10.4 shows return loss versus input power. The match is optimized for

low input power levels where the tag requires the most efficiency in order to operate. At

high input power levels, efficiency is not important because there is an abundance of input

power.

Output voltage versus input power for a 1.2 µA load is shown in Figure 10.5, and

Figure 10.6 shows rectifier efficiency at varying output voltages. This data was taken by

connecting the rectifier output to a power supply at the desired output voltage. An ammeter

was connected in series to measure current delivered from the rectifier into the power supply.

In the efficiency calculation, rectifier back leakage was counted as loss, while power delivered

to the power supply and to the analog core was counted as delivered power. Lastly, the data

Page 173: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

151

-25

-20

-15

-10

-5

0

850 900 950 1000

Frequency (MHz)

| S11 (dB)|

Modulator Off

Modulator On

Figure 10.3: Measured rectifier return loss versus input frequency with modulation transistoron and off. Measurements were taken at -15 dBm input power with frequency ranging from850 MHz to 1 GHz.

-25

-20

-15

-10

-5

0

-15 -10 -5 0

Input Power (dBm)

| S11 (dB)|

Figure 10.4: Measured rectifier return loss versus input power at 915 MHz.

Page 174: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

152

0.50

0.70

0.90

1.10

1.30

1.50

1.70

1.90

2.10

2.30

-23 -21 -19 -17 -15

Input Power (dBm)

Output Voltage (V)

Figure 10.5: Measured rectifier output voltage with analog core (approx. 1.2 µA) load.

was taken at room temperature. Increases in temperature reduce the transistor Vt, which

increases the efficiency but also increases the rectifier back-leakage.

10.4.2 Oscillator Performance

The oscillator frequency distribution at 0.7 V supply voltage is plotted in Figure 10.7. The

mean is 2.96 MHz and the standard deviation is 387 kHz. Greater than 3 standard deviations

of measured oscillator variation is tolerable for the tag to meet timing specifications.

Oscillator frequency variation over temperature is shown in Figure 10.8. Note that the

oscillation frequency increases with both supply voltage and bias current. However, the bias

is PTAT while the supply is CTAT. This data includes the PTAT bias generator variation

but not the CTAT supply voltage variation because it was powered from a power supply.

Regardless, the temperature coefficient of the oscillator is tolerable. Without the oscillator

temperature compensation circuit, the variation would be much higher. With additional

compensation from the CTAT supply, temperature variation will reduce further.

Page 175: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

153

0%

5%

10%

15%

20%

25%

30%

35%

40%

-25 -20 -15 -10 -5 0

Input Power (dBm)

Rectifier Efficiency

0.5 V

1.0 V

1.5 V

2.0 V

WISP 1.8 V

Figure 10.6: Measured rectifier power efficiency at various output voltages.

2

2.2

2.4

2.6

2.8

3

3.2

3.4

3.6

3.8

4

1 2 3 4 5 6 7 8 9 10

Board #

Oscillation Frequency (MHz)

Figure 10.7: Measured oscillator frequency distribution at 0.7 V supply voltage.

Page 176: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

154

3

3.1

3.2

3.3

3.4

3.5

0 25 50 75 100

Temp. (C)

Freq. (M

Hz)

Measured Data

Simulated Slow-Slow Corner

Simulated Fast-Slow Corner

Figure 10.8: Measured ring oscillator frequency versus temperature.

10.4.3 Bandgap and Regulators

The bandgap voltage is well-represented by the unloaded output voltage of the 1.2 V regula-

tor due to its unity-gain feedback. Both versions have identical circuitry, so their measure-

ments can be combined into one statistic. Mean output voltage is 1.20 V, with a standard

deviation of 1.05%. The mean of both versions are identical, but version 2 has a higher

standard deviation; this is expected given the small sample size of 10. The 1.8 V regulator

only exists on version 1 and shares the bandgap reference. Mean output voltage is 1.78 V,

and the standard deviation is 0.64%. The 1.8 V regulator standard deviation is identical

to that of 1.2 V regulator samples from version 1. This suggests that the resistor ratio

accuracy is very good, and variation is originating from the bandgap reference.

Page 177: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

155

10.4.4 Bias

The bias current generator cannot be measured directly due to pin constraints. However,

the total analog core quiescent current can be used as a rough proxy (the total current

should be a multiple of the unit-reference current). The mean current for version 1 is

1.23 µA with 3.07% standard deviation. The mean current for version 2 is 0.92 µA with

3.66% standard deviation. Referring to Table 9.6 and excluding oscillator current, simulated

current consumptions for version 1 and version 2 are 1.195 µA and 745 nA, respectively.

Measured mean current consumption for version 1 is in line with expectations from simula-

tion; however, the high measured current consumption compared to simulation for version

2 is unexpected.

10.4.5 Digital Core Performance

A limited set of test conditions are available from COTS readers because settings may not

be manipulated individually. Instead, preset groups of settings are available which optimize

for speed or read range. For example, Figure 10.9 demonstrates communication using FM0

uplink modulation and Tari = 12.5 µs. Bit-level traces are inset. Figure 10.10 shows Miller

4 uplink modulation and Tari = 25 µs. These traces were captured by an oscilloscope

with RF power delivered through a coax cable in order to reduce RF-induced noise in the

captured waveform.

Figure 10.11 shows power consumption versus supply voltage and clock rate for the

digital core with no communication occuring. Active power increases by approximately

400 nA at 80 Hz tag read rate, and by approximately 500 nA at 400 Hz read rate. This

indicates that the logic design does an effective job at reducing clock rates, especially in the

transmit logic, to the minimum possible. However, the rise in current with clock rate when

communication is idle indicates that additional clock gating could significantly reduce power

consumption below present levels. Ideally, current consumption would not increase with

clock rate but with useful activity (communication) performed by the logic. Nevertheless,

performance is competitive with other designs reported in the literature (see Table 9.7).

Page 178: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

156

Reader Command: QueryRep Tag Response: RN16

Figure 10.9: FM0-modulated communication voltage trace, with the QueryRep commandenlarged to show modulation details.

