60-GHz CMOS MICRO-RADAR SYSTEM-IN-PACKAGE FOR VITAL SIGN AND
VIBRATION DETECTION
By
TE-YU KAO
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
2013
1
© 2013 Te-Yu Kao
2
To my parents and fiancée
3
ACKNOWLEDGMENTS
I would like to express my sincere appreciation to my advisor Dr. Jenshan Lin for his
guidance, mentoring, and encouragement throughout my Ph.D. life. I am also thankful to my
Ph.D. committee members, Dr. William Eisenstadt, Dr. Jane Gu, and Dr. Jennifer Rice for the
valuable advice and feedback from many different points of view.
Being a member of Radio Frequency Circuits and Systems Research (RFCSR) group at
UF, Dr. Lin has always been so patient and supportive to every one of us, providing a very
helpful and friendly environment for graduate study. I would also like to thank my best
colleagues, Dr. Yan Yan, Dr. Xiaogang Yu, Dr. Tze-Min Shen, Dr. Austin Chen, Dr. Minqi
Chen, Gabriel Reyes, Jianxuan Tu, Chien-Ming Nieh, Ron-Chi Kuo, Taesong Hwang, Raul
Chinga, Changyu Wei, and Jaime Garnica for the useful discussion and brainstorming together in
the lab. In addition, I am thankful to our previous group member Prof. Changzhi Li from Texas
Tech University for the helpful guidance and discussion. Finally, I would also like to thank Dr.
Kenneth O, Dr. Dongha Shim, Dr. Ning Zhang, Dr. Chuying Mao, Dr. Hsin-Ta Wu, Dr.
Wuttichai Lerdsitsomboon, Dr. Tie Sun, Rounan Han, Dr. Chieh-Lin Wu, Jason Seol, and YangHun Yun for their guidance and help in the early stage of my Ph.D. life.
For the fabrication of the 60-GHz micro-radar system-in-package, I would like to
acknowledge the single-die bumping and flip-chip process sponsored by Mr. Terence Collier
from CVInc, Richardson, Texas, USA. We also acknowledge the 90-nm CMOS chip fabrication
by United Microelectronics Corporation (UMC), Hsin-Chu, Taiwan R.O.C., and the microwave
laminates provided by Rogers Corporation, Rogers, CT, USA.
4
TABLE OF CONTENTS
page
ACKNOWLEDGMENTS ...............................................................................................................4
LIST OF TABLES ...........................................................................................................................7
LIST OF FIGURES .........................................................................................................................8
LIST OF ABBREVIATIONS ........................................................................................................12
ABSTRACT ...................................................................................................................................14
CHAPTER
1
INTRODUCTION ..................................................................................................................16
1.1
Millimeter-wave Doppler Radar in CMOS..................................................................16
1.1.1
Doppler Radar...................................................................................................16
1.1.2
System Implementation ....................................................................................19
1.2 Vibration Detection and Quadrature Architecture .......................................................21
1.2.1
Optimal and Null Detection Points...................................................................21
1.2.2
Complex Signal Demodulation ........................................................................23
1.3 Vital Sign Detection .....................................................................................................23
1.4 Millimeter-wave Packaging and Integration ................................................................26
2
SYSTEM DESIGN AND INTEGRATION ...........................................................................29
2.1
2.2
2.3
2.4
3
Overview ......................................................................................................................29
Sensitivity and Radar Received Power ........................................................................30
Design Consideration for IF Stage ...............................................................................32
Floor Plan and Flip-Chip Transition ............................................................................33
CIRCUIT COMPONENT DESIGN .......................................................................................34
3.1
3.2
3.3
3.4
3.5
3.6
Inductor ........................................................................................................................34
Radar Receiver Front-end Design ................................................................................37
3.2.1
LNA ..................................................................................................................37
3.2.2
Active Mixer and RF VCO ...............................................................................39
Radar Transmitter Front-end Design ...........................................................................41
3.3.1
Passive Balun ....................................................................................................42
3.3.2
TX Driver .........................................................................................................43
IF Quadrature VCO and Passive Mixer .......................................................................44
CMOS Radar Chip Overview ......................................................................................46
Flip-chip Integration and PCB Patch Antenna.............................................................47
3.6.1
Transition Design and Impedance Match .........................................................48
3.6.2
Patch Antenna ...................................................................................................51
5
4
EXPERIMENTAL RESUTLS ...............................................................................................53
4.1
4.2
4.3
4.4
4.5
Millimeter-wave CMOS Transceiver Measurement ....................................................53
IF Quadrature Ring VCO .............................................................................................54
Patch Antenna Test ......................................................................................................56
Radar Transmitted Power Test.....................................................................................59
Mechanical Vibration Detection ..................................................................................63
4.5.1
Quadrature Channel Test ..................................................................................63
4.5.2
Sensitivity to Small Vibration ..........................................................................64
4.6 Heartbeat and Respiration Detection ...........................................................................65
5
ANALYSIS AND IMPROVEMENT.....................................................................................69
5.1
Analysis on 60-GHz Vital Sign Detection ...................................................................69
5.1.1
Respiration Detection Improvement by Two-tone Monitoring ........................71
5.1.2
Heartbeat Detection ..........................................................................................75
5.2 Proposed Time-domain Recovery Algorithm ..............................................................75
5.2.1
Analysis on Quadrature Baseband Outputs ......................................................76
5.2.2
MATLAB Program Implementation ..............................................................78
5.2.3
Experimental Results ........................................................................................81
5.2.4
Discussion .........................................................................................................83
5.3 Broadband Antenna on LTCC System-in-Package .....................................................86
5.3.1
Introduction ......................................................................................................86
5.3.2
Sequential Rotation Patch Antenna Array ........................................................88
5.3.3
Vital Sign Detection .........................................................................................93
SUMMARY ...................................................................................................................................95
APPENDIX: MATLAB CODING OF TIME-DOMAIN RECOVERY ALGORITHM ...........97
LIST OF REFERENCES .............................................................................................................103
BIOGRAPHICAL SKETCH .......................................................................................................109
6
LIST OF TABLES
Table
3-1
page
Simulated inductor performance at 60 GHz ......................................................................35
7
LIST OF FIGURES
Figure
page
1-1
Illustration of typical vital sign detection using Doppler radar system. ............................17
1-2
A 5.8-GHz Doppler radar system integrated on a PCB with two 22 patch antenna
arrays for TX and RX. .......................................................................................................19
1-3
The concept of small displacement detection at optimum and null points of Doppler
radar system. ......................................................................................................................22
1-4
Comparison of vital sign detection using different radar frequencies ...............................25
1-5
Typical wire-bonding packaging configuration. ................................................................26
2-1
Block diagram of the 60 GHz CMOS micro-radar system including transceiver chip
in 90nm CMOS, TX and RX patch antennas, and flip-chip integration. ...........................29
2-2
Sensitivity estimation of the Doppler radar receiver .........................................................30
3-1
Simulation of a 1.5-turn, 95-pH inductor along with the surrounding ground plane ........34
3-2
Microphotograph of the on-chip inductors used in the 60-GHz front-end ........................35
3-3
Simulation of the isolation between two closely placed inductors as they are in the
actual on-chip situation. .....................................................................................................36
3-4
RX front-end (60 GHz to 6 GHz) including the 5-stage LNA, single-ended mixer,
and 54-GHz VCO (Bias and LO distribution details not shown). .....................................38
3-5
Microphotograph showing the cascode portion of the layout and vertical access to
the power grid. ...................................................................................................................38
3-6
54-GHz RF LO generation and distribution ......................................................................40
3-7
Microphotograph showing the 54-GHz LO distribution network from VCO to upand down-convert mixers. ..................................................................................................40
3-8
TX front-end (6 GHz to 60 GHz) using the double-balanced up-convert mixer,
balanced loads (balun), and three-stage driver at 60 GHz. ................................................42
3-9
Lumped-element-modeled transformer balun with differential to single-ended
impedance conversion. .......................................................................................................42
3-10
IF stage (6 GHz to dc) including the quadrature ring VCO, IF LO buffers, and
passive mixers. ...................................................................................................................44
8
3-11
Design of the delay cells (D1 and D2 in Figure 3-10) with two tuning mechanisms
(Vp and Vb). ........................................................................................................................45
3-12
Simulated four output phases (LOI+, LOI-, LOQ+, and LOQ- in Figure 3-10)
generated by the quadrature ring VCO. .............................................................................46
3-13
Microphotograph of the 60 GHz CMOS micro-radar........................................................47
3-14
Flip-chip transition design between the 60-GHz CMOS radar chip and PCB patch
antennas on RT/duroid 5870 laminate. ..............................................................................48
3-15
Impedance analysis of the transition at 55 GHz before and after the flip-chip process. ...49
3-16
Microphotograph of the flip-chip area on RT/duriod 5870 surface ...................................50
3-17
Simulated patch antenna s-parameter after the flip-chip packaging ..................................51
3-18
Simulated patch antenna pattern after the flip-chip packaging. .........................................52
4-1
Measured down-conversion gain (60 GHz to 6 GHz) versus RF input frequency ............53
4-2
Measured up-conversion (6 GHz to 60 GHz) gain compression and Pin (differential)
versus Pout (single-ended) curve .........................................................................................54
4-3
Measured single-ended output spectrum of the quadrature ring VCO ..............................55
4-4
Antenna return loss (S11) measurement. ............................................................................56
4-5
Patch antenna test structure with probing/flip-chip G-S-G-S-G area zoomed-in. .............57
4-6
Probe-based measurement setup for the broadside radiation patterns. ..............................58
4-7
Radiation patterns of the single-patch antenna ..................................................................58
4-8
Measured and simulated realized gain spectrums at zenith. ..............................................59
4-9
The final system configuration of the 60-GHz micro-radar system-in-package
including the CMOS transceiver chip, two PCB patch antennas, and dc biasing
through blue wires..............................................................................................................60
4-10
Experimental setup for the TX output power of CMOS transceiver chip. ........................61
4-11
Photo of the experimental setup to test TX transmitted power of CMOS micro-radar
chip. ....................................................................................................................................61
4-12
The screenshot of the received power Pr = -82.24 dBm on the spectrum analyzer ...........62
4-13
I and Q baseband outputs test of the micro-radar system ..................................................63
9
4-14
The experimental results of small mechanical vibration detection ....................................64
4-15
Heartbeat detection using the 60-GHz radar when the target holds the breath at 0.3 m
away. ..................................................................................................................................66
4-16
Measurement results of the respiration detection at 60 GHz as fr = 15 beat/minute, D
= 0.3, and mr is slightly varied in the two tests. .................................................................67
5-1
Theoretical plots of Jn(a) = Jn(4πm/λ) versus vibration amplitude m ................................71
5-2
Simulated output spectrum of vital sign detection at 60 GHz ...........................................72
5-3
Vital sign detection results as the person breathes shallowly at 0.3 m in front of the
radar ...................................................................................................................................73
5-4
Vital sign detection results as the person breathes deeply at 0.3 m in front of the
radar ...................................................................................................................................74
5-5
Non-linear input-output mapping when the vibration is comparable to λ at 60 GHz. .......76
5-6
Non-linear input-output mapping of I and Q channels when the vibration is
comparable to λ at 60 GHz.................................................................................................77
5-7
Time-domain recovery technique by simply monitoring I and Q baseband outputs
when the vibration is comparable to λ at 60 GHz. .............................................................78
5-8
Illustration of the continuous Flip and Follow operations. ................................................79
5-9
Simplified flow chart of the time-domain recovery algorithm. .........................................80
5-10
Respiration detection outputs before and after the recovery algorithm is applied. ...........82
5-11
Recovered respiration peak compared to the original spectrum in Figure 4-16 (B) ..........82
5-12
Respiration detection outputs before and after the recovery algorithm is applied. The
subject inhaled for 2 s, exhaled for 2 s, paused for 3 s, and repeated the cycle.................83
5-13
CSD spectrum outputs before and after the recovery algorithm is applied. ......................84
5-14
Duplicate of Figure 5-13 showing the consecutive Follow periods indicated by the
recovery algorithm .............................................................................................................85
5-15
Spectrum of the waveform marked by blue circle in Figure 5-14 which shows the
correct heartbeat rate detection result. ...............................................................................85
5-16
Measured and simulated return loss of a single patch PCB antenna as the
manufacturing variation is present. ....................................................................................88
10
5-17
Layer profile presents the LTCC system-in-package including 11 metal layers (L1–
L11) and the FR4 board with a slot for CMOS chip. ..........................................................89
5-18
Broadband antenna design on LTCC .................................................................................90
5-19
Measured and simulated return losses (S11) and TX/RX isolation (S12) of the antenna
array. ..................................................................................................................................91
5-20
Realized gain and axial ratio (AR) spectrums of the broadband sequential rotation
patch antenna array at zenith..............................................................................................92
5-21
Radiation patterns of the patch antenna array.. ..................................................................92
5-22
Top-view of the final system assembly. The flip-chip-integrated CMOS radar chip
and surface-mounted bypass capacitors are placed on the other side of LTCC. ...............93
5-23
CSD output spectrum of vital sign detection using the broadband patch antenna array
on LTCC ............................................................................................................................94
11
LIST OF ABBREVIATIONS
λ
Wavelength
ADC
Analog-to-digital conversion
BGA
Ball grid array
BPM
Beat per minute
CMOS
Complementary-symmetry metal-oxide-semiconductor
CPW
Coplanar waveguide
CS
Common-source
DSP
Digital signal processing
EM
Electromagnetic
FFT
Fast Fourier transform
fMAX
Power-gain cutoff frequency
fT
Unity-current-gain frequency
IC
Integrated circuit
IF
Intermediate frequency
LNA
Low noise amplifier
LO
Local oscillator
MEMS
Micro-electro-mechanical systems
MM-wave
Millimeter-wave
NF
Noise figure
P1dB
1-dB gain compression point
PA
Power amplifier
PCB
Printed circuit board
Psat
Saturated output power of PA/driver in TX
RX
Receiver
12
RF
Radio frequency
SGH
Standard gain horn antenna
SiP
System-in-package
SoC
System-on-chip
TX
Transmitter
VCO
Voltage controlled oscillator
13
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
60-GHz CMOS MICRO-RADAR SYSTEM-IN-PACKAGE FOR VITAL SIGN AND
VIBRATION DETECTION
By
Te-Yu Kao
May 2013
Chair: Jenshan Lin
Major: Electrical and Computer Engineering
The dissertation begins with basic concepts of Doppler radar and motivations of 60-GHz
design. Compared to previous works at lower frequencies, the benefits of shorter wavelength are
explained by theoretical analysis. State-of-the-art Doppler radar systems are discussed. Chapter 1
also lists challenges associated with high operating frequency such as CMOS circuit performance
and loss, package and antenna transition, and strong non-linear Doppler phase modulation from
both hardware and signal processing points of view.
Chapter 2 describes the system design considerations such as receiver architecture,
sensitivity, and layout floor plan. The detail of each circuit block is investigated in Chapter 3,
introducing inductor EM modeling, RF transceiver front-end, IF stages, and flip-chip integration
with PCB patch antennas. Chapter 4 shows the experimental results including on-wafer
measurement and antenna tests. Detection of small mechanical vibration and human vital sign
are successfully demonstrated. In Chapter 5, theoretical analysis is provided in detail to explain
the difficulties vital sign detection at 60 GHz. A detection technique monitoring both the
fundamental and second harmonic of respiration is proposed to improve the detection accuracy
of respiration. In addition, a time-domain signal recovery algorithm is proposed and tested to
help the detection of target movement comparable to wavelength. Finally, a circularly polarized
14
sequential-rotation antenna array is implemented on LTCC (low-temperature co-fired ceramic) to
increase the antenna bandwidth. As the process and manufacturing variations are often present in
mm-wave systems, wide antenna bandwidth is able to cover the possible frequency drift and
increase the system yield in mass production.
The summary is provided in Chapter 6. This work demonstrates the first vital sign
detection by the flip-chip-integrated CMOS micro-radar at 60 GHz. The shorter wavelength
offers significant area reduction and flexibility in system integration. The compact, low-cost
CMOS system can be embedded in portable devices such as the smart phone and tablet for daily
healthcare and vibration monitoring, as well as deployed in a large sensor network for many
other applications.
15
CHAPTER 1
INTRODUCTION
1.1 Millimeter-wave Doppler Radar in CMOS
Recent progress on non-contact vital sign and vibration detections have been made based
on microwave Doppler radar system [1]-[6]. Compared to other methods of detection such as
laser-based sensors and interferometers [7], simple architecture of Doppler radar realized by
integrated circuit and system usually makes it a cost-effective and low-power solution. It is also
useful in various situations such as longer distance, low visibility and through-wall detections
[8]. Current laser displacement sensors [9]-[11] offer excellent precision down to tens of
nanometers at a higher cost and within short detection range (usually less than 1 m), and a
measured displacement of 10 μm and vibration velocity of 27.7 mm/s have been reported by
utilizing microwave interferometry and sophisticated signal processing methods [7]. Lately the
frequency and amplitude of two-tone sinusoidal vibration can be distinguished and accurately
measured by using non-linear Doppler phase modulation [8] [12].
1.1.1 Doppler Radar
Non-contact vital sign detection has been found to be the important application of Doppler
radar system and drawn increasing attention. It proves to be a safe, low-cost, and effective way
to monitor the heartbeat and respiration without physical contact. A typical vital sign detection
system using Doppler radar is shown in Figure 1-1 [3] where an un-modulated signal T(t) is
transmitted with the amplitude normalized to unity:
T (t )  cos(2 ft  vco )
(1-1)
where f and ϕvco are the frequency and phase noise. T(t) is reflected and phase modulated by the
target displacement x(t), which is the chest-wall movement in this case. The baseband output B(t)
can be expressed as [13]:
16
ADC
FFT
&
DSP
x(t)
d0
TX
T(t)
B(t)
Radar
Transceiver
R(t)
RX
Figure 1-1. Illustration of typical vital sign detection using Doppler radar system.
 4 x(t )

