a wireless mri system using millimeter wave …

A WIRELESS MRI SYSTEM USING MILLIMETER WAVE

TRANSMISSION

A DISSERTATION

SUBMITTED TO THE DEPARTMENT OF ELECTRICAL ENGINEERING

AND THE COMMITTEE ON GRADUATE STUDIES

OF STANFORD UNIVERSITY

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

KAMAL AGGARWAL

MARCH 2016

http://creativecommons.org/licenses/by-nc/3.0/us/

This dissertation is online at: http://purl.stanford.edu/mg090pk7640

© 2016 by Kamal Aggarwal. All Rights Reserved.

Re-distributed by Stanford University under license with the author.

This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 United States License.

ii



http://purl.stanford.edu/mg090pk7640

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Ada S Y Poon, Primary Adviser


John Pauly


S Wong

Approved for the Stanford University Committee on Graduate Studies.

Patricia J. Gumport, Vice Provost for Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.

iii

v

Abstract

Conventional MRI relies on a wired connection between a receiver coil array

and an external processing circuitry to generate accurate images. To improve image

quality, the number of receiver coil elements are increased and separate receiver coil

arrays are used for different parts of the body. This while improving image quality

also increases cabling complexity. Furthermore, baluns and radio frequency (RF) traps

are required for each channel, and cables must be routed carefully to minimize coil

interactions. This increases the operation and maintenance costs. Moreover, these

receiver coil arrays are heavy and cumbersome and can be intimidating and ill-fitting

for children. The coil setup time can occupy a significant fraction of the total exam

time. Consequently, removing these cables from the receiver coils will lead to a more

cost effective and time efficient system.

In the past, a number of architectures have been proposed to enable wireless

MRI for minimizing or eliminating the use of cables. All of these past efforts used

microwave frequencies up to 3 GHz, and generic protocols such as 802.11b or MIMO

that are intended for long-range communication over distances of 10 m to 100 m. Such

generic long range communication protocols are sub-optimal solutions for wireless

MRI in terms of power consumption and size. This is because typical MRI bore

diameters vary from 60 cm to 70 cm. And, depending on a patient’s physical attributes

and the part of the body to be imaged, the distance between the coil array and the

magnet bore/edge can vary from 10 cm to 50 cm.

A millimeter (mm) wave radio for wireless MRI data transmission is presented

in this work. High path loss and availability of wide bandwidth make millimeter

(mm)-waves ideal for short range, high data rata communication required for wireless

MRI. The proposed system uses a custom designed integrated chip (IC) mm-wave

radio with 60 GHz as radio frequency carrier. We assess performance in a 1.5T MRI

field, with the addition of optical links between the console room and magnet. The

system uses ON-OFF keying (OOK) modulation for data transmission and supports

data rates from 200 Mb/s to 2.5 Gb/s for distances up-to 65 cm. The presence of

highly directional, linearly polarized, on-chip dipole antennas on the mm-wave radio

vi

along with time division multiplexing (TDM) circuitry allows multiple wireless links

to be created simultaneously with minimal inter-channel interference. This leads to a

highly scalable solution for wireless MRI.

vii

Acknowledgments

Life and the universe have a mysterious way of connecting us with people,

places and circumstances at the right time which benefit us in the long run. I always

thought that I would be a mechanical engineer, but lack of opportunities in my

university of choice led me to opt for electronics at the National Institute of

Technology, Jalandhar. I was fortunate to graduate having several full time job offers

from software companies. Instead, I decided to join Infineon Technology, India as an

intern to get some experience in the field of my graduate coursework in electrical

engineering. This is where I met Kaushik Saiprasad, a friend who would play a pivotal

role in my life.

We were traveling by bus in Bangalore in January of 2006 when Kaushik

suggested that I should pursue my interests and apply for a Ph.D. program abroad.

Even though, I was doubtful and unconvinced about his suggestion, like a true friend

he never gave up on me. Over the course of next year and a half, he was my friend, my

motivator, my guide and my biggest critique. He had more faith in me than I had in

myself and even offered to pay for my application expenses if I was not accepted by

Stanford.

Come September 2007 and I start my MS/Ph.D. program at Stanford. As a

masters’ candidate, I had no funding and pursuing a Ph.D. seemed like a distant dream.

That’s when Natasha Newson from Stanford EE recommended me for a teaching

assistant position to Prof. Boris Murmann. One assignment led to another and over the

course of next 4 years, I had the privilege of working as a teaching assistant with Prof.

Boris Murmann, Prof. Robert Dutton, Prof. Bruce Wooley, Prof. Thomas Lee, and

Prof. Hamid Rategh. I could not have asked for better mentors to introduce me to the

field of teaching and I will always be grateful to them for giving me the opportunity to

work with them.

Then in the summer of 2008, I met Dr. Amir Amirkhany while pursuing

internship program at Rambus. He agreed to secure funding for my Ph.D. research

project from Rambus Inc. As luck would have it, great crash of 2008 happened and by

the end of the summer of 2010, I was in search of another project and an advisor.

viii

I took a course with Prof. Ada Poon in 2009 and was really impressed by her

enthusiastic spirit and supportive nature. She encouraged me to explore different

projects before finalizing my PhD research topic. Without her guidance and persistent

help, this doctoral research would not have been possible. I would always be grateful

to her for showing faith in me.

I would like to extend my special thanks to my PhD oral exam committee

members, Prof. Antony Frazer-Smith (Chair), Prof. Simon Wong (dissertation reader),

Prof. John Pauly (dissertation reader), Dr. Greig Scott and Dr. Shreyas Vasanavala. I

am especially indebted to Prof. Simon Wong and Prof. Joh Pauly for agreeing to be

my dissertation reader and for their continued support. I am grateful to Dr. Shreyas

Vasanavala, Prof. John Pauly and Dr. Grieg Scott for giving me the opportunity to

work on the wireless MRI project. I am thankful to Dr. Greig Scott for spending

countless hours with me in the lab and introducing me to the wonderful world of MRI.

I would like to thank Mazhareddin Taghivand and Yashar Rajavi for their

friendship, assistance, and many months of hard work and sleepless nights. We

worked together and helped each other throughout our research work. Our group effort

enabled us to develop a system which would have been a challenge to accomplish

otherwise. I would like to express my gratitude to my good friends Kiran Raj Joshi,

Lenin Patra and Vipul Chawla for their help in testing the designed system and

valuable feedback.

I would like to express my heartfelt thanks to the wonderful staff at Stanford -

Pauline Prather for her help with wire-bonding, Joe Little, and John Desilva for their

support on weekends and holidays, Amy Duncan for making sure that I was on track

with my graduation program, Rolando Villalobos and Junko Perry from the

international student center for helping me navigate through the complex web of

immigration and our exceptional admins June Wang, Ann Guerra and Douglas

Chaffee for their help in everything, big and small, regardless of the time of day.

I would like to thank my friends in Prof. Poon’s group, Andrew Anatoly,

Bryan, Chris, John, Ming, Sanghoek, Saihua, Stephanie for all those wonderful

conversations and dinners. I am grateful to friends from CIS circuit and device groups

ix

for teaching me about device fabrication and circuit analysis. Sincere thanks to Amir,

Ashwin, Pedram, Drew, Edward, Jayant, Jasmine, Kasra, Mahmoud, Maryam,

Mohammad, Nick, and Ryan.

My graduate years at Stanford have been amazing and wonderful because of

my friends in the bay area who were always there to motivate and encourage me. My

gratitude goes to Abhishek, Adnan, Aneesh, Anirudh, Anoma, Dinesh, Jayesh,

Kalpana, Kanupriya, Khusboo, Lavina, Manu, Mounir, Nagaphani, Neha, Pankaj,

Prashant, Priyanka, Sakshi, Shweta, Siddharth, Suheil, Swadesh, Shivam, Vinod,

Vipul, and Vijay Uncle. I am especially thankful to Ashima, Chaitanya, Gaurav, and

Lenin for being my family away from home. I feel blessed to have Vaibhav Triapathi

and Siddharth Panwar as my friends and mentors all these years. I am grateful to

Siddharth Panwar and Nagaphani Ateukuri for sacrificing their time and reviewing my

thesis.

I would like to express my sincere gratitude to my parents for their constant

support and understanding because of which I was able to achieve my goals. I am the

first engineer in the family and will always appreciate all their sacrifices to ensure my

success. I am forever indebted to my brother Vikas, who dropped out of college so that

I could pursue my studies at Stanford. Special thanks to sister-in law, Mona, for

unfailingly being supportive during this process. I would not be writing this thesis if

not for the love, patience, support and understanding of my dear wife Chandini. She

has always been there as my pillar of strength and given me hope in times of distress. I

am grateful to my in-laws who had confidence in me and agreed to marry their

daughter while I was still a student. And finally, I bow in reverence to the almighty

God for bringing all these people in my life and making me what I am today.

xi

Table of Contents

Abstract ............................................................................................................. v

Acknowledgments ........................................................................................... vii

List of Tables .................................................................................................. xiii

List of Figures ................................................................................................. xv

CHAPTER 1 Introduction ............................................................................... 1

1.1 Magnetic Resonance Imaging .................................................................. 1

1.1.1 Working Principle ........................................................................... 1

1.1.2 System Components ....................................................................... 3

1.1.3 Drawbacks of Existing MRI Systems ............................................. 8

1.1.4 Redesign MRI Receiver Coil .......................................................... 9

1.2 Organization ........................................................................................... 13

CHAPTER 2 Wireless Receiver Coil ............................................................ 15

2.1 Wireless Receiver Coil Architectures .................................................... 15

2.1.1 Single-Element Module ................................................................ 15

2.1.2 Multi Element Module .................................................................. 15

2.2 Wireless Technologies ............................................................................ 17

2.2.1 Wi-Fi 802.11ac ............................................................................. 18

2.2.2 802.11ad ........................................................................................ 20

2.2.3 Custom Millimeter (mm)-Wave Solution ..................................... 22

CHAPTER 3 Millimeter-Wave Wireless Transceiver ................................ 23

3.1 Proposed mm-Wave Transceiver v/s Prior Art ...................................... 23

3.2 Transceiver Architecture ........................................................................ 25

3.3 Transmitter Design ................................................................................. 28

3.3.1 Voltage-Controlled Oscillator (VCO) .......................................... 29

3.3.2 Power Amplifier (PA) ................................................................... 32

3.3.3 Transmit-Receive (TR) Switch ..................................................... 35

3.4 Receiver Design ...................................................................................... 37

3.5 Dipole Antenna ....................................................................................... 41

3.5.1 Method of Images ......................................................................... 47

xii

3.6 Energy Harvesting Circuit design .......................................................... 49

3.7 Measurements ......................................................................................... 51

3.8 Summary ................................................................................................. 57

CHAPTER 4 Design and Evaluation of Wireless MRI System ................. 59

4.1 Background ............................................................................................. 59

4.2 System Design Challenges ..................................................................... 60

4.3 System Design ........................................................................................ 62

4.3.1 Design Overview .......................................................................... 62

4.3.2 60-GHz Radio ............................................................................... 63

4.3.3 The Fiber Optic Link .................................................................... 65

4.3.4 System Link Budget ...................................................................... 65

4.4 System Evaluation .................................................................................. 67

4.4.1 System Measurements inside the MRI Room ............................... 67

4.4.2 System Measurements outside the MRI Room............................. 75

4.4.3 Power Consumption for Different Signaling Schemes ................. 80

4.5 Discussion ............................................................................................... 82

CHAPTER 5 Conclusions .............................................................................. 86

5.1 Conclusions ............................................................................................ 86

5.2 Future Work ............................................................................................ 87

Bibliography .................................................................................................... 89

xiii

List of Tables

Table 1-1: Comparison of different magnet types ............................................................ 4

Table 2-1: 802.11ac data rates for different modulations and spatial streams [8]. ........ 19

Table 2-2: 802.11AD MODULATION AND CODING SUMMARY [13] .................. 21

Table 3-1: Performance Comparison .............................................................................. 58

Table 4-1: BER for Different Distance and Data Rates ................................................. 73

xv

List of Figures

Fig. 1.1: Impact of external magnetic field and radio frequency signal on the millions

of hydrogen proton in the human body during MRI imaging. ............................. 2

Fig. 1.2: MRI scanner cutaway showing the permanent magnet, gradient coils and RF

coils along with the patient location. .................................................................... 3

Fig. 1.3: Different coils inside the MRI scanner. ............................................................. 5

Fig. 1.4: Impact of different gradient coils on static magnetic field. ............................... 6

Fig. 1.5: Different RF coils (a) volume coil, (b) single surface coils, and (c) phased

array surface coils. (Image courtesy: Siemens) .................................................... 7

Fig. 1.6: Separate receiver coils for different body part. (Image courtesy: Siemens) ...... 9

Fig. 1.7: A single four element phased array coils used for imaging (coil image

courtesy: Siemens). ............................................................................................. 10

Fig. 1.8: Multiple four element phased array coils combined together to create a single

image (coil image courtesy: Siemens). ............................................................... 10

Fig. 1.9: Bulky receiver coils cables with RF traps and cable connecter ports on the

MRI scanner (image courtesy: GE). ................................................................... 11

Fig. 1.10: Block diagram of the proposed wireless MRI system ................................... 12

Fig. 1.11: Receiver coil with wireless transmitter. ......................................................... 13

Fig. 2.1: (a) Proposed four-element wireless receiver coil module, (b) Four four-

element modules placed together to create a sixteen element module. (Coil

image courtesy: Siemens) ................................................................................... 16

Fig. 2.2: Channels defined for 5GHz band [8]. .............................................................. 18

Fig. 2.3: 60-GHz band channel plan and frequency allocation by region [14]. ............. 20

Fig. 3.1: (a) Linear relationship between the data rate and transmit power consumption.

(b) Application in point of sale advertisement. (c) Application in medium to

high data rate: neural data transmission of small beings. ................................... 25

Fig. 3.2: Transceiver architecture and corresponding waveforms.................................. 26

Fig. 3.3: Average power consumption of a single transmitter pulse. ............................. 27

xvi

Fig. 3.4: (a) Transmitter ON time waveform for different data rates, and (b)

corresponding power consumption for a RZ-OOK modulation. ........................ 27

Fig. 3.5: (a) Transmitter ON time waveform for different data rates, and (b)

corresponding power consumption for a RZ-PWC-OOK modulation. .............. 28

Fig. 3.6: 2×2 Transceiver RF blocks. ............................................................................. 28

Fig. 3.7: The two transmit VCOs with fast startup. ........................................................ 29

Fig. 3.8: Beat frequency generation due to transmit VCOs' mismatch. ......................... 31

Fig. 3.9: Coherence time vs capacitance mismatch. ....................................................... 32

Fig. 3.10: (a) Standard class F-1 PA. (b) Current and voltage waveforms for a class F-1

PA with only one tank at third harmonic (solid) and ideal (dotted). .................. 33

Fig. 3.11: The implemented class E/F2, odd PA. .............................................................. 34

Fig. 3.12: Drain Voltage (solid) and current (dotted) waveforms for (a) ideal class E/F2,

odd PA, and (b) implemented class E/F2, odd PA. ................................................. 34

Fig. 3.13: (a) TR switch interface with the PA and LNA. (b) TR switch when TX is on

(c) TR switch when RX is on. ............................................................................ 35

Fig. 3.14: Receiver chain (RF and BB). ......................................................................... 39

Fig. 3.15: Simulated NF, gain and return loss of the LNA. ........................................... 40

Fig. 3.16: Total NF of RX chain vs. LNA input power. ................................................. 40

Fig. 3.17: Dual dipole antenna with patterned shield. .................................................... 43

Fig. 3.18: HFSS simulation for antenna to antenna coupling. ........................................ 44

Fig. 3.19: Impact of substrate thickness on antenna gain and normalized power [47]. . 45

Fig. 3.20: Simulated radiation pattern (a) Dual dipole with metal reflector. (b) Dual

dipole without metal reflector. (c) Single dipole with metal reflector. (d) Single

dipole without metal reflector. ........................................................................... 46

Fig. 3.21: Normalized measured radiation patterns (a) Elevation (b) Azimuth. ............ 46

Fig. 3.22: Simulated S11 for the dual dipole antenna. ..................................................... 47

Fig. 3.23: Image of a unit positive charge, and (b) image of a current carrying wire. ... 48

Fig. 3.24: Image of a dual dipole antenna over a ground plane. .................................... 48

Fig. 3.25: Energy harvesting front-end circuit. .............................................................. 49

xvii

Fig. 3.26: Energy harvesting. (a) Supply detection mechanism. (b) Packet mode. (c)

Continuous mode. ............................................................................................... 50

Fig. 3.27: Supply detection. (a) Bandgap reference. (b) Vdd level detection. (c) Extra

high Vdd safety switch. ....................................................................................... 51

Fig. 3.28: Die photo (a) first version for NFC application, (b) the second version for

wireless MRI application. The silicon is 1.8 mm × 0.9 mm for both versions. . 52

Fig. 3.29: (a) Metal reflector facing the front-side. The radiation is through the PCB on

the back-side. (b) The plastic fixture used to hold the metal reflector. The

harvest antenna connector ports into the chip. (c) The silicon radio wire-

bonded on a FR4 PCB material. ......................................................................... 52

Fig. 3.30: Measurement setup (a) Harvesting efficiency. (b) BER. (c) Pulse-width. (d)

Coherent time. .................................................................................................... 54

Fig. 3.31: (a) Harvest efficiency. (b) TX power vs. data rate. (c) Bit-error rate vs. data

rate. ..................................................................................................................... 54

Fig. 3.32: Oscilloscope eye diagram at 2.45 Gb/s at 10 cm (voltage scale: 100 mV/div,

time scale: 100 ps/div). ....................................................................................... 55

Fig. 3.33: Power spectral density of a long PRBS. ........................................................ 56

Fig. 3.34: Beat frequency and coherence time measurements. (Voltage scale: 250

mV/div, time scale: 10 ns/div). ........................................................................... 57

Fig. 4.1: Proposed architecture for the mm-wave wireless MRI system along with the

baseband required for digitization of MRI coil data. ......................................... 59

Fig. 4.2: Image demonstrating the strength of magnetic fields in the MRI room. ......... 62

Fig. 4.3: (a) Metallic holder for the board, and (b) custom designed Lego® holder with

the designed mm-wave transceiver. ................................................................... 62

Fig. 4.4: (a) mm-wave radio architecture, and (b) signal waveforms at different points

inside the TX and RX. ........................................................................................ 64

Fig. 4.5: A block diagram showing the test setup for link verification inside the MRI

room at a distance of 10 cm. ............................................................................... 68

Fig. 4.6: (a) The test setup showing TX and RX alignment, and (b) a magnified view

of PCB mounting showing the 60-GHz TX and the 60-GHz RX chip. ............. 68

xviii

Fig. 4.7: MRI test setup placed (a) in-line with the direction of the static magnetic field,

and (b) perpendicular to the direction of the static magnetic field. .................... 68

Fig. 4.8: Differential 7-bit PRBS sequence as captured on the sampling scope at 500

Mb/s for a distance of 10 cm (voltage scale: 200 mV/div, time scale: 50 ns/div).