0 5 10 15-0.2

0

0.2

0.4

0.6

0.8

1

1.2

Time (ms)

Voltage (V)

Reader

Tag

ReadAckQueryRep ReqRN

Figure 10.10: Oscilloscope trace full communication sequence (QueryRep-Ack-ReqRN-Read) to retrieve sensor data using M4-modulated communication.

Page 179: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

157

0

2

4

6

8

10

12

14

0.5 0.6 0.7 0.8 0.9 1 1.1 1.2

Supply Voltage

Po

we

r C

on

su

mp

tio

n (

µW

)

Leakage

1 MHz

2 MHz

3 MHz

4 MHz

5 MHz

Figure 10.11: Measured digital core power consumption versus supply voltage and clockfrequency. The typical operating point is 3 MHz at 0.7 V, which consumes 3 µW

Bit Number

Chip Number

5 10 15 20 25 30 35 40 45

1

2

3

4

5

6

Figure 10.12: Sample IDs read from different tags.

10.4.6 UID Performance

The UID has been extensively characterized in [93] including analysis of the bit distributions.

However, to demonstrate system functionality, sample tag IDs generated by the UID and

read through the COTS reader are shown in Figure 10.12. Due to an interface error between

the digital core and the UID readout circuitry, the ID space was reduced by a factor of two.

However, all chips are uniquely addressable, and this bug is easily fixed.

Page 180: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

158

204.5

205

205.5

206

206.5

207

207.5

208

208.5

209

209.5

1 101 201 301 401 501 601

Sample Number

ADC Code

Figure 10.13: Fully-wirelessly-powered ADC samples reported through a COTS RFIDreader. The DC input shows a low peak noise level of ± 1.5 bits peak to peak.

10.4.7 ADC Performance

Figure 10.13 shows an ADC time series captured through a COTS RFID reader using a

DC input. This test was performed on the moth deployment board, which is described in

Chapter 11. The board was approximately 1 m from the reader and fully wirelessly powered.

This demonstrates the low system noise floor of ± 1.5 bits peak to peak.

10.4.8 Conclusions

Full system functionality was achieved in a single tapeout through the use of several impor-

tant design verification techniques. For example, testing of the digital core with a COTS

RFID reader through synthesis and implementation on an FPGA proved crucial to design

success. In particular, the author’s experience with WISP firmware design indicated that ro-

bustness against communication errors is near impossible to achieve without bench testing.

In fact, the WISP analog front end was used to interface the COTS reader to the FPGA.

Other non-obvious verification that helped ensure success included mixed-signal cosimula-

tion of the analog components (UID, ADC) and digital core as well as transient analysis of

Page 181: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

159

the startup current draw of the analog components. Specifically, all components must start

up and shut down gracefully without any large current consumption spikes. Components

must also be robust to both slow and fast power supply ramp ups. One are of verification

that was overlooked was the power consumption of the digital core. Industry standard tools

can provide insight into the power consumption of the digital core and thus allow estimation

and optimization of the power consumption during the design stage.

Page 182: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

160

Chapter 11

IN-FLIGHT INSECT TEMPERATURE RECORDING

11.1 Introduction

This chapter presents the deployment of socWISP for muscle temperature recording of a

Manduca Sexta moth. We are studying the biomechanics and temperature dependence of

Manduca flight muscle with Professor Tom Daniels and Nicole George at the University of

Washington. This work enables untethered, in-flight recording with a minimum of weight

and unlimited sensor lifespan.

Temperature is sensed via an implanted thermocouple and wirelessly reported by the

socWISP, described in Chapter 9. A minimum number of off-chip components are used for

thermocouple signal amplification before digitization by the on-chip ADC and transmission

to the RFID reader. The entire system is wirelessly-powered by a COTS UHF RFID

reader, consumes less than 1 cm2 in PCB area, and weighs only 0.35 g including antenna,

thermocouple and assembled PCB.

The design of this system is discussed in Section 11.2. Experimental setup and methods

are given in Section 11.3. Results are presented in Section 11.4.1 and Section 11.4.2. Finally,

Section 11.6 concludes.

11.2 Design

The socWISP provides much of the circuitry necessary for the design. However, the ther-

mocouple signal is too small for direct digitization by the ADC. A DC-coupled differential

amplifier with adjustable virtual ground provides 1000x gain. The schematic for the ampli-

fier is shown in Figure 11.1. Two micropower precision opamps (OPA369) are used in the

design. The first is used to create a low impedance virtual ground via unity-gain feedback

and a voltage-divider reference. The second is used to amplify the thermocouple signal via

non-inverting resistive feedback. The non-inverting gain configuration and CMOS input

Page 183: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

161

Figure 11.1: Off-chip thermocouple amplifier circuit schematic.

stage provide high input impedance for the thermocouple, which cannot drive a load.

The CMOS input stage of the OPA369 has a maximum 0.75 mV input offset voltage,

which is small for micropower opamps. However, this maximum offset is on the scale of the

input signal (2 mV swing for zero to 40C). Input offset has a random distribution as it

comes from transistor mismatch; most designs will have tolerable offsets. Those that are

not tolerable can be adjusted by trimming the reference voltage for the virtual ground. A

fully-integrated design could easily adjust the reference voltage wirelessly; CMOS switches

can short out resistor segments to trim the resistor values. The Gen2 Write command could

be used to write the appropriate switch states to the tag. Flash, EEPROM or e-fuse could

be used to store the value once programmed, or the application could remember the settings

for each tag and program them appropriately.

The PCB is shown in Figure 11.2. Unlike the test board used in Chapter 10, the PCB

weight and size must be minimal in order to not affect the moth’s flight. To minimize

size, small 0402 surface mount components are used and routing is performed on internal

Page 184: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

162

(a) socWISP photo (front). (b) socWISP photo (back).

layers of a 4-layer board. Blind vias are not used because they increase the manufacturing

cost and do not significantly affect the board size. The PCB weight is also reduced by

minimizing board thickness to 25 mils. Further reduction may compromise board stiffness,

which would in turn endanger the COB bonding during deployment. The IC and bondwires

are encapsulated with epoxy to protect against physical damage (not shown in Figure 11.2

to illustrate COB features). The board, components and IC weigh 0.25 g, and the board

measures 0.91 cm by 0.73 cm. The board size does not provide sufficient ground plane

area for use of a monopole antenna; a dipole antenna made of 30 AWG wrapping wire

is used instead. The type-T thermocouple is made from copper and constantan, and the

thermocouple junction is formed by twisting the bare wire together. The fully-assembled

board, including thermocouple and antenna weighs 0.35 g.