B(t )  Gs  cos 
 t 
 

(1-2)
where λ is wavelength of T(t), variable Gs is defined as total system gain, and ϕt is the total
residue phase accumulated in the circuit and transmission path. Assuming x(t) is much smaller
than λ and ϕt is at odd multiples of /2, the system shows approximately linear transfer function
near optimal detection points. For example, as ϕt = -/2, B(t) can be simplified by small angle
approximation and expressed as :
B(t )  Gs
4 x(t )

.
(1-3)
As x(t) << λ, the baseband output B(t) is proportional to the target displacement x(t) , and it can
be sampled by an analog-to-digital convertor (ADC) for further digital signal processing.
Equation (1-3) indicates one of the motivations to design the system at 60 GHz. The
shorter λ in the denominator provides a higher “system demodulation gain” to distinguish small
displacement at a longer distance away. Ideally the linear input-output mapping can track
arbitrary movement of x(t) as long as x(t) is much smaller than λ. It is potentially useful to detect
small (m range) vibration of objects such as insects, acoustic devices, and micro-electro-
17
mechanical systems (MEMS). It can serve as a low-power, low-cost monitoring network for
industrial and security applications.
From the viewpoint of total system gain Gs which relates (1-1) and (1-2), it is usually
determined by antenna gains, power loss during the reflection and propagation (distance), and
transceiver circuits. In radar design, moving the operating frequency up to millimeter-wave (mmwave) range theoretically achieves larger radar received power by using less antenna area, which
can be seen in the analysis as follows. Generally the antenna gain G increase with frequency for
the same antenna effective area Ae [14]:
G
4 Ae
2
.
(1-4)
For a simple flat-plate reflector perpendicular to the line of sight (LOS) at far-field, the radar
cross section σ is approximately [15]:

4 T 2
(1-5)
2
where T is the actual area of the plate. Radar range equation can be used to estimate the received
power under far-filed condition for simple analysis [14]:
GG   
Pr
 t r 
Pt
4  4 R 2 
2
(1-6)
where Pt and Pr are the transmitted and received power. Gt and Gr are the gain of TX
(transmitter) and RX (receiver) antennas, and R is the distance between the target and radar. If
(1-4) and (1-5) are plugged into (1-6), the received power can be estimated by:
At ArT 2
Pr  Pt
 4 R4
18
(1-7)
where At and Ar are the effective area of TX and RX antennas. The equation shows apparent
advantage of short  if other parameters remain the same and air absorption is negligible in a few
meters range. Comparing 6-GHz and 60-GHz radar systems for example, Pr at 60 GHz is
theoretically 102 times higher than that at 6 GHz, and 1/10 antenna area is used for TX and RX.
1.1.2 System Implementation
Doppler radar system can be implemented by discrete components on printed circuit board
(PCB) or instrument level for testing and proof of concepts [4] [8] [16]. Figure 1-2 shows a 5.8GHz Doppler radar system integrated on a PCB with two 22 patch antennas array for TX and
RX [8], and it can be used for both non-contact vital sign and vibration detection. Compared to
the board level implementation, SoC (system-on-chip) or SiP (system-in-package) realization are
usually desired in terms of cost and system integration. Several single-chip Doppler radar
transceivers on CMOS have been successfully developed for non-contact vital sign detection [3]
[17] [18]. To increase the circuit operating frequency above tens of GHz, III-V compound
semiconductor (GaAs) is traditionally used for its superior performance at high frequencies, and
SiGe hetrojunction bipolar transistors (HBT) process offers another high-performance siliconbased alternative whose fT/fMAX can be nearly 200/300 GHz in the 130 nm process.
Figure 1-2. A 5.8-GHz Doppler radar system integrated on a PCB with two 22 patch antenna
arrays for TX and RX.
19
The state-of-the-art, single-chip Doppler radar systems are implemented on SiGe based
processes to achieve high conversion gain and better noise performance [6] [19]. Nevertheless,
the economies-of-scale of CMOS technology with the ability to integrate DSP circuits makes it a
very low-cost and highly desired platform. For a single transistor, fT and fMAX are usually used to
evaluate the performance of the device. FT is the unity-current-gain frequency which is often
used to estimate the analog circuit bandwidth at low frequency. FMAX is called the power-gain
cutoff frequency, representing the transistor limitation at high operating frequency. At the
frequencies above fMAX, the device is basically passive. Attributing to the improved transistors
performance in scaled CMOS process, fT/fMAX reaches above 120/150 GHz in 90-nm technology
and around 200/300 GHz in 45-nm technology, and it is expected to be further improved in more
advanced CMOS technologies such as 32-nm and 16-nm technologies [20].
Increasing number of research works and significant progress on mm-wave CMOS circuits
and systems have been reported for wireless communication, radar, and imaging [22]-[26], and
the unlicensed frequency band near 60 GHz is one of the interests. Single-chip, mm-wave
Doppler radar transceiver proves to be feasible in CMOS technology [21], however, vital sign
detection by a low-cost Doppler radar SiP fully-integrated with antennas has not been
investigated at this frequency range. Other performance limitations such as TX transmitted
power, components loss, and high flicker inherent in CMOS process need to be considered and
overcome especially from the radar point of view. In this work, the design considerations in
system architecture, circuit components, packaging, and baseband demodulation techniques are
studied to realize the radar system at 60 GHz.
20
1.2 Vibration Detection and Quadrature Architecture
As mentioned in the previous Chapter 1.1,  of radar signal is a key factor in determining
the sensitivity to small vibration. It is shown in (1-3) that the smaller  gives higher level of
baseband output B(t) and improves the “demodulation gain” of the Doppler radar system. As x(t)
and other conditions remain the same, the use of shorter  helps distinguish small displacement
at a longer distance. A single-channel, 94-GHz CMOS Doppler transceiver chip with two horn
antennas is potentially to detect small movement [21], however the system has limit of null
detection point issue which will be explained in the following paragraphs.
1.2.1 Optimal and Null Detection Points
Figure 1-3 illustrates the concept of small displacement detection using Doppler radar
system. When the total residue phase in (1-2) is at odd multiples of /2, the vibrating target is at
the optimal detection point as x1(t). As long as x1(t) is small enough compared to the  of T(t),
the system shows approximately linear transfer function, and the baseband output B1(t) can be
easily demodulated by FFT to obtain the spectrum target vibration. To detect the displacement
(peak-to-peak) value of the vibration x1(t), calibration is usually needed to determine Gs in (1-2)
at a fixed frequency and distance. Again ideally the system is able to map any movement of x1(t)
to baseband output B1(t) with minimal distortion in the approximately linear region.
If the displacement of x1(t) becomes larger to a point that the small-angle approximation
in (1-3) is no longer valid, Bessel function is used to analyze the non-linear system response
assuming x(t) is sinusoidal [1]. The non-linear phase modulation effect results in harmonics and
intermodulation terms on the output spectrum which will be discussed in the following chapters.
21
Baseband
output
T(t)
B1(t) =
cos(4πx1(t)/λ)
t
(Not in scale)
t
t
B2(t) =
cos(4πx2(t)/λ)
4πx2(t)/λ
Target
displacement
4πx1(t)/λ
Figure 1-3. The concept of small displacement detection at optimum and null points of Doppler
radar system. For illustration purpose, x1(t) and x2(t) are not to scale.
The ratio of harmonics proves to be useful to obtain the amplitude of sinusoidal x(t) without any
distance or gain calibration, and the method is not limited to single-tone sinusoidal vibration.
Recently a two-tone mechanical vibration with the amplitude of a few millimeters can be
accurately measured by 6-GHz Doppler radar system at 1-2 m away [8] [12]. If higher radar
operating frequency such as 60 GHz is used, the minimum measurable displacement is expected
to be further improved as mentioned earlier.
Near the null detection point as illustrated by x2(t) in Figure 1-3, the total residue phase in
(1-2) is at even multiples of /2. It can be observed that the baseband output B2(t) does not
contain the fundamental tone of the original vibration. The alternating optimal and null detection
points occurs every λ/8 as the distance d0 varies ϕt [13], and the issue becomes more problematic
for the mm-wave Doppler radar as λ is inversely proportional to the radar frequency. Doublesideband architecture [18] or frequency [28] tuning is possible to alleviate the null detection
point issue, but it also becomes more and more impractical as λ is getting small.
22
1.2.2 Complex Signal Demodulation
To solve the null point issue without extra tuning, complex signal demodulation (CSD) [5]
is utilized in the 5.8 GHz CMOS quadrature receiver [3]. In-phase (I) and quadrature-phase (Q)
baseband outputs can be generated by the quadrature architecture of the receiver circuits, and the
CSD baseband output can be software-reconstructed in real time:
 4 x ( t )
 4 x(t )

 4 x(t )
 j
S (t )  I (t )  j Q(t )  cos 
 t  +j sin 
 t  = e 
 

 



t 

=e
j
4 x ( t )

e jt .
(1-8)
Since exp(jϕt) has a constant magnitude, the method eliminates the effect of total residue phase ϕt
when Fourier transform is applied to S(t) to obtain the CSD spectrum. CSD makes the Doppler
radar detection independent of the residue phase shift, which is mainly determined by the
distance d0 between the radar and target. The quadrature architecture is essential to solve null
detection point issue for the mm-wave Doppler radar design.
I/Q generation is one of the design challenges for the mm-wave system. It can be realized
by a RF quadrature VCO in a direct-conversion system [21], but the loss (power consumption)
and mismatch of mm-wave quadrature LO distribution needed to be concerned [29]. Quadrature
separation is also possible in the RF signal path using a current-domain method or a poly-phase
filter as introduced in [26] [30], however, high flicker noise of the CMOS active mixers used in
the system is not preferable as the vital sign output is nearly dc (1-2 Hz). In this work, I/Q
generation is designed at IF stage (6 GHz) by utilizing a compact ring oscillator, and it is able to
drive two large passive mixers for low flicker noise requirement.
1.3 Vital Sign Detection
Many successful human heartbeat and respiration detections by Doppler radar systems
have been demonstrated [1] [3] [13] [18] [27] [31] at the frequency range from a few to hundreds
of GHz. As mentioned in the previous sections, the increase in frequency reduces the component
23
size and improves the sensitivity to small displacement. A 228-GHz heterodyne radar system
reports the detection of vital sign at 50 m away [27]. However, as the λ becomes too small
compared to target displacement such as chest-wall movement, the harmonics of respiration
starts to degrade the accuracy of heartbeat detection. For the vital sign detection of typical
human, the optimal radar frequency is found to be around 27 GHz [32]. In this work, techniques
have been developed to overcome the challenges and implement the compact SiP for human vital
sign detection. In addition, the 60-GHz radar frequency is potentially optimal to detect the vital
sign of small animals, for example, which normally have smaller chest-wall movement and
different ratio of heartbeat and respiration rate.
Human vital sign sensing is a special case of two-tone sinusoidal vibration analyzed in [8],
where the respiration amplitude (mr  1-6 mm) is normally one order of magnitude larger than
that of heartbeat (mh  0.2 mm). The chest-wall movement due to human respiration and
heartbeat can be approximated by a two-tone sinusoidal vibration:
xr (t )  xh (t )  mr sin(2 f r t )  mh sin(2 f ht )
(1-9)
where fr and fh are the frequency of the respiration and heartbeat. As presented in Figure 3 of [1],
it shows the detection results of the Ka-band Doppler radar system using the frequency of 27
GHz at 2 m in front of the target. The upper-right corner is the baseband output signal in time
domain, and the spectrum clearly shows the vital sign signals. It should be noticed that the
magnitude of peaks are not always proportional to the heartbeat and respiration amplitude. They
are determined by the coefficients of Bessel functions and will be discussed in the later chapters.
Based on the illustration in Figure 1-3, the displacement of respiration is likely to be out of the
linear region depending on  of T(t) and results in harmonics on the baseband output spectrum.
When the operating frequency of the radar is increased, the heartbeat peak is possibly blocked by
24
the 3rd or higher-order harmonics of respiration, for example, and result in the difficulty of
Normalized
CSD spectrum
heartbeat detection.
1
Heartbeat (H1)
0.33
0
Normalized
CSD spectrum
Respiration (R1)
0.67
0
1
50
100
150
Frequency (beat/minute)
(A)
R1 R2
0.5
0
0
50
100
150
Frequency (beat/minute)
(B)
Figure 1-4. Comparison of vital sign detection using different radar frequencies. The simulation
results of (A) 6-GHz and (B) 60-GHz Doppler radar system are plotted.
The radar frequency at 60 GHz is far beyond the optimal carrier frequency [32] for vital
sign detection, and thus further detection techniques need to be developed. The  at 60 GHz (5
mm) is comparable to mr and the non-linear phase modulation becomes much more serious.
Figure 1-4 shows a simulated comparison between 6-GHz and 60-GHz detection results after
CSD. In Figure 1-4 (B), the relatively small heartbeat peak is overwhelmed by the harmonics of
respiration even without the presence of system and environmental noise. In some cases, the
fundamental respiration peak (R1) is too small to be distinguished and results in detection failure,
which will be discussed in Chapter 5. For this reason, the instrument-based 60-GHz millimeter25
wave life detection system (MLDS) [31] measures the heartbeat signal while the target person is
holding the breath, and even the accuracy of respiration detection itself is limited because of the
harmonics, intermodulation, and high sensitivity to environmental noise. In this work, further
analysis based on Bessel functions and signal recovery techniques are investigated to detect the
vibration comparable to .
1.4 Millimeter-wave Packaging and Integration
Increasing the radar operating frequency poses challenges in the transition design between
chip and antenna. The length of typical bonding wire could be several hundreds of micrometers
depending on the chip thickness as illustrated in Figure 1-5.
Bonding wire (~ 850 μm)
Chip pad
350 μm
Die
Board pad
PCB
Figure 1-5. Typical wire-bonding packaging configuration. The parasitic inductance of the wire
is usually in nanohenry range at around 10 GHz.
Consequently the parasitic inductance of the traditional packaging configuration is unacceptably
high (nH range) for this frequency range, and thus the research on low-loss, cost-effective
transition has drawn increasing attention in the mm-wave ICs. On-chip antennas presented in
[19] [33] [34] are alternatives to avoid the lossy mm-wave transition as the antenna size is
usually small enough in the mm-wave applications. However, the on-chip antenna generally
suffers from low radiation efficiency due to the silicon substrate [22] [33], and thus the limited
patch antenna gain from -10 dBi to -2 dBi is usually reported at 60 GHz. Another drawback of
26
the on-chip antenna is typically the narrow bandwidth which is limited by the relatively short
distance between the top and bottom metal layers.
With a low-loss antenna transition, a PCB antenna with a superior performance compared
to an on-chip counterpart can be used, achieving system-in-package integration. Several custom
mm-wave packaging techniques [22] [35] successfully achieve the low-loss transition by
designing supporting structure to horizontally align the chip and antenna, which greatly reduces
the parasitic inductance. As shown in Figure 9 (b) of [22], one technique is to use BGA (ball grid
array) tin ball underneath to raise the chip to the same plane of the antenna. In this case three
very short bond wires are used in parallel to minimize the parasitic resistance and inductance
effects. The parasitic inductance of the antenna transition is estimated to be less than 100 pH at
60 GHz [22].
In addition, bumping and flip-chip process widely used in IC industry is known to be
another promising solution for mm-wave SiP applications without the need of extra supporting
structure. Analysis and tests have been made to characterize the mm-wave flip-chip transitions,
and techniques such as high-impedance compensation are proposed to optimize the transitions
based on CPW-fed structure [36]-[39]. This work proposes a compact flip-chip transition for
mm-wave radars, in which TX/RX isolation and optimal LO distribution path are considered.
Single-die bumping and flip-chip process provided by Collier Ventures Inc. is evaluated and
adopted to integrate the 60-GHz CMOS radar chip with two closed placed PCB patch antennas.
The issues including TX/RX isolation, impedance matching, and loss of the structure will be
investigated and presented in Chapter 3.6.
The Doppler radar SiP in this work demonstrates the first vital sign detection by the CMOS
flip-chip-integrated radar at 60 GHz. The shorter  offers significant area reduction and
27
flexibility in system integration, and the CMOS flip-chip implementation provides the low-cost
potential for mass production. It can be readily embedded into one of the smartphone functions,
for example, making it a pervasive first-aid tool for non-contact vital sign monitoring. The
system can also be applied to a large sensor network for many other applications.
28
CHAPTER 2
SYSTEM DESIGN AND INTEGRATION
2.1 Overview
Top View (Not to scale)
31.3 mm
GND
Patch
Antenna
LNA
175 μm
TX
G
6 GHz IF
60 GHz RF
Flip-chip
Attach
G
G
S
S
Passive
Mixer
Active
Mixer
G
GND
LOI
150 μm
G
G
S
S
G
G
λ/4
S
1.8 mm
RX
G
1.66 mm
50 Ω
Microstrip
GND
BI(t)
0.96 mm
45 mm
S
150
μm
Baseband
Zoom-in
View
54 GHz
RF VCO
6 GHz IF
QVCO
Balun
LOI
Driver
2.35 mm
RF Pads
BQ(t)
LOQ
On-chip
RT/duroid 5870 Laminate
Figure 2-1. Block diagram of the 60 GHz CMOS micro-radar system including transceiver chip
in 90nm CMOS, TX and RX patch antennas, and flip-chip integration.
Figure 2-1 shows an overview of the system including the CMOS radar chip (0.96 mm by
2.35 mm) and two PCB patch antennas on RT/duroid 5870 laminate (31.3 mm by 45 mm) for
TX and RX. RF pads on-chip and the metal traces on the laminate (zoom-in area) are designed
for mm-wave flip-chip integration. The integrated transmitter containing two VCOs, an upconversion mixer, a balun, and a driver is designed to transmit an un-modulated, 60-GHz
continuous wave (CW) signal through the TX antenna. The signal is reflected and phase
modulated by a small vibration of the target due to Doppler effect, and then received by the RX
antenna. For the integrated receiver, the weak received signal is amplified by a 60-GHz LNA and
down-converted by the same VCOs. Since the same VCOs are used for TX and RX and the
phase noise are correlated for short distance detection, this range correlation effect [13] [40]
results in significant reduction of VCO phase noise at radar baseband output, and thus freerunning VCOs can be adopted in the system.
29
2.2 Sensitivity and Radar Received Power
Sensitivity estimation provides an idea of margin that certain RX received power is targeted
to achieve required output signal-to-noise ratio (SNRre). For vital sign applications, the frequency
of interest is near dc (1 Hz) which results in several situations quite different from typical
communication systems. As shown in Figure 2-2, the noise figure (NF) F4 and F5 are estimated
to be as high as 60 dB [3] due to flicker noise, and total NF is around 31.2 dB even with a highgain (35 dB) first stage. The FFT observation time window (TW) at baseband is usually about 20
s to obtain enough cycles and maintain good spectrum resolution bandwidth (RBW = 1/TW =
0.05 Hz). The use of small RBW significantly reduces the overall noise level at output. However,
a low sampling frequency (fs) of ADC around 50 Hz is normally chosen to have a reasonable
FFT bin size (TW  fs) for real-time computation on portable devices. After sampling, the actual
noise level near 1 Hz increases due to aliasing since fs is far below baseband output bandwidth B
( 1 MHz) and flicker noise corner. Experiment results in [3] shows the noise level is often
dominated by the folded white noise when fs is low, and the radar RX sensitivity (S) is estimated
as (in dB scale):
S  kT  RBW (dBm)  NFwh (dB) 
B
(dB)  SNRre (dB)
fs
dc
dc
LNA
RF
Mixer
IF Buf
IF Passive
Mixer
Baseband
Amp
Pin
Pout
Gain: G1=35
NF:
(2-1)
F1=7
G2=0
G3=0
G4=-5
G5=30
(dB)
F2=25
F3=5
F4=60
F5=60
(dB)
Figure 2-2. Sensitivity estimation of the Doppler radar receiver. F4 and F5 are estimated to have
high noise figure (NF) at 1 Hz due to flicker noise.
30
where kT is thermal noise floor per Hz at input, and SNRre is the requirement at output. Here
kTRBW (dBm) + NFwh (dB) represents the white noise level before aliasing and NFwh is around
7.1 dB dominated by the first two stages in Figure 2-2. The corresponding sensitivity is -117
dBm at SNRre = 20 dB. It should be noticed that if fs is increased or RX has a higher flicker noise
corner, the folded flicker noise might become dominant and degrade the sensitivity.
The estimation of radar received power is nontrivial since it involves radar cross section
of human body, and sometimes the short range results in near-field detection where (1-7) is
invalid. However, the maximum possible received power can be estimated by assuming an
infinite perfect reflector and using [14]:
  