............................................................................................................................ 69


room at a distance of 25 cm. ............................................................................... 70

Fig. 4.10: The test setup showing the TX and RX alignment at 25 cm. ......................... 70

Fig. 4.11: (a) Magnified view of RX aligned to the output of the horn antenna. (b) The

test setup for 25 cm link placed inside the MRI bore. ........................................ 71


room at a distance of 50 cm, and 65 cm. ............................................................ 71

Fig. 4.13: The test setup showing the TX and RX alignment at 50 cm and 65 cm. ....... 72

Fig. 4.14: (a) Magnified view of RX aligned to the output of the LNA-horn antenna

assembly. (b) The test setup for 50 cm and 65 cm link placed inside the MRI

bore. .................................................................................................................... 72

Fig. 4.15: (a) The baseband processing unit implemented on the transmitter side, and

(b) the baseband processing unit implemented on the receiver side for image

processing. .......................................................................................................... 74

Fig. 4.16: (a) The MRI image broken down into 9 image blocks before transmitting

through the system. (b) The received image obtained by assembling the

individually transmitted blocks. . ....................................................................... 74

Fig. 4.17: Bit-error rate versus data rate for 10 cm, 25 cm, 50 cm and 65 cm. .............. 76

Fig. 4.18: A block diagram showing the test setup verifying horn antenna’s field of

view. ................................................................................................................... 77

Fig. 4.19: Bit-error rate versus data rate as the transmitter is moved sideways with TX-

RX distance of 50 cm. ........................................................................................ 77

Fig. 4.20: (a) Real time eye diagram measured using the BERTScope at 2 Gb/s, and (b)

the measured BER at 2 Gb/s using PRBS-7. ...................................................... 78

xix

Fig. 4.21: (a) Real time eye diagram measured using the BERTScope at 2.5 Gb/s, and

(b) the measured BER at 2.5 Gb/s using PRBS-7. ............................................. 78

Fig. 4.22: The block diagram showing the test setup for multiple transmitters at a

distance of 50 cm from the receiver. .................................................................. 79

Fig. 4.23: The test setup for multiple TX to demonstrate time division multiplexing

(TDM) at a data rate of 250 Mb/s and distance of 50cm. ................................... 80

Fig. 4.24: Different shaped markers showing the received data corresponding to

different transmitters when the (a) TDM block is turned OFF (voltage scale:

100 mV/div, time scale: 1 ns/div), and (b) when the TDM block is turned ON

(voltage scale: 100 mV/div, time scale: 2 ns/div). ............................................. 80

Fig. 4.25: Transmitter dc power consumption versus data rate for different signaling

schemes. .............................................................................................................. 81

Fig. 4.26: A 32-element receiver coil with 4-coil module sharing a single processing

unit. Multiple 4 coil-modules are placed such that the RF transmitters in

adjacent modules are orthogonal to each other, enabling multiple spatial

streams. ............................................................................................................... 84

Fig. 4.27: (a) 60-GHz radio with on-chip dipole placed inside an MRI safe package,

and (b) its HFSS simulated radiation pattern with maximum gain of 9.1dBi.

The metal acts as a reflector and the dielectric as a lens for enhanced gain. ..... 85

Fig. 5.1: Proposed implementation for the designed wireless MRI system. .................. 86

Chapter 1 - Introduction

1

CHAPTER 1

Introduction

1.1 Magnetic Resonance Imaging

Magnetic Resonance Imaging (MRI) is one of the most accurate medical

diagnostic techniques available today. MRI systems generate very detailed images of

the human body tissues and operate on the principle of Nuclear Magnetic Resonance

(NMR). After its inception in the 1970’s, the name of the system was changed from

NMR to MRI because of the negative connotation associated with the word nuclear.

Since then there has been a rapid growth in the field of MRI. MRI started as a

tomographic technique where it took hours to image one thin slice of human body.

Compared to that, modern MRI machines can image the whole human body in less

than an hour. Unlike CT scans that use ionizing x-rays, MRI systems rely on non-

ionizing radio waves and magnetic fields, hence they are much safer to use. Further,

compared to CT scans, MRI systems not only provide a superior soft tissue contrast

but also provide the flexibility to image the body from any plane. Over the years MRI

has emerged as a very powerful and popular medical diagnostic tool. Today MRI is

used to diagnose pinched nerves in the spinal column, various heart diseases, multiple

sclerosis and other diseases of central nervous system. As a result the number of MRI

machines have grown from 12 in 1980 to around 25,000 in 2012 [1]. At present

around 2,000 MRI imaging units are sold worldwide annually

1.1.1 Working Principle

Seventy percent of human body is made up of water. Each water molecule has

one oxygen atom and two hydrogen atoms. Each hydrogen atom has a positively

charged particle called proton in its nucleus. The proton has a fundamental property of

spin associated with it. Due to this spin, each hydrogen atom develops a finite

magnetic moment, just like a tiny magnet with a north and a south pole. At any point

in time, millions of these protons in a human body are randomly aligned such that the


2

net magnetic moment of the human body is equal to zero. In the presence of an

external magnetic field B0 few of these protons re-align so that a net magnetic moment

M appears along the direction of the external applied field, as shown in Fig. 1.1(b).

During the course of imaging an external field B1 is applied in the direction

perpendicular to the existing field B0. As a result the magnetic moment vector M is

displaced from its initial position as shown in Fig. 1.1(c). When this external field B1

is turned off, the magnetic moment vector M precesses along the initial field B0 before

coming to rest at its initial position. The radio frequency (RF) signal generated during

the precession of M is used to generate the final MRI image. The precession frequency

of the proton is directly proportional to the external magnetic field and is given by

Larmor Equation:

𝜔 = 𝛾 ∙ 𝐵 , (1.1)

where ω is the Larmor frequency in MHz, γ is the gyromagnetic ratio in MHz/T and B

is the strength of the static magnetic field in tesla. For hydrogen H-1 isotope, the

gyromagnetic ratio is 42.58 MHz/T [2].

Fig. 1.1: Impact of external magnetic field and radio frequency signal on the millions of

hydrogen proton in the human body during MRI imaging (image courtesy: mri-q.com).


3

1.1.2 System Components

MRI system comprises of five main components: 1) static magnetic field

generator, 2) gradient coils to generate variable magnetic fields, 3) RF coils to

generate the RF excitation signal and to capture the MRI signals generated during

precession of magnetic moment M, 4) RF receiver to process the signal from RF coils,

and 5) finally an image processing system to convert the captured RF signal into

visible image. The static magnet, gradient coils and the radio frequency transmitter

coils are part of the MRI scanner as shown in Fig. 1.2.

Fig. 1.2: MRI scanner cutaway showing the permanent magnet, gradient coils and RF coils

along with the patient location (image courtesy: losangeles-mri.com).

1.1.2.1 Static Magnetic Field Generator

As discussed in section 1.1.1, a strong static magnetic field is required to align

the hydrogen protons in the human body. This field is called the primary magnetic

field of the magnet. There are three typical methods to generate this field: permanent

magnets, resistive magnets (current passing through a coil) and super-conducting

magnets. Permanents magnets are made of alloys that possess ferromagnetic properties.

Though they are very heavy, they don’t need any power to maintain the field strength

thus resulting in low capital and maintenance cost. Resistive magnets consist a

collection of coils through which a strong electric current is passed. They are much


4

lighter than the permanent magnets but need current to maintain their field strength.

Like resistive magnets, super conducting magnets also have coils through which

current is passed to generate the magnetic fields. However, these coils are made up of

super conducting material that is cooled to absolute zero using liquid helium. This

results in a very strong, stable and homogenous magnetic field around these coils.

To achieve a high resolution image, the magnetic fields generated by these

magnets need to be very strong, homogenous in space and stable in time. Permanent

magnets and resistive magnets are generally restricted to field strengths below 0.4T

and hence cannot be used for high-resolution imaging. Super conducting magnets on

the other hand have produced fields as high 14.1T and hence are primarily used for the

majority of MRI imaging. A brief comparison of different magnet types is given in

Table 1-1 [1].

Table 1-1: Comparison of different magnet types

Magnet

Type

Advantages Disadvantages

Permanent No electricity, refrigeration

required,

Open architecture;

Limited fringe field;

Fields cannot be switched off;

Limited field strength;

Sensitive to temperature

changes;

Limited signal to noise ratio.

Resistive No refrigeration required;

Open architecture;

Fields can be switched off.

High electricity requirements;

Limited field strength;

Limited signal to noise ratio

Super-

Conducting

High signal to noise ratio;

Strong. homogenous and

stable magnetic fields;

High field strength.

High running cost, electricity

requirements;

Need special cooling;

Closed architecture, very noisy

and causes claustrophobia;

Switching off complicated and


5

costly.

1.1.2.2 Gradient coils

Gradient coils are loops of wires of thin conductive sheets and are embedded

inside the cylindrical bore of the MRI scanner as shown in Fig. 1.3. When electric

current is passed through these coils, a secondary magnetic field is generated. The

interaction between the existing primary field and this secondary field creates a linear

magnetic field gradient inside the MRI bore. This causes the precession frequency of

the proton to vary as the function of position inside the MRI bore. Thus these gradient

coils results in a spatial encoding of the MRI signal.

Fig. 1.3: Different coils inside the MRI scanner (image courtesy: howtolearn.com).

An MRI system has x, y and z gradient coils (Fig. 1.3) to produce gradients in

three dimensions and hence an image slice can be created over any place within the

patient’s body. Each coil set is controlled independently and creates a gradient field

whose z- component varies linearly along the x, y and z directions respectively as

shown in Fig. 1.4. The z-gradient coils are usually laid out as circular (Maxwell) coils,

while the x- and y- gradient coils are designed as saddle (Golay) coils [3].


6

Fig. 1.4: Impact of different gradient coils on static magnetic field (image courtesy: mri-

q.com.

1.1.2.3 RF coils

RF coils can be divided into three main categories: transmit coils, receive coils

and transceiver coils, and they are chosen depending on the area of the body that needs

to be imaged. Transmit RF coils create the B1 field that rotates the net magnetization

vector during MRI imaging as shown in Fig. 1.1(c), whereas the receive coil is used to

capture the RF signal that is generated during precession and is hence used for MRI

image creation. The transceiver coil on the other hand can both generate the RF signal

and also capture the precession signal.

The RF coils are designed to resonate at the Larmor frequency, hence they

comprise of an inductive element (L) and a capacitive element (C). The inductive

element is provided by one or multiple windings of low-resistance metallic wire,

usually copper due to its non–magnetic properties. As the coil may need to be tuned

once it is placed on the patient’s body, the capacitive element comprises of a variable

capacitor. The resonant frequency of the coils is given by

𝑓𝑐𝑜𝑖𝑙 =

1

2𝜋√𝐿𝐶 .

(1.2)


7

Depending on their geometry, the RF coils or the imaging coils are further

categorized into volume coils and surface coils. As the name implies, volume coils

surround the imaged object, with solenoid, saddle-shape, bird-cage and slotted

resonator being the popular geometries. These are primarily used for the head, the

knees and the neck. Surface coils on the other hand are placed adjacent to the imaged

object and are primarily used as receiver coils. Surface coils are primarily designed to

detect magnetic resonance signal from a small region of the imaged object. They are

very popular because they provide a very high signal-to-noise ratio for the small

volume of tissue close to the coil as compared to a standard volume RF coil. However,

the sensitivity of the surface coils drops off as the distance from the coil increases. In

modern MRI systems multiple surface coils are grouped together to create phased

array coils. The signal received by these coils is collected simultaneously and

combined to construct a single image of the object. Furthermore, phased array coil

provides a superior signal-to-noise ratio as compared to a single large surface coils and

enables parallel imaging of multiple body parts.

Fig. 1.5: Different RF coils (a) volume coil, (b) single surface coils, and (c) phased array

surface coils. (Image courtesy: Siemens)

1.1.2.4 RF Receiver

The RF signal generated by the precessing proton is very weak. Hence the

surface coils are placed very close to the patient’s body to pick up those weak signals.

The captured weak signal is then passed on to an RF receiver for further processing.

To increase the amplitude of the received signal, the first stage of an RF receiver

consists of multiple stages of low noise, high gain amplifiers. A mixer is then used to

bring the signal to a low frequency intermediate frequency (IF) frequency which is


8

then converted to digital domain by a high resolution, low speed, 12-bit to 16-bit

analog-to-digital converter (ADC) [4]. This architecture uses low bandwidth ADC

with sampling rate less than 1MHz. However, signal can be directly sampled by using

a high-bandwidth, high-resolution, 12-bit to 16-bit ADC with sample rates up to

100MHz, thereby eliminating the analog mixer from the chain.

1.1.2.5 Image Processing

The raw data captured during the MRI imaging contains both the spatial

frequency and phase information of the precessing protons and is called as the k-space

data [3]. This k-space data is then processed using Fourier transform and converted to

a grey scale image of the object.

1.1.3 Drawbacks of Existing MRI Systems

With rapid advancements in the imaging technology and its wide ranging

benefits, MRI has widely emerged as one of the most accurate medical diagnostic

techniques available to physicians. Despite this, the cost of MRI is still prohibitively

high as compared to CT scan or normal X-ray. Apart from initial equipment cost, one

of the main contributors to the operating cost is the cost associated with the RF coils

used for imaging. As discussed in section 1.1.2.3, to get a high resolution image, the

surface receiver coils need to be placed very close to the imaged object. Furthermore,

phased-array coils are used to improve the resolution even further. This implies that

separate receiver coils are required to image different parts of the human body so that

the coils conforms to the human body as shown in Fig. 1.6. Thus, hospitals and

imaging centers need to purchase and maintain separate sets of coils not only for

different body parts but also for people with different height, weight, and body types.

Moreover, these coils are not cheap. The cost of each coil can vary from $12,000 to

$120,000.

Apart from the cost, these coils can be quite ill-fitted and can be intimidating,

especially for children. As a result hospitals regularly administer anesthesia to the

children before performing an MRI which adds to the costs associated with MRI. Thus


9

in order to reduce the cost of MRI, there is an urgent need to redesign the MRI

receiver coils.

Fig. 1.6: Separate receiver coils for different body part. (Image courtesy: Siemens)

1.1.4 Redesign MRI Receiver Coil

1.1.4.1 Modular Receiver Coil

As discussed in section 1.1.3, in order to a get a high resolution image via an

MRI system, the hospitals need to buy and maintain a lot of different types of receiver

coils. Being expensive, this adds to the overall cost of the MRI. One possible solution

is to use a modular coil instead of a single big coil. For example, a set of four element

phased- array coil can form a single MRI receiver coil module. If one needs to image a

smaller area of human body, one of these module can be used as shown in Fig. 1.7. To

image a bigger part of the body, multiple of these coils can be joined together like

Lego® bricks as shown in Fig. 1.8, thereby eliminating the need to buy different coils

for different body parts. Furthermore, to ensure that the four element unit can be used

as a building block for all the patients, these coils can be made even smaller.


10

Fig. 1.7: A single four element phased array coils used for imaging (coil image courtesy:

Siemens).

Fig. 1.8: Multiple four element phased array coils combined together to create a single

image (coil image courtesy: Siemens).


11

While the approach of using small diameter phased array coils as a building

block for a modular coil may seem feasible, it has its own drawbacks and limitations.

First, each of these module rely on cables to transfer the captured signal from the

precessing proton to the image processing system located inside the MRI console

room. As shown in Fig. 1.9, these cable are heavy and cumbersome due to the RF

traps that are added to filter out the stray RF signals that corrupt the MRI signal.

Furthermore, these cables have to be aligned properly on the patient’s body so that

they do not interfere with the imaging process. If multiple modules are used at the

same time, then the placement of these heavy cables would pose a significant

challenge as the area occupied by these cables may become comparable to the area of

very small coils. Moreover, each element in the coil corresponds to one RF receiver

channel. A typical MRI machine has four 32-channel connectors thus supporting a

maximum of 128 channels. Due to the limitations imposed by data transfer and

processing circuitry, only 32 of those channels can be accessed at any instance. Thus

even if cables placement problem is resolved, the maximum imaged area would be

limited by the maximum number of cables that the MRI scanner connector supports.