11.3 Experimental Setup

The thermocouple’s linearity was verified using the on-chip chopper amplifier. A two-

point calibration using a mineral oil bath was performed to scale the measurements. It is

important to take the mineral oil bath measurements quickly and keep the board away from

the hot air which rises from the oil bath. The hot air warms the board and thus affects

the cold-junction temperature. This rise in cold-junction temperature causes a small but

noticeable error in the measurement.

Page 185: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

163

Attaching the board to the moth, without permanently affecting the board, is a non-

trivial exercise. The moth’s scales render tape useless, but any form of glue is difficult to

remove from the board. Instead, electrical tape was cut to the size of the board and affixed

to the board. Then, the electrical tape (with board attached) was glued to the moth. This

allows removal of the board after testing.

The thermocouple is implanted in the moth’s wing muscle tissue. Preparations are well

documented in the literature [123]. First, a small hole is punctured in the exoskeleton above

the dorsal longitudinal muscle using a fine needle. The hole should be placed to one side

of the midline to avoid damaging the heart. This does not disturb the moth and may be

performed without sedation. Then, the thermocouple is inserted into the hole to 3 mm to

4 mm depth. Optionally, the thermocouple may be glued in place to avoid displacement

during moth flight.

11.4 Results

Two in vivo experiments were performed using a Manduca Sexta moth. In the first test,

the moth was tethered to a metal rod. In the second test, the moth was free to fly in a

plexiglass box of approximately 1 m per side.

11.4.1 Tethered Test

Figure 11.2 shows the setup for the tethered experiment. The exoskeleton hole is visible to

the right of the midline; the thermocouple is not yet implanted. The reader antenna (not

shown) is placed 2 m to 3 m away from the experiment. A COTS Impinj Speedway reader

is used with custom the application software (described in Chapter 8).

Time series data is shown in Figure 11.3. At 15 minutes, the moth begins flying. By 20

minutes, the moth has reached steady-state temperature. At 25 minutes, the moth stops

flying and gradually cools off.

Page 186: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

164

Figure 11.2: Photograph of the socWISP attached to a moth, with thermocouple implantedin muscle tissue.

Figure 11.3: Temperature time series data from the socWISP on a tethered moth.

Page 187: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

165

Figure 11.4: Photograph of an untethered, instrumented moth hovering on a flower.

11.4.2 Untethered Test

The value of this work is enabling untethered, in-flight recording. Figure 11.4 shows an

untethered moth wearing the socWISP measurement system. The moth is placed in a 1-m

per side plexiglass enclosure with various flowers for stimulation. One reader antenna was

used in this test, which was nearly sufficient for good coverage of the enclosure. Future tests

will use two antennas to avoid the null due to dipole antenna directionality.

Time series data from a test is shown in Figure 11.5. The temperature begins warming

at 70 minutes when the moth begins flying. From 83 minutes to 87 minutes, the moth is

resting and the temperature cools off. The moth then flies for 4 minutes before resting

again. These trends are annotated in the figure. Two segments of several minutes duration

are lost (“No Data”) due to the moth being out of the antenna’s field of view.

11.5 Thermocouple Characterization Using a Low Power On-Chip Amplifier

Figure 11.6(a) shows thermocouple characterization time series using a mineral oil bath,

hotplate, and ground truth sensor. Measurements were performed using a chopper amplifier

designed by Fan Zhang at the University of Washington. Figure 11.6(b) shows a linear fit,

and Figure 11.6(c) shows the residual error of the linear fit.

Page 188: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

166

Figure 11.5: Temperature time series data from the socWISP on an untethered moth.

11.6 Conclusion

These experiments demonstrate the feasibility of in vivo, untethered, in-flight temperature

recording of insects. Future work on antenna size reduction will enable deployment on

smaller insects. Experiments measuring neural or muscle signals are feasible as described

in Chapter 6 provided a FIFO can by successfully synthesized on-chip or if a custom RFID

reader with less overhead (however, still Gen2 compliant) is employed. Specifically, the

reader must not resingulate tags for each read operation; the tag handle can be used to

read the ADC value repeatedly. Using this method and the 640 kbps uplink rate, neural

and muscle signals could be easily sampled for a small insect population. If there is 100%

overhead due to downlink communications and uplink packets are 48 bits in length (16-bit

handle + 16-bit word + 16-bit CRC), 1 insect could be sampled at 6.6 kHz or 2 insects

could be sampled at 3.3 kHz. If ADC samples are queued in a FIFO, overhead can be

significantly reduced. For example, reading four 8-bit samples require only 64 bits versus

48 bits for one sample in the un-queued mode. Furthermore, downlink overhead does not

increase with uplink packet size. This would triple or quadruple the number of insects for a

given sample rate. Still, it is desirable to instrument much larger population sizes. For these

applications, the tags would need to compress and/or filter the data to reduce transmission

Page 189: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

167

(a) Thermocouple characterization time series.

(b) Thermocouple linear fit.

(c) Thermocouple linear fit residual.

Page 190: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

168

of redundant data. The WISP is an ideal platform to prototype these protocol and tag

behavior changes due to its software programmability; testing with silicon designs require

several months design iteration time.

Page 191: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

169

Chapter 12

CONCLUSIONS

For many wirelessly-deployed sensors, periodic battery replacement is costly at best and

infeasible at worst. Two key features of RFID make it attractive for wireless sensor deploy-

ment: power is wirelessly and deliberately supplied to the tag, and backscatter modulation

allows for near-zero-power communication by the tag. These features do not preclude the

use of (a) additional power sources such as thermal, solar, and mechanical power harvesting,

(b) application-specific timescale energy storage ranging from a capacitor (milliseconds to

hours) to a rechargeable battery (days to years), and (c) additional long range radios to

augment the communication capabilities. Applications that require sensing while the tag is

out of range of the reader will require at least some non-conventional features in order to

operate. However, the core features of RFID will continue to enable tag lifespans which are

unachievable for conventional radio-based sensor nodes.