Pr  Pt  Gt  Gr  20  log 
.
 4 R 
(2-2)
where Gt and Gr are both 5 dB, Pt is set at 0 dBm, and the target is at 2 m away. The maximum
received power Pr is then calculated to be -70 dBm at 60 GHz while the travel distance R is 4 m
based on the image theory. In the real case, the received power is lower since the actual target
has smaller reflection area and other sources of loss are present. The estimation above reveals
possible margin between the received power and sensitivity, and it indicates the need of a highgain LNA for noise suppression and minimized flicker noise of baseband circuit blocks.
For the detection within a few meters, moderate transmitted power around 0 dBm in (2-2)
is targeted to reduce TX power consumption, as long as the received power meets sensitivity
requirement. Gain stages may be placed at RX to satisfy ADC input specifications. In fact,
Doppler radar transmitting un-modulated radar signal allows TX operating in nonlinear region to
have high efficiency. For comparison, generally one stage of a LNA in 90-nm CMOS achieves a
gain around 7 dB with a power consumption less than 15mW at 60 GHz [41]-[43], on the other
hand, state-of-art PAs achieving 20-dB gain and 10-dBm saturated power (Psat) usually require
31
more than 150 mW [41] [45]. Thus, a TX driver amplifier operating near Psat is adopted in the
system for the short-range detection.
2.3 Design Consideration for IF Stage
Quadrature channels at RX are required for the use of CSD to eliminate null detection points.
I/Q separation is realized at the IF stage of heterodyne receiver instead of a direct-conversion
topology, which avoids the loss, mismatch, and power consumption of 60-GHz I/Q separation
and distribution. Because the phase noise is correlated at short detection range and significantly
reduced by range correlation effect [13] [40], a free-running quadrature ring VCO can be used at
IF to provide a compact and power-efficient choice for driving two large passive mixers. The
compact ring oscillator without the use of inductors is able to drive the large passive mixers and
achieves a wide tuning range (2-8 GHz).
Passive mixers are used for low flicker noise because this noise from mixer is not cancelled
through range correlation. Flicker noise is one of the major considerations for the Doppler radar
application since the baseband outputs is very close to dc (0-2 Hz for heartbeat and respiration).
The flicker noise can be represented by [46]:
vn 2 
K fk
1
CoxWL f

(2-3)
where Kfk is a process-dependent constant and Cox is the total gate capacitance of the transistor.
Based on the equation, larger transistors tend to give lower flicker noise at a fix frequency. Thus
two passive mixers with large transistor sizes are used in I and Q path to minimize the flicker
noise presented in the baseband [3].
The wide tunable IF provided by the ring VCO compensates the possible RF drift due to
mm-wave circuit modeling uncertainty and improves the system robustness. In addition, the IF
frequency chosen roughly a decade away from the RF LO (54 GHz) makes the LO feed-through
32
be effectively attenuated by the output tank of the RF mixer. At the output of IF stage, two sets
of baseband amplifiers are also integrated on the CMOS chip, enabling the baseband outputs
BI(t) and BQ(t) to be directly sampled for software CSD [5].
2.4 Floor Plan and Flip-Chip Transition
For the system floor plan shown in Figure 2-2, the mixers are easily reached by the RF
VCO, and single-ended antennas are adopted and placed closely on the same side of the chip to
minimize loss and power consumption of 54-GHz LO distribution. It also prevents the
dc/baseband connections from interfering with antennas. The proposed G-S-G-S-G transition
achieves impedance match and provides enough TX/RX isolation to reduce the direct coupling
of signal from TX. By using the 150-m pitch and 50- impedance interface, on-wafer (for
chip) and on-board (for antennas) probing measurement can be conducted separately. The
antennas and flip-chip transition were designed at the measured optimal operating frequency of
the chip after fabrication, which helps to ensure the system performance against the modeling
uncertainty, under-estimation of parasitics, and manufacturing variation.
33
CHAPTER 3
CIRCUIT COMPONENT DESIGN
3.1 Inductor
Lumped-element-modeled inductors are extensively used throughout the 60 GHz RF
portion of the system for their low area consumption compared to transmission lines [41]. To
achieve high effective quality factor Q [47] and compact layout, different designs of inductors
are evaluated by 3-D EM simulation which captures important mechanisms such as skin effect,
proximity effect [48], substrate loss, and return current path [49]. In the 90 nm CMOS process,
the inductor is implemented on the top metal 9 (M9) layer, surrounded by thick ground (GND)
walls stacking from substrate all the way to M9 to minimize the parasitic resistance and
inductance on the chip ground. Figure 3-1 shows a 95-pH inductor at 60 GHz along with the
interaction between the ground structure and the inductor itself. As excited from port 1, the
return current Ir seeks the lowest impedance path back to the source [49] [50] and contributes
negative mutual inductance. The displacement current Id due to capacitive (E-filed) coupling
flows through the substrate and back to the source ground.
Ir
Ir
20 μm
M1-M9 GND
Id
` Cp
Port1
20 μm
Id
Ir
Port2
M1-M9 GND
Figure 3-1. Simulation of a 1.5-turn, 95-pH inductor along with the surrounding ground plane.
The return current (Ir) and displacement current (Id) illustrate the EM interactions
between the ground plane and the inductor.
34
Table 3-1. Simulated inductor performance at 60 GHz
Number of Turns
Inductance (nH)
Cp (fF)
Q
Rs ()
1
124.2
17.2
1.7
18.6
2
122.1
10.4
2.4
15.7
3-μm trace width and 2-μm trace spacing in UMC 90-nm CMOS process with 9 metal layers.
If the ground plane is moved closer to the inductor, the inductance density drops and parasitic
capacitance to the ground (Cp) increases which degrade the Q of the inductor. Thus there is no
metal ground underneath the inductor in the structure.
Increasing the number of turns achieves the same inductance with smaller footprint and
thus Cp as listed in Table 3-1. However, the series resistance (Rs) of 2-turn inductor is higher
than that of the 1-turn inductor, which is largely due to the current crowding effect at the high
frequency which can be observed in the current distribution plot similar in [48]. The Q of the
single-turn inductor is 20% higher than that of the two-turn inductors, despite the increase in area
and Cp. The distance between M9 and substrate in the CMOS process used is about 6 μm which
helps lower the impact of Cp. Considering the tradeoff between the area and Q of the inductor,
the L-shaped, 1.5-turn structure as in Figure 3-1 is adopted in the mm-wave transceiver design
utilizing the double-π inductor model [47] in the circuit simulation.
Port 1 Port 2
Port 2
Silicon
substrate
Port 1
Inductor
on M9
Metal stack
from M1 to M9
120 μm
120 μm
(A)
(B)
Figure 3-2. Microphotograph of the on-chip inductors used in the 60-GHz front-end. (A) 95-pH,
L-shaped inductor (simulated Q  18). (B) 128-pH circular inductor (simulated Q 
19.5).
35
The 95-pH inductor achieves a simulated Q around 18 at 60 GHz, and the die photo of the
L-shaped inductor is shown in Figure 3-2 (A). Figure 3-2 (B) presents another circular inductor
configuration which is useful in the layout situation that port 1 and port 2 are very close to each
other. Similarly the metal ground underneath the inductor is removed, and the single-turn
structure minimizes the current crowding effect mentioned previously to reduce the resistive
loss. The circular inductor provides the simulated inductance of 128 pH and Q of 19.5 at 60
GHz. This configuration can also be used as a high-Q differential inductor for VCO resonant
tank if the center tap is applied [52].
M1-M2 GND
120 μm
Port3
Port4
64 μm
M1-M9 GND
M1-M9 GND
Port1
Port2
Substrate
(dB)
-40
-50
-60
S12
S14
-70
-80
2
22
42
Frequency (GHz)
62
80
Figure 3-3. Simulation of the isolation between two closely placed inductors as they are in the
actual on-chip situation.
36
Minimizing the coupling between adjacent inductors facilitates a compact chip layout and
increases the flexibility of the system floor plan. As previously mentioned the stacked M1
through M9 ground walls surrounding the inductor reduces the parasitics on the ground, and they
also increases the level of isolation between closely placed inductors [43]. Figure 3-3 shows the
3-D EM simulation of two 95-pH inductors placed closely together (distance  64 m) to
emulate the real on-chip situation. It can be seen from the plot that the L-shaped inductor
configuration provides the advantage in isolation since the adjacent metal traces of two inductors
are orthogonal, which helps to reduce the magnetic coupling [44]. The simulated coupling level
between the two inductors is low as S12 and S14 are both around -50 dB over the 40 GHz to 80
GHz frequency range.
3.2 Radar Receiver Front-end Design
In the mm-wave radar application, analysis in Chapter 2.2 indicates the need of a high-gain
LNA in RX front-end. Illustrated in Figure 3-4, as Vin is from the single-ended RX patch
antenna, the RX front-end is composed of a 60-GHz five-stage cascode LNA, a single-balanced
mixer, and a 54-GHz VCO. Transistors are biased at the current density around 0.2 mA/μm for
the optimal fMAX and noise performance [41], and extra parasitic capacitance across the gate,
drain, and source due to interconnects was estimated by a 3-port extraction in the 3-D EM
simulator and included in the circuits simulation [51].
3.2.1 LNA
The 5-stage cascode topology with series inductors (Lp1 - Lp5) is chosen for its superior
performance in gain and isolation. Lg1 and Ls1 are 190 pH and 52 pH respectively as matched to
the 50-Ω flip-chip transition. Stage 2 - 4 are identical while inter-stage conjugate matching was
achieved by designing the values of Cg2 to Cg5.
37
5 stages
Ld6
Ld2
Ld1
Ld7
6 GHz +
IF
LO -
N12
N13
54 GHz
LO +
N4
N2
Cg3
Cg2
`
Lp1
Lp6
`
Lp2
Lg1
N11
N3
N1
Cg1
`
Ls1
N1 & N2: 30 μm
Vin
N3 – N8: 35 μm
N9 & N10: 56 μm
Figure 3-4. RX front-end (60 GHz to 6 GHz) including the 5-stage LNA, single-ended mixer,
and 54-GHz VCO (Bias and LO distribution details not shown).
Access to Vdd grid
0.52 mm
Ld4
Ld2
Lp2
Lp4
N8 N9
0.3 mm
N4 N5
Ld1
N3
N2
Lp5
Lp3
N10
N6 N7
Ld5
N1
Lp1
Ld3
Access to Vdd grid
Figure 3-5. Microphotograph showing the cascode portion of the layout and vertical access to
the power grid.
In this manner the 95-pH inductor in Figure 3-1 can be extensively reused in Lp1 – Lp4 and Ld1 –
Ld4 which greatly reduce the custom layout cycle. Lp5 and Ld5 are both set at 45 pH for the output
matching between the LNA and following single-balanced active mixer. As presented in Figure
3-5, the 1.5-turn, L-shape inductor can be arranged to achieve a highly compact layout at the low
38
coupling level between adjacent inductors as previously mentioned. The 5-stage cascode layout
consumes an area of 0.16 mm2. Simulation shows the LNA provides a 38-dB gain and 5.2-dB
NF at 60 GHz while consuming 38 mA at a 1.2 V power supply.
3.2.2 Active Mixer and RF VCO
The single-balanced mixer with inductive loads (Ld6 = Ld7 = 3.4 nH) is adequate for the
application based on the simulation. It shows a gain of 0 dB and also serves as a single-ended to
differential conversion for the following IF mixers. The 54-GHz LO feed-through is far away
from the down-converted signal on the spectrum and also greatly attenuated by the load resonant
tank designed at the IF. A source follower buffer is necessary at the mixer output [3] to provide
low output impedance and drive the large passive mixers.
Figure 3-6 shows the LC cross-coupled, 54-GHz VCO tuned by accumulation-mode
varactors [52] with pseudo-differential LO buffers [53] to drive up- and down-convert mixers.
The simulated tuning range covers from 51.6 to 54.9 GHz, and the phase noise is -101 dBc/Hz
(at 1-MHz offset) at 54 GHz. The VCO core consumes around 18 mW. The first stage of the
differential LO distribution buffers utilizes cascode topology to provide better isolation to the
core, which prevents frequency shift due to possible loading effects. Common-source (CS) is
used in the rest of the stages for the larger voltage headroom and avoiding the pole of cascode
transistors (between N2+ and N3+).
Figure 3-7 shows the microphotograph of LO distribution network design. L1 to L12
represent the differential inductors and each of them contains the positive (+) and negative (-)
parts. The differential inductors require the co-design of conjugate matching and adequate
physical length to reach the next stage. In some cases the inductor length as long as 100 m to
200 m is required.
39
+
-
Down-convert mixer
Vbias
Vbuf
L1-
96 pH
96 pH
L1 +
2 stages
Vbuf
L6 +
L4 +
L2 +
L3 +
Vc
Vout-
Vout+
50 fF
50 fF
54 GHz
VCO Core
1.3×20
μm
N1-
L10+
L7 +
N4 +
1.3×20
μm
L9 +
L5 +
N3 +
N5 +
N2 +
+
L8 +
+
Up-convert
mixer
N1 +
(A)
(B)
Figure 3-6. 54-GHz RF LO generation and distribution. (A) 54-GHz VCO core. (B) Differential
LO distribution buffer network to up- and down-convert mixers.
L4Down-convert
mixer
L2L1-
L6
L3
L5
L1+
L11-
L12
L9-
L10
L4+
L7
Buffer
(N2 & N3)
L8
Up-convert
mixer
L11+
~ 500
μm
L2+
VCO Core
(N1)
M1 to M9 metal
stack GND plane
L9+
~ 650 μm
Figure 3-7. Microphotograph showing the 54-GHz LO distribution network from VCO to upand down-convert mixers.
To accurately model every trace as a lumped inductor, the physical size has to be smaller than
1/10 of the on-chip effective wavelength (λeff) which can be expressed as
40
eff 