Fig. 1.9: Bulky receiver coils cables with RF traps and cable connecter ports on the MRI

scanner (image courtesy: GE).


12

1.1.4.2 Proposed Modular Wireless Receiver Coil

The main challenge towards the adoption of modular receiver coils is the

presence of heavy cables and its associated limitations. The coil setup time can be a

significant fraction of the total exam time. Consequently, removing these cables from

the receiver coil will lead to a more cost effective and time efficient system. In the past,

a number of architectures have been proposed to enable wireless MRI [5][6][7][8][9] -

[10] for minimizing or removing the cables. All these efforts use microwave

frequencies up to 3GHz, and protocols such as 802.11b or MIMO that are intended for

long-range communication over distances of 10 m to 100 m. This results in a sub-

optimal solution for wireless MRI in terms of power consumption and size.

Here, we propose a custom millimeter (mm) wave transceiver architecture that

meets the requirements for wireless MRI at minimum power consumption and size. A

block diagram is shown in Fig. 1.10, and a possible architecture for the proposed

wireless receiver coils is shown in Fig. 1.11. An earlier version of this transceiver has

been used for short range, high data-rate, near field communication (NFC) system [11].

The mm-wave transmitter (TX) would be located on the receiver coil jacket while the

mm-wave receiver (RX) can be embedded inside the MRI system’s bore tube. Once

the data is received at the bore tube over the mm-wave wireless link, a fiber optic

transceiver converts electrical signal into the optical domain. A fiber optic cable

transfers this data from the MRI bore to the scanner console room. A second fiber

optic transceiver converts this signal back to electrical domain. The data is then

further processed to obtain the final image. The transceiver operates in mm-wave

frequencies with an RF carrier at 60 GHz using on-off key (OOK) modulation.

Fig. 1.10: Block diagram of the proposed wireless MRI system


13

Fig. 1.11: Receiver coil with wireless transmitter.

1.2 Organization

In chapter 2, different architectures for the proposed wireless receiver coil are

presented along with a brief survey of competing wireless technologies. The

architecture and design details for the proposed wireless MRI system are then

discussed in chapter 3. Chapter 4 describes the test setup and the measurement results

for the proposed system and chapter 5 concludes the discussion along with the

direction for future research.

Chapter 2 - Wireless Receiver Coil

14


15

CHAPTER 2

Wireless Receiver Coil

2.1 Wireless Receiver Coil Architectures

The number of elements in a receiver coil can vary from 1 to 32, depending on

the area of the body to be imaged. Furthermore, to ensure conformity to the body part

to be imaged, coils with same number of elements may have different shapes and sizes.

For improved SNR, parallel imaging performance, and field of view (FOV), the

maximum number of elements will go to 128 elements in the near future. Thus, the

first step to eliminate separate receiver coils for different scenarios and replace it with

a universal receiver coil design is to pick up an architecture for modular receiver coil.

2.1.1 Single-Element Module

A single element module would offer maximum flexibility in terms of its reuse,

as multiple single element modules can be placed simultaneously to image different

parts of human body. Each receiver coil would have a separate data processing unit

and a wireless transceiver which may result in a high system power consumption. As

the coils may have to be placed in closed proximity during imaging, the signal from

adjacent wireless channels may interfere with each other. This will impose limitations

on the minimum distance between the coils. Furthermore, the choice of wireless

technology may also restrict the maximum number of coils that can be used

simultaneously.

2.1.2 Multi Element Module

A multi-element module can have anywhere between 2 to 32 elements. In a

multi-element module, data from different elements can be collated and processed

using a single data processing unit. It can then be transmitted using a single wireless

transmitter (TX). To ensure data integrity, the data and clock signal between different

elements would have to be synchronized. This would require a careful design of


16

routing between individual coils and shared data processing unit. The MRI signal may

couple to these connections thereby corrupting the data being processed. Furthermore,

the routing complexity would increase with the number of elements in the module.

2.1.2.1 Proposed four-element module

A four-element module is symmetric as compared to other multi-element

modules with number of elements less than ten. By placing the data processing unit at

the intersection of elements, as shown in Fig. 2.1(a), the symmetry ensures that the

signals from different elements have identical delay, thus ensuring clock and data

synchronization. The symmetry also ensures that any two data processing units are

separated by a minimum distance equal to the diameter of two coil elements, as shown

in Fig. 2.1(b). As compared to single element module, this separation would

considerably reduce inter-channel interference between wireless transceivers located

at each module.

(a) (b)

Fig. 2.1: (a) Proposed four-element wireless receiver coil module, (b) Four four-element

modules placed together to create a sixteen element module. (Coil image courtesy: Siemens)

During imaging, the data at each module can be digitized using a quad-channel

ADC like TI ADS5263 [12] and serialized using a high speed quad-channel serializer

like TI DS32EL042 [13]. The serialized data can then be transmitted using the

wireless transceiver located at each module. The availability of MRI safe quad-

channel data processing modules adds to the commercial viability of a four-element

module.


17

Even though a four-element coil has been proposed as a preferred architecture,

a single-element coil architecture cannot be discarded. A single-element coil might be

useful for scenarios where a very small area needs to be imaged that might be too big

for a multi-element coil. Furthermore, to generate high resolution images, future MRI

systems may require access to raw data generated by the sampling ADCs which is of

the order of gigabits per second. In such a scenario, a single coil architecture might be

the only feasible solution due to the data rate limitations imposed by the commercially

available wireless technologies.

2.2 Wireless Technologies

During imaging, the receiver coil captures the signal generated by the

precessing proton as discussed in Chapter 1. As this signal is very weak, it is amplified

using low noise, high-gain amplifiers and sent to the MRI console room via cables for

further processing. In case of wireless receiver coils, after initial amplification, the

receiver coil data can be digitized using MRI compatible analog-to-digital convertor

(ADC) like Texas Instruments (TI) ADS5263 [12]. ADS5263 is a 4 channel, 16-bit,

100 MS/s ADC with in-built decimation. This results in a raw data of 1.6 Gb/s for

each element of the receiver coil. As the MRI data is very narrow band, it can be

decimated to get an effective data rate of 20 Mb/s for each element. Thus depending

on receiver coil architecture and digital data processing, the data rates for wireless

receiver coil module may vary from megabits per second to gigabits per second.

Therefore, the wireless TX placed on the receiver coil should support high data rates.

For wireless receiver coils, the TX would be powered by a non-magnetic battery or by

energy harvesting circuits. This makes TX power consumption an important criteria

for system design. Moreover, in a modular approach to receiver coil design, several of

these TXs might be placed in close proximity, resulting in high inter channel

interference. In such a scenario, these TXs need to be placed and orientated to

minimize this interference. Finally, if the wireless receiver is embedded inside the

MRI system bore tube, the distance between the TX and the receiver would be less

than one meter. Thus wireless MRI requires a low power, high-data rate, and scalable


18

solution that can work up to one meter and some of the competing technologies are

discussed in the next section.

2.2.1 Wi-Fi 802.11ac

802.11ac is the latest standard from IEEE and is the first one to support Gb/s

data rates in the traditional 5 GHz band. 802.11ac supports multiple channel

bandwidths (20 MHz, 40 MHz, 80 MHz, and 160 MHz), multiple modulation schemes

(binary phase shift keying (BPSK), quadrature phase shift keying (QPSK), 16-

quadrature amplitude modulation (QAM), 64-QAM, and 256-QAM) and up to eight

spatial streams for multiple input, multiple output (MIMO) [14]. The system provides

twenty five 20MHz channels which can be joined to form twelve 40MHz channels or

six 80MHz channels or two 160 MHz channels as shown in Fig. 2.2 [15]. Depending

on the configuration, the system can support data rates from 7.2 Mb/s to 6933.3 Mb/s,

as shown in Table 2-1 [15]. However, the system throughput is approximately one-

half of the supported data rate [16] and decreases as the distance between the two

wireless transceivers is increased [17].

Fig. 2.2: Channels defined for 5GHz band [15].

With 20 Mb/s of decimated data for each element, the proposed four-element

receiver coil module would generate data at the rate of 80 Mb/s. This can be supported


19

by either using two parallel streams of 20 MHz or a single 40 MHz stream. For a

MIMO system, each stream requires its own antenna with a minimum physical

separation of lambda-by-2 between them and a suggested separation of four lambda

[18] for optimal performance. This would lead to an antenna separation of ~1.1ʺ to ~9ʺ

for the 5 GHz band. The size of a single element in the receiver coil can vary from

2.5ʺ to 8ʺ. Therefore, having multiple antennas on each module would increase the

design complexity in terms of routing to different antennas, as well as the required

abutting of multiple modules to image a wider section of human body. Hence a single

antenna, with 256-QAM single stream 40 MHz channel, supporting 200 Mb/s data rate

would be a preferred solution.

Table 2-1: 802.11ac data rates for different modulations and spatial streams [15].

With twelve 40 MHz channels and a dedicated TX and receiver (RX) for every

four-element module, the 802.11ac can support a maximum of 48 elements

simultaneously. This is assuming that each TX-RX pair occupies a separate channel.

More modules can be supported by using time-division multiplexing. Here data from

different TXs would go to a single RX. However, this would require the MRI data to

be buffered at each TX while it waits for its turn to transmit. With data rates of 80

Mb/s for each module, this could result in a significant power and memory overhead.


20

On the other hand, for a single-element module, even 160MHz bandwidth channel

doesn’t support 1.6 Gb/s of raw data generated at the coil.

In terms of power consumption, with a complex modulation schemes of 256-

QAM and orthogonal frequency division multiplexing (OFDM) signaling schemes, the

average power consumption of an 802.11ac system at maximum data rates is ~10

nJ/bit. This results in a power consumption of 2W for each module [19] and does not

scale linearly with data rate.

2.2.2 802.11ad

The IEEE 802.11ad standard uses the 8 GHz unlicensed band at 60-GHz and is

aimed at providing data rates up to 7 Gb/s. High path loss at 60-GHz and the

availability of a wider bandwidth makes 802.11ad an ideal candidate for short range,

high data rate communication system with distances up to 10 m. The international

telecommunication union (ITU) recommends the use of four channels, each 2.16 GHz

wide with center frequencies of 58.32, 60.48, 62.64, and 64.80 GHz [20]. It can,

however, be seen from Fig. 2.3 that only channel 2 with its center frequency of 60.48

GHz is available globally. This is recommended to be the default channel.

Fig. 2.3: 60-GHz band channel plan and frequency allocation by region [21].

802.11ad supports multiple coding and modulation schemes as shown in Table

2-2. There are several options that support data rates greater than 80 Mb/s required by

four-element module and even 1.6 Gb/s required for the raw data rate transfer by a


21

single element module. There are multiple companies [22][23] - [24] providing low

power 802.11ad chips with energy efficiency of ~50pJ/bit at the highest data rate,

which is orders of magnitude better than that offered by 802.11ac systems. With

support for required data rate and superior energy efficiency, 802.11ad seems like a

viable solution for wireless receiver coils. However, all the commercial solutions are

designed to support distances up to 10 meter and operate at the maximum data rate of

around 4 Gb/s. As they need to adhere to linear modulation schemes like OFDM, the

system power remains constant irrespective of the data rate. Thus even while

transmitting data at 80 Mb/s, the 802.11 ad system would consume ~200mW of power.

There are few low power solutions offered by SiBEAM™ [22] as a replacement for a

physical universal serial bus (USB) connector. However, this low power solution

works for distances less than 1 cm and hence can’t be used for wireless receiver coils.

Wireless USB may become a viable alternative if its range is extended to 1 m in near

future. Availability of only one single channel may significantly increase the

interference between multiple TXs as they are placed in closed proximity inside the

MRI system bore tube while imaging. This can be mitigated by using beam forming at

each TX and time division multiplexing between multiple TXs.

Table 2-2: 802.11AD MODULATION AND CODING SUMMARY [20]


22

2.2.3 Custom Millimeter (mm)-Wave Solution

Based on the discussion in previous sections, 60-GHz seems to be a possible

solution for modular wireless MRI receiver coils. However, the limitations imposed by

existing 802.11ad standard warrants the need of a custom solution. The processing

overheads and limitations imposed by complex modulation schemes such as OFDM

results in a high power consumption in existing 60-GHz solutions. This can be

resolved by using simpler modulation schemes such as on-off keying (OOK). In OOK,

the TX consumes power while it is transmitting ‘1’ and is OFF when it is transmitting

‘0’. This can reduce the system power consumption by a factor of two as compared to

conventional TXs that are ON all the time. The use of non-coherent modulation like

OOK simplifies the system architecture as the TX and the RX don’t need to be phase

synchronized. As the absolute phase of the system is not a concern, it relaxes the

linearity constraints for the power amplifier (PA) design, thus allowing the use of a

more efficient non-linear PA. This reduces system power consumption so that they can

be powered using tiny non-magnetic batteries. With OOK modulation at its core, a

custom mm-wave transceiver was designed that meets the requirements of wireless

MRI system. The system uses 60-GHz as the RF carrier frequency. The TX power of

the proposed system scales from 1.3 mW to 14.0 mW as the data rate is varied from

200 Mb/s to 2500 Mb/s, while the RX consumes a fixed DC power of 76 mW. The 60-

GHz radio occupies an area of 1.62 mm2 in TSMC 40 nm CMOS GP process and

would be discussed next.

Chapter 3 - Millimeter-Wave Wireless Transceiver

23

CHAPTER 3

Millimeter-Wave Wireless Transceiver

3.1 Proposed mm-Wave Transceiver v/s Prior Art

Numerous transceiver architectures have been proposed since the inception of

60-GHz as a viable band for civilian application. All these architectures can be

broadly classified into two categories. The first category focuses on providing a

reliable wireless link at long range ( 1 - 10 m), and maximizes data rate by enhancing

the spectral efficiency at the expense of high power consumption and costly antenna-

chip packaging technology [25] - [26][27][28][29]. Examples are wireless docking [30]

and wireless HD [31]. The long range requirement necessitates the use of high gain

off-chip antennas in two-dimensional arrays for beam-forming [32], [33]. In order to

achieve high spectral efficiency, more symbols need to be packed into a limited band-

width (BW), hence forcing the system to apply more complex modulations such as 16

QAM and 64 QAM. These modulation schemes in turn increase the system linearity

and noise requirements. The design of linear radio frequency (RF) front-ends, low

noise RF and base-band (BB) blocks, and low phase noise local oscillator (LO) further

increases the power consumption and complexity of these radios [25]- [28]. Since

beam-forming is an integral part of this category of transceivers, additional challenges

arise in the implementation of phase shifters in either RF, LO, BB, or any other

combination [32] - [33][34][35][36][37][38][39]. These complexities increase the cost,

size, and power consumption of a product, hence limiting the use of 60 GHz radios in

mass markets.

The second category of 60 GHz radios, to which this work belongs, targets at

applications that require high data rate, low power, and short range (< 1 m) wireless

link. In general, these systems use relatively simpler modulation schemes such as

QPSK (coherent) and OOK (non-coherent). When coherent modulations are used, the

relative phase between the transmitter and receiver needs to be maintained. The


24

requirement of phase alignment necessitates the use of a phase lock loop (PLL) in both

the transmitter and receiver. For 60-GHz systems, the PLL and the LO in-phase (I)

and quadrature (Q) generation consume substantial power. In [40], a 4 × 4 QPSK

transceiver with on-chip antennas is demonstrated with high data rate and energy

efficiency. However, its transmitter consists of four free running VCOs and it assumes

that the receiver LO knows the frequency of these VCOs. This architecture with its

assumption will severely complicate demodulation in a practical system. On the other

hand, in systems using non-coherent modulation, the transmitter carrier frequency

does not need to be synchronized with the receiver LO. In fact, OOK with the

envelope detection in the receiver can be implemented without the LO requiring a PLL

in either the receiver or transmitter. Since LO generation consumes significant portion

of the total power, its elimination substantially reduces the power consumption [41],

[42]. A 1 × 1 OOK transceiver with on-chip antennas is reported in [43] where it uses

return-to-zero (RZ) signaling with pulse-width control (PWC) to reduce the power

consumption.

To meet the requirements of the wireless MRI systems, a very low cost, low

power and fully integrated 2 × 2 60-GHz transceiver with on-chip antennas has been

designed. The transceiver has a small silicon area of 1.62 mm2, including energy

harvesting circuits, and is suitable for near-range communication. The 2 × 2 system

improves the link budget and consequently the communication distance. Utilizing the

large bandwidth available at 60-GHz, this impulse-radio ultra-wideband transceiver

uses RZ-OOK modulation in conjunction with PWC to significantly reduce its power

consumption. A low-loss transmit-receive (TR) switch allows for the sharing of two

dipole antennas between the transmitter and receiver, allowing for a very small silicon

area. Furthermore, a low-power transmitter ensures that the energy harvested at the

unlicensed ISM band of 2.45 GHz is sufficient to power up the transceiver in the

transmit mode.

Harvesting energy at 2.45 GHz for a radio link at 60-GHz has important

advantages. Because of the large separation between the 60-GHz and 2.45 GHz

frequencies, a potentially high power harvesting signal would not desensitize the 60-


25

GHz receiver. This is because the 60-GHz antenna and the receiver would sufficiently

attenuate an out of band jammer such as the one at 2.45 GHz. If the harvest and the

radio link frequencies were close to each other, extra filtering would have been

necessary to eliminate a desensitization scenario. Such extra filtering would add cost

and further insertion loss to the system.