As demonstrated throughout this thesis, RFID is an exciting technology for both ultra-

low-power communication as well as wireless power delivery. Coupling this technology

with recent advances in brain-machine interfaces and biomedical research has enabled sig-

nificant progress towards practical application and deployment. For example, Chapter 6

discussed the design of a wirelessly-powered neural spike recorder. As neuroscientists better

learn to interface with neural signals, exciting advances in prosthetic control can leverage

this technology to make the neural interface both seamless and unobtrusive. Additionally,

Chapter 9 presented a fully integrated tag for more general biomedical sensing. Results of

in vivo deployment on a live, untethered moth are shown in Chapter 11. The key novelty is

that long-term wireless recording was previously unattainable due to the size, weight, and

lifespan of conventional sensors.

Collaboration has also produced exciting results in other, relatively unrelated fields.

For example, next generation aerospace sensors are discussed in Chapter 5, and wirelessly

Page 192: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

170

charged tags for shipment tracking are presented in Chapter 7. Lastly, Chapter 4 discusses

the implementation of strong cryptography on a tag. As RFID is employed in passports

and credit cards, this research is imperative for consumer privacy and security.

It is clear that while a wide variety of applications are explored in this thesis, there are

a great deal of topics yet to be explored. As discussed in Chapter 3, the configurability

and programmability of the WISP platform make it a valuable research tool across many

disciplines. Through a generous donation by Intel Corporation, over 150 WISPs have been

donated to 23 academic research groups. It will be exciting to see what innovative research

can be enabled through this technology.

Page 193: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

171

BIBLIOGRAPHY

[1] Nikola Tesla. The Tesla Papers: Nikola Tesla on Free Energy & Wireless Transmissionof Power. Adventures Unlimited Press, 2000.

[2] WC Brown. A Survey of the Elements of Power Transmission by Microwave Beam.In IRE International Convention Record, pages 93–106, 1961.

[3] P.E. Glaser. Power from the sun: Its future. Science, 162(3856):857–861, 1968.

[4] J.C. Schuder. Powering an artificial heart: Birth of the inductively coupled-radiofrequency system in 1960. Artif Organs, 26(11):909–915, 2002.

[5] B.P. Wells. Series resonant inductive charging circuit, 6 2005. US Patent 6,972,543.

[6] S. Salimian, C.M. Heller, L. Li, et al. Dual-frequency capacitively-coupled plasmareactor for materials processing, 12 1997. US Patent 5,656,123.

[7] Gordon E. Moore. Cramming more components onto integrated circuits. Electronics,38(8), April 1965.

[8] C. Patel. Lecture 11: Scaling. Technical report, UMBC, October 2005.

[9] Joshua R. Smith and Daniel J. Yeager. Range scaling of far field wireless powertransfer. Technical report, Intel Labs Seattle, 2008.

[10] H. Nakamoto, D. Yamazaki, T. Yamamoto, H. Kurata, S. Yamada, K. Mukaida,T. Ninomiya, T. Ohkawa, S. Masui, and K. Gotoh. A passive UHF RFID tag LSIwith 36.6% efficiency CMOS-only rectifier and current-mode demodulator in 0.35 µmFeRAM technology. In Solid-State Circuits Conference, 2006. ISSCC 2006. Digest ofTechnical Papers. IEEE International, pages 1201–1210, Feb. 2006.

[11] Triet T. Le. Efficient power conversion interface circuits for energy harvesting appli-cations. PhD thesis, Oregon State University, 2008.

[12] E. Carlson, K. Strunz, and B. Otis. 20 mV input boost converter for thermoelectricenergy harvesting. In 22nd International Converence on VLSI Design, 2009.

[13] R. Weinstein. RFID: a technical overview and its application to the enterprise. ITProfessional, 7(3):27–33, May-June 2005.

Page 194: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

172

[14] D.C. Ranasinghe, K.S. Leong, M.L. Ng, D.W. Engels, and P.H. Cole. A distributedarchitecture for a ubiquitous RFID sensing network. In Intelligent Sensors, SensorNetworks and Information Processing Conference, 2005. Proceedings of the 2005 In-ternational Conference on, pages 7–12, Dec. 2005.

[15] M. Philipose, K.P. Fishkin, M. Perkowitz, D.J. Patterson, D. Fox, H. Kautz, andD. Hahnel. Inferring activities from interactions with objects. Pervasive Computing,IEEE, 3(4):50–57, Oct.-Dec. 2004.

[16] M. Philipose, J.R. Smith, B. Jiang, A. Mamishev, Sumit Roy, and K. Sundara-Rajan.Battery-free wireless identification and sensing. Pervasive Computing, IEEE, 4(1):37–45, Jan.-March 2005.

[17] S. Johan, Xuezhi Zeng, T. Unander, A. Koptyug, and H.-E. Nilsson. Remote moisturesensing utilizing ordinary RFID tags. In Sensors, 2007 IEEE, pages 308–311, Oct.2007.

[18] D. G. Watters, P. Jayaweera, A. J. Bahr, and D. L. Huestis. Design and performanceof wireless sensors for structural health monitoring. In D. O. Thompson and D. E.Chimenti, editors, Quantitative Nondestructive Evaluation, volume 615 of AmericanInstitute of Physics Conference Series, pages 969–976, may 2002.

[19] R. Want. Enabling ubiquitous sensing with RFID. Computer, 37(4):84–86, April 2004.

[20] Michael Buettner and David Wetherall. An empirical study of UHF RFID perfor-mance. In MobiCom ’08: Proceedings of the 14th ACM international conference onMobile computing and networking, pages 223–234, New York, NY, USA, 2008. ACM.

[21] Alanson P. Sample, Daniel J. Yeager, Pauline S. Powledge, Alexander V. Mamishev,and Joshua R. Smith. Design of an RFID-based battery-free programmable sensingplatform. In IEEE Transactions on Instrumentation and Measurement, 2008.

[22] Hee-Jin Chae, Daniel J. Yeager, Joshua R. Smith, and Kevin Fu. Maximalist cryptog-raphy and computation on the WISP UHF RFID tag. In Proceedings of the Conferenceon RFID Security, July 2007.

[23] D.J. Yeager, G.R. Quetin, S.H. Kim, J.Duncan, M. Miller, J.R. Smith, and P. Feraboli.Development of a wirelessly-powered strain gage for aerospace applications. Technicalreport, Intel Research Seattle, 2007.