r
(3-1)
where λ◦ is the wavelength at 60 GHz in free space and εr is the relative permittivity. Since the
inductors in the LO distribution network usually have longer physical length and do not require
compactness and high inductance density, metal ground shielding (on M1 and M2) can be placed
right underneath the inductors to block the high-permittivity silicon substrate (εr  12). This is
different from the inductor design concept mentioned in Chapter 3.1. By this arrangement, the
inductors only see the silicon oxide (εr  4) assuming the effect of air above the silicon is
neglected for simple estimation, and thus εr is locally increased which assures λeff is around 2500
μm for the accurate lumped-element modeling.
It can be observed from Figure 3-7 that the L-shaped inductor configuration facilitates the
compact LO distribution design connecting the buffer stages, and the GND plane filling
increases the isolation between the closely placed components as mentioned in Chapter 3.1. At
the power consumption of 110 mW, the receiver front-end is designed to provide a single-ended
conversion gain of 38 dB when one of the IF outputs terminated by 50-Ω. Simulated input 1-dB
compression point (P1dB) is at -44 dBm which is much higher than the radar received power
estimated in (2-2).
3.3 Radar Transmitter Front-end Design
As explained in Chapter 2, the TX shares the same IF and RF VCOs with RX to utilize range
correlation effect [13] [40] in the radar system. Figure 3-8 shows the TX front-end design. A
double-balanced mixer is used to accommodate the differential signals from the RF and IF VCOs
and convert the 60-GHz outputs to single-ended by the passive balun load. A three-stage driver
41
(tuned amplifier) at 60 GHz following the balun boosts the power to around 2 dBm at Vout and
drives the single-ended TX patch antenna.
3.3.1 Passive Balun
The passive balun can be realized by coupled transmission line structure [54] or transformerbased coupled spiral inductors [41]. The lump-element-modeled transformer balun shown in
Figure 3-9 is designed to provide the balanced output loads for the up-convert mixer while
minimizing the area overhead.
Ls
k
`
Lp
Vout
54 GHz
LO+
LO+
LO-
6 GHz
LO+
6GHz
LO-
Figure 3-8. TX front-end (6 GHz to 60 GHz) using the double-balanced up-convert mixer,
balanced loads (balun), and three-stage driver at 60 GHz.
k
Port1
GND
Port1
Zp
Lp
Ls
Zs
Port3
M9 (Ls)
Center
tap
Port2
Al Layer (Lp)
50 μm
50 μm
Port3
Substrate
Port2
M1-M2 GND
Figure 3-9. Lumped-element-modeled transformer balun with differential to single-ended
impedance conversion.
42
The primary inductor (Lp) is implemented on the high aluminum pad layer to reduce the
loss to substrate, while M9 is used for the secondary inductor (Ls) to form a stacked coupling
structure. Based on the CMOS metal design rules, the vertically stacked transformer shows less
magnetic flux leakage and higher coupling coefficient (k) compared to a planar transformer. The
metal ground plane underneath the transformer is removed similar to that of inductor design in
Chapter 3.1. The transformer serves as part of the matching network between the mixer and the
TX driver, and the differential impedance Zs and single-ended impedance Zp follows [55]:
Z s n2
L

 s.
Zp
1
Lp
(3-2)
Here n is conventionally defined as the turn ratio. At a given k which normally ranges from 0.3
to 0.9, varying the value of Lp and Ls achieves different conversion of the impedances. The Lp
and Ls designed here are 108 pH and 88.5 pH respectively considering the tradeoff among area,
matching, and parasitic loss. The fitting k of the balun is around 0.65, and area is 0.07 μm by
0.07μm. The area overhead is much smaller than that of the transmission-line-based balun which
usually has a size comparable to wavelength. In 3-D EM simulation, the differential to singleended insertion loss of the balun is around 5 dB at 60 GHz, which is dominated by the limited
mutual inductance. Multi-turn inductors may be used to improve k at the cost of higher parasitic
loss such as series resistance and parallel capacitance to the ground.
3.3.2 TX Driver
The three-stage TX driver at 60 GHz uses two cascode stages and a CS tuned amplifier as
the last output stage. In the simulation, the LO from IF quadrature VCO is -10.7 dBm
(differential) at 6 GHz, and it is up-converted to 60 GHz with the power level boosted to 2 dBm
(single-ended) by the mixer and driver. The output power of the TX driver is designed to operate
near its Psat as mentioned Chapter 2.2. The conversion gain from IF to RF is 12.7 dB, and the
43
driver output is matched to 50  for the flip-chip transition and antenna or on-wafer probing
measurement. Overall, the TX front-end consumes 35-mW power and 0.13-mm2 area, which are
both less than 10% of the overall system consumption.
3.4 IF Quadrature VCO and Passive Mixer
The I/Q separation is performed at IF stage as the compact quadrature ring VCO is used
to drive two passive mixers for low flicker noise at baseband outputs in Figure 3-10. he twostage ring C is able to generate four output phases 0 90 1 0 and 270 based on the
Barkhausen criteria [56] [57]. The differential I and Q LO signals with two-stage resistive-load
buffers are able to drive large passive mixers, while I LO signals are also needed by the upconvert mixer in the TX front-end. The buffers (AI and AQ) have separate power supply (VI and
VQ) to compensate possible I/Q amplitude mismatch due to the different loads. The two passive
mixers with large transistor sizes 75 μm avoid transconductor stage and dc bias current to
minimize the flicker noise at the baseband outputs [3].
-
+
-
D1
To up-convert mixer
VI
LOI-
+
AI
IF+
D2
LOI+
BI -
LOQ-
BQ +
LOI-
+
LOQ+
VQ
AQ
BI +
IF-
-
+
-
To baseband buffers
LOI+
IF+
IFLOQ+
BQ -
LOQ-
Figure 3-10. IF stage (6 GHz to dc) including the quadrature ring VCO, IF LO buffers, and
passive mixers.
44
Figure 3-11 shows the design of the delay cell (D1 and D2) which includes two tuning
mechanisms to improve the precision and tuning range. By only looking into the left-hand side of
the delay cell, the oscillation frequency can be represented as [57]:
f os
1