Fig. 3.1: (a) Linear relationship between the data rate and transmit power consumption. (b)

Application in point of sale advertisement. (c) Application in medium to high data rate:

neural data transmission of small beings.

Finally, apart from wireless MRI, the designed transceiver could be used for

different applications in consumer electronics and the future internet of things (IoT),

depending on the power and data rate requirements. As shown in Fig. 3.1(a), data rate

linearly scales with the transmitter power consumption, hence providing an agile

platform to support various applications. This radio could be used as an alternative to

the near field communication (NFC), enabling substantially higher data rate and

energy harvesting capability. Fig. 3.1(b) shows another application of this radio as an

enabler of point of sale advertisement, which could become a major trend in the

emerging IoT markets. As illustrated in Fig. 3.1(c), the small size of the radio can

enable neural data transmission of small beings and insects where wireless

communication at longer wavelengths could present a mechanical challenge to the

biological experimentation due to the large size of the antenna.

3.2 Transceiver Architecture

The 60-GHz transceiver architecture along with the TX and RX timing

diagram is depicted in Fig. 3.2 . In a time-division duplexing (TDD) communication, a

TR switch enables the two dipole antennas to be shared between the RX and TX


26

modes. The transmitter baseband generates an RZ bit stream from a non-return- to-

zero (NRZ) data and clock as illustrated in Fig. 3.2. The RZ signal pulse-width is then

programmed to assume a value between 250 ps to 1200 ps using a 6-bit Pulse Width

Controller (PWC). The signal at the output of the PWC switches the two VCOs and

the PAs on and off simultaneously. When switched on, the VCOs oscillate at 60-GHz,

and the PAs transmit power to the two dipole antennas at 60-GHz. In the receiver, the

voltage outputs of the two LNAs are summed and fed into a self-mixer, as shown in

Fig. 3.2. The self-mixer output point feeds the baseband gain stages. The output of the

RX baseband is an RZ signal that can drive a load such as an oscilloscope.

Fig. 3.2: Transceiver architecture and corresponding waveforms.

The transmitter baseband comprising of RZ and PWC circuit is essential to

enable linear scaling of power with the data rate. In RZ-OOK modulation, sending a

data bit of value “0” consumes no energy as TX is off. A data bit of value “1” is sent

by the TX in the form of a short 60-GHz pulse as shown in Fig. 3.2 and the TX

consumes power only for that duration. As shown in Fig. 3.3, the transmitter

consumes power only during the pulse-width T when both the VCO and PA are turned


27

on. It consumes only a leakage power outside the pulse-width T. The average DC

power consumption of a single pulse is:

𝑃𝑝𝑢𝑙𝑠𝑒 =

𝑇

𝑇′ 𝑃𝑝𝑢𝑙𝑠𝑒 = 𝑇 ∙ 𝑑𝑎𝑡𝑎𝑟𝑎𝑡𝑒 ∙ 𝑃𝑝𝑢𝑙𝑠𝑒

(3.1)

where 𝑃𝑝𝑢𝑙𝑠𝑒 is the sum of both the VCO and PA DC power consumptions. Equation

(3.1) suggests that the average DC transmitter power consumption is a linear function

of both the pulse width and the data rate.

Fig. 3.3: Average power consumption of a single transmitter pulse.

Fig. 3.4: (a) Transmitter ON time waveform for different data rates, and (b) corresponding

power consumption for a RZ-OOK modulation.


28

In case of RZ-OOK modulation, the transmit pulse width increases

proportionally with the data rate as demonstrated in Fig. 3.4(a), hence transmitter

power consumption is constant irrespective of the data rate as shown in Fig. 3.4(b).

However, for RZ-PWC-OOK, the transmit pulse width remains constant for different

data rates as shown in Fig. 3.5(a). As a result, the transmitter power consumption

scales linearly with data rate as shown in Fig. 3.5(b).

Fig. 3.5: (a) Transmitter ON time waveform for different data rates, and (b) corresponding

power consumption for a RZ-PWC-OOK modulation.

3.3 Transmitter Design

In the proposed transceiver, the transmitter has two separate transmitting

elements that are fully symmetric. As shown in Fig. 3.6, each transmit element

consists of a VCO and a PA. The VCO drives the PA input and the PA drives a

differential dipole antenna.

Fig. 3.6: 2×2 Transceiver RF blocks.


29

3.3.1 Voltage-Controlled Oscillator (VCO)

The two VCOs in this design are cross-coupled NMOS pairs as shown in Fig.

3.7. Since the phase in OOK modulation does not convey information, there is no

phase noise requirement for the VCO design. In this OOK design, there are two main

considerations for the VCO, namely, phase coherency and frequency matching. The

phase coherency requirement demands that the two VCOs stay in-phase while they are

operating. When the VCOs start up from the circuit noise or other small initial

perturbations, the two VCO phases would be random and hence not guaranteed to be

aligned. Because of this randomness, the electromagnetic waves emanating from the

VCOs, after radiating through the antennas by the PAs, could undergo a partial or a

complete destructive interference. The two VCOs can be phase aligned by starting

them up with a large voltage initial condition [44].

Fig. 3.7: The two transmit VCOs with fast startup.

This design modifies the start-up technique of [44] to make it more suitable for the

OOK modulation and to create a larger voltage initial condition across the LC tank of

the VCOs. As shown in Fig. 3.7, at the rising edge of the RZ transmit base-band data,

M1 goes into strong triode and pulls the source of M3 to the ground. A sudden rush of

current through M3 results in a large voltage perturbation across the VCO LC tank.


30

The same voltage rising edge arrives at the gate of M2 after a time delay, which is

provided by the two cascaded inverters. Therefore, M2 turns on and enters into a

strong triode. To ensure proper differential operation, M2 should turn on after a delay

of half a period. Thus the two cascaded invertors provide a delay of 1/2𝑓0 = 8 ps for

the 60 GHz VCO. At this point the differential voltage across the VCO LC tank is

large, for example, 200 mV. This ensures both VCOs start at the same phase and

quickly reach a maximum swing.

The other requirement is the frequency matching between the VCOs. Since the VCOs

are open loop, their exact frequencies cannot be determined. There is a coarse 2 bit

digital-to-analog (DAC) that is connected to the control voltage of an NMOS varactor

pair, but it is only used to center the VCO frequency at a desirable channel within the

57–64 GHz range.


31

Fig. 3.8: Beat frequency generation due to transmit VCOs' mismatch.

Fig. 3.8 illustrates the case in which the two VCOs have a frequency difference

of 2∆𝑓 . The spatial power of the EM waves emanating from the two VCOs will

combine to form a beat with a frequency of ∆𝑓. Here the coherence time, 𝑡𝑐𝑜ℎ𝑜𝑟𝑒𝑛𝑐𝑒,

of the two VCOs is defined as the time in which the spatially combined signal loses

half of its power i.e.,

cos(2𝜋∆𝑓𝑡𝑐𝑜ℎ𝑜𝑟𝑒𝑛𝑐𝑒) =

1

√2

⇒ 𝑡𝑐𝑜ℎ𝑜𝑟𝑒𝑛𝑐𝑒 =

1

8∆𝑓.

(3.2)

The frequency of an LC oscillator is given by

𝑓 =

1

2𝜋√𝐿𝐶 , (3.3)


32

where 𝐿 and 𝐶 are the effective inductance and capacitance of the oscillator

respectively. Considering the variation in the effective inductance, ∆𝐿, is negligible

compared to the variation in the effective capacitance, ∆𝐶, the frequency variation can

be simplified as:

∆𝑓 =

1

2 ∙

Δ𝐶

C ∙ 𝑓 ,

(3.4)

By substituting (3.4) in (3.2), the coherence time of the VCOs is found to be:

𝑡𝑐𝑜ℎ𝑜𝑟𝑒𝑛𝑐𝑒 =

1

4𝑓 (

Δ𝐶

C)

−1

. (3.5)

The coherence time as a function of the total variation of the VCO tank capacitance is

shown in Fig. 3.9. This VCO capacitance is exclusively due to the NMOS transistors,

which have a lower statistical variation of less than 5%, compared to metal capacitors

(about 20%).

Fig. 3.9: Coherence time vs capacitance mismatch.

3.3.2 Power Amplifier (PA)

Since OOK modulation has no linearity requirement, high-efficiency switching

PAs can be used in the transmitter. A number of class E designs have been reported at

60 GHz in both SiGe and CMOS processes [45] - [46][47][48]. In a class E design, the

active device is treated as an ideal switch which gives an open circuit during its “off”

mode and a perfect short-circuit during its “on” mode. However, with the operating


33

frequency being a significant portion of the device transit frequency (fT), this ideal

switching is not possible thus leading to a sub-optimal class E operation [46]. On the

other hand, a class F-1 PA relies on harmonic tuned loads to shape the output voltage

and current waveforms.

Fig. 3.10: (a) Standard class F-1 PA. (b) Current and voltage waveforms for a class F-1 PA

with only one tank at third harmonic (solid) and ideal (dotted).

For example, Fig. 3.10(a) shows an ideal F-1 PA implementation. The high Q

resonators provide a short for third, fifth (odd) voltage harmonics while an open for all

even harmonics. This results in non-overlapping transistor voltage and current

waveforms as shown in Fig. 3.10(b). In a practical implementation, these high Q

resonators result in a significant area penalty and also introduce extra loss in the

output matching network.

The implemented PA is a combination of class E and class F-1 PA and is shown

in Fig. 3.11. Similar to class E PA, the drain cap Cdd is resonated out with drain

inductors L1A, 1B at the fundamental frequency of 60 GHz. At the same time, the drain

matching network consisting of capacitors Cdd, L1A, 1B, Cs and the dipole antenna is

designed as a harmonic tuned load akin to class F-1 PA. As a result its operation

resembles a class E/F2, odd PA [45]. This can be seen by comparing the simulated drain

voltage and current of the PA to the ideal class E/F2, odd PA as shown in Fig. 3.12.


34

Fig. 3.11: The implemented class E/F2, odd PA.

The gate of transistors M1A, 1B is biased at a nominal voltage of 700 mV by a 3

bit resistor DAC, while the drain is connected to a supply voltage of 1 V. A tail

transistor, Mtail, enables quick turn on/off of the PA during OOK operation. When

turned on, the drain of Mtail is pulled to a lower potential (about 100 mV). Thus the

impact of Mtail on the output swing is minimized. With an average dc power

consumption of 13.5 mW, the PA delivers 4 mW of power at 60 GHz to a load, thus

achieving a drain efficiency of 29.6%.

Fig. 3.12: Drain Voltage (solid) and current (dotted) waveforms for (a) ideal class E/F2, odd

PA, and (b) implemented class E/F2, odd PA.


35

3.3.3 Transmit-Receive (TR) Switch

To maintain low cost and small silicon area, it is very important to share the

antennas between the receiver and transmitter blocks. The TR switch network shown

in Fig. 3.13(a) is used to isolate the differential input and output front-ends from each

other [28].

Fig. 3.13: (a) TR switch interface with the PA and LNA. (b) TR switch when TX is on (c)

TR switch when RX is on.

Inductor LCH, as part of the TR switch, acts as an electrostatic discharge (ESD)

protection device, hence, eliminating the need for a diode ESD protection that would

present a prohibitively large parasitic capacitance at 60GHz [49]. As shown in Fig.

3.13(a), LCH self-resonates at 60GHz and presents a parallel resistance of RCH = 3kΩ

between the dipole antenna differential input ports. Since the dipole antenna has an

equivalent parallel radiation resistance of 100Ω, RCH is too large to load the dipole

antenna.

3.3.3.1 Transmit Mode:

As shown in Fig. 3.13(b), when the radio is in the transmitting mode,

minimum channel length NMOS transistors SW1, SW2, and SW3 are turned on by

pulling their gate voltage to a high voltage value such as VDD = 1V. In this mode, the

drain and source DC voltages of SW1 and SW2 are at ground through the DC path

provided by the choke inductor LCH. Ideally these switches should be large so their on-


36

resistance, RTSW, become much smaller than the antenna impedance. But, in the

receiving mode, it is desirable to minimize the source-drain capacitances of SW1 and

SW2. So an optimum switch size of (32µm/40nm) is chosen to satisfy both cases. This

switch size is large enough to meet the metal electro-migration rule for at least 20mA

of DC current. The parasitic model of SW1 is shown in Fig. 3.13(a). Other switches

have the same parasitic model. It is important to use a large resistor, for example, 5kΩ,

at the gates of SW1 and SW2 so their gate-source and gate-drain capacitances would

not load the PA. For SW3, the drain and source voltages are also pulled to ground

through a biasing network that pulls the voltage at nodes B1 and B2 to ground. As

shown in Fig. 3.13(b), when SW3 is on, the differential impedance on the right seen at

the differential terminals of the dipole antenna is:

𝑍𝑟𝑖𝑔ℎ𝑡 = 2𝐿𝐺𝜔𝑗 + 2𝑅𝐺 +

2

𝐶𝐶𝜔𝑗+ 𝑅𝑅𝑆𝑊|| (

2

𝐶𝐺𝑆𝜔𝑗+ 𝑅𝐺𝑆),

(3.6)

where RRSW is the on-resistance of SW3, RG is the parasitic resistance of gate

inductance LG, Rgs and Cgs are the parasitic resistance and capacitance of the LNA

input, and CC=220fF is an AC coupling cap. With an SW3 size of (4µm/40nm), the

overall impedance of 2𝑗

𝐶𝐶𝜔+ 𝑅𝑅𝑆𝑊|| (

2

𝐶𝐺𝑆𝜔𝑗+ 𝑅𝐺𝑆) equals (78 +

1

(72𝑓𝐹)𝜔𝑗) Ω. With LG

= 580pH and a quality factor of 5, 𝑍𝑟𝑖𝑔ℎ𝑡 = 1129Ω||400𝑗Ω at 60GHz. The simplified

equivalent circuit for Zright and the dipole parallel differential resistance of 100Ω is

shown in Fig. 3.13(b). The effective impedance seen by the PA is:

100Ω||1129Ω||400𝑗Ω = 91Ω||400𝑗Ω. (3.7)

The imaginary part of (3.7) can be absorbed by the PA matching network.

Therefore, the TR switch loss due to the loading of the LNA is:10 log(91Ω/100Ω) =

−0.85𝑑𝐵. An extra simulated loss of 1dB is due to the on-resistance of SW1/SW2. So

the total TR switch loss is -1.85dB.

3.3.3.2 Receive Mode:

As shown in Fig. 3.13(c), in receive mode, NMOS transistors SW1, SW2, and

SW3 are turned off by pulling their gate voltages to ground. The drain and source

voltages of SW1 and SW2 are DC-coupled to ground through LCH. When turned off,


37

SW1 and SW2 present an effective parasitic drain-source capacitance, CTSW, and an

effective parasitic series resistance, RTX. As shown in Fig. 3.13(c), looking at the left

of the dipole antenna, simulated at 60GHz, the effective impedance is lower bounded

as:

𝑍𝑙𝑒𝑓𝑡 > 2𝑅𝑇𝑋 +

2

𝐶𝑇𝑆𝑊𝜔𝑗= (36.3 − 328𝑗)Ω

= 3𝑘Ω||(−332𝑗)Ω

(3.8)

Therefore, the effective antenna impedance seen by the PA is lower bounded

by: 100Ω||𝑍𝑙𝑒𝑓𝑡 = 97Ω||(−332𝑗)Ω. The imaginary part of (3.8) can be absorbed by

the LNA input matching network. The loss due to the TR switch during RX on is

therefore:10 log(97Ω/100Ω) = −0.3𝑑𝐵.

3.4 Receiver Design

The receiver architecture is shown in Fig. 3.14. There are two LNAs for the

two dipole antennas. Each LNA consists of two stages that are coupled by

transformers. The third stage is common between the two paths and sums the output

voltages of the two LNAs. This summing improves the receiver sensitivity by 3 dB. It

also provides additional 6 dB of voltage gain. A common source amplifier is used for

the LNA's front-end stage. The choice is partially driven by the fact that the antenna is

being shared between the LNA and PA. In this case, the overall link budget was

improved as the common source LNA improved the TR switch loss in transmit mode.

There are two other reasons for using a common source LNA.

First, for the same power consumption, inductor degeneration reduces the gain

of the first stage, increasing the noise figure (NF) contribution of the later stages. In

this design, the NF of the self-mixing stage and the chain of the baseband amplifiers is

significantly higher than the LNA. Therefore, it is best to have the highest gain at the

LNA stages as the noise contribution of the post LNA stages will be divided by the

overall LNA gain.

Second, for low power LNA design, where the input NMOS transistors are

small, the effective resistance of the NMOS input transistor in series with its gate


38

capacitance is sizable compared to the antenna radiation resistance. At an operating

gate voltage of 0.7 V, the series gate resistance and capacitance of a differential

NMOS pair are 67 Ω and 6.1 fF ( 33.5 Ω and 12.2 fF single-ended). This impedance

should be matched to a 100 Ω differential antenna. A gate inductance, LG, of 580 pH

resonates out the gate capacitance at 60 GHz. With a Q of 5, the parasitic series

resistance of LG equals 43 Ω. Therefore, the total impedance seen by the antenna at 60

GHz is 67 Ω + 2 × 43 Ω = 153 Ω. This results in an S11 of lower than -10 dB.

Fig. 3.15 shows the simulation results for the LNA NF, gain and S11. The

return loss, S11, maintains lower than -10 dB from 52 GHz to 66 GHz. The NF is

below 6 dB from 54 GHz to 66 GHz with a minimum of 5.2 dB at around 60 GHz.