[24] Jeremy Holleman, Daniel J. Yeager, Richa Prasad, Joshua R. Smith, and Brian Otis.NeuralWISP: An energy-harvesting wireless neural interface with 1-m range. In Pro-ceedings of Biological Circuits and Systems Conference (BioCAS), November 2008.

Page 195: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

173

[25] Daniel J. Yeager, Pauline S. Powledge, Richa Prasad, David Wetherall, and Joshua R.Smith. Wirelessly-charged UHF tags for sensor data collection. In RFID, IEEEInternational Conference on, pages 320–327, April 2008.

[26] Daniel J Yeager, Fan Zhang, Azin Zarrasvond, and Brian Otis. A 9.2µA Gen 2compatible UHF RFID sensing tag with -12dBm sensitivity and 1.25µVrms input-referred noise floor. In IEEE International Solid-State Circuits Conference (ISSCC)Technical Digest, Feb. 2010.

[27] EPCglobal. Class 1 Generation 2 UHF Air Interface Protocol Standard, 2009.

[28] EPCglobal. EPC Tag Data Standard (TDS), 2009.

[29] EPCglobal. Low Level Reader Protocol (LLRP), 2007.

[30] R. Glidden and J. Schroeter. Bringing long-range UHF RFID tags into mainstreamsupply chain applications. RF DESIGN, 28(7):24, 2005.

[31] You-Chang Ko, S. Roy, J.R. Smith, Hyong-Woo Lee, and Choong-Ho Cho. RFIDMAC performance evaluation based on ISO/IEC 18000-6 Type C. CommunicationsLetters, IEEE, 12(6):426–428, June 2008.

[32] Joshua R. Smith, Bing Jiang, Sumit Roy, Matthai Philipose, Kishore Sundara-Rajan,and Alexander Mamishev. ID modulation: Embedding sensor data in an RFID time-series. In Information Hiding, pages 234–246, 2005.

[33] Joshua R. Smith, Alanson P. Sample, Pauline S. Powledge, Sumit Roy, and AlexanderMamishev. A wirelessly-powered platform for sensing and computation. In Ubicomp,pages 495–506, 2006.

[34] Daniel J. Yeager, Alanson P. Sample, and Joshua R. Smith. WISP: A passively pow-ered UHF RFID tag with sensing and computation. RFID Handbook: Applications,Technology, Security, and Privacy, 2008.

[35] Benjamin Ransford, Shane Clark, Mastooreh Salajegheh, and Kevin Fu. Gettingthings done on computational RFIDs with energy-aware checkpointing and voltage-aware scheduling. In Proceedings of USENIX Workshop on Power Aware Computingand Systems (HotPower), December 2008. to appear.

[36] Richa Prasad. Energy debugging for RFID sensor networks. UW CSE Masters Project,2009.

Page 196: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

174

[37] Alexei Czeskis, Karl Koscher, Joshua R. Smith, and Tadayoshi Kohno. RFIDs andsecret handshakes: defending against ghost-and-leech attacks and unauthorized readswith context-aware communications. In ACM Conference on Computer and Commu-nications Security, pages 479–490, 2008.

[38] Ari Juels. Minimalist cryptography for low-cost RFID tags. In C. Blundo andS. Cimato, editors, The Fourth International Conference on Security in Communi-cation Networks – SCN 2004, volume 3352 of LNCS, pages 149–164. Springer-Verlag,2004.

[39] Benoit Calmels, Se’bastien Canard, Marc Girault, and Herve’ Sibert. Low-cost cryp-tography for privacy in RFID systems. In Josep Domingo-Ferrer, Joachim Posegga,and Daniel Schreckling, editors, International Conference on Smart Card Researchand Advanced Applications – CARDIS, LNCS, Tarragona, Spain, April 2006. IFIP,Springer-Verlag.

[40] Yang Cui, Kazukuni Kobara, Kanta Matsuura, and Hideki Imai. Lightweight asym-metric privacy-preserving authentication protocols secure against active attack. In In-ternational Workshop on Pervasive Computing and Communication Security – PerSec2007, pages 223–228, New York, USA, March 2007. IEEE.

[41] Pedro Peris-Lopez, Julio Cesar Hernandez-Castro, Juan Estevez-Tapiador, and ArturoRibagorda. LMAP: A real lightweight mutual authentication protocol for low-costRFID tags. In Workshop on RFID Security 2006(RFIDSec 06), Graz, Austria, July2006. Ecrypt.

[42] Pedro Peris-Lopez, Julio Cesar Hernandez-Castro, Juan Estevez-Tapiador, and ArturoRibagorda. M2AP: A minimalist mutual-authentication protocol for low-cost RFIDtags. In International Conference on Ubiquitous Intelligence and Computing – UIC06,volume 4159 of LNCS, pages 912–923. Springer-Verlag, September 2006.

[43] Pedro Peris-Lopez, Julio Cesar Hernandez-Castro, Juan M. Estevez-Tapiador, andArturo Ribagorda. EMAP: An efficient mutual authentication protocol for low-costRFID tags. In OTM Federated Conferences and Workshop: IS Workshop – IS’06,volume 4277 of LNCS, pages 352–361. Springer-Verlag, November 2006.

[44] Selwyn Piramuthu. HB and related lightweight authentication protocols for secureRFID tag/reader authentication. In Collaborative Electronic Commerce Technologyand Research – CollECTeR 2006, Basel, Switzerland, June 2006.

[45] Axel Poschmann, Gregor Leander, Kai Schramm, and Christof Paar. DESL: Anefficient block cipher for lightweight cryptosystems. In Workshop on RFID Security2006(RFIDSec 06), Graz, Austria, July 2006. Ecrypt.

Page 197: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

175

[46] Benessa Defend, Kevin Fu, and Ari Juels. Cryptanalysis of two lightweight RFIDauthentication schemes. In International Workshop on Pervasive Computing andCommunication Security – PerSec 2007, pages 211–216, New York, USA, March 2007.IEEE.