2
g mn12   GL  g mp1  g mp 3 
2
CL 2
(3-3)
where gm is the transconductance of the transistor. GL stands for the total resistive load of Mn1,
Mp1, and Mp3, and CL is the total capacitance seen at the output node (Vout-).
Vp
Mp5
Mp3
Mp4
Mp2
Mp1
VoutVin+
Vout+
Mn1
Vc
Mn2
Vin-
Vb
Figure 3-11. Design of the delay cells (D1 and D2 in Figure 3-10) with two tuning mechanisms
(Vp and Vb).
By controlling VP, turning on Mp3 achieves the highest oscillating frequency fos (GL +
gmp3 = gmp1) at a fix CL, while turning off Mp3 results in the lowest fos. The accumulation-mode
varactors at output nodes [52] [58] provide extra flexibility to tune CL at the cost of increased
total capacitive load and reduced maximum fos. The width (WV) and number of finger (FV) are
chosen to be the minimum allowed by the process (WV = 1.6 μm FV = 4) to reduce the impact on
maximum fos, while the channel length (LV) is set to be maximum (LV = 2 μm to achieve highest
tuning range. In the simulation the tuning range is 63 % (5.2 – 9.96 GHz) without the varactors,
45
and is increased to 89% (3.28 – 8.5 GHz) by adding the varactors. Over the entire tuning range,
the common-mode output level of the ring VCO stays roughly the same at 0.6 V which eases the
design of the next stage. The simulated phase noise is -83 dBc/Hz (at 1-MHz offset) when the fos
is at 6.4 GHz, and total power consumption of the IF stage is 99 mW.
Figure 3-12 shows an example of the quadrature LO outputs (LOI+, LOI-, LOQ+, and
LOQ- in Figure 3-10). In the simulation, the oscillating frequency is tuned at 6.375 GHz, and the
four output phases 0 90 1 0 and 270 are shown in the plot. he output common-mode level is
maintained around 0.6 V which is about half of the power supply voltage (1.2 V).
LOI+
LOI-
LOQ+
LOQ-
Figure 3-12. Simulated four output phases (LOI+, LOI-, LOQ+, and LOQ- in Figure 3-10)
generated by the quadrature ring VCO.
3.5 CMOS Radar Chip Overview
Figure 3-13 shows the microphotograph of the micro-radar in 90 nm CMOS technology.
On the left are the 60-GHz RF outputs and on the right are the baseband I/Q signal outputs which
can be directly sampled by the oscilloscope or ADC. All the dc biases are accessible from the dc
46
pins on the top and bottom to determine the optimal bias points. The area of the 60 GHz RF core
is 0.73 mm2. The area reduction of the prototype chip is limited by the number of dc pads.
dc bias
2.35 mm
G
G
S
S
G
G
S
S
BI -
RX
BI +
0.96 mm
TX
BQ 60 GHz RF
G
G
6 GHz
IF
Baseband
BQ +
dc bias
Figure 3-13. Microphotograph of the 60 GHz CMOS micro-radar.
3.6 Flip-chip Integration and PCB Patch Antenna
The flip-chip transition and antennas are designed at the optimal operating frequency (55
GHz) of the transceiver circuits measured in Chapter 4. The flip-chip process is able to provide a
low-loss, impedance-matched transition for the mm-wave system-in-package applications. By
proper design of the solder bump, on-chip RF pads, and the G-S-G (Ground-Signal-Ground)
trances on the PCB, the parasitic capacitance and inductance can be estimated by 3-D EM
simulator and compensated [36]-[39]. Different from the conventional wire-bonding package
which is usually 1-2 mm long and introduces unacceptably large series inductance (a few nH) at
mm-wave frequencies, flip-chip process places small solder bump between the chip and PCB
substrate, and the bonding is established by the re-flow of solder bumps under heat and pressure.
The series inductance of the transition is usually smaller than 150 pH at 60 GHz [38].
47
3.6.1 Transition Design and Impedance Match
Figure 3-14 shows the flip-chip transition design between the 60-GHz CMOS radar
transceiver chip and PCB patch antennas on the RT/duriod 5870 laminate. Two bumps are used
on each RF signal pad (S) to reduce the series inductance and resistance, and similarly three pairs
of bumps are placed between the RF ground pads (G) and ground traces on upper metal of PCB,
which is also connected to the bottom ground through PCB vias. Since the width of the 50-Ω
feed-line on PCB is around 350 μm and the pitch of flip-chip G-S-G traces is only 150 μm the
taper structure is designed in-between for the conversion. The increased inductance due to the
taper structure is capacitively compensated by controlling the length d1 and d2 to maintain 50-
impedance match. The flip-chip transition can be modeled as a two-port network [38] as shown
in Figure 3-14 while port 1 is on the PCB and port 2 is on the chip.
Side-view (Not to scale)
Top-view (Without Chip)
Port2
d1
S
G
RF Pad
Chip Silicon
300
μm
DC Pad
175 μm
d2
150
μm
G
Bump
S
50 μm
G
30 μm
Bump
20
μm
Upper Metal
PCB Via
(GND)
127
μm
Port1
RT/duroid 5870 Laminate
Bumps
Bottom GND
FR4 Supporting Board
600
μm
Figure 3-14. Flip-chip transition design between the 60-GHz CMOS radar chip and PCB patch
antennas on RT/duroid 5870 laminate.
The inductance and capacitance values in the two-port  model are extracted from the 3D EM simulation at 55 GHz and shown in Figure 3-15. Based on the simulation, the transition is
not solely dominated by the series inductance since the nearby ground structure contributes
48
significant amount of shunt capacitance, and the resistance is small enough to be neglected. The
value of the parasitic inductance and capacitance can be controlled in certain degree by designing
0.8
6
0.
0.
4
G
C
2.
0
B
0
3.
27 fF
150 μm
0
4.
5.0
0.2
D
10.0
5.0
4.0
A
3.0
10.0
2.0
1.0
0.8
0.4
0.2
0
RX
Before
flip-chip
0.6
C
G
D
B
50 Ω
feed-line
2
-0.
-4
.0
-5.
0
S
At 55 GHz:
-2
.0
B: 0.9 - 0.3j
D: 1 - 0.2 j
-1.0
-0
S
A: 0.95 + 0.2j
C: 0.9 + 0.4j
After
flip-chip
-0.8
-0
.6
Chip.4
-3
.0
G
36 fF
-10.0
S
λ/4
D
100 pH
Zoom-in
view
`
G
S
50 Ω
A
`
A
TX
1.0
the number/dimension of the bump, on-chip pad size, and the metal traces on PCB.
Figure 3-15. Impedance analysis of the transition at 55 GHz before and after the flip-chip
process.
Figure 3-15 demonstrates the impedance analysis before and after the flip-chip transition.
The TX and RX patch antennas are matched to 50 Ω respectively using the cut on the edge and
λ/4 transformer. Looking into point A is the impedance before the flip-chip process, and point D
is the impedance seen by the chip output after flip-chip transition. As shown on the Smith chart
from point A, B, C, and to D, by properly designing the flip-chip transition, the parasitic
capacitance and inductance can be used to largely cancel the effect of each other. The impedance
remains very close to the center of the smith chart and achieve adequate match. The insertion
loss of the flip-chip transition is around 1.5 dB at 55 GHz.
49
he thin RF laminate 127 μm is chosen as considering the trade-off between the
antenna bandwidth and surface wave loss [22] [59]. In the Doppler radar application, the antenna
is not expected to achieve high bandwidth since single frequency is transmitted and the
frequency selection of antennas also removes the undesired harmonics and images. The use of
thin PCB substrate also achieves the adequate width of 50-Ω feed-line for the flip-chip transition
design and reduces the spurious radiation. A thick FR4 PCB is needed underneath the RF
laminate to support the soft RT/duriod 5870 laminate.
dc bias copper traces
To TX
antenna
Solder
mask
G
S
Through
hole
G
S
G
0.96 mm
Solder
mask
To RX
antenna
dc bias copper traces
Figure 3-16. Microphotograph of the flip-chip area on RT/duriod 5870 surface. The solder mask
in dark green is deigned to control the reflow of solder bumps during the flip-chip
process. The hole goes through both the RT/duriod 5870 and FR4 supporting board.
Figure 3-16 shows the microphotograph of flip-chip area on RT/duriod 5870 laminate
surface. In the flip-chip process, two solder bumps will be placed on each RF copper trace (G-SG-S-G), and one solder bump will be placed on each dc bias trace. All the copper traces are
50
partially covered by the green solder mask which is designed to control the reflow of the solder
bumps as the heat and pressure are applied. After the chip is attached to the board, the through
hole underneath the chip allows the RF circuits contacting the air, which helps to maintain the
EM characteristics previously simulated.
3.6.2 Patch Antenna
Figure 3-17 shows the simulated s-parameter of the 55 GHz patch antenna after
integrated with the flip-chip transition (point D in Figure 3-15). Looking into port 1 is the TX
antenna and port 2 is the RX antenna. By using the orthogonal feed-lines and G-S-G-S-G
arrangement, the isolation between port 1 and port 2 reaches -34 dB at 55 GHz even the TX and
RX antennas are placed closely to each other in the compact floor plan. Minimizing the unmodulated signal directly coupled from TX to RX reduces the dc offset in the system. The
simulated gain of the single patch antenna is around 5 dBi as shown in Figure 3-18.
0
|Sij| (dB)
-10
S11
S12
-20
-30
G
S
-40
Port 1
G
-50
53
54
55
Frequency (GHz)
56
57
S
Port 2
Chip
G
Figure 3-17. Simulated patch antenna s-parameter after the flip-chip packaging. S11 shows the
input matching of the antenna and S12 represents the isolation between two ports.
51
E-plane
H-plane
(dBi)
Figure 3-18. Simulated patch antenna pattern after the flip-chip packaging.
52
CHAPTER 4
EXPERIMENTAL RESUTLS
4.1 Millimeter-wave CMOS Transceiver Measurement
A separate test structure which duplicates the 60 GHz RF core in Figure 2-1 was measured
on-wafer. At a total 190 mW power consumption (1.2 V power supply), the receiver provides
more than 30 dB down-conversion gain from 52 GHz to 56.5 GHz as plotted in Figure 4-1. The
peak down-conversion gain is 36 dB from RF to IF (single-ended) with the RF VCO operating at
48.4 GHz. The tuning range of the RF VCO is from 48 to 51 GHz. In the first pass, the reduced
peak gain and shifted peak frequency from 60 GHz to 55 GHz are possibly due to several
reasons. For example, the modeling of active and passive devices might under-estimates some of
the parasitic capacitance and inductance at this frequency. Also the current return path assumed
in the EM simulation setup (Figure 3-1) is different from the complex situation on the actual
chip, resulting in the discrepancy between simulated and actual inductance values. The input P1dB
of the receiver is measured at -42 dBm.
30
20
40
30
Simulated
Measured
20
10
10
Simulated
Measured
0
-10
-20
40
S11 (dB)
Conversion gain (dB)
40
0
-10
50
60
RF input frequency (GHz)
-20
67
Figure 4-1. Measured down-conversion gain (60 GHz to 6 GHz) versus RF input frequency. The
RF input power was set at -60 dBm to emulate the weak reflected radar signal.
53
Figure 4-2 shows the measurement results of the TX front-end. As the IF input is fixed at 6.6
GHz emulating the LO signal from the quadrature VCO and the RF VCO is at 48.4 GHz, the
output P1dB of the transmitter is about -0.3 dBm and Psat is at 1.5 dBm. In this test structure an
extra on-chip balun was used to convert the single-ended IF input signal to differential signal for
the up-convert mixer, and the simulated balun loss was de-embedded in the experiment. The 6GHz on-chip balun is not used in the final system since the IF VCO already provided the
differential signals. If the IF VCO power is estimated to be -10 dBm (differential) in Figure 4-2,
the radar TX output power reaches around 1 dBm which is very close to its Psat as previously
20
20
15
15
10
10
5
5
0
0
-5
-5
-10
-10
-30
-20
Pin (dBm)
-10
Pout (dBm)
Conversion gain (dB)
anticipated in Chapter 3.
0
Figure 4-2. Measured up-conversion (6 GHz to 60 GHz) gain compression and Pin (differential)
versus Pout (single-ended) curve. The IF input frequency was at 6.6 GHz emulating
the LO signal from the IF quadrature VCO and the RF VCO is at 48.4 GHz.
4.2 IF Quadrature Ring VCO
A separate test structure of the IF quadrature ring VCO is fabricated and tested to verify
the tuning range and output power. Figure 4-3 presents the measured spectrum of the ring VCO.
When both Vp and Vc in Figure 3-11 are at the highest voltage level (1.2 V), Mp3 is off and the
54
two varactors provide the largest CL. The lowest fos = 1.8 GHz is measured as shown in Figure 43 (A). On the contrary, as both Vp and Vc are at the lowest voltage level (0 V), Mp3 is on and the
CL from the varactors is minimized. The highest fos = 7.84 GHz is measured as presented in
Figure 4-3 (C). Figure 4-3 (B) shows an intermediate fos = 6 GHz which can be achieved by
tuning Vp and Vc. Amplified by the two-stage buffer showing in Figure 3-11, the VCO output
power level is around -3 dBm to -1 dBm (single-ended) in those three cases after a 2-dB cable
loss is de-embedded. In the experiments, the total power consumption including the two-stage
buffers is around 90 mW.
(B)
(A)
(C)
Figure 4-3. Measured single-ended output spectrum of the quadrature ring VCO. (A) At lowest
fos = 1.8 GHz. (B) At intermediate fos = 6 GHz. (C) At highest fos = 7.48 GHz.
55
4.3 Patch Antenna Test
After the optimal operating frequency of the RF transceiver is determined to be around 55
GHz from the measurement, the patch antennas on PCB can be designed accordingly. The design
of G-S-G metal traces with 150-m pitch shown in Figure 2-1 facilitates both the probing
measurement of the antennas and flip-chip bumping. Figure 4-4 (A) shows an example of the
antenna test structure and 67 GHz G-S-G probe setup. Figure 4-4 (B) plots the measured S11 of
the patch antenna before flip-chip process similar to point A in Figure 3-15. The good agreement
between 3-D EM simulation and measurement indicates the feasibility to design the antennas
right at the measured optimal frequency of the transceiver chip, even the antenna bandwidth is
limited by the thin RT/duroid 5870 laminate, which will be further discussed in Chapter 5.3.
Antenna test
structure
GGB 67A-GSG
probe
(A)
0
-5
S11 (dB)
-10
-15
Simulated
Measured
-20
-25
-30
45
50
55
Frequency (GHz)
60
65
(B)
Figure 4-4. Antenna return loss (S11) measurement. (A) Test structure and G-S-G probe setup
(B) Measured and 3-D EM simulated S11 of the PCB single patch antenna.
56
A separate patch antenna test structure as shown in Figure 4-5 was used to measure the
radiation pattern. Two identical linearly polarized antennas surrounded by ground plane were
fabricated on the RT/duroid 5870 laminate for TX and RX, and the zoom-in area shows the G-SG-S-G structure which can be used for both probing measurement and flip-chip integration.
Figure 4-6 shows the experimental setup for the measurement of the radiation patterns [61]. The
laminate was placed on a probe station and a G-S-G probe was used to excite one of the
antennas. The metallic chuck right underneath the antennas is covered by the absorber to reduce
its influence on the radiation pattern. A standard gain horn antenna (SGH) was set on a sliding
track which covers an azimuthal range of ±25° at the broadside. The probe and the SGH were
connected to the network analyzer to measure |S21|. A calibration is performed in advance to
locate the reference planes at the ends of two cables. Then, Friis transmission equation was then
employed to calculate the realized gain of the patch antenna, which is written as
Gr  S21
2
L  4 R 


Gt  0 
2
(4-1)
GND
GND
Via
6 mm
1.8 mm
1.63 mm
RX
TX
8.1 mm
λ/4
50-Ω
Microstrip
S
X
Z
G
G
Zoom- in
View
S
G
150 μm
Y
GND
Figure 4-5. Patch antenna test structure with probing/flip-chip G-S-G-S-G area zoomed-in.
57
where 0 is a free space wavelength, L is the insertion loss of the probe including the interfaces
between cable and probe coaxial end and between antenna and probe tip, Gt is the gain of the
SGH, which has been characterized previously, and R is the distance between the SGH and the
patch antenna.
Cable
Cable
Sliding track
SGH
Gt
R
(Reference plane)
Gr
L
GSG Probe
Figure 4-6. Probe-based measurement setup for the broadside radiation patterns. The metallic
chuck right underneath the antennas is covered by the absorber to reduce its influence
on the radiation pattern.
(Z)
0
(dB)
10
330
(dB)
10
30
0
-10
300
60
-10
300
60
-20
90 (X) -30 270
-30 270
-20
90 (Y)
-20
240
0
10
30
0
-20
-10
330
(Z)
0
210
150
180
-10
Co-pol (Sim.)
0 (Sim.)
Cross-pol
Co-pol (Mea.)
10
Cross-pol (Mea.)
120
210
150
180
Figure 4-7. Radiation patterns of the single-patch antenna. (A) XZ-plane. (B) YZ-plane. The
cross-polarization shows the 90 polarization mismatch pattern.
The measured and simulated gain patterns in XZ- and YZ-plane at 55 GHz are shown in
Figure 4-7. The peak gain of the main beam is about 4.86 dBi. The co-polarization pattern shows
the results when the orientation of linearly-polarized SGH matched to that of the single-patch
58
antenna, and the cross-polarization pattern was obtained by rotating the SGH by 90 and resulted
in very little antenna reception which is ideally zero due to the 90 polarization mismatch. The
realized gain at zenith against frequency is shown in Figure 4-8. The measured results show a
good agreement with the simulations.
8
Realized Gain (dBi)
6
4
2
0
-2
-4
Simulated
Measured
-6
-8
50
51
52
53
54
55
56
57
58
59
60
Frequency (GHz)
Figure 4-8. Measured and simulated realized gain spectrums at zenith.
4.4 Radar Transmitted Power Test
Figure 4-9 shows the photo of the final 60-GHz micro-radar SiP assembly including the
CMOS transceiver chip, two PCB patch antennas, and the dc bias through the blue wires. Bypass
capacitors with large capacitance value (22 μF) are used on the PCB to suppress the power
supply noise. The weight of the system shown in the photo is less than 10 gram (0.3 ounce) and
can be easily pasted to an upright cardboard facing the target in the experiments.
To test the actual power level transmitted by the CMOS micro-radar chip right before the
flip-chip transition, Figure 4-10 and Figure 4-11 show the experimental setup including the
micro-radar with TX antenna gain Gt  4 dBi, a horn antenna with Gr of 23 dBi, waveguide-tocoaxial-cable adapters, a 50 – 75 GHz down-convert mixers, and E4448A spectrum analyzer.
The spectrum analyzer provides built-in LO to down-convert the 55 GHz signal to IF.
59
31.3 mm
45 mm
Figure 4-9. The final system configuration of the 60-GHz micro-radar system-in-package
including the CMOS transceiver chip, two PCB patch antennas, and dc biasing
through blue wires.
A 55-GHz power source was used to calibrate the loss of the down-convert path, and the total
loss (L) from the adapters, cable, and mixer was estimated to be 42 dB. The experiment setup
needs to maintain the far-field condition, so that Friis transmission equation can be used to
calculate the transmitted power. The far-field condition is summarized as follows [62]:
R
2D 2

(4-2.a)
R
D
(4-2.b)
R

(4-2.c)
where D is the maximum dimension of the antennas. For the single-patch antenna on PCB, Dt is
around 1.8 mm as shown in Figure 4-5, and the horn antenna Dr is 40 mm. Based on (4-2), R was
chosen to be 0.75 m to ensure the far-field condition in the experiment. It should be noticed that
in many of the indoor, short-distance vital sign detection applications, the distance often results
in the near-field condition due to the relatively large reflector (human body) at this frequency. It
is non-trivial to use equations to estimate the actual reflected power, for example, due to nearfield effects such as multi-path reflection. In these cases, many of the system performance
60
parameters such as detection distance usually do not have a theoretical value based on
calculations as a reference, and it mainly relies on experiments to test the system.
R
λ = 5.5 mm
WR15 to 1.85mm(M)
Coax Adapter
Gt
Dr
Antenna
55GHz
Gr
GGB 67GHz Cable
(1.85 mm F to M)
Chip
55GHz
Agilent E4448A
Spectrum Analyzer
1.85mm(F) to
1.85mm(F) Adapter
1.85mm(M) Coax to
WR15 Adapter
OML M15HDW
50 to 75 GHz Mixer
(WR15 to SMA)
IF
LO
Figure 4-10. Experimental setup for the TX output power of CMOS transceiver chip.
Micro-radar
system
Spectrum
Analyzer
0.75m
Horn antenna
Figure 4-11. Photo of the experimental setup to test TX transmitted power of CMOS microradar chip. The distance R is chosen to be 0.75 m to satisfied far-field condition.
Figure 4-12 shows the screenshot of the received power Pr = -82.24 dBm on spectrum
analyzer. The resolution bandwidth was lowered from default 3 MHz to 100 KHz to reduce the
61
noise floor. This helps to distinguish the small received signal on the spectrum at a cost of longer
refresh time. Friis transmission equation including the down-convert loss L can be used to
calculate the transmitted power Pt:
Pr  Pt  Gr  Gt  L  20  log(