The maximum overall gain of the two LNAs is 40.5 dB simulated at the output of the

voltage summation circuit at above 60 GHz. Each LNA path has a simulated gain of

34.5 dB. The first, second and third LNA stages have voltage gains of 10.6 dB, 11.5

dB, and 9.4 dB. There is additional 3 dB gain due to the input matching Q boosting.

All input transistors of the LNA stages are biased by a 3 bit resistor DAC. The cascode

transistors are biased to the voltage supply. The total power consumption of the two

LNAs is programmable from 38 mW to 72 mW. The simulated NF was achieved at a

48 mW total power consumption.

As shown in Fig. 3.14, each LNA stage included stacked cross-coupled NMOS

transistors to provide negative resistance. This was intended to boost the LNA gain.

As explained earlier, the NF of the receiver chain is largely dominated by the envelope

detector mixer and the following BB amplifiers, so maximizing the gain of the LNA

stages results in the lowest overall NF. However, during the measurement, the

negative resistance caused oscillations so the cross-coupled NMOS were kept off.


39

Fig. 3.14: Receiver chain (RF and BB).

The output of the voltage summation stage feeds a passive AC-coupled self-

mixer to extract the OOK modulation envelope, as shown in Fig. 3.14. The drain and

source bias voltage for the mixer transistors is provided by the self-biased inverter that

follows the mixer. The gate bias is at one threshold voltage above the source-drain

voltage. This ensures the transistors are biased at near threshold for maximum RF to

BB gain. Because of the non-linear nature of the envelope detector mixer, its NF is a

very strong function of the RF input voltage. The receiver baseband is a 3-stage DC-

coupled inverter chain with resistive feedback followed by a common-source amplifier

with programmable resistive load. The last stage drives an external 50 Ω load, for

example, an oscilloscope.


40

Fig. 3.15: Simulated NF, gain and return loss of the LNA.

Fig. 3.16: Total NF of RX chain vs. LNA input power.

Fig. 3.16 shows the integrated noise figure of the total RX chain over the

signal bandwidth of interest versus the input signal power to the LNA. The integrated

NF (10 kHz to 8 GHz) of the receiver including the mixer and the baseband chain is


41

31 dB in simulations when the RF input at 60 GHz is at the sensitivity level. The

overall receiver NF is overwhelmed by the mixer and the BB chain. Because of the

very non-linear nature of the mixer, the integrated noise figure is a very strong

function of the input signal power. The VCOs in the TX were measured to oscillate at

around 56.5 GHz, as a result of extra parasitics that were not captured in simulations.

Therefore, the LNA gain at this frequency is 10 dB lower than its intended peak at 60

GHz. As a result, the overall RX NF is worse than expected.

3.5 Dipole Antenna

At mm-wave frequencies, on-chip antennas become a viable option as the

wavelength is comparable to the die size of a complex transceiver design [50], [51],

[52], [53]. Furthermore, on-chip antennas eliminate the loss due to the interconnect

techniques such as wire-bond, wafer-level-packaging (WLP), ball-grid-array (BGA),

and so on. However, the low resistivity of the silicon substrate (10Ω-cm) introduces

electric field losses leading to a reduction in antenna efficiency and radiation

resistance. Also, the high permittivity of the silicon substrate (εr = 11.9) results in the

absorption of the fields in the substrate thus giving rise to undesirable substrate modes

that adversely impacts the antenna efficiency [54], [55]. The two most commonly used

techniques to improve antenna efficiency involve using a substrate dielectric lens [50],

[56] and substrate thinning [42], [53], [57]. The di-electric lens is large, expensive,

and needs accurate placement with respect to the on-chip antenna. On the other hand,

die thinning is a one-time process and can be performed at the wafer-level to reduce

the cost for mass fabrication.

The implemented on-chip antennas are shown in Fig. 3.17. They consists of

two dipole elements on top of aluminum layer with a separation of 0.12 λ, where

λ=5mm is the free space wavelength at 60 GHz. The assumption for the mm-Wave

radios with off-chip antenna arrays is that the antenna spacing is 0.5 λ. One concern in

choosing the appropriate antenna spacing is the mutual coupling between the antennas.

It is defined as the ratio of the power received from one antenna to the power that the

other antenna is transmitting. This mutual coupling is typically undesirable as the


42

power that should be transmitted is wasted as the result of absorption in another

antenna. It could also change the pattern of each individual antenna. Furthermore, a

high coupling between the adjacent antennas could exacerbate the voltage wave

standing ratio (VWSR) in the transmitter PA. The VSWR at the output of the PA

needs to be as small as possible for two reasons. The first reason is the reliability of

the CMOS devices. CMOS transistors have a reliability voltage guideline. Exceeding

these limits would either destroy the device or reduce its lifetime. The second reason is

to minimize the error vector magnitude (EVM). The EVM is the inverse of the SNR

and factors such as noise, linearity and mismatch could degrade it. A large VSWR

appears at the PA output and makes it nonlinear, hence increasing the EVM. On the

receiver side, the mutual coupling would make the inputs correlated. The correlation

adversely affects the array beamforming and resolution.


43

Fig. 3.17: Dual dipole antenna with patterned shield.

Fig. 3.18 shows the coupling between the two dipole antennas as simulated in

HFSS. The simulation shows surprisingly low coupling between the dipole antennas.

The coupling is less than −14dB over a bandwidth of more than 20 GHz (50GHz –

70GHz). If the dipoles were located at the same distance of 0.12 λ over air, the

coupling would have been significantly more. Since the dipoles are embedded on-chip,

and due to the high dielectric constant of silicon (𝜖𝑟=11.9) and the PCB layer (𝜖𝑟=3.8)

under it, most of the EM energy is absorbed into the silicon and subsequently into the

PCB material. Qualitatively, once in silicon, the EM wave is slowed by a factor of

√𝜖𝑟=11.9=3.45. This means the effective wavelength inside the silicon is shortened

by a factor of 3.45. Therefore, the spacing between the antennas is 3.45×0.12

𝜆𝑠𝑖𝑙𝑖𝑐𝑜𝑛=0.41 𝜆𝑠𝑖𝑙𝑖𝑐𝑜𝑛. This is not a rigorous explanation of the phenomenon, but rather


44

an intuitive and easy to understand qualitative explanation. It is very desirable to be

able to place the array antennas very close to each other, without suffering from

excessive coupling between them.

Fig. 3.18: HFSS simulation for antenna to antenna coupling.

To meet the metal density rule, a patterned floating shield on the two top

copper metal layers was placed below the dipole antennas, as shown in Fig. 3.17. The

area under the shield is used for digital circuits as well as energy storing capacitors for

the harvest block. To minimize the loss due to the substrate, the dipole elements were

placed very close to the edge of the chip [58], and the silicon was thinned to 100μm

[55]. At 60 GHz, λ0 = 5 mm, and with ℎ= 100 μm, ℎ/λ0 = 0.02. This is well within the

suggested range illustrated in Fig. 3.19.


45

Fig. 3.19: Impact of substrate thickness on antenna gain and normalized power [54].

Furthermore, a metal reflector was placed at a distance of quarter wavelength

from the antenna. Simulations suggest that metal reflector improves the overall

antenna gain by about 4dB (Fig. 3.20), which was also noticed in the measurement, as

shown in Fig. 3.21. The antenna gain is a measure of how much of the

transmitted/received power is concentrated in a particular direction. Intuitively, the

electromagnetic waves are blocked from radiating or passing through the metal

reflector. Thus the half-space behind the metal reflector has no EM power in it and the

half-space in front of it has twice as much power and hence, ideally a 3 dB gain

improvement. This theoretical 3 dB gain improvement can also be explained by

method of images and would be discussed in next section. The radiation patterns for

the dual dipole antennas and the single dipole antenna are shown in Fig. 3.20 with and

without the metal reflector.

The simulated S11 from HFSS indicates a bandwidth of greater than 20GHz as

shown in Fig. 3.22. The silicon chip with the dipole antennas sits over a FR4 PCB

material. The PCB is 600μm thick. The radiation is taken from the PCB side. This is

because the electromagnetic fields tend to go through the materials with higher

dielectric constant than that of air. The choice of the FR4 material is due to its lower

cost. There are other PCB materials with lower loss at 60-GHz but are more expensive


46

and hence not so desirable for low cost and mass markets. The overall efficiency of

the dipole antenna over the PCB is 50.11%. For silicon the dielectric constant and the

loss tangent is assumed to be 11.9 and 0.65 respectively. For the PCB, the dielectric

constant and loss tangent is 3.8 and 0.02 respectively [59].

Fig. 3.20: Simulated radiation pattern (a) Dual dipole with metal reflector. (b) Dual dipole

without metal reflector. (c) Single dipole with metal reflector. (d) Single dipole without

metal reflector.

Fig. 3.21: Normalized measured radiation patterns (a) Elevation (b) Azimuth.


47

Fig. 3.22: Simulated S11 for the dual dipole antenna.

3.5.1 Method of Images

Consider a unit positive charge at a distance of 𝑑 from an infinitely large

perfect electrical conductor (PEC). The boundary condition imposed by a perfectly

conducting sheet (𝜎𝑒→∞), is that the electric fields intersecting it should be

perpendicular, and the tangential component of the electric fields should be zero.

Since 𝜎𝑒→∞, the current density inside the PEC cannot be finite except when the

electric field inside the PEC is zero. The boundary condition is:

𝑛 × = 0, (3.9)

where is the unit vector normal to the PEC and is the electric field at the

boundary. These boundary conditions are similar to having a unit negative charge

placed at a distance 2𝑑 from the unit positive charge as shown in Fig. 3.23(a).

Similarly for a wire carrying electrical current 𝐼 at a distance of 𝑑 from an infinitely

large PEC, the magnetic field should be parallel to the surface of the PEC as shown in

Fig. 3.23(b). From Faraday’s law, zero electric field in a PEC implies that the time-

varying component of the magnetic field also vanishes. The boundary condition for

the magnetic field, , is


48

𝑛 × = 𝐽 , (3.10)

where, 𝐽 is the surface current density. This condition is similar to having a wire with

current flowing in opposite direction placed at a distance of 2𝑑 from the wire. Thus

method of images states that the fields produced by a current carrying conductor at a

distance 𝑑 over an infinite ground plane is identical to that of two current carrying

conductors separated by a distance of 2𝑑 and having currents in opposite directions.

Fig. 3.23: Image of a unit positive charge, and (b) image of a current carrying wire.

The method of images can hence be used to calculate the array factor of a

dipole antenna located over a ground plane. A dual dipole antenna placed at a distance

of a λ/4 from the ground plane is equivalent to two dual dipole antennas separated by a

distance of λ/2 as shown in Fig. 3.24. This results in an array gain of two, hence a 3

dB expected gain increment [60]. The overall array gain of four antennas is expected

to be 6 dB over a single dipole antenna, which matches with the simulated gain shown

in Fig. 3.20.

Fig. 3.24: Image of a dual dipole antenna over a ground plane.


49

3.6 Energy Harvesting Circuit design

The energy harvesting circuit uses four stages of cascaded CMOS rectifiers as

shown in Fig. 3.25 [61]. The input of the rectifier is connected to a loop antenna on the

PCB through a shunt-series capacitor matching network that harvests RF energy at

2.45 GHz. The impedance looking into the harvesting circuit varies as a function of

the RF input power. This design uses a fixed off-chip matching network for all levels

of the input RF power. Therefore, for some RF input powers, the harvesting efficiency

is not optimized because of the impedance mismatch. The output of the four-stage

CMOS rectifier is connected to an on-chip 3 nF capacitor for energy storage. With

four stages of rectification, a nominal output voltage in excess of 1.1 V is achieved.

Each rectifier stage consists of two NMOS/PMOS transistor pairs which are

successively turned on and off by the incoming RF sinusoid thereby pumping the

charge to the storage cap.

Fig. 3.25: Energy harvesting front-end circuit.

Since the rectifier output voltage can vary with 2.45 GHz RF input power level

as well as load current, a supply detection mechanism is necessary. This is to ensure

that the supply remains at a level high enough for the transmitter to operate properly.

Fig. 3.26(a) shows the supply detection mechanism state diagram. The transmitter

turns on when the harvested supply has reached 1.1 V and keeps transmitting until the

supply falls below 0.8 V, as shown in Fig. 3.26(b). This hysteresis allows sufficient

time for transmit operation, before the supply drops below the 0.8 V level due to the


50

storage capacitor discharge. In the case where enough harvested energy is available,

the transmitter can operate continuously so long as the harvested supply remains

above 0.8 V as shown in Fig. 3.26(c).

Fig. 3.26: Energy harvesting. (a) Supply detection mechanism. (b) Packet mode. (c)

Continuous mode.

The supply detection circuit is shown in Fig. 3.27. It consists of a band-gap

reference, Vdd level detector and extra high Vdd safety switch, all operating from the

rectifier output. Using two bandgap references, as shown in Fig. 3.27(a), we can detect

reference voltage settling. The first bandgap reference (BG1) directly connects to Vdd

and generates Vref = 0.5 V, whereas the second bandgap reference (BG2) operates

from 0.75 Vdd and generates 0.55 V output. Since BG2 is running from a lower supply,

it is guaranteed to settle slower than BG1 when Vdd is ramping up. Once BG2 output

exceeds Vref, a comparator issues ‘Ref ready’ signal. The Schmitt trigger circuit shown


51

in Fig. 3.27(b) is responsible for the supply level detection. The ‘Vdd ready’ output of

the comparator is asserted whenever 0.8 V < Vdd < 1.1 V. The circuit shown in Fig.

3.27(c) applies a leakage load current to the supply whenever Vdd exceeds 1.3 V. This

is to lower the harvested VDD voltage for the case when too much harvested energy is

available and the supply voltage is outside the reliability region.

Fig. 3.27: Supply detection. (a) Bandgap reference. (b) Vdd level detection. (c) Extra high

Vdd safety switch.

3.7 Measurements

A 0.9×1.8mm2 chip, including two on-chip dipole antennas, was fabricated in

TSMC 40nm GP technology with 9 metal layers. Two different versions of the chip

were fabricated with different baseband circuits. The first version focused on

achieving ultra-low power operation for NFC communication and was tested using

harvested energy on the transmitter side. For the second version of the chip, the digital

baseband was redesigned to meet the requirements of wireless MRI systems. As a

result, second version has more pins as compared to the first version. This version

was tested using non-magnetic batteries and its performance inside the MRI would be

discussed in detail in the next chapter. The RF performance of both the versions was

similar and the die micrographs are shown in Fig. 3.28.


52

Fig. 3.28: Die photo (a) first version for NFC application, (b) the second version for

wireless MRI application. The silicon is 1.8 mm × 0.9 mm for both versions.

Fig. 3.29: (a) Metal reflector facing the front-side. The radiation is through the PCB on the

back-side. (b) The plastic fixture used to hold the metal reflector. The harvest antenna

connector ports into the chip. (c) The silicon radio wire-bonded on a FR4 PCB material.

All the measurements include the metal reflector shown in Fig. 3.29 for both

the RX and TX. With NFC as the primary application, the transmitter and receiver

were placed at two distances of 5 and 10cm, as shown in Fig. 3.30(a) and (b). In all


53

link measurements, the transmitter energy was harvested at 2.45GHz using a loop

antenna differentially connected through an impedance matching network to the on-

chip harvesting circuit as shown in Fig. 3.25. Since the 5cm and 10cm links have

different power consumptions, the on-chip loading seen by the harvest loop antenna is

not constant. For some on-chip loading conditions the power transfer and hence

harvest efficiency (η) is near the optimum value. Since we used the same matching

network for all loading conditions, η changes as a function of the TX power

consumption. η was measured using the setup shown in Fig. 3.30(a) and measured η of

the matching network, PCB interconnects and the 4-stage CMOS rectifier is shown in

Fig. 3.31(a). The RF power source used for energy harvesting is a signal generator at

2.45GHz, which feeds a loop antenna that radiates RF energy to the harvesting loop

antenna that is connected to the harvesting block through off-chip impedance

matching.

The harvested power was measured by probing the voltage of the supply pin at

the storage cap, Cstorage, as shown in Fig. 3.25, using an oscilloscope. For example, at a

certain data rate the oscilloscope measured a harvested voltage of 0.9V. We could

deduce the TX average DC power consumption by disconnecting the harvesting

antenna and instead providing a DC power supply of 0.9V at the same pin that was

probed by the oscilloscope. Now we can easily measure the DC power consumption

by reading the DC current through the 0.9V supply. This would be identical to the

power obtained by harvesting that resulted in a 0.9V supply. The loss, Lharvest, between

the harvest antenna of the transmitter and the RF power source antenna was measured

to be 10dB and 8dB at 10cm and 5cm respectively. Therefore, the RF power needed at

2.45GHz at the source loop antenna to provide an average DC power of Ptx (dBm) for

the transmitter operation is:

𝑃RF_Source = 𝑃𝑡𝑥(dBm) +

1

η(dB) + 𝐿ℎ𝑎𝑟𝑣𝑒𝑠𝑡.

(3.

11)

For example, for a transmitter average DC power consumption of 1 mW (0

dBm), by looking at Fig. 3.31(a), the harvest efficiency is approximately 30% (- 4.5dB)

and therefore an RF source power of 14.5 and 12.5 dBm for 10 and 5 cm distance is


54

required. This power can be readily available from the PAs in handheld devices with

WLAN capabilities operating at ISM band of 2.45GHz [62].

Fig. 3.30: Measurement setup (a) Harvesting efficiency. (b) BER. (c) Pulse-width. (d)

Coherent time.