[47] Daesung Kwon, Daewan Han, Jooyoung Lee, and Yongjin Yeom. Vulnerability of anRFID authentication protocol proposed at SecUbiq 2005. In International Workshopon Security in Ubiquitous Computing Systems – Secubiq 2006, LNCS, Seoul, Korea,August 2006. Springer-Verlag.

[48] Tieyan Li and Guilin Wang. Security analysis of two ultra-lightweight RFID authen-tication protocols. In IFIP SEC 2007, Sandton, Gauteng, South Africa, May 2007.IFIP.

[49] Ron Rivest. The RC5 encryption algorithm. In Fast Software Encryption, pages86–96, 1994. http://theory.lcs.mit.edu/ rivest/Rivest-rc5rev.pdf.

[50] Joshua R. Smith, Alason P. Sample, Pauline S. Powledge, Sumit Roy, and Alexan-der Mamishev. A wirelessly-powered platform for sensing and computation. In 8thInternational Conference on Ubiquitous Computing (Ubicomp 2006), pages 495–506,Orange Country, CA, USA, September 2006.

[51] Adrian Perrig, Robert Szewczyk, Victor Wen, David E. Culler, and J. D. Tygar.SPINS: security protocols for sensor networks. In Mobile Computing and Networking,pages 189–199, 2001.

[52] Chris Karlof, Naveen Sastry, and David Wagner. TinySec: A link layer securityarchitecture for wireless sensor networks. In Second ACM Conference on EmbeddedNetworked Sensor Systems (SenSys 2004), pages 162–175, November 2004.

[53] Damith C. Ranasinghe, Daihyun Lim, Peter H Cole, and Srinivas Devadas. Whitepaper: A low cost solution to authentication in passive RFID systems. TechnicalReport WP-HARDWARE-029, Auto-ID Labs, The University of Adelaide, Adelaide,Australia, September 2006.

[54] ThomasNet. Short range UHF EPC tag is designed for item-level tagging. http://news.thomasnet.com/fullstory/482149, May 2007. Last viewed May 12, 2007.

[55] Alanson P. Sample, Daniel J. Yeager, Pauline S. Powledge, and Joshua R. Smith.Design of a passively-powered, programmable platform for UHF RFID systems. InIEEE International Conference on RFID 2007, March 2007.

[56] Manfred Aigner. DemoTag, 2006. Last viewed May 16, 2007.

Page 198: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

176

[57] Crossbow. MICA2 data sheet. http://www.xbow.com, 2009.

[58] RFID Spotlight. 5 things must know for RFID starters. http://www.innovez-one.com/blogs/, August 2006. Blog entry, Last viewed May 15, 2007.

[59] Jonathan Westhues. Proxmark3. http://cq.cx/proxmark3.pl, 2007. Last viewedMay 16, 2007.

[60] Melanie Rieback, Georgi Gaydadjiev, Bruno Crispo, Rutger Hofman, and An-drew Tanenbaum. A platform for RFID security and privacy administration. InUSENIX/SAGE Large Installation System Administration conference – LISA’06,Washington DC, USA, December 2006.

[61] EM Microelectronic. EM 4035 datasheet. http://www.emmicroelectronic.com/webfiles/product/RFID/ds/EM4035_FS.pdf, 2006.

[62] Pasin Israsena. Securing ubiquitous and low-cost RFID using tiny encryption al-gorithm. In International Symposium on Wireless Pervasive Computing, Phuket,Thailand, January 2006. IEEE.

[63] Manfred Aigner and Martin Feldhofer. Secure symmetric authentication for RFIDtags. In Telecommunication and Mobile Computing – TCMC 2005, Graz, Austria,March 2005.

[64] Mark Buccini. New MSP430 family from Texas Instruments runs at 12 RISC MIPSand standby of 1uA. http://www.microcotroller.com/news/ti_msp4302xx.asp,October 2004. Last viewed May 14, 2007.

[65] Sara Black. Composite rib structure for Airbus A380 vertical tail. High PerformanceComposites, 2004.

[66] Jeff Hawk. The Boeing 787 Dreamliner: More Than an Airplane. Presentation toAIAA/AAAF aircraft noise and emissions reduction symposium, American Instituteof Aeronautics and Astronautics, 2005.

[67] Eric Greene. Marine Composites. Davis Consulting Group, second edition, 1997.

[68] Steven J. Hooper, editor. Composite Materials: Testing & Design (ASTM SpecialTechnical Publication). ASTM International, 1997.

[69] W. H. Craig and T. H. Courtney. On the tension test as a means of characterizingfibre composite failure mode. Journal of Materials Science, 10:1119–1126, 1975.

Page 199: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

177

[70] R. P. Michelson, M. M. Merzenich, C. R. Pettit, and R. A. Schindler. A cochlearprosthesis: Further clinical observations; preliminary results of physiological studies.Laryngoscope, 83(7):1116–1122, 1973.

[71] A. Jackson, J. Mavoori, and E. E. Fetz. Long-term motor cortex plasticity inducedby an electronic neural implant. Nature, 444:56–60, 2006.

[72] D. M. Taylor, S. I. Helms Tillery, and A. B. Schwartz. Direct cortical control of 3Dneuroprosthetic devices. Science, 296:1829–1832, 2002.

[73] R. R. Harrison, P. T. Watkins, R. J. Kier, R. O. Lovejoy, D. J. Black, B. Greger,and F. Solzbacher. A low-power integrated circuit for a wireless 100-electrode neuralrecording system. Solid-State Circuits, IEEE Journal of, 42(1):123–133, 2007.

[74] R. H. Olsson and K. D. Wise. A three-dimensional neural recording microsystemwith implantable data compression circuitry. Solid-State Circuits, IEEE Journal of,40(12):2796–2804, 2005.

[75] Daniel Halperin, Thomas S. Heydt-Benjamin, Benjamin Ransford, Shane S. Clark,Benessa Defend, Will Morgan, Kevin Fu, Tadayoshi Kohno, and William H. Maisel.Pacemakers and implantable cardiac defibrillators: Software radio attacks and zero-power defenses. In Proceedings of the 29th Annual IEEE Symposium on Security andPrivacy, May 2008.

[76] J. Holleman and B. Otis. A sub-microwatt low-noise amplifier for neural recording.Engineering in Medicine and Biology Society, 2007. EMBS 2007. 29th Annual Inter-national Conference of the IEEE, pages 3930–3933, 2007.