).
4 R
(4-3)
Figure 4-12. The screenshot of the received power Pr = -82.24 dBm on the spectrum analyzer.
The resolution bandwidth was lowered from default 3 MHz to 100KHz.
By plugging in the known parameters, the transmitted power Pt of the micro-radar chip is
estimated to be about -2.56 dBm. This measured power value is lower than that was designed in
Chapter 3.3 ( 2 dBm). Referred to the measured plot in Figure 4-2, the test indicates the actual
IF VCO power fed into the TX up-convert mixer might be lower than -10 dBm (differential)
which was previously estimated. The discrepancy might be also due to other unexpected
estimation error in each parameter of Friss transmission equation. However, from the system
application point of view, the test verifies that the TX of CMOS micro-radar chip works properly
and is ready to be used for the following experiments.
62
4.5 Mechanical Vibration Detection
To verify the I/Q separation by the quadrature receiver and explore the detection range and
resolution, a 0.15 m by 0.15 m flat metal plate was attached to a Zaber T-LA60A-S actuator and
placed at a distance D from the radar. A 1-Hz mechanical vibration was generated by the
actuator to test the proper I and Q baseband outputs at the optimal and null detection points.
4.5.1 Quadrature Channel Test
Figure 4-13 shows the detection results directly sampled by the oscilloscope at D = 0.3 m.
In Figure 4-13 (A), the differential Q channels are near the optimal detection point as the
baseband output waveforms (BQ+ and BQ-) show the successful detection and the vibration
frequency (1 Hz) can be easily read from the time axis. On the other hand, the differential I
channels (BI+ and BI-) are near the null detection point. The output waveforms near null
detection point have smaller amplitude and do not contain the fundamental tone of the original
vibration as explained in Figure 1-3. Figure 4-13 (B) shows the opposite scenario at a slightly
different D, which results in the change of total residue phase ϕt.
0.6
0.6
BQ -
BQ 0.5
BI Voltage (V)
Voltage (V)
0.5
0.4
0.3
BI -
B I+
0.2
0.4
0.3
BQ+
0.2
B I+
B Q+
0.1
-2
-1
0
Time (s)
1
0.1
2
-2
-1
0
Time (s)
1
2
(B)
(A)
Figure 4-13. I and Q baseband outputs test of the micro-radar system. (A) Q is near optimum
detection point and I is near null detection point. (B) I is near optimum detection
point and Q is near null detection point.
63
In this case, the differential I channels (BI+ and BI-) are near the optimal detection point, and
differential Q channels (BQ+ and BQ-) are near the null detection point. This result indicates the
I/Q generation of IF quadrature VCO with passive mixers are working properly, and at least one
of the I/Q channels carries the valid detection results for CSD in (1-8) as the detection distance D
varies.
4.5.2 Sensitivity to Small Vibration
In the second experiment to test the detection range and resolution, the vibration
displacement A was varied at a fixed D = 0.3 m, and the received CSD spectrum was normalized
CSD Spectrum
by the largest displacement (A = 1 mm) as presented in Figure 4-14 (A).
1
0.8
0.6
0.4
0.2
0
A=1mm
A=0.2mm
A=0.02mm
0
1
2
3
4
Frequency (Hz)
5
6
(A)
D=0.3m
D=0.6m
D=0.9m
D=1.2m
D=2.1m
CSD Spectrum Peak at 1Hz
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1
Vibration Displacement (A) (mm)
0
(B)
Figure 4-14. The experimental results of small mechanical vibration detection. (A) Various
vibration displacement A at D = 0.3 m (B) Spectrum peak at 1 Hz versus vibration
displacement A at various distances. All the data points are normalized to the largest
spectrum peak at A = 1 mm, D = 0.3 m.
64
It shows the vibration frequency (1 Hz) can still be detected when A is as small as 20 m. Figure
4-14 (B) displays a more complete comparison through the same normalization. Two
observations are made from the plot. First, the received CSD spectrum peak decreases with
distance as expected. A small vibration with A = 0.2 mm can be detected at D = 2.1 m away.
Second at a fixed distance the sensitivity to “displacement change” increases when the vibration
is getting smaller. For example, at D = 0.3 m, as A is reduced by half from 1 mm to 0.5 mm, the
CSD spectrum peak drops about 18% ( 1 to 0.82 ); however, as A is reduced by half from 0.2
mm to 0.1 mm, the drop of CSD spectrum peak increases to 42% ( 0.53 to 0.31 ). This implies
the detection has an optimal range of target vibration displacement in term of the detection
sensitivity, which is worth to be investigated further.
4.6 Heartbeat and Respiration Detection
The chest-wall movement due to human respiration and heartbeat can be approximated by
a two-tone sinusoidal vibration as described in (1-9). In the first test, the CMOS micro-radar
system was pasted to a vertically placed cardboard facing the human target. The heartbeat signal
in Figure 4-15 (A) and (B) was directly sampled by the oscilloscope as the person sitting on a
chair with his chest wall 0.3 m in front of the radar. The breath was held to avoid the blocking
from the harmonics of respiration signal as mentioned in Chapter 1.3. The result shows the
robustness of the quadrature architecture to avoid the null detection points, which occur every
1.25 mm (λ/4) in detection distance. At the time t = 0 ~ 7 s, Q channel was around the optimal
point while I channel was near the null point. However, the target entered the null point of Q
channel after t = 9 s due to some slight body movement, and I channel started to take over the
detection. Figure 4-15 (C) shows the spectrum of detected signal and the heartbeat rate at 69
beats/minute (BPM) after the CSD, which agrees with the result of human counting. The Doppler
65
frequency shift due to body movement also shows up in the spectrum. The total power
consumption of the micro-radar SiP is 377 mW at 1.2 V power supply.
I (v)
0.57
0.56
0.55
0.54
0
2
4
6
8
Time (s)
10
12
14
0
2
4
6
8
Time (s)
10
12
14
Q (v)
0.22
0.21
0.2
CSD Spectrum
0.19
1
0
Heartbeat
Body
Movement
0.5
0
20
40
60
80
Beat / Minute
100
120
Figure 4-15. Heartbeat detection using the 60-GHz radar when the target holds the breath at 0.3
m away. The baseband output signal is shown in (A) I channel and (B) Q channel. (C)
CSD spectrum showing accurate heartbeat rate at 69 BPM.
The total baseband output noise of the system is band-limited by the sampling rate of the
baseband ADC. In this case (Figure 4-15) the sampling rate is 50 Hz and the measured baseband
noise voltage is around 1 mVrms, corresponding to a baseband SNR around 5 (14 dB). This result
is consistent with [3] showing that even the noise figure of the system could be high around the
band of interest due to the large CMOS flicker noise, the vital sign detection is still achievable by
proper design of the system.
The detection of the respiration and heartbeat simultaneously at 60 GH proves to be
challenging due to the severe non-linear phase modulation as discussed in Chapter 1.3. As mr is
66
comparable to or larger than , and the modulated phase 4x(t)/ in (1-2) travels through
multiples of 2, and B(t) is no longer monotonic during inhale or exhale. This shows on the
spectrum as harmonics and intermodulation which seriously degrades the detection accuracy.
Depending on the value or mr, in some cases the frequency of the respiration can be
distinguished on the spectrum, but it is overwhelmed by the harmonics and noise in some other
cases. The weaker heartbeat signal is usually not able to be read from the CSD spectrum. Figure
4-16 shows two measurement results of the respiration detections as fr is around 15 beat/minute
and mr is slightly varied in the two tests.
Normalized
CSD spectrum
1
0.75
Possible
respiration peak
0.5
0.25
0
50
Beat/Minute
(A)
100
150
100
150
Normalized
CSD spectrum
1
0.8
0.6
Respiration
detection fails
0.4
0.2
0
50
Beat/Minute
(B)
Figure 4-16. Measurement results of the respiration detection at 60 GHz as fr = 15 beat/minute, D
= 0.3, and mr is slightly varied in the two tests.
In the top plot of Figure 4-16, the possible respiration peak around 15 beat/minute (BPM)
shows on the CSD spectrum, however, there are also many other peaks with larger magnitude
which make the detection result in doubt. Sometimes after mr is varied, the main respiration peak
67
disappears as shown in the bottom plot of Figure 4-16. The weaker heartbeat signal at 80 BPM
could not be identified in both cases. The 60-GHz vital sign detection difficulty will be discuss in
detail is Chapter 5. Theoretical analysis and simulation will be provided to address these issues,
and new techniques such a time-domain recovery algorithm will be proposed to improve the
accuracy of respiration detection.
68
CHAPTER 5
ANALYSIS AND IMPROVEMENT
5.1 Analysis on 60-GHz Vital Sign Detection
The nature of human heartbeat and respiration raises demodulation issues for 60-GHz
vital sign detection as described in Chapter 1.3, and the detection results shown in Chapter 4.5
indicates further theoretical analysis is needed to understand and improve the system. Typically
the amplitude of chest-wall movement mr (1-6 mm) due to respiration is at least one order of
magnitude larger than mh (0.2 mm) due to heartbeat, and the ratio between mr and  (5 mm) no
longer satisfies the small angle approximation in (1-3). The large amplitude difference and strong
nonlinear phase modulation [1] significantly increase the complexity of output spectrum and
detection difficulty.
Human chest-wall movement due to heartbeat and respiration can be modeled as a twotone, sinusoidal vibration as in shown in (1-9). The baseband outputs of I and Q channels can be
modified from (1-3):
 4  mr sin(2 f r t )  mh sin(2 f ht ) 

 4 x(t )

BI (t )  cos 
 t   cos 
 t  .

 



 4  mr sin(2 f r t )  mh sin(2 f ht ) 

BQ (t )  sin 
 t  .



(5-1.a)
(5-1.b)
In CSD process, complex S(t) is generated by combining I and Q baseband outputs similar to (18), and it can be expanded by Bessel function as the summation of its frequency components [8]:
S (t )  BI (t )  j  BQ (t )
 4  mr sin(2 f r t )  mh sin(2 f ht ) 

=exp  j 
 t 






 J
k  p 
k
(ar )  J p (ah )  exp  j  2 (kf r  pf h )t   exp( jt )
69
(5-2)
where ar = 4mr/, ah = 4mh/, and Jn(x) is the nth-order Bessel function of the first kind. The
CSD spectrum is obtained by taking FFT of S(t), and the constant magnitude of exp(jϕt) no
longer affects the detection.
The heartbeat and respiration amplitudes with respective to wavelength determine the
magnitude of harmonics and intermodulation. A peak at x Hz is proportional to [8]:
Hx 