Fig. 3.31: (a) Harvest efficiency. (b) TX power vs. data rate. (c) Bit-error rate vs. data rate.

For the BER measurements, the transmitter and receiver were aligned at the

broadside direction of their respective dual-dipole, as shown in Fig. 3.30(b). A 127-bit

on-chip pseudo random bit sequence (PRBS) generator was used to generate 38Mb/s

to 2450Mb/s data stream with 6-bit programmable (N=0, … ,5) divide by 2N. For a

BER < 5×10–5, the two-element transmitter consumes 260µW to 11.9mW at a distance

of 10cm. At 5cm, a BER < 5×10-4 was achieved by turning on only one transmit


55

element, which consumes 100µW to 6.3mW. Fig. 3.31(b) shows the linear relationship

between power and data rate for 5cm and 10cm distances. The receiver, including the

baseband drivers, consumes 74mW. The BER was measured by saving the receiver

baseband output to an Agilent DSA90804A oscilloscope and processing the data in

Matlab. Fig. 3.31(c) shows the measured BER for different data rates at 5cm and 10cm

distances. An eye-diagram view of the received baseband pulses is shown in Fig. 3.32.

Since the clock recovery was not implemented on-chip, the 2.45GHz clock from the

RF power source was directly fed into the oscilloscope. In addition to the oscilloscope,

an Agilent N4903B J-BERT was used to perform BER measurement. Since J-BERT

had a built-in clock recovery, there was no need to provide an external clock.

Fig. 3.32: Oscilloscope eye diagram at 2.45 Gb/s at 10 cm (voltage scale: 100 mV/div, time

scale: 100 ps/div).

Fig. 3.30 (c) shows the measurement setup of the power spectral density of the

PRBS signal. The TX carrier frequency is around 56.5GHz. Fig. 3.33 shows an

approximate sinc function with a BW of 3.5GHz. This corresponds to a pulse width of

570ps. The next maxima on the left is 13dB lower in power which matches that of an

ideal sinc function. This is not the case for the maxima on the right, and that is because

of extra losses of the cables and the down-converter at higher frequencies. The


56

measurement confirms the nearly rectangular envelope of the 60GHz signal. Narrower

pulse widths were also measured, and the increased BW could be seen, but due to the

limitations of the measurement setup they did not compare well with sinc function

created with wider pulse widths.

Fig. 3.33: Power spectral density of a long PRBS.


57

Fig. 3.34: Beat frequency and coherence time measurements. (Voltage scale: 250 mV/div,

time scale: 10 ns/div).

The coherence time measurement setup is shown in Fig. 3.30(d). In this

measurement, the two transmit VCOs are continuously on without any modulation.

The two VCOs oscillate at a carrier frequency close to 56.5GHz. A horn antenna

captures the spatially combined VCO waveforms and feeds it to a down-converter. As

evident from the measurement shown in Fig. 3.34, the two VCOs did not run at the

same frequency. A beat frequency of roughly 1

104𝑛𝑠= 9.6𝑀𝐻𝑧 means the two VCOs

have a frequency difference of 19.2MHz. The measured coherence time is 13ns, which

is significantly larger than the minimum required. The system performance as

compared to other published work is shown in Table 3-1.

3.8 Summary

In this chapter, the design and measurements of fully integrated 2 × 2 OOK 60-

GHz transceiver with on-chip dipole antennas, while using energy harvesting in the

transmit mode, have been presented. The energy is harvested at 2.45 GHz with an

average efficiency of 30%. For a data rate of 2.45 Gb/s, the transceiver provides a

BER < 5 × 10-4 with an energy efficiency of 2.6 pJ/bit at a distance of 5 cm. At 10 cm,


58

with a BER < 5 × 10-5, the transmit energy efficiency is 4.9 pJ/bit for a data rate of

2.45 Gb/s. This radio has a small silicon area of 1.62 mm2. Therefore, this is a very

low cost replacement for NFC-type applications with high data rate and low power

requirements.

In the next chapter, the design details and the measurement results for wireless

MRI system using this transceiver would be discussed.

Table 3-1: Performance Comparison [62]

VLSI 2010

[29]

ISSCC

2012

[63]

JSSCC

2010

[43]

RFIC 2013

This Work

Technology 90nm 65nm 90nm 65nm SOI 40nm GP

Carrier Frequency

(GHz)

60 60 60 60 60

Modulation FSK QPSK OOK-NRZ OOK-RZ OOK-RZ

Antenna On-board,

folded

dipole

Bond wire On-board,

folded

dipole

On-chip,

folded

dipole

On-chip, dipole

Data Rate (Mb/s) 2000 2620 3000 500 to 2200 38 to 2450

Distance for max

Data Rate (cm)

41 5 4.5 4 5 10

BER < 10-12 NA 10-3 10-5 5 × 10-4 5 × 10-5

Power Consumption

(mW)

TX: 280

RX 150

TX: 160

RX: 233

TX: 183

RX: 103

TX+RX: 98 TX: 0.10-

6.26

RX: 74

TX: 0.26-

11.9

RX: 74

Energy/bit for max

Data Rate (pJ/bit)

215 150 95 45 32.8 35.1

Tx Energy/bit (pJ/bit) 140 61 61 NA 2.6-2.56 6.8- 4.9

Die Area (mm2) 1.26 2.86 1.1 5.89 1.62

TX×RX /Shared

Antenna

1×1 / No 1×1 / Yes 1×1 / No 1×1 / No 2×2 / Yes

Energy Harvesting No No No No Yes

ScalableTX Power No No No Yes Yes

RX/TX Non-

Coherence Tolerant

Yes No Yes Yes Yes

Chapter 4 – Design and Evaluation of Wireless MRI System

59

CHAPTER 4

Design and Evaluation of Wireless MRI

System

4.1 Background

As discussed in Chapter 1, conventional MRI relies on a wired connection

between the receiver coil array and the external processing circuitry to generate

accurate images. To improve image quality, the number of receiver coil elements are

increased and separate receiver coil arrays are used for different parts of the body.

This results in increased cabling complexity. Furthermore, baluns and radio frequency

(RF) traps are required for each channel, and cables must be routed to minimize coil

interactions. This increases the operation and maintenance costs. Moreover, these

receiver coil arrays are heavy and cumbersome and can be intimidating and ill-fitting

for children. The coil setup time can occupy a significant fraction of the total exam

time. Consequently, removing these cables from the receiver coils will lead to a more

cost effective and time efficient system.

Fig. 4.1: Proposed architecture for the mm-wave wireless MRI system along with the

baseband required for digitization of MRI coil data.

In the past, a number of architectures have been proposed to enable wireless

MRI [5] - [6][7][8][9][10] for minimizing or eliminating the use of cables. All of these

past efforts used microwave frequencies up to 3 GHz, and generic protocols such as

802.11b or MIMO that are intended for long-range communication over distances of


60

10 m to 100 m. Such generic long range communication protocols are sub-optimal

solutions for wireless MRI in terms of power consumption and size. This is because

typical MRI bore diameters vary from 60 cm to 70 cm. And, depending on a patient’s

physical attributes and the part of the body to be imaged, the distance between the coil

array and the magnet bore/edge can vary from 10 cm to 50 cm.

In this work, we propose a custom millimeter (mm) wave transceiver

architecture that is specifically designed to meet the requirements of wireless MRI

data rates while minimizing power consumption and size. A block diagram is shown

in Fig. 4.1 in which the mm wave radio provides a short-range (10-50 cm) link within

the MRI bore, and a fiber optic link transports data beyond the magnet to the console

for image reconstruction. The mm-wave data transmitter (TX) is located on the MRI

receiver coil assembly and the mm-wave data receiver (RX) is embedded inside the

MRI system bore. With the on-chip dipole antennas, the designed 60-GHz radio

achieves a raw bit error rate (BER) of 10-6 for a distance of 10 cm, which is much

better than the raw BER specification of 10-2 for a typical WiFi, 802.11n system [64],

[65], or 10-3 for a Bluetooth system [66]. For a TX-RX separation of 50 cm, the

receiver is coupled with an external horn antenna and an LNA to compensate for

higher signal loss at a longer transmission distance.

4.2 System Design Challenges

The static magnetic field inside the MRI bore tube (~1.5T) is ten thousand

times stronger than the earth’s magnetic field. As a result, any magnetic material in the

vicinity of the MRI bore tube experiences a very strong magnetic pull and may even

damage the MRI machine as illustrated in Fig. 4.2. Thus designing a system for an

MRI room poses certain challenges that are normally not encountered in a normal lab

environment. For example:

Power Supplies: Conventional power supplies cannot be used inside the

MRI room as their consoles are made up of iron/steel which are highly

magnetic. The proposed MRI system solves this problem by using non-

magnetic batteries (PowerStream - PGEB-NMO53040).


61

Metallic Board Holders: During the initial evaluation phase of a system in

a lab, metallic holders are used to hold the test boards of the design as

shown in Fig. 4.3(a). Typically, these holders are made out of iron/steel

and hence can’t be used inside the MRI room. This was resolved by

designing a custom holder made up of Lego® bricks which is MRI safe as

shown in Fig. 4.3(b).

Proximity between Oscilloscopes and Design under Test (DUT): To

evaluate a system’s true performance, it is desirable to minimize the

distance between the measurement equipment such as an oscilloscope and

the design under test. Commercially available oscilloscopes can’t be taken

inside the MRI room because they have iron/steel enclosures. As a

consequence, the distance between the DUT placed inside the MRI bore

tube and the measurement equipment in the MRI console room is ~5 m.

Signal transmission over such long distances using standard SMA cables

results in a high transmission loss which is undesirable. Furthermore,

magnetic field gradients inside the MRI room might corrupt the signal. To

address these issues, a fiber optic transceiver (Firecomms 1Gbps LC

transceiver) system is used to transfer data between the DUT and the

measurement equipment.


62

Fig. 4.2: Image demonstrating the strength of magnetic fields in the MRI room.

Fig. 4.3: (a) Metallic holder for the board, and (b) custom designed Lego® holder with the

designed mm-wave transceiver.

4.3 System Design

4.3.1 Design Overview

Wireless MRI data link consists of two main parts: a 60-GHz radio, and a fiber

optic link. The radio transceiver operates in mm-wave frequencies with an RF carrier

at 60 GHz using on-off key (OOK) modulation. The use of non-coherent modulation


63

like OOK simplifies the system architecture as the TX and RX are not required to be

phase synchronized. As the absolute phase of the system is not a concern, it relaxes the

linearity constraints for the power amplifier (PA) design, thus allowing the use of a

more efficient non-linear PA. This reduces system power consumption allowing

operation with small non-magnetic batteries. An earlier version of this transceiver was

used for short range, high data-rate, near field communication (NFC) system [11] and

was discussed in detail in chapter 3. The TX power of the new design scales from 1.3

mW to 14.0 mW as the data rate is varied from 200 Mb/s to 2500 Mb/s, while the RX

consumes a fixed DC power of 76 mW and works up to 2500 Mb/s. The 60-GHz radio

occupies a die area of 0.9 × 1.8 mm2 (1.62 mm2) and was designed and fabricated

using a TSMC 40-nm 9-metal CMOS GP process.

4.3.2 60-GHz Radio

The 60-GHz radio IC is comprised of a TX and an RX block. An on-chip

dipole antenna is shared between the TX and RX blocks via a transmit-receive (TR)

switch. The radio architecture along with signal waveforms at different points inside

the TX and RX blocks is shown in Fig. 4.4. Significant changes were made from the

prior design [11] to target MRI systems. As most of the commercially available analog

to digital convertors (ADCs) for MRI provide a low voltage differential signaling

(LVDS) output, an LVDS receiver was added to the radio baseband. Further, the

baseband was modified to provide support for both return-to-zero (RZ) and non-

return-to-zero (NRZ) signaling protocols. The previous version only supported RZ.

This was done as the ADCs output is normally in NRZ format. Additional circuitry

was added to enable clock and data synchronization between multiple transmitters.

This allows multiple transmitters to be operated at the same time without interference,

using built-in time division multiplexing (TDM) logic. Changes were also made to the

power management circuit to minimize the leakage power. On the RF side, a dedicated

TX was designed without the TR switch thus leading to a higher drain efficiency for

the PA. For the benefit of the readers a brief overview of different circuit blocks is

provided here. For a detailed circuit analysis of different blocks the readers are

advised to refer to chapter 3.


64

4.3.2.1 Transmitter Block

The transmitter consists of a digital baseband followed by an RF front end. The

digital data is fed to the transmitter baseband circuitry which supports both NRZ and

RZ signaling protocols. The width of the RZ pulse can be programmed to assume any

value from 250 ps to 1200 ps using a 6-bit pulse width controller (PWC). This RZ

pulse is then fed to the transmitter radio RF front-end. The RF front-end consists of

two identical TX elements each consisting of a voltage controlled oscillator (VCO), a

power amplifier (PA), and an on-chip dipole antenna. To improve the on-chip antenna

gain, the fabricated die was thinned down to 100 µm to minimize substrate losses and

surface waves. The thinned die is mounted on a 600 µm thick FR4 printed circuit

board (PCB) for measurements.

The VCO is a cross-coupled NMOS LC-oscillator and drives a class E/F2, odd

PA with a simulated drain efficiency of 29%. The two VCOs and PAs are switched on

and off, simultaneously, by the PWC modified RZ bit stream to generate impulse radio

ultra-wide-band (IR-UWB) waveforms at the two dipole antennas.

Fig. 4.4: (a) mm-wave radio architecture, and (b) signal waveforms at different points

inside the TX and RX.

4.3.2.2 Receiver Block

The receiver consist of an RF front end followed by a digital baseband. The RF

front end consists of an on-chip dipole antenna followed by a 3-stage transformer-

coupled low noise amplifier (LNA). The LNA is an NMOS common source amplifier.


65

The third stage of the LNA sums the received signals from the two dipole antennas,

resulting in an improved signal sensitivity (point E). A passive AC-coupled self-mixer

is used to extract the OOK modulation envelope (point F) which is then fed to the RX

baseband. The RX baseband is a 3-stage DC-coupled inverter chain followed by a

common-source amplifier with programmable resistive load. The last stage of the

baseband is designed to drive an external 50 Ω load.

4.3.3 The Fiber Optic Link

The baseband data from the 60-GHz receiver in the MRI bore is transmitted

over a fiber optic link to the MRI console room. This removes any possible RF

coupling to the signal and also eliminates the need for RF traps on the signal path. The

CMOS level single-ended baseband data from the 60-GHz RX, is converted to LVDS

using a CMOS to LVDS translator (ON Semiconductor NB4N527). The LVDS data is

then fed to an optical transmitter (Firecomms 1-Gbps LC transmitter). A plastic optical

fiber (POF) carries the optical signal to the MRI console room. An optical receiver

(Firecomms 1-Gbps LC receiver) recovers the electrical signal which is captured using

a high speed sampling scope (Keysight Infiniium DSA91304A). This data is processed

in MATLAB to recover the final test data. Though Firecomms system is designed to

support data rates up-to 1 Gb/s, initial testing revealed that there was considerable

inter-symbol interference for data rates beyond 500 Mb/s. Thus testing inside the

magnet was limited to 500 Mb/s data rates.

4.3.4 System Link Budget

According to the Friis transmission equation [67], the received power 𝑃𝑟 at

distance R, for carrier wavelength λ, and transmitted power 𝑃𝑡 is given by:

𝑃𝑟 = 𝑃𝑡𝐺𝑡𝐺𝑟 (

𝜆

4𝜋𝑅)

2

, (4.1)

where 𝐺𝑡 and 𝐺𝑟 are the transmit and receive antenna gains, respectively. The

minimum received power required for a desired signal-to-noise ratio, SNR, is given by:

𝑃𝑟 = 𝑘𝑇 ∙ 𝐵 ∙ 𝐹𝑟 ∙ 𝑆𝑁𝑅, (4.2)


66

where 𝑘, 𝑇, 𝐵 and 𝐹𝑟 are Boltzman constant, temperature in Kelvin, bandwidth of the

receiver (bandwidth of the transmit pulse should be smaller than or equal to 𝐵), and

the overall receiver noise factor, respectively. By substituting (4.1) into (4.2), the

communication distance R becomes:

𝑅 = √𝑃𝑡 ∙ 𝐺𝑡 ∙ 𝐺𝑟

𝑘𝑇 ∙ 𝐵 ∙ 𝐹𝑟 ∙ 𝑆𝑁𝑅 (

𝜆

4𝜋) .

(4.3)

The choice of SNR is dictated by the required bit-error rate (BER). For non-

coherent OOK modulation, the receiver SNR should be better than 14 dB to achieve a

raw BER < 10-6 [68] (excludes error correction or encoding). In our implementation,

the Tx/Rx antenna gains of the 2-dipole system are simulated to be 5.2 dB, hence by

reciprocity 𝐺𝑡 = 𝐺𝑟 = 3.31 (5.2 dB). Since each PA radiates 4 mW, the total

transmitted power through both PAs gives 𝑃𝑡 = 8 mW. The receiver has noise factor

𝐹𝑟 = 1479 (31.7 dB), and bandwidth B = 8 GHz. The TX bandwidth is programmable

through the PWC with the maximum bandwidth at 8 GHz (corresponding to the

narrowest pulse-width of 250 ps). Using 𝑘 = 1.38 x 10-23 J/K, T= 300 K, 𝑆𝑁𝑅 =25.1

(14 dB), and 𝜆 = 5 mm, in (4.3) results in a link distance of 𝑅 = 10.6 cm which

barely meets the lower end of coil to magnet-bore distance requirement.