[77] R. R. Harrison and C. Charles. A low-power low-noise CMOS amplifier for neuralrecording applications. Solid-State Circuits, IEEE Journal of, 38(6):958–965, 2003.

[78] T. Horiuchi, T. Swindell, D. Sander, and P. Abshier. A low-power CMOS neuralamplifier with amplitude measurements for spike sorting. In Circuits and Systems,2004. ISCAS’04. Proceedings of the 2004 International Symposium on, volume 4, 2004.

[79] P. Mohseni and K. Najafi. A fully integrated neural recording amplifier with DC inputstabilization. IEEE Transactions on Biomedical Engineering, 51(5):832–837, 2004.

[80] R. J. Vogelstein et al. Spike sorting with support vector machines. Engineeringin Medicine and Biology Society, 2004. EMBC 2004. Conference Proceedings. 26thAnnual International Conference of the, 1:546–549, 2004.

[81] I. Obeid and P.D. Wolf. Evaluation of spike-detection algorithms for a brain-machineinterface application. Biomedical Engineering, IEEE Transactions on, 51(6):905–911,2004.

Page 200: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

178

[82] J. Holleman and B. Otis. A sub-microwatt low-noise amplifier for neural recording.Engineering in Medicine and Biology Society, 2007. EMBS 2007. 29th Annual Inter-national Conference of the IEEE, pages 3930–3933, 2007.

[83] S. Rai, J. Holleman, J.N. Pandey, F. Zhang, and B. Otis. A 500 µW neural tag with2µvrms AFE and frequency multiplying MICS/ISM FSK transmitter. In Solid-StateCircuits Conference, 2009. Digest of Technical Papers. ISSCC. 2009 IEEE, 2009.

[84] T Denison, K Consoer, A Kelly, A Hachenburg, and W Santa. A 2.2 µW 94 nV/√Hzchopper-stabilized instrumentation amplifier for EEG detection in chronic implants.In International Solid-State Circuits Conference Digest of Technical Papers, 2007.

[85] H. Wu and Y.P. Xu. A 1V 2.3 µW biomedical signal acquisition IC. Solid-StateCircuits, 2006 IEEE International Conference Digest of Technical Papers, pages 119–128, 2006.

[86] Woradorn Wattanapanitch, Michale Fee, and Rahul Sarpeshkar. An energy-efficientmicropower neural recording amplifier. IEEE Transactions on Biomedical Circuitsand Systems, 1(2):136–147, June 2007.

[87] D. C. Daly, P. P. Mercier, M. Bhardwaj, A. L. Stone, J. Voldman, R. B. Levine, J. G.Hildebrand, and A. P. Chandrakasan. A pulsed UWB receiver SoC for insect motioncontrol. In IEEE International Solid-State Circuits Conference (ISSCC) TechnicalDigest, pages 200–201, Feb 2009.

[88] O.Y. Dergacheva, I.E. Khachikova, and A.A. Burikov. Dynamics of neuron spike activ-ity in the oral nucleus of the pons during the sleep-waking cycle in cats. Neuroscienceand Behavioral Physiology, 34(5):485–489, 2004.

[89] X. Jia, M.A. Koenig, R. Nickl, G. Zhen, N.V. Thakor, and R.G. Geocadin. Earlyelectrophysiologic markers predict functional outcome associated with temperaturemanipulation after cardiac arrest in rats. Critical Care Medicine, 36(6):1909, 2008.

[90] M. Beckner, M. Simms, and R. Venkatesh. Pro RFID in BizTalk Server 2009. ApressBerkely, CA, USA, 2009.

[91] J. Champion. Zedgraph, 2007.

[92] U. Karthaus and M. Fischer. Fully integrated passive UHF RFID transponder ICwith 16.7-W minimum RF input power. Solid-State Circuits, IEEE Journal of,38(10):1602–1608, Oct. 2003.

[93] Ying Su, J. Holleman, and B.P. Otis. A digital 1.6 pJ/bit chip identification circuitusing process variations. Solid-State Circuits, IEEE Journal of, 43(1):69–77, Jan.2008.

Page 201: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

179

[94] Shiho Kim, Jung-Hyun Cho, Hyun-Sik Kim, Haksoo Kim, Hee-Bok Kang, and Suk-Kyung Hong. An EPC Gen 2 compatible passive/semi-active UHF RFID transponderwith embedded FeRAM and temperature sensor. In Solid-State Circuits Conference,2007. ASSCC ’07. IEEE Asian, pages 135–138, Nov. 2007.

[95] F. Kocer and M.P. Flynn. A new transponder architecture with on-chip ADC for long-range telemetry applications. Solid-State Circuits, IEEE Journal of, 41(5):1142–1148,May 2006.

[96] H. Nakamoto, D. Yamazaki, T. Yamamoto, H. Kurata, S. Yamada, K. Mukaida,T. Ninomiya, T. Ohkawa, S. Masui, and K. Gotoh. A passive UHF RF identificationCMOS tag IC using ferroelectric RAM in 0.35-µm technology. Solid-State Circuits,IEEE Journal of, 42(1):101–110, Jan. 2007.

[97] V. Pillai, H. Heinrich, D. Dieska, P.V. Nikitin, R. Martinez, and K.V.S. Rao. Anultra-low-power long range battery/passive RFID tag for UHF and microwave bandswith a current consumption of 700 nA at 1.5 V. Circuits and Systems I: RegularPapers, IEEE Transactions on, 54(7):1500–1512, July 2007.

[98] R. Barnett, G. Balachandran, S. Lazar, B. Kramer, G. Konnail, S. Rajasekhar, andV. Drobny. A passive UHF RFID transponder for EPC Gen 2 with -14dBm sensitivityin 0.13µm CMOS. In Solid-State Circuits Conference, 2007. ISSCC 2007. Digest ofTechnical Papers. IEEE International, pages 582–623, Feb. 2007.

[99] Hongwei Shen, Lilan Li, and Yumei Zhou. Fully integrated passive UHF RFID tagwith temperature sensor for environment monitoring. In ASIC, 2007. ASICON ’07.7th International Conference on, pages 360–363, Oct. 2007.