 J
k  p 
k
(ar )  J p (ah )
(5-3)
where k and p are integers satisfying k∙fr + p∙fh = x. For example, the fundamental respiration
peak (R1) located at x = fr (k = 1 and p = 0) is represented by J1(ar ∙J0(ah), and the fundamental
heartbeat peak (H1) at x = fh (k = 0 and p = 1) is determined by J0(ar ∙J1(ah). As illustrated in the
Bessel function plot in Figure 5-1 (A), the small mr and mh compared to  at 6 GHz (50 mm) stay
quite close to the origin on x-axis. Ideally any harmonic and intermodulation Hx consists of Jn(a)
with |n|  3 is small enough to be neglected, and this explains the spectrum in Figure 1-4 (A)
showing clear R1 and H1. It should be noted that Figure 5-1 only plots Jn(a) with positive n, and
Jn(a) with n < 0 follows the symmetry of Bessel functions [8]:
J n (a) 
J n (a), for even n  0
J n (a), for odd n  0.
(5-4)
On the other hand, the detection scenario at 60 GHz is quite different as presented in
Figure 5-1 (B). The heartbeat amplitude mh still stays near the origin on x-axis, but the highorder terms of Jn(ar) emerge as mr comparable to λ, resulting in a more complex spectrum. For
example, as mr near 5 mm, even J10(ar) is not negligible and thus generates a prominent peak of
R10. The vital sign spectrum of various mr is simulated in Figure 5-2 with the respiration rate at
15 beat/minute. The follows discuss the 60-GHz system’s detection difficulties in terms of
respiration and heartbeat, respectively.
70
J0 (a)
λ
J1 (a)
n
Bessel Function J (a)
1
J2 (a)
0.5
J (a)
3
0
mr & m h
-0.5
0
20
40
60
Amplitude m (mm)
(A)
80
J0 (a)
λ
J (a)
n
Bessel Function J (a)
1
1
0.5
J (a)
2
J10 (a)
0
mh
-0.5
mr ≈ 2~6 mm
0
2
4
6
Amplitude m (mm)
(B)
8
Figure 5-1. Theoretical plots of Jn(a) = Jn(4πm/λ) versus vibration amplitude m. (A) At 6 GHz.
(B) At 60 GHz. Typical value of respiration mr ranges from 1-6 mm, and the
heartbeat mh is usually around 0.2 mm.
5.1.1 Respiration Detection Improvement by Two-tone Monitoring
The fundamental and harmonics of respiration (R1, R2, R3, and etc) can be express as
Jn(ar ∙J0(ah) with n = 1, 2, 3, and etc. Observed from Figure 5-1 (B), as the value of J0(ah) is
always close to unity due to small ah , the respiration harmonics are generally larger than other
intermodulation terms and can be easily identified on the spectrum in Figure 5-2. However, the
detection relying on R1 is not robust since J1(ar) shows multiple zero-crossing points.
71
Normalized CSD
spectrum
Normalized CSD
spectrum
Normalized CSD
spectrum
1
R1
R2
Weak
Heartbeat (H1)
0.67
0.33
0
0
30
1
0.67
R1
60
90
120
Frequency (beat/minute)
(A)
150
60
90
120
Frequency (beat/minute)
(B)
150
60
90
120
Frequency (beat/minute)
(C)
150
R2
0.33
0
0
30
1
R1
R2
0.5
0
0
30
Figure 5-2. Simulated output spectrum of vital sign detection at 60 GHz. (A) At mr = 1.5 mm.
(B) At mr = 2.1mm. (C) At mr = 4 mm. The fr is at 15 beat/minute. Heartbeat mh is
fixed at 0.2 mm with fh = 72 beat/minute.
As shown in Figure 5-2 (A) and (C), R1 vanishes as mr near 1.5 mm and 4 mm, which can be
verified by the plot in Figure 5-1 (B) showing the zero-crossing points of J1(ar). Similarly, R2
disappears as mr near 2.1 mm in Figure 5-2 (B). Observed from Figure 5-1 (B), J2(ar) approaches
local maximum while J1(ar) is near zero-crossing points, and vice versa. The important property
from the simulation leads to the conclusion that the respiration detection can be improved by
monitoring R1 and R2 simultaneously. Theoretically the first prominent peak is either R1 or R2 as
the frequency swept from low to high on the output spectrum, and this detection method is valid
72
for all values of mr. The frequency of other higher-order respiration harmonics can be used to
distinguish between R1 and R2 by division which will be demonstrated in the following
experiments.
Voltage(V)
0.8
BI (t)
BQ (t)
0.6
0.4
0.2
-10
-8
-6
-4
-2
0
2
Time(s)
4
6
8
10
(A)
Normalized CSD
spectrum
1
R1
R2
0.8
0.6
0.4
0.2
0
0
30
60
90
Beat/Minute
120
150
(B)
Figure 5-3. Vital sign detection results as the person breathes shallowly at 0.3 m in front of the
radar. (A) Time-domain waveforms. (B) CSD spectrum.
Figure 5-3 shows the vital sign detection of a person sitting 0.3 m in front of the radar
and breathing shallowly. BI(t) and BQ(t) are displayed on an oscilloscope (fs = 25 Hz) and the
observation time window is 20 s. The results in Figure 5-3 (B) shows the sallow breath mainly
generates the fundamental (R1 at 13.8 beat/minute) and second harmonic (R2 at 27.2
beat/minute), and the detected respiration rate agrees with human counting. The higher-order
harmonics are not prominent since Jn(ar ∙J0(ah) with |n| > 3 are all small as predicted in Figure 51 (B). From the plot mr can be estimated to be around 1 mm in this test.
As the target person breathing deeply at a same rate of 15 beat/minute, Figure 5-4 (A)
and (B) shows the I and Q baseband outputs. The observation time window was increased to 50 s
73
and fs was 10 Hz. It is noted that in BI(t) and BQ(t), the modulated phase term 4x(t)/ in (1-2)
travels through multiples of 2 due to the large chest-wall movement mr, resulting in much more
complex time-domain waveforms compared to previous experiment of shallow breath.
Voltage(V)
0.6
BI (t)
0.5
0.4
0.3
0.2
-10
-5
0
Time(s)
5
10
(A)
Voltage(V)
0.6
BQ (t)
0.5
0.4
0.3
0.2
-10
-5
0
Time(s)
5
10
Normalized CSD
spectrum
(B)
R7
1
R8
R9
0.8
R2
0.6
0.4
0.2
0
0
30
60
90
Beat/Minute
120
150
(C)
Figure 5-4. Vital sign detection results as the person breathes deeply at 0.3 m in front of the
radar. (A) Time-domain waveform of I channel. (B) Time-domain waveform of Q
channel. (C) CSD spectrum. The observation time is increased to 50 s, but only 20 s
is shown here for the comparison with Figure 5-3.
Figure 5-4 (C) shows the output CSD spectrum where the frequency resolution is improved by
the longer observation time. As discussed earlier in this chapter, ideally the prominent peak at 30
beat/minute is either fundamental (R1) or second harmonic (R2) of respiration. Since the
74
frequency of higher-order harmonics around 105 beat/minute (R7) and 135 beat/minute (R9) are
not dividable by 30 beat/minute, it is concluded that the peak around 30 beat/minute is R2 and
fundamental respiration frequency (R1) is at fr =15 beat/minute. However, as presented in (5-3),
theoretically there are expected to be more prominent harmonics such as J4(ar ∙J0(ah) and
J5(ar ∙J0(ah) on the spectrum, which are not shown in the result of Figure 5-4 (C). One of the
possible reasons is that human respiration and heartbeat movements are not purely sinusoidal as
modeled in (1-9) for simple analysis. The respiration signal itself is actually composed of
fundamental tones and certain higher order harmonics, which results in the discrepancy of
higher-order behavior of the Bessel function.
5.1.2 Heartbeat Detection
The fundamental and harmonics of heartbeat (H1, H2, H3, and etc.) can be expressed as
J0(ar ∙Jn(ah) with n = 1, 2, 3, and etc. Normally the relatively small value of mh near the origin on
x-axis in Figure 5-1 (B) makes all Jn(ah) too small to be distinguished as respiration is present. In
addition, the heartbeat peaks are also affected by respiration term J0(ar). In some rare cases as
J0(ar) is near peak value (mr ≈ 1.5 mm and the nearby respiration harmonics are weak small
peak of H1 can be seen on the spectrum as in Figure 5-2 (A). If J0(ar) is near its zero-crossing
points, it makes the already weak H1 more unlikely to be read from the output spectrum.
Currently the heartbeat detection at 60 GHz is obtained by holding the breath to avoid respiration
harmonics on the output spectrum as explained in Chapter 1.3. Detection from the back is an
alternative way to reduce the interference from respiration.
5.2 Proposed Time-domain Recovery Algorithm
The technique monitoring the first and second harmonic peaks on the spectrum helps to
improve the success rate and accuracy of respiration detection. However, observed in some
extreme cases of the experiments, both the first and second respiration peaks may not be
75
distinguishable on the spectrum due to the environmental noise and system non-ideality as
captured in Figure 4-16 (B). Especially when the chest-wall movement has larger deviation from
the ideal sinusoidal waveform as modeled in (1-9), the magnitude of each harmonic and
intermodulation term may not follow the theoretical prediction by Bessel function curves plotted
in Figure 5-1 (B). In another word, the target displacement comparable to or larger than  such
as respiration movement generates numerous peaks on the output spectrum, and the
demodulation based on the recognition of Bessel harmonic peaks on the spectrum is susceptible
to any small system non-ideality and noise. This makes the algorithm improvement in frequency
domain difficulty and not robust. One thought is to unwrap the non-linear modulated phase term
from time domain first, and then the accuracy on frequency-domain peak recognition might be
further improved.
5.2.1 Analysis on Quadrature Baseband Outputs
Doppler radar system has a non-linear input-output mapping when the displacement is
comparable to λ. The relation can be seen by the plot in Figure 5-5 modified from Figure 1-3. On
baseband output B(t), two types of peaks (transition between positive and negative slope) can be
identified.
BB
B
C
T(t)
B(t)
π
0
2π
A
A
A
t
C
Baseband output
B(t) = cos(4πx(t)/λ)
Target
displacement
4πx(t)/λ
t
Figure 5-5. Non-linear input-output mapping when the vibration is comparable to λ at 60 GHz.
76
Type-I peaks are due to the distortion of nonlinear system mapping (peak A and B), and
Type-II peak is corresponding to the peak of real displacement (peak C). Plotting both I and Q
channels in Figure 5-6, the following relations can be found:
1.
As a Type-I peak happens on BI(t), the sign of slope on BQ(t) remains unchanged (and
vice versa).
2.
As a Type-II peak happens, the sign of slope on BI(t) and BQ(t) change simultaneously.
This indicates the peak is due to the original displacement.
3.
For any adjacent Type-I peaks (peak A and B) on BI(t), the corresponding sign of slope
on BQ(t) is opposite (and vice versa).
Thus if the two types of peaks on BI(t) and BQ(t) can be distinguished by software, the baseband
output can be un-folded to show the real target displacement, and the peak of interest such as
respiration can be recovered on the final CSD spectrum.
BB
B
C
BI(t)
π
0
2π
I
A
A
A
Q
t
C
Baseband output
BI(t) = cos(4πx(t)/λ)
t
Target
displacement
4πx(t)/λ
Figure 5-6. Non-linear input-output mapping of I and Q channels when the vibration is
comparable to λ at 60 GHz (BQ(t) is not shown in the figure).
77
5.2.2 MATLAB Program Implementation
The time-domain signal recovery algorithm based on the above observation was
implemented in MATLAB to improve the accuracy of respiration detection. By simply
monitoring the system baseband outputs BI(t) and BQ(t), the algorithm keeps the non-distorted
portion of waveform from original vibration (in Follow mode) and recovers the distorted portion
of waveform (in Flip mode) as demonstrated in Figure 5-7 . No other intermediate reference
signals are needed as the algorithm converts BI(t) and BQ(t) directly to BI(t)’ and BQ(t)’.
Conceptually the algorithm is trying to “un-wrap” the non-linear input-output mapping to a
approximately linear transfer function. Although the system mapping after the phase unwrapping is still not completely “linear” due to the nature of sinusoidal shape, the main
frequency component of the original displacement x(t) can be largely recovered in the final
baseband outputs which will be shown in the following sections.
Flip mode
I channel
Follow mode
Q channel
BB
B
C
T(t)
A
A
C
t
Recovered
baseband output
BI(t)’
A
t
Baseband
output BI(t)
Target
displacement
4πx(t)/λ
Figure 5-7. Time-domain recovery technique by simply monitoring I and Q baseband outputs
when the vibration is comparable to λ at 60 GHz (BQ(t) and BQ(t ’ are not shown in
the figure).
78
Figure 5-8 illustrates a simple technique to perform the continuous Flip and Follow
operations in the code. It can be observed from Figure 5-7 that the switch between Flip mode and
Follow mode is only triggered by Type-I peaks (peak A and B), and these two modes are always
“alternating” regardless of the direction of the displacement x(t). In Figure 5-8 (A), the algorithm
follows every point in segment A1-B1 makes no change to the waveform. As soon as the
algorithm detects peak B1, it starts to vertically flip the first point of segment B1-C with respect
to the previous value and update all the points after this current point. In this manner the
algorithm maintains the continuity of the waveform. It repeats the same Flip operation point by
point in segment B1-C and segment C-B2 as illustrated in Figure 5-8 (B) and Figure 5-8 (C).
Since peak C is not a Type-I peak, it does not trigger the switch between Flip and Flow modes.
After the algorithm detecting Type-I peak B2, the program switch back to Follow mode, and the
two Type-I peaks (B1 and B2) generated by the non-linear system transfer function are eliminated
from the original waveform.
B2
C
C
B1 B2
B1
B1
B2
C
A2
A1
A2
A1
A1
(A)
(B)
A2
(C)
Figure 5-8. Illustration of the continuous Flip and Follow operations. (A) Follow segment A1-B1
(B) Flip segments B1-C and C-B2 (C) Follow segment B2-A2.
79
Figure 5-9 presents the simplified flow chart of the time-domain recovery algorithm. After
importing BI(t) and BQ(t), moving average is applied to remove the glitches of the waveform due
to noise and reduce the chance of peak misjudgment. In the same loop which go through every
point of the waveform, BI(t) and BQ(t) are processed in parallel to increase the efficiency.
Start
Import BI(t) and BQ(t)
(I_p and Q_p)
Moving average
(Low-pass filter)
I <= length(I_p)
I <= length(I_p)
N
N
Y
Y
N
Flip stage
i=i+1
Follow stage
i=i+1
Call functions
“Peak check” & “Trend check”
Type-I Peak?
Call functions
“Peak check” & “Trend check”
Type-I Peak?
Y
Y
Output BI(t)’ and BQ(t)’
(I_p_fix and Q_p_fix)
End
Figure 5-9. Simplified flow chart of the time-domain recovery algorithm. Similarly BI(t) and
BQ(t) represents the original waveforms and BI(t)’ and BQ(t ’ are the recovered
waveforms. he details of function “Peak check” and function “ rend check” are
shown in Appendix.
80
N
Function “Peak check” is used to identify any valid peak on the waveform and function “ rend
check” is designed to check if the waveform has a consecutive trend on the waveform no peak .
The details of the two sub functions are presented in Appendix. Based on the three relations
described in Chapter 5.2.1, once a valid peak is detected on BI(t) and there is a consecutive trend
(no peak) on BQ(t), it is identified as a Type-I peak on BI(t) which triggers the switch between the
Flip and Follow mode on BI(t) (and vice versa on BQ(t)). The algorithm makes no change to the
Type-II peaks, which preserves the original target displacement.
5.2.3 Experimental Results
Referred to the previous outputs in Figure 4-16 (B), the respiration detection failed and
no prominent fundamental or second harmonic can be read from the spectrum. In the experiment,
the same baseband signals BI(t) and BQ(t) of Figure 4-16 (B) were used to test the effectiveness
of the time-domain recovery algorithm. Figure 5-10 compares the baseband outputs before and
after the proposed time-domain recovery algorithm, and as expected the respiration waveforms
close to the original chest-wall movement are recovered on BI(t ’ and BQ t ’. It should be notice
that from the waveforms of recovered BI(t ’ and BQ t ’, it verifies that the respiration movement
in this case is not very close to the single-tone, pure sinusoidal waveform as modeled in (1-9).
Thus the harmonics and intermodulation terms on the original spectrum in Figure 4-16 (B) do
not completely follow the Bessel function analysis in Chapter 5.1.
Figure 5-11 shows the CSD spectrum after applying the time-domain recovery algorithm.
Compared to the original spectrum in Figure 4-16 (B), it shows a prominent respiration peak at
15.11 beat/minute, which closely agrees with the results of human counting. This demonstrates
the recovery algorithm can significantly increase the accuracy and robustness of 60 GHz Doppler
micro-radar detection.
81
BQ(t) BI(t)
0.2
Voltage(V)
0
-0.2
-0.4
-0.6
-0.8
-1
-10
BI(t)’
-8
-6
BQ(t)’
-4
-2
0
2
Time(s)
4
6
8
10
Normalized
CSD spectrum
Figure 5-10. Respiration detection outputs before and after the recovery algorithm is applied.
The respiration waveforms close to the original chest-wall movement are recovered
on BI(t)’ and BQ(t ’by the algorithm.
1
0.75
Recoverd respration peak
at 15.11 beat/minute
0.5
0.25
0
50
100
Beat/Minute
150
Figure 5-11. Recovered respiration peak compared to the original spectrum in Figure 4-16 (B).
The prominent peak at 15.11 beat/minute shows the accuracy and robustness of the
detection have been improved.
82
5.2.4 Discussion
Figure 5-12 shows the results of recovery algorithm test on different respiration pattern. In
the experiment, the respiration was controlled by the subject as he inhaled for 2 s, exhaled for 2
s, paused for 3 s, and repeated the cycle. The overall respiration rate should be 60/(2+2+3) = 8.57
beat/minute. This respiration pattern is no longer single-tone, pure sinusoidal movement, and the
harmonic and intermodulation terms on the spectrum cannot be predicted by the Bessel function
analysis in Chapter 5.1. The experimental results show the respiration movement can
successfully be recovered by the time-domain algorithm, even the target has an arbitrary
movement pattern.
0.8
0.7
0.6
Voltage(V)
0.5
0.4
0.3
BI(t)’
BQ(t)
’
0.2
BI(t) BQ(t)
0.1
0
-0.1
-10
-8
-6
-4
-2
0
2
4
6
8
10
Time(s)
Figure 5-12. Respiration detection outputs before and after the recovery algorithm is applied.
The subject inhaled for 2 s, exhaled for 2 s, paused for 3 s, and repeated the cycle.
The heartbeat signal can be seen in the intervals between the respirations.
Figure 5-13 presents the CSD spectrum before and after applying the recovery algorithm.
In Figure 5-13 (A), the result shows a relatively low signal-to-noise ratio (SNR) on the spectrum,
and unknown peaks such as the one at 6.6 BPM appear due to the environmental and system
83
noise. After the time-domain recovery algorithm, the prominent respiration peak emerges at 8.65
BPM as shown in Figure 5-13 (B). The SNR on the spectrum is greatly enhanced, which proves
to increase the accuracy and robustness of the respiration detection.
Normalized
CSD spectrum
1
0.75
Ambiguous peak
at 6.6 beat/minute
0.5
0.25
0
50
100
150
Beat/Minute
(A)
Normalized
CSD spectrum
1
0.75
0.5
Recoverd respration peak
at 8.65 beat/minute
0.25
0
50
Beat/Minute
100
150
(B)
Figure 5-13. CSD spectrum outputs before and after the recovery algorithm is applied. The
prominent peak at 8.65 beat/minute is fairly close the real respiration rate (8.57 beat/
minute), which shows the detection accuracy is improved.
From the viewpoint of the recovery algorithm in Figure 5-12, around the time period of 7 s to -4 s, 0.5 s to 3.5 s, and after 7.5 s, the program indicates consecutive Follow mode rather
than alternating Flip and Follow modes. The result implies the fluctuations of Doppler modulated
phase in (1-2) during these time periods is less than , indicating the target does not have a large
displacement comparable to . As mentioned in Chapter 1.3, the respiration movement is
comparable or larger than , and thus the small target movement in these consecutive Follow
periods only captures the heartbeat signal with some small random body movement. Figure 5-14
duplicates Figure 5-13 to show the consecutive Follow periods indicated by the recovery
84
algorithm. One of the time periods marked by the blue circle captures only the heartbeat with
small random body movement, corresponding to the paused respiration period. If FFT is applied
to the waveform in the blue circle, the heartbeat rate at 72 beat/minute can be obtain correctly as
shown in Figure 5-15. This detection result agrees well with the heartbeat rate measured by a
wrist-band monitor as a reference.
0.8
0.7
Consecutive
Follow
Consecutive
Follow
0.6
Consecutive
Follow
Voltage(V)
0.5
0.4
0.3
0.2
Heartbeat
signal
0.1
0
-0.1
-10
-8
-6
-4
-2
0
2
Time(s)
4
6
8
10
Figure 5-14. Duplicate of Figure 5-13 showing the consecutive Follow periods indicated by the
recovery algorithm. The time period marked by blue circle contains only the heartbeat
signal with small random body movement.
2
Heartbeat at
72 beat/Minute
Spectrum
1.5
1
0.5
0
0
50
100
Beat/Minute
150
Figure 5-15. Spectrum of the waveform marked by blue circle in Figure 5-14 which shows the
correct heartbeat rate detection result.
85
As analyzed in Chapter 5.1.2, the simultaneous detection of respiration and heartbeat using
60 GHz Doppler radar proves to be difficult due to the intermodulation terms of the Bessel
function. However, the discussion above indicates the possible detection in a normal breathing
situation. In fact, except for some extreme cases such as heavy breathing right after excises, the
normal human breathing contains certain time interval between each respiration cycle, especially
for the vital sign monitoring during sleep. The time-domain recovery algorithm indicates those
paused respiration intervals by consecutive “Follow” periods, which capture only the small
movement including heartbeat and body movement. Even if each interval is as short as 2-3
seconds, the heartbeat rate can be extracted from the waveform fragment as shown in the above
experiments. In the real-time vital sign monitoring, the heartbeat rate detection result can be
updated at every time the interval is detected. This method provides a potential solution for the
60-GHz Doppler micro-radar to simultaneously detect the heartbeat and respiration in most of
the normal situations.
5.3 Broadband Antenna on LTCC System-in-Package
5.3.1 Introduction
As antenna size shrinks with radar frequency, previous chapters present the 60-GHz
CMOS micro-radar with PCB patch antennas to achieve a low-cost, fully-integrated vital sign
sensor. The use of planar microstrip antennas facilitates the mm-wave flip-chip attachment, and
the low profile is suitable for many portable applications such as tablets and smartphones. For a
typical edge-fed patch antenna, the choice of dielectric thickness faces the tradeoff between
antenna bandwidth and adequate width of 50- feed line [60]. Although the bandwidth
requirement is not critical in terms of using a single-tone, fixed-frequency radar signal, a
wideband antenna is desired to improve yield when modeling and manufacturing variations are
present. The LTCC (low-temperature co-fired ceramic) substrate with multiple metal layers is
86
used to make a better compromise among the bandwidth, radiation efficiency, and proper
microstrip width for flip-chip transition design.
For vital sign detection systems utilizing multiple radars facing each other such as for
random body movement cancellation [5] [63], direct coupling from the TX of one radar to RX of
another needs to be minimized for interference and DC offset considerations. Orthogonal linearly
polarized antennas can be used to alleviate the issue [5], however, as the number of radars
increased, circularly polarized antennas are needed. The right-hand circularly polarized (RHCP)
antenna at RX provides better isolation against other radars with left-hand circularly polarized
(LHCP) TX antennas.
As mentioned in Chapter 3.6.2 and 4.3, the thinnest substrate (127 m) available was
chosen to minimize the 50- microstrip width ( 360 m) shown in Figure 4-5, which sacrifices
the antenna bandwidth [60]. Since the required flip-chip G-S-G-S-G pitch is only 150 m,
narrow 50- feed line is highly preferable for the flip-chip transition design, and it also reduces
the spurious radiation which disturbs the antenna pattern. As shown in Figure 4-4 and Figure 4-8
of previous chapter, the 10-dB return loss bandwidth of the single patch antenna is limited to
around 2.4%, and the simulated 3-dB gain bandwidth is only about 5%. Figure 5-16 shows a case
where the measured return loss deviates from the targeted frequency around 55 GHz due to the
manufacturing variation of antenna fabrication. In the example, the narrow 10-dB antenna
bandwidth barely covers the targeted frequency around 55 GHz. At the PCB etching precision
around 5%, the antenna center frequency shift of 2-3 GHz was observed in the worst case.
Taking into account the process variation of CMOS chip, the actual radar operating frequency is
possibly to fall out of the optimal performance range of the patch antenna in some worst-case
scenarios. In this section, the circularly polarized sequential-rotation 22 patch antenna array is
87
introduced to increase the gain and bandwidth performance compared to the linearly polarized
single-patch antenna.
0
Magnitude (dB)
-5
-10
-15
-20
-25
-30
50
Simulated
Measured
51
52
53
54
55
56
57
58
59
60
Frequency (GHz)
Figure 5-16. Measured and simulated return loss of a single patch PCB antenna as the
manufacturing variation is present.
5.3.2 Sequential Rotation Patch Antenna Array
The multilayer LTCC substrate is studied for the design of wide band patch antenna. The
proposed layer profile is shown in Figure 5-17. The process provides 11 metal layers (L1-L11)
with a thickness of 10 m each, and the relative dielectric constant is 3.9. The antenna elements
utilize L1 and the ground (GND) is on L6, which forms a total dielectric thickness of 328 m to
improve the antenna bandwidth. However, as the surface wave loss and coupling increases with
the dielectric thickness [60], the design compromise has to be made between bandwidth and
radiation efficiency. The microstrip feed line and its GND are on L1 and L2 with a dielectric
thickness of 65 m to keep a narrow feed line width. All GND layers are connected by metallic
via fences. For the RF signal, a through substrate via is employed to provide a vertical transition
between the chip and the antenna. Six grounded vias are located around the through substrate via
to mimic a coaxial transmission line effect [64]. In addition, the LTCC serves as the compact
88
packaging for the micro-radar chip. The chip and dc-bias lines are located at the bottom (L11),
reducing the interference to the antenna and making the area usage more efficient.
(Not in scale)
Radiation
50-Ω Vertical
transition
RF signal
(Patch Ant.)
LTCC
L1
L2
263 μm
GND
L6
6 grounded
vias
dc bias
500 μm
Via
Via
FR4
65 μm
CMOS
100 μm
L10
L11
FR4 board
Figure 5-17. Layer profile presents the LTCC system-in-package including 11 metal layers (L1–
L11) and the FR4 board with a slot for CMOS chip.
In addition to increasing the substrate thickness of the antenna, sequential-rotation
techniques [65] are exploited to further improve the operation bandwidth. Figure 5-18 (A) shows
the configuration of two sequential-rotation 2×2 arrays in LTCC with opposite polarizations for
transmitting (LHCP) and receiving (RHCP), respectively. The radiating elements are single-feed
corner-truncated square patches which are sequentially rotated and fed by the sequential phase
networks. The zoom-in area shows the G-S-G-S-G pads for the flip-chip integration which is at
the bottom (L11). Figure 5-18 (B) shows the detail topology of the multi-ports feed network. The
input port has a characteristic impedance of Z0 = 50 . The four output ports are designed to
provide balanced signals with equal magnitude and an incremental 90 phase delay. Each output
port is connected to the antenna element through a /4 transformer, which results in an input
impedance of ZA. Two meander lines with electrical lengths of 90 and 270 not only provide
89
180 phase difference, but also perform the quarter-wavelength transformation to achieve the
impedance matching between the input port and the antennas.
19 mm
RX
0.93 mm
TX
Feed line
GND (L2)
λ/4
0.35 mm
Z
X
Zoom-in
View
9 mm
Y
G
G
S
50Ω
Microstrip
G
S
Vertical
transition
Antenna ground (L6)
50-Ω Microstrip (L11)
(A)
Antenna 2
Antenna 3
(ZA)
λ/4
(ZA)
ZA ; θ
ZA; θ+90°
ZA; θ+90°
ZA; θ
(ZA)
Zt; 90° Zt; 270°
Z0
λ/4
(ZA)
Antenna 1
Antenna 4
(B)
Figure 5-18. Broadband antenna design on LTCC. (A) Circularly polarized sequential-rotation
arrays (B) Detail topology of the sequential-phase feed network.
Thus, the characteristic impedance Zt is determined by
Zt2   2  Z0    Z A 2   Z0  Z A
(5-5)
Figure 5-19 shows the return loss and TX/RX isolation of the sequential-rotation array. To
allow on-wafer measurement, the probing pads and the antenna arrays need to be located on the
90
same side of the LTCC substrate. Therefore, additional vertical transitions are used to connect
the bottom microstrip lines to the top layer as shown in the inset in Figure 9. The effects of the
additional transitions are included in both simulated and measured results. The measured 10-dB
return loss bandwidth achieves 12.8%. Figure 9 also shows the simulated isolation around 30 dB
at 55 GHz without the additional transitions.
0
Magnitude (dB)
-10
-20
Return loss
-30
-40
Isolation
(Ant.)
Probing pad
-50
(Probe)
Additional
transition
LTCC
S11 (Sim.)
S11 (Mea.)
S12 (Sim.)
-60
50 51 52 53 54 55 56 57 58 59 60
Frequency (GHz)
Figure 5-19. Measured and simulated return losses (S11) and TX/RX isolation (S12) of the
antenna array.
The experimental setup for the radiation patterns is the same as the previous antenna
design. The circularly polarized gain and axial ratio were obtained by measuring the complex
fields of two orthogonal linearly polarized components followed by post processing [66]. Figure
5-20 shows the LHCP realized gain and the axial ratio at zenith against frequencies. It is
observed that the 3-dB axial ratio bandwidth is about 10%. The peak gain of the array is around
9.7 dBi at 54 GHz with 3-dB gain bandwidth around 12%. The radiation patterns in XZ- and YZplane at 54 GHz are shown in Figure 5-21. The measured LHCP gain of the main beam is 9.7
dBi with a cross polarization discrimination of about 27 dB at zenith. Compared to the single-
91
patch antenna, the wider bandwidth gives the system a better chance to operate in the optimal
antenna performance range as modeling and manufacturing variations are present. Besides, as the
circularly polarized antenna arrays applied in multiple-sensor network facing each other, it
provides better isolation against multiple interference sources, which is not achievable by the
Gain (dBi) / Axial Ratio (dB)
orthogonal linearly polarized antennas.
12
10
8
6
Gain (Sim.)
Gain (Mea.)
AR (Sim.)
AR (Mea.)
4
2
0
50
52
54
56
58
Frequency (GHz)
60
Figure 5-20. Realized gain and axial ratio (AR) spectrums of the broadband sequential rotation
patch antenna array at zenith.
(Z)
0
(dB)
20
330
30
10
0
300
0
60
30
300
60
-10
-20
-20
90 (X) -30 270
-30 270
-20
90 (Y)
-20
-10
240
10
20
330
10
-10
0
(Z)
0
(dB)
20
210
150
180
-10
Co-pol0(Sim.)
Cross-pol (Sim.)
Co-pol10(Mea.)
Cross-pol
20 (Mea.)
120
210
150
180
Figure 5-21. Radiation patterns of the patch antenna array. (A) XZ-plane. (B) YZ-plane.
92
5.3.3 Vital Sign Detection
Figure 5-22 shows the final system assembly attached to an upright cardboard facing the
target. The CMOS radar chip is on the other side of LTCC, and DC bias is provided by the FR4
board through the wire and solder bump underneath the LTCC. Compared to the previous SiP
shown in Figure 4-18, the LTCC substrate provides a more area-efficient and flexible mm-wave
integration, since all the surface components including flip-chip-integrated CMOS chip and
bypass capacitors are on the bottom side of the antenna. The entire top side area is used for the
antenna array, which potentially can be integrated with an antenna array with more number of
elements in the future applications.
FR4
LTCC substrate
9 mm
19 mm
Figure 5-22. Top-view of the final system assembly. The flip-chip-integrated CMOS radar chip
and surface-mounted bypass capacitors are placed on the other side of LTCC.
As a person sitting in front of the radar at D = 0.3 m, the baseband outputs BI(t) and BQ(t)
as previously shown in Figure 3-13 are directly sampled by an analog-to-digital converter (ADC)
at a sampling frequency fs for a period of time t, and CSD is used to eliminate the null detection
points. Figure 5-23 (A) shows the successful heartbeat detection by the LTCC radar system-inpackage at fs = 50 Hz and t = 20 s, where the person has to hold the breath to avoid the
interference by large respiration harmonics. The detected heartbeat rate at 71 beat/minute agrees
with human counting. The respiration detection is plotted in Figure 5-23 (B) with fs = 10 Hz and
93
t = 50 s. It can be seen that the spectrum resolution is improved by the longer measurement time.
Due to the nonlinear Doppler phase modulation, some harmonics of respiration signal are
prominent, but the fundamental respiration rate at 16 beat/minute can be successfully read from
CSD Spectrum
the spectrum.
1.5
Heartbeat at
71 beat/min
1
0.5
0
0
50
100
Beat/Minute
150
(A)
0.8
CSD Spectrum
Respiration at 16 beat/min
0.6
2nd harmonic at 32 beat/min
0.4
0.2
0
0
50
100
Beat/Minute
150
(B)
Figure 5-23. CSD output spectrum of vital sign detection using the broadband patch antenna
array on LTCC. (A) Heartbeat detection. (B) Respiration Detection
94
CHAPTER 6
SUMMARY
The 60-GHz micro-radar SiP including the CMOS transceiver chip, flip-chip integration,
and antennas was designed and tested. The system achieves high level of integration by the flipchip transition and small antenna size at mm-wave frequency range. The down-conversion gain
of the transceiver chip is measured to be 36 dB at 55 GHz. Successful detection of small
mechanical vibration with a displacement of 20 μm can be achieved at 0.3 m away, and the
detection range reaches 2.1 m if the displacement increases to 200 μm. The quadrature system
architecture utilizing CSD solves the null detection point problem of Doppler radar without extra
frequency tuning, which assures the robust detection against detection distance change.
The use of 60-GHz radar frequency offers various advantages such as higher sensitivity
and smaller antenna size compared to lower-frequency systems, however, the respiration
amplitude comparable to wavelength causes strong non-linear phase modulation, and relatively
small heartbeat amplitude results in demodulation difficulties. Theoretical analysis and
simulation of 60-GHz detection are provided to address these issues. Both shallow and deep
breathings are tested in the experiments, and the detection technique monitoring both the
fundamental and second harmonic of respiration is proposed. In addition, the signal recovery
algorithm is proposed to improve the accuracy of vital sign detection at 60 GHz. The algorithm
demonstrates the respiration movement can be successfully recovered in time domain, even it is
an arbitrary respiration pattern whose harmonics on spectrum could not be predicted by the
simple Bessel function analysis. .
The circularly polarized sequential-rotation antenna array integrated in the 60-GHz SiP is
implemented on the LTCC substrate. The LTCC technology with multiple metal-layer structure
is able to satisfy the different dielectric thickness requirements of antennas and microstrip feed
95
lines, and it also provides a compact mm-wave packaging option for the CMOS chip. Compared
to linearly polarized single-patch antenna, the 10-dB bandwidth of the antenna array on LTCC
increases from 1.3 GHz (2.4 %) to 7 GHz (12.8 %) at 55 GHz, and narrow 50- feed line is
obtained to realize the flip-chip transition. As the process and manufacturing variations are often
present in the mm-wave systems, wide antenna bandwidth is able to cover the possible frequency
drift and increase the system yield.
This work demonstrates the first vital sign detection by the flip-chip-integrated CMOS
radar at 60 GHz. The shorter wavelength offers significant area reduction and flexibility in
system integration. It can be readily embedded into one of the smartphone functions, for
example, making it a pervasive first-aid tool for noncontact vital sign monitoring, or applied to a
large sensor network for different vibration monitoring.
96
APPENDIX
MATLAB CODING OF TIME-DOMAIN RECOVERY ALGORITHM