Even if the transmit antenna and power are maintained, link distance can still

be increased by a higher gain RX antenna and preamplifier. A circular horn antenna

(Ducommun ACH-14115-02) has 15 dB gain, 3-dB bandwidth of 10 GHz (58 GHz -

68 GHz) and 3-dB beam width of 24 degrees. By aligning the RX antenna with the

horn waveguide, the link distance increases to R=33 cm, but is still short of our 50 cm

goal. Next, consider adding a 60-GHz LNA (Spacek Labs SL6010). This LNA has a

gain of 𝐺𝐿𝑁𝐴 = 31.6 (15dB), noise factor, 𝐹𝐿𝑁𝐴 = 3.98 (6 dB) and a 10 GHz (55 GHz-

65 GHz) 3-dB bandwidth. With the external horn antenna and LNA, the receiver noise

factor is now given by

𝐹 = 𝐹𝐿𝑁𝐴 +

𝐹𝑟 − 1

𝐺𝐿𝑁𝐴 ∙ 𝐺𝑟 ,

(4.4)

resulting in the effective receiver noise factor F=18.1 (12.6 dB), and yields a link

distance of R=297 cm, well beyond the 50 cm goal.


67

4.4 System Evaluation

4.4.1 System Measurements inside the MRI Room

The MRI experiments were carried out on a GE 1.5 T Signa scanner. The aim

of these experiments is to demonstrate that a high data rate, mm-wave link can be

established reliably inside the MRI magnetic bore in the presence of a strong static

magnetic field. Gradient coils and RF pulses are not enabled during these tests. Initial

link tests employed an on-chip 7-bit pseudo random bit sequence (PRBS) generator to

generate sufficient bit error statistics. Subsequent tests demonstrated image transfer

with specialized hardware.

4.4.1.1 Short Range Link Verification for 10 cm

Short range data integrity was tested by conducting experiments with the MRI

system comprising of the on-chip dipole antennas alone (without the horn antenna).

The short link setup and relative positioning of the TX and RX are shown in Fig. 4.5

and Fig. 4.6(a), (b) respectively. An RF signal generator (KEYSIGHT N9310A)

located in the console suite provided the clock signal (via a plastic optical fiber) for

baseband PRBS generation in the 60-GHz transmitter. . For the proof-of-concept

experiments, the 60-GHz transmitter and receiver were placed either parallel or

perpendicular to the axis of the magnets as shown in Fig. 4.7(a) or Fig. 4.7 (b),

respectively. Experiments in the two orthogonal orientations were conducted to

evaluate performance degradation due to Hall Effect [69] in static magnetic fields.

Data from these experiments is transmitted by the receiver via the fiber optic link to

the scanner console room for further processing and BER calculations. Both

orientations achieved BER < 0.6 × 10-6 thus demonstrating that the system

performance is independent of its orientation inside the magnet. Fig. 4.8 shows the

screen image of the 7-bit PRBS sequence as captured on the sampling scope. The

measured raw BER for different data rates and orientations at 10 cm is included in

Table 4-1.


68

Fig. 4.5: A block diagram showing the test setup for link verification inside the MRI room

at a distance of 10 cm.

Fig. 4.6: (a) The test setup showing TX and RX alignment, and (b) a magnified view of

PCB mounting showing the 60-GHz TX and the 60-GHz RX chip.

Fig. 4.7: MRI test setup placed (a) in-line with the direction of the static magnetic field, and

(b) perpendicular to the direction of the static magnetic field.


69

Fig. 4.8: Differential 7-bit PRBS sequence as captured on the sampling scope at 500 Mb/s

for a distance of 10 cm (voltage scale: 200 mV/div, time scale: 50 ns/div).

4.4.1.2 Link Verification at 25 cm, 50 cm and 65 cm

For link distances greater than 10 cm, additional gain elements were added to

the receiver chain as discussed in section 4.3.4. A horn antenna placed before the 60-

GHz radio receiver improved the link distance to 25 cm. However, the measured worst

case BER of 9.25 × 10-5 is higher than the expected BER of 10-6. We speculate that

this is due to the misalignment between the horn antenna, TX and RX. Furthermore,

the transmitter center frequency of 56.5 GHz falls outside the antenna 3-dB gain

bandwidth. This reduces antenna gain and increases receiver noise figure, and

therefore adversely impacts the BER.


at a distance of 25 cm.

Fig. 4.10: The test setup showing the TX and RX alignment at 25 cm.


71

Fig. 4.11: (a) Magnified view of RX aligned to the output of the horn antenna. (b) The test

setup for 25 cm link placed inside the MRI bore.

The block diagram of the setup is shown in Fig. 4.9. The TX, RX and Horn

assembly were mounted on a custom designed slider made of MRI safe materials (Fig.

4.10). The slider allowed for a variable distance between the TX and RX without

disturbing the system component alignment. Only axial orientations were tested due to

space restrictions (Fig. 4.11). We note that there was no detectable performance

degradation due to Hall Effect in the short range experiments. We assume that this

holds for long range experiments as well.


at a distance of 50 cm, and 65 cm.

The TX-RX distance was further increased to 50 cm and then to 65 cm by

adding an external LNA to the horn antenna output as shown in Fig. 4.12, Fig. 4.13,


72

and Fig. 4.14. The measured BER for different date rates and link distances is

summarized in Table 4-1. The sampling scope used for capturing the BER had a

limited memory depth resulting in one million data points. Even after multiple data

captures, no error was observed for any chunk of one million points suggesting that

the BER is <1.0 × 10-6.

Fig. 4.13: The test setup showing the TX and RX alignment at 50 cm and 65 cm.

Fig. 4.14: (a) Magnified view of RX aligned to the output of the LNA-horn antenna

assembly. (b) The test setup for 50 cm and 65 cm link placed inside the MRI bore.


73

Table 4-1: BER for Different Distance and Data Rates

TX-RX

Distance

BER

@200 Mb/s

BER

@250 Mb/s

BER

@ 500 Mb/s

10 cm 1x10-6 1x10-6 6.4x10-6

10 cm* 1x10-6 1x10-6 4.8x10-6

25 cm 3.2x10-5 3.9x10-5 9.2x10-5

50 cm 1x10-6 1x10-6 1x10-6

65 cm 1x10-6 1x10-6 1x10-6

* Setup was placed perpendicular to the external magnetic field as shown in Fig. 4.7(b). All other

measurements were taken by placing the setup along the external static magnetic field.

4.4.1.3 Image Transfer over 50 cm

Having established link reliability, synthesized MRI image transmission was

attempted over a link distance of 50 cm. The PRBS function was replaced by a base-

band processing unit implemented in a FPGA board (Xilinx Kintex-7 FPGA KC705),

and the data recovery processing was implemented in MATLAB. The baseband

architectural details are highlighted in Fig. 4.15. A previously acquired 256 × 256

MRI image was reformatted to 9 sub-blocks of fixed precision 16-bit real and

imaginary pixels. The image blocks were serialized to a bit sequence, and converted to

the RZ format. Lastly, run length limited (RLL) coding was applied. The RLL

operations prevent a DC offset build-up in the analog RF receiver circuitry which

could corrupt the data.

After the image encoding, the encoded bit streams were downloaded to the

FPGA board memory. Upon external triggering, the binary data was transferred at

200MHz over POF to the mm-wave link inside the magnet, where it was envelope

detected in the 60-GHz receiver at 50 cm separation, returned over a second POF link

to the console room, and captured by a high speed sampling oscilloscope (Infinium

DSA91304A).

In MATLAB post-processing of the scope data, clock information was

recovered from the oversampled digital samples, allowing correct bit alignment, and


74

subsequent RLL decoding of the bit stream and reformatting into image blocks. The

segmented MRI image and the reconstructed image are shown in Fig. 4.16. The

received image is an exact replica of the original image and no pixel errors were

observed.

(a)

(b)

Fig. 4.15: (a) The baseband processing unit implemented on the transmitter side, and (b)

the baseband processing unit implemented on the receiver side for image processing.

Fig. 4.16: (a) The MRI image broken down into 9 image blocks before transmitting

through the system. (b) The received image obtained by assembling the individually

transmitted blocks. .


75

4.4.2 System Measurements outside the MRI Room

The 60-GHz radio can achieve much higher data rates, but was limited in-bore

by the fiber optic channel limit of 500 Mb/s. To assess the true limits, the 60-GHz link

was deployed outside the MRI, and the fiber optic cables were replaced by low loss

SMA cables (Samtec RF316-01SP1-01SP1-0607) and high bandwidth baluns (ETS

PI-102). Under these optimized signal transmission conditions, data rates up to 2.5

Gbps were analyzed, and time domain multiplexing was assessed. These experiments

suggest that the 60-GHz link we have designed is capable of data transfer rates of up

to 2.5 Gbps.

4.4.2.1 Link Verification at 10 cm, 25 cm, 50 cm and 65 cm

BER tests were first repeated for 10-65 cm link distances using identical

methods to those discussed in sections 4.4.1.1 - 4.4.1.2 with on-chip PRBS. The BER

versus data rate for different distances is shown in Fig. 4.17. The measured BER is

identical to that measured inside the MRI bore at the lower data rates. For all

distances, BER increases as the data rate is increased. This is primarily due to finite

receiver bandwidth and a higher inter-symbol interference at higher data rates.


76

Fig. 4.17: Bit-error rate versus data rate for 10 cm, 25 cm, 50 cm and 65 cm.

4.4.2.2 Field of View of Horn Antenna

As mentioned in section 4.3.4, the horn antenna used for measurement at 50

cm and above has a 3 dB beamwidth of 24 degree. Hence, even if the transmitter is not

aligned to the receiver horn antenna, the system should still meet the BER

requirements. This is important for an MRI system because a perfect alignment

between the transmitter on a patient’s body and the receiver embedded inside the MRI

bore tube may not always be possible. Thus, with a fixed distance of 50 cm from the

receiver, the transmitter was moved sideways as shown in Fig. 4.18 and measured

BER is shown in Fig. 4.19. Beyond a lateral distance of 7.5 cm, a significant

degradation in BER was observed and it went below the required threshold value of

1×10-3 even for lower data rates.


77

Fig. 4.18: A block diagram showing the test setup verifying horn antenna’s field of view.

Fig. 4.19: Bit-error rate versus data rate as the transmitter is moved sideways with TX-RX

distance of 50 cm.


78

4.4.2.3 Stress Test of the Link at 50 cm

Our BER statistical floor was limited to 10-6 by our instrumentation. As

discussed in the section 4.3.4, the available link margin could possibly result in a

much superior BER than the target value of 10-6. Thus BER measurements were

carried out using a bit error rate tester (Tektronix BERTSscope Si 17500C, courtesy

Marvell). The experiments were carried out using NRZ PRBS-7 at 2 Gb/s and 2.5

Gb/s. The real time eye diagram corresponding to a measured BER of 8.25×10-12 at 2

Gb/s for a distance of 50 cm is shown in Fig. 4.20.

Fig. 4.20: (a) Real time eye diagram measured using the BERTScope at 2 Gb/s, and (b) the

measured BER at 2 Gb/s using PRBS-7.

Fig. 4.21: (a) Real time eye diagram measured using the BERTScope at 2.5 Gb/s, and (b)

the measured BER at 2.5 Gb/s using PRBS-7.


79

At 2.5 Gb/s, the BER dropped to 2.42×10-5 as shown in Fig. 4.21. The drastic

degradation in BER after 2 Gb/s matches with the general waterfall curves for BER

v/s energy per bit to noise spectral density 𝐸𝑏/𝑁0 [68]. The 𝐸𝑏/𝑁0 is related to SNR as

𝐸𝑏

𝑁𝑜= 𝑆𝑁𝑅 ∙

𝐵𝑤

𝐹𝑠 ,

(4.5)

where 𝐹𝑠 is the system data rate, 𝐵𝑤 is the channel bandwidth. As the system data rate

goes up, 𝐸𝑏/𝑁0 falls, resulting in a higher BER. As data rates increase, the errors

increases due to inter symbol interference, since more bits are packed closer and sent

through the channel.

4.4.2.4 Time Domain Multiplexing (TDM)

As mentioned in section 4.3.2.1, the TX baseband supports TDM. Data from

multiple transmitters can be received at a single RX without any interference. The test

setup is shown in Fig. 4.22, and Fig. 4.23, where two TXs are placed with a lateral gap

of 5 cm between them such that both are at a distance of 50 cm from the RX.

Fig. 4.22: The block diagram showing the test setup for multiple transmitters at a distance

of 50 cm from the receiver.

Fig. 4.24(a) shows the screen shot of the initial received data for a data rate of

250 Mb/s. There is a significant overlap in the received data resulting in severe data

corruption. Fig. 4.24(b) shows the same data after TDM is enabled in the TX baseband.


80

There is no overlap between the data from different TX, thus the data from each TX

can be reliably deciphered at the RX.

Fig. 4.23: The test setup for multiple TX to demonstrate time division multiplexing (TDM)

at a data rate of 250 Mb/s and distance of 50cm.

Fig. 4.24: Different shaped markers showing the received data corresponding to different

transmitters when the (a) TDM block is turned OFF (voltage scale: 100 mV/div, time scale:

1 ns/div), and (b) when the TDM block is turned ON (voltage scale: 100 mV/div, time scale:

2 ns/div).

4.4.3 Power Consumption for Different Signaling Schemes

The 60-GHz radio supports multiple signaling schemes as mentioned in section

4.3.2.1. Fig. 4.25 shows the TX power consumption for different modulation schemes

as the data rate is varied from 200 Mb/s – 2500 Mb/s. The 60-GHz TX consumes DC

power only while transmitting a data bit value of “1” because the TX is off while it is


81

transmitting a data value of “0”. Thus, by using the PWC with RZ OOK, the power

consumption of the TX scales with the data rate.

Fig. 4.25: Transmitter dc power consumption versus data rate for different signaling

schemes.

In addition to the data rate, the TX power consumption scales linearly with the

pulse width. As shown in Fig. 4.4(b), the transmitter consumes power only during

pulse width 𝑇 when both the VCO and PA are turned on. It consumes only leakage

power outside the pulse width 𝑇. The average DC power consumption of a single

pulse is

𝑃𝑠_𝑝𝑢𝑙𝑠𝑒 =

𝑇

𝜏𝑃𝑝𝑢𝑙𝑠𝑒 = 𝑇 ∙ 𝑑𝑎𝑡𝑎𝑟𝑎𝑡𝑒 ∙ 𝑃𝑝𝑢𝑙𝑠𝑒 ,

(4

.6)


82

where 𝑃𝑝𝑢𝑙𝑠𝑒 is the sum of both VCO and PA power consumptions and is the bit

duration. A pulse width of 400 ps was used for measurements outside the MRI room

resulting in a TX power consumption of 14mW from a 1.1 V supply at 2500 Mb/s.

This corresponds to an energy per bit of 5.6 pJ/bit for the TX. When the PWC is

turned off, the TX power consumption is almost constant for different data rates. The

slight increase in power consumption with data rate is due to an increase in dynamic

power consumption of the digital circuits.

As the receiver is always on, the power consumption is independent of the

modulation type and the data rate. The RX consumes 76.8 mW of power from a 1.3 V

supply.

4.5 Discussion

This work demonstrated that a high data rate, mm-wave link can be established

reliably inside the MRI magnetic bore in the presence of strong static magnetic fields.

Gradients and RF pulses were not enabled because the cabling needed for testing the

wireless MRI system would expose the mm-wave radio to induced voltages if

gradients and RF pulses were present. We note that these cables would not exist whilst

the wireless system is in use. In agreement with this, when a gradient echo (GRE)

sequence was run with all cabling removed, the 60-GHz radio performance remained

unaltered after later reconnection. We also expect that the OOK modulation, which

does not require phase synchronous detection, will be more robust in the presence of

gradient field perturbations, and optionally, data could be timed to transmit only in

MRI RF-silent intervals if RF overload were an issue.

The minimum SNR is dictated by the required bit-error rate (BER). For non-

coherent OOK, the receiver SNR must exceed 14 dB to get a raw BER < 10-6 [68]

(before error correction or encoding). By comparison, the raw BER specification for

802.11n WiFi and Bluetooth are 10-2 and 10-3 respectively [64] - [66]. From our link

analysis, this was achieved with a 15dB link margin at 50 cm using the horn/LNA

additions to the receiver. A 6 dB increase in receive antenna gain or decrease in

receiver noise figure increases the link distance by a factor of 2. The target BER below


83

10-6 provides extra robustness for frequency mismatch or antenna misalignment,

flexibility in trading link margin for lower TX power, and simplified error correction

coding.

The mm-wave radio offers significant flexibility in deployment. The design

uses highly directional, linearly polarized, on-chip dipole antennas. Fig. 4.26 shows a

concept in which one radio services a 4-coil module with quad ADCs and preamps. As

a result, a 32-element coil could be formed using eight of these modules. The dipoles

can be placed orthogonal to each other in nearest neighbor radios to significantly

reduce cross talk. HFSS simulation predicts a coupling of -35dB between two

orthogonal transmitters separated by a distance of 5mm. This coupling reduces to -

43dB as the distance between the transmitters is increased to 1cm. At 10 Mb/s for 500

kHz decimated MRI BW, each module would stream 40 Mb/s, but channel encoding

and error correction will increase this rate further. For example, a ½ convolution code

would result in an effective data rate of 80 Mb/s.


84

Fig. 4.26: A 32-element receiver coil with 4-coil module sharing a single processing unit.