[100] Namjun Cho, Seong-Jun Song, Sunyoung Kim, Shiho Kim, and Hoi-Jun Yoo. A5.1-µW UHF RFID tag chip integrated with sensors for wireless environmental mon-itoring. In Solid-State Circuits Conference, 2005. ESSCIRC 2005. Proceedings of the31st European, pages 279–282, Sept. 2005.

[101] Zhou Shenghua and Wu Nanjian. A novel ultra low power temperature sensor forUHF RFID tag chip. In Solid-State Circuits Conference, 2007. ASSCC ’07. IEEEAsian, pages 464–467, Nov. 2007.

[102] Azin Zarrasvand. Ultra-Low Power Circuit Design for Wireless Systems and Sensors.Master’s thesis, University of Washington, 2009.

[103] Alanson Sample and Joshua Smith. Wireless power from ambient RF signals. IntelTechnical Report, 2008.

Page 202: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

180

[104] Joseph Polastre, Robert Szewczyk, and David Culler. Telos: enabling ultra-low powerwireless research. In Fourth International Symposium Information Processing in Sen-sor Networks, Piscataway, NJ, USA, 2005. IEEE Press.

[105] T. Umeda, H. Yoshida, S. Sekine, Y. Fujita, T. Suzuki, and S. Otaka. A 950 MHzrectifier circuit for sensor networks with 10 m-distance. In Solid-State Circuits Con-ference, 2005. Digest of Technical Papers. ISSCC. 2005 IEEE International, pages256–597 Vol. 1, Feb. 2005.

[106] G. De Vita and G. Iannaccone. Design criteria for the RF section of UHF and mi-crowave passive RFID transponders. Microwave Theory and Techniques, IEEE Trans-actions on, 53(9):2978–2990, Sept. 2005.

[107] R. Barnett, S. Lazar, and Jin Liu. Design of multistage rectifiers with low-costimpedance matching for passive RFID tags. In Radio Frequency Integrated Circuits(RFIC) Symposium, 2006 IEEE, pages 4 pp.–, June 2006.

[108] Jun Yi, Wing-Hung Ki, and Chi-Ying Tsui. Analysis and design strategy of UHFmicro-power CMOS rectifiers for micro-sensor and RFID applications. Circuits andSystems I: Regular Papers, IEEE Transactions on, 54(1):153–166, Jan. 2007.

[109] J. Yi, W.-H. Ki, and C.-Y. Tsui. Corrections to analysis and design strategy of UHFmicro-power CMOS rectifiers for micro-sensor and RFID applications [jan 07 153-166].Circuits and Systems I: Regular Papers, IEEE Transactions on, 54(6):1406–1406, June2007.

[110] S. Mandal and R. Sarpeshkar. Low-power CMOS rectifier design for RFID applica-tions. Circuits and Systems I: Regular Papers, IEEE Transactions on, 54(6):1177–1188, June 2007.

[111] Soumyajit Mandal. Far field RF power extraction circuits and systems. Master’sthesis, Massachusetts Institute of Technology, 2004.

[112] T. Le, K. Mayaram, and T. Fiez. Efficient far-field radio frequency energy harvest-ing for passively powered sensor networks. Solid-State Circuits, IEEE Journal of,43(5):1287–1302, May 2008.

[113] R. Glidden, C. Bockorick, S. Cooper, C. Diorio, D. Dressler, V. Gutnik, C. Hagen,D. Hara, T. Hass, T. Humes, J. Hyde, R. Oliver, O. Onen, A. Pesavento, K. Sund-strom, and M. Thomas. Design of ultra-low-cost UHF RFID tags for supply chainapplications. Communications Magazine, IEEE, 42(8):140–151, Aug. 2004.

[114] F. Cilek, K. Seemann, D. Brenk, J. Essel, J. Heidrich, R. Weigel, and G. Holweg. Ultralow power oscillator for UHF RFID transponder. In Frequency Control Symposium,2008 IEEE International, pages 418–421, May 2008.

Page 203: Development and Application of Wirelessly-Powered Sensor …wisp.wikispaces.com/file/view/yeager-thesis-final.pdf/162645953/... · Development and Application of Wirelessly-Powered

181

[115] F. Song, J. Yin, H.L. Liao, and R. Huang. Ultra-low-power clock generation circuitfor EPC standard UHF RFID transponders. Electronics Letters, 44(3):199–201, 312008.

[116] C. Klapf, A. Missoni, W. Pribyl, G. Holweg, and G. Hofer. Analyses and designof low power clock generators for RFID TAGs. In Research in Microelectronics andElectronics, 2008. PRIME 2008. Ph.D., pages 181–184, 22 2008-April 25 2008.

[117] R. Barnett and Jin Liu. A 0.8V 1.52MHz MSVC relaxation oscillator with invertedmirror feedback reference for UHF RFID. In Custom Integrated Circuits Conference,2006. CICC ’06. IEEE, pages 769–772, Sept. 2006.

[118] L. Lincoln, K. Leung, and H.C. Luong. A 7-µW clock generator in 0.18-µm CMOS forpassive UHF RFID EPC G2 tags. In Solid State Circuits Conference, 2007. ESSCIRC2007. 33rd European, pages 412–415, Sept. 2007.

[119] I. Zalbide, J. Vicario, and I. Velez. Power and energy optimization of the digital coreof a Gen2 long range full passive RFID sensor tag. In RFID, 2008 IEEE InternationalConference on, pages 125–133, April 2008.

[120] A. Ricci, M. Grisanti, I. De Munari, and P. Ciampolini. Design of a 2 µW RFIDbaseband processor featuring an AES cryptography primitive. In Electronics, Circuitsand Systems, 2008. ICECS 2008. 15th IEEE International Conference on, pages 376–379, 31 2008-Sept. 3 2008.

[121] F. Cilek, K. Seemann, G. Holweg, and R. Weigel. Impact of the local oscillator onbaseband processing in RFID transponder. In Signals, Systems and Electronics, 2007.ISSSE ’07. International Symposium on, pages 231–234, 30 2007-Aug. 2 2007.

[122] David Johns and Ken Martin. Analog Integrated Circuit Design. Wiley, 1996.

[123] DA Dorsett. Preparation for flight by hawk-moths. Journal of Experimental Biology,39(4):579–588, 1962.