Multiple 4 coil-modules are placed such that the RF transmitters in adjacent modules are

orthogonal to each other, enabling multiple spatial streams.

Furthermore, the 60-GHz radio supports time division multiplexing (TDM) so

multiple transmitters can be time interleaved to avoid data corruption in scenarios

where the data from multiple transmitters is being collected at a single receiver. The

radio also provides multiple options for clock and data synchronization between

different transmitters. CMOS level clock signals can be physically routed to multiple

transmitters and a 10-bit delay element in the baseband can be programmed to

synchronize the data exiting different transmitters. This could be done as a one-time

calibration for a particular set of transmitter receiver pairs. Finally, by connecting a

loop antenna to the LVDS input port in the transmitter, a reference clock can be

harvested by sending a pilot signal at any frequency, for example 2.4 GHz, inside the

bore [11].

The proposed system should be easy to integrate in MRI systems. The small

size transmitter with on-chip dipole antennas can be embedded inside the receiver coil


85

array without influencing the MRI performance. When the radio chip is encapsulated

in a dielectric lens assembly with a metal reflector as shown in Fig. 4.27 , the radio

antenna gain is simultaneously enhanced to 9.1 dBi [70]. On the receiver side, the

horn-LNA-60-GHz radio assembly can be easily embedded inside the bore magnet,

but similar gains and steerable sensitivity are ultimately best achieved by a beam

steering 60-GHz receive array. Finally as the signal from the magnet to the console

can be transferred over the fiber optic link, it will not couple and corrupt any existing

RF signal in the magnet.

Fig. 4.27: (a) 60-GHz radio with on-chip dipole placed inside an MRI safe package, and (b)

its HFSS simulated radiation pattern with maximum gain of 9.1dBi. The metal acts as a

reflector and the dielectric as a lens for enhanced gain.

Chapter 5 – Conclusions

86

CHAPTER 5

Conclusions

5.1 Conclusions

We have proposed a 60-GHz short-range radio system that offers a low power

and scalable solution for wireless MRI digital links. As compared to analog

transmission, digital transmission offers better noise immunity, stability and flexibility.

The 60-GHz radio supports high data rates generated by commercially available ADCs

designed to meet MRI dynamic range and SNR requirements. The use of fiber optic

link to transport the data beyond the magnet to the console for image reconstruction

minimizes any possible data corruption from existing MRI signals. Fig. 5.1 illustrates

a potential implementation of the proposed wireless MRI system with wireless

receiver coil placed on the patient’s body and its associated wireless receiver with

fiber optic connections embedded inside the MRI bore tube.

Fig. 5.1: Proposed implementation for the designed wireless MRI system.

Chapter 5 – Conclusions

87

5.2 Future Work

The 60-GHz radio is just a first step towards the development of a wireless

MRI system. Significant effort is needed for low power digitizing and decimation of

the MRI coil data before transmission. A conventional MRI receiver chain of preamp,

quad pipeline ADC and serializer (e.g. TI-TL5500, ADS5263, DS32EL0421) alone

results in a DC power budget of 600mW/element, and could easily reach 1W/channel

once other control electronics are added. If a receive chain <100mW/channel were

realizable by passive mixer down-conversion and SAR or continuous time sigma delta

ADCs, one could envision a 32-128 channel wearable array supplied by non-magnetic

battery or even wireless power delivery [71].

In present systems, the above 100 MS/s pipeline ADC sample rate results in

raw data of 1.6 Gb/s for a single array element – a rate that would overwhelm the

802.11ac standard. By locally decimating to 500 KS/s (a high speed MRI case), 20 bit

I/Q and 50% Rx duty cycle, a much more reasonable data rate of 10 Mb/s per element

is achieved. Thus the required data throughput for a receiver array would vary from

320 Mb/s to 1280 Mb/s for 32 to 128 elements. These rates are borderline feasible for

single-stream 802.11ac (5 GHz) but achievable in the 60 GHz 802.11ad standard and

by our technology. An open question is whether the WiFi standards, which employ

phase-coherent modulations, are robust to MRI RF bursts, potential gradient

waveform perturbations of VCOs, and whether they can co-ordinate data streaming

with the MRI pulse sequences.

Although it has been over 2 decades since wireless MRI coils were first

proposed, our on-chip MRI system is a first-of-a-kind demonstration that proves

beyond doubt that wireless MRI is very much feasible. However, widespread adoption

of wireless MRI would be possible only if the subsystems such as pre-amps,

ADCs and serializer become more power efficient than today. We believe that our

experimental demonstration of wireless MRI will encourage future research in

improving the power efficiency of these subsystems and thereby contribute to the

development of wireless MRI systems.

Bibliography

[1] [Online]. Available: http://www.magnetic-resonance.org.

[2] [Online]. Available: http://radiopaedia.org/.

[3] [Online]. Available: http://mri-q.com/.

[4] [Online]. Available: https://www.maximintegrated.com/en/app-

notes/index.mvp/id/4681.

[5] Y. Murakami, T. Tetsuhiko and Y. Etsuji, "Nuclear magnetic resonanace

inspection apparatus and its method.". US Patent 5,384,536, 24 Jan 1995.

[6] E. Boskamp et al., "Wireless RF Module for an MR imaging system.". US Patent

2003/020619A1, Nov 2003.

[7] G. Scott and K. Yu, "Wireless transponders for RF coils: systems issues," in

Proceedings of the 13th Annual Meeting of ISMRM, Miami, Florida, USA,

2005.

[8] M. J. Riffe et al., "Using on-board microprocessors to control a wireless MR

receiver array," in Proceedings of the 17th Annual Meeting of ISMRM,

Honolulu, Hawaii, USA, 2009.

[9] J. Wei et al., "A realization of digital wireless transmission for MRI signals based

on 802.11 b," Journal of Magnetic Resonance , vol. 186, no. 2, pp. 358-

363, 2007.

[10] O. Heid et al., "CUTTING THE CORD-WIRELESS COILS FOR MRI," in Proc.

Intl. Soc. Mag. Reson. Med, 2009.

[11] M. Taghivand, K. Aggarwal, Y. Rajavi and A. S. Y. Poon, "An Energy

Harvesting 2 × 2 60 GHz Transceiver With Scalable Data Rate of 38–2450

Mb/s for Near-Range Communication," IEEE Journal of Solid-State

Circuits, vol. 50, no. 8, pp. 1889-1902, Aug. 2015.

[12] TEXAS INSTRUMENTS, "Quad Channel 16-Bit, 100MSPS High-SNR ADC,"

ADS5263 datasheet, May. 2011-Rev. Jan. 2013.

[13] TEXAS INSTRUMENTS, "DS32EL0421, DS32ELX0421 125 - 312.5 MHz

FPGA-Link Serializer with DDR LVDS Parallel Interface," DS32EL0421

datasheet, May. 2008-Rev. Apr. 2013.

[14] [Online]. Available:

http://www.cisco.com/c/en/us/products/collateral/wireless/aironet-3600-

series/white_paper_c11-713103.html.

[15] Aruba Networks, "802.11AC IN-DEPTH," Sunnyvale, 2014.

[16] [Online]. Available: http://www.dell.com/Support/Article/us/en/19/QNA43752.

[17] Y. Zeng et al., "A first look at 802.11ac in action: Energy efficiency and

interference characterization," in 2014 IFIP Networking Conference, 2-4

Bibliography

90

June 2014.

[18] S. Joshi et al., "Effect of antenna spacing on the performance of multiple input

multiple output LTE downlink in an urban microcell," International

Journal of Wireless & Mobile Networks, vol. 4, no. 6, pp. 175 - 188, Dec.

2012.

[19] S. K. Saha et al., "Power-throughput tradeoffs of 802.11n/ac in smartphones," in

IEEE Conference on Computer Communications (INFOCOM), 2015, April

26 2015-May 1 2015.

[20] [Online]. Available: http://www.radio-electronics.com/info/wireless/wi-fi/ieee-

802-11ad-microwave.php.

[21] Agilent Technologies, "Wireless LAN at 60 GHz - IEEE 802.11ad Explained,"

Agilent Technologies, May 2013.

[22] [Online]. Available: http://www.sibeam.com/.

[23] [Online]. Available: http://wilocity.com/.

[24] [Online]. Available: http://www.tensorcom.com/.

[25] M. Boers et al., "A 16 TX/16 RX 60 GHz 802.11ad chipset with single coaxial

interface and polarization diversity," in IEEE ISSCC Dig. Tech. Papers,

Feb. 2014.

[26] T. Tsukizawa et al., "A fully integrated 60 GHz CMOS transceiver chipset based

on WiGig/IEEE802.11ad with built-in self calibration for mobile

applications," in IEEE ISSCC Dig. Tech. Papers, Feb. 2013.

[27] J. Lee et al., "A low-power fully integrated 60 GHz transceiver system with OOK

modulation and on-board antenna assembly," in IEEE ISSCC Dig. Tech.

Papers, Feb. 2009.

[28] K. Okada et al., "A full 4-Channel 6.3 Gb/s 60 GHz direct-conversion transceiver

with low-power analog and digital baseband circuitry," in IEEE ISSCC

Dig. Tech. Papers, Feb. 2012.

[29] T. Mitomo et al., "A 2 Gb/s-Throughput CMOS transceiver chipset with

inpackage antenna for 60 GHz short-range wireless communication," in

IEEE ISSCC Dig. Tech. Papers, Feb. 2012.

[30] E. Cohen et al., "A CMOS bidirectional 32-element phased-array transceiver at

60 GHz with LTCC antenna," in IEEE Radio Frequency Integrated

Circuits Symp. (RFIC), Jun. 2012.

[31] J. M. Gilbert et al., "A 4-Gbps uncompressed wireless HD A/V transceiver

chipset," IEEE. Micro, vol. 28, no. 2, pp. 56-64, 2008.

[32] M. Taghivand and A. S. Y. Poon, "Supporting and enabling circuits for antenna

arrays in wireless communications," Proc. IEEE, vol. 100, p. 2207–2218,

2012.

[33] S. Emami et al., "A 60 GHz CMOS phased-array transceiver pair for multi-Gb/s

wireless communications," in IEEE ISSCC Dig. Tech. Papers, Feb. 2011.

[34] A. Balankutty et al., "A 12-element 60 GHz CMOS phased array transmitter on

LTCC package with integrated antennas," in IEEE ASSCC Dig.Tech.

Bibliography

91

Papers, Nov. 2011.

[35] H. Krishnaswamy and H. Hashemi, Integrated Beamforming Arrays in mm-Wave

Silicon Technology: 60 GHz and Beyond, Berlin, Germany: Springer,

2008.

[36] A. Natarajan et al., "A fully-integrated 16-Element phased-array receiver in siGe

BiCMOS for 60-Ghz communications," IEEE J. Solid-State Circuits, vol.

46, no. 5, pp. 1059-1075, May 2011.

[37] K. -J. Koh et al., "A millimeter-wave (40-45 Ghz) 16-Elementvphased-array

transmitter in m SiGe BiCMOS technology," IEEE J. Solid-State Circuits,

vol. 44, no. 5, pp. 1498-1509, May 2009.

[38] A. Valdes-Garcia et al., "A SiGe BiCMOS 16-element phased-array transmitter

for 60 GHz communications," in IEEE ISSCC Dig. Tech. Papers, Feb.

2010.

[39] Y. Yu et al., "A 60 GHz phase shifter integrated with LNA and PA in 65 nm

CMOS for phased array systems," IEEE J. Solid-State Circuits, vol. 45, no.

9, pp. 1697-1709, Sep. 2010.

[40] E. Cohen et al., "A bidirectional TX/RX four-element phased array at 60 GHz

with RF-IF conversion block in 90-nm CMOS process," IEEE Trans.

Microw. Theory Techn., vol. 58, no. 5, pp. 1438-1446, May 2010.

[41] X. Yi et al., "A 57.9-to-68.3 GHz 24.6 mW frequency synthesizer with in-phase

injection-coupled QVCO in 65 nm CMOS," in IEEE ISSCC Dig. Tech.

Papers, Feb. 2013.

[42] L. Kong et al., "A 50 mW-TX 65 mW-RX 60 GHz 4-element phasedarray

transceiver with integrated antennas in 65 nm CMOS," in IEEE ISSCC

Dig. Tech. Papers, Feb. 2013.

[43] A. Siligaris et al., "A low power 60-GHz 2.2-Gbps UWB transceiver with

integrated antennas for short range communications," in Proc.IEEE RFIC

Symp., Jun. 2013.

[44] L. Kong et al., "A mW mm-Wave phased-array transmitter in 65 nm CMOS," in

Proc. Symp. VLSI Circuits, Jun. 2012.

[45] S. D. Kee et al., "The class E/F family of ZVS switching amplifiers," IEEE Trans.

Microw. Theory Techn., vol. 51, no. 6, pp. 1677-1690, Jun. 2003.

[46] A. Valdes-Garcia et al., "A 60 GHz class-E power amplifier in SiGe," in IEEE

ASSCC Dig. Tech. Papers, Nov. 2006.

[47] N. Kalantari et al., "A 19.4 dBm, Q-band class-E power amplifier in a 0.12 um

SiGe BiCMOS process," IEEE MWCL, vol. 20, pp. 283-285, May 2010.

[48] O. T. Ogunnika et al., "A 60 GHz class-E tuned power amplifier with in 32 nm

SOI CMOS," in IEEE Radio Frequency Integrated Circuits Symp. (RFIC),

Jun. 2012.

[49] N. Talwalkar et al., "An integrated 5.2 GHZ CMOS T/R switch with LC-tuned

substrate bias," in IEEE ISSCCDig. Tech. Papers, Feb. 2003.

[50] A. Babakhani et al., "A 77 GHz phased array transceiver with on-chip dipole

Bibliography

92

antennas: receiver and antennas," IEEE J. Solid-State Circuits, vol. 41, no.

12, pp. 2795-2806, Dec. 2006.

[51] M. H. Barakat et al., "On the design of 60 GHz integrated antennas on m SOI

technology," in Proc. of IEEE Int. SOI Conf., 2007.

[52] C. -H. Wang et al., "A 60 GHz transmitter with integrated antenna in m SiGe

BiCMOS technology," in IEEE ISSCC Dig. Tech. Papers, Feb. 2006.

[53] A. Arbabian et al., "A 94 GHz mm-Wave-to-baseband pulsed-radar transceiver

with applications in imaging and gesture recognition," IEEE J. Solid-State

Circuits, vol. 48, no. 4, pp. 1055-1071, Apr. 2013.

[54] N. G. Alexopoulos et al., "Substrate optimization for integrated circuit antennas,"

IEEE Trans. Microw. Theory Techn., vol. 83, p. 550, 1983.

[55] G. M. Rebeiz et al., "Millimeter-wave and terahertz integrated circuit antennas,"

Proc. IEEE, vol. 80, no. 11, pp. 1748-1770, 1992.

[56] D. B. Rutledge et al., "Imaging antenna arrays," IEEE Trans. Antennas Propag.,

vol. 30, pp. 535-540, Jul. 1982.

[57] F. Gutierrez et al., "On-chip integrated antenna structures in CMOS for 60 GHz

WPAN systems," IEEE J. Sel. Areas Commun., vol. 27, no. 8, pp. 1367-

1378, Oct. 2009.

[58] D. P. Neikirk et al., "Far-infrared imaging antenna arrays," Appl. Phys.Lett., pp.

203-205, Feb. 1982.

[59] A. P. Toda et al., "60-GHz substrate materials characterization using the covered

transmission-line method," IEEE Trans. Microw. Theory Techn., vol. 63,

no. 99, pp. 1-13, Mar. 2015.

[60] C. A. Balanis, Antenna Theory Analysis and Design, Third edition, John Wiley &

Sons Inc., 2005.

[61] S. Mandal et al., "Low-power CMOS rectifier design for RFID applications,"

IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 54, no. 6, pp. 1177-1188,

Jun. 2007.

[62] H. Wang et al., "A 60-GHz FSK transceiver with automatically-calibrated

demodulator in 90-nm CMOS," in VLSI Circuits Symp. Dig., Jun. 2010.

[63] J. Lee et al., "A Low-Power Low-Cost Fully-Integrated 60-GHz Transceiver

System with OOK Modulation and On-Board Antenna Assembly," IEEE J

Solid-State Circuits, vol. 45, no. 2, pp. 264-275, Feb. 2010.

[64] F. Peng, J. Zhang and W. E. Ryan, "Adaptive Modulation and Coding for IEEE

802.11n," in IEEE Wireless Communications and Networking Conference,

11-15 March 2007.

[65] A. Goldsmith, Wireless communications, Cambridge university press, 2005.

[66] C. Gehrmann, P. Joakim and B. Smeets, Bluetooth Security (Artech House

Computer Security Series), Colorado: Artech House Publishers, 2004.

[67] H. T. Friis, "A note on a simple transmission formula," in Proc. IRE, 1946.

[68] Q. Tang et al., "BER performance analysis of an on-off keying based minimum

Bibliography

93

energy coding for energy constrained wireless sensor applications," in

IEEE International Conference on Communications, 2005.

[69] E. Hall, "On a New Action of the Magnet on Electric Currents," American

Journal of Mathematics, vol. 2, no. 3, pp. 287-292, 1879.

[70] L. Shafai, "Dielectric Loaded Antennas," in Encyclopedia of RF and Microwave

Engineering, John Wiley & Sons, Apr. 2005.

[71] K. Byron et al., "RF Gated Wireless Power Transfer System," in Proceedings of

the 23rd Annual Meeting of ISMRM, Toronto, Ontario, Canada, 2015.

a wireless mri system using millimeter wave …

Documents