clc; clear all; close all;
Data =
csvread('C:\Users\Jason\Desktop\Measurement\Full_system_0308\scope_255.csv',2
,0);
fs = 1/(Data(2,1) - Data(1,1)) %Extract sampling freuquency fs.
% ---------------------------------------------------------------------% Remove dc component. Q_p and I_p: positive baseband outputs of the Q
% and I channels. Q_m and I_m:
% negtive baseband outputs of the Q and I channels.
Q_p
Q_m
I_p
I_m
=
=
=
=
Data(:,2)-mean(Data(:,2));
Data(:,3)-mean(Data(:,3));
Data(:,4)-mean(Data(:,4));
Data(:,5)-mean(Data(:,5));
Q_p_raw = Q_p;
I_p_raw = I_p;
%Store the raw data for later comparison.
%---------------------------------------------------------------------% Choose to turn on or off the 3 levels of data smoothing. The simple
% moving average acts like a low-pass filter to reduce the effect of noise
% and prevent the misjudgment of recovery algorithm. Depending on the
% quality of raw data, more filters can be turned on to improve the
% recovery accuracy.
for i = 2:length(I_p)-1
% Low-pass filter for I_p
I_p(i) = ( I_p(i-1)+I_p(i+1) )/2;
end
for i = 2:length(I_p)-1
I_p(i) = ( I_p(i-1)+I_p(i+1) )/2;
end
for i = 2:length(I_p)-1
I_p(i) = ( I_p(i-1)+I_p(i+1) )/2;
end
for i = 2:length(Q_p)-1
% Low-pass filter for Q_p
Q_p(i) = ( Q_p(i-1)+Q_p(i+1) )/2;
end
for i = 2:length(Q_p)-1
Q_p(i) = ( Q_p(i-1)+Q_p(i+1) )/2;
end
for i = 2:length(Q_p)-1
Q_p(i) = ( Q_p(i-1)+Q_p(i+1) )/2;
end
%---------------------------------------------------------------------figure
plot(Data(:,1), I_p, '-r'); % Plot I_p before the recovery algorithm
97
hold on;
h_axis=gca;
get(h_axis,'FontSize'); % displays the default Font size
set(h_axis,'FontSize',16); % sets the font size of axis
AX = legend('B_I(t)');
LEG = findobj(AX,'type','text');
set(LEG,'FontSize',16)
grid;
ylabel('Voltage(V)');
h_ylabel = get(gca,'YLabel');
set(h_ylabel,'FontSize',16);
xlabel('Time(s)');
h_xlabel = get(gca,'XLabel');
set(h_xlabel,'FontSize',16);
title('Time-domain Vital Sign Recovery');
h = get(gca, 'title');
set(h, 'FontSize', 16);
plot(Data(:,1), Q_p, '-b'); % Plot Q_p before the recovery algorithm
%
%
%
%
---------------------------------------------------------------------Find the trend of I_p and Q_p signal and save the value to Trd_Ip and
Trd_Qp. 1 represents the voltage level is rising, and 0 represents the
voltage level is falling.
Trd_Ip(1) = 1; %Assign arbitrary trend to the first index
for i = 2:length(I_p)
if I_p(i) > I_p(i-1)
Trd_Ip(i) = 1;
elseif I_p(i) < I_p(i-1)
Trd_Ip(i) = 0;
else
Trd_Ip(i) = ~Trd_Ip(i-1);
%If I_p(i) = I_p(i-1), inverse Trd_Ip
%to prevent a false trend. It
%eliminates the slope = 0 case.
end
end
Trd_Qp(1) = 1; %Assign arbitrary trend to the first one
for i = 2:length(Q_p)
if Q_p(i) > Q_p(i-1)
Trd_Qp(i) = 1;
elseif Q_p(i) < Q_p(i-1)
Trd_Qp(i) = 0;
else
Trd_Qp(i) = ~Trd_Qp(i-1);
%If Q_p(i) = Q_p(i-1), inverse Trd_Qp
%to prevent a false trend. It
%eliminates the slope = 0 case.
end
end
% ---------------------------------------------------------------------% Start of the recovery process.
98
% I_p_fix and Q_p_fix represent the signal after recovery. Initially
% I_p_fix = I_p and Q_p_fix = Q_p;
I_p_fix = I_p;
Q_p_fix = Q_p;
% IF flip == 1, it is in flip status. IF flip == 0, it is in follow status.
% Initially set flip = 1.
flip_I = 1;
flip_Q = 1;
% Assign arbitrary initial value to previos peak and slope type as long as
% they are not equal to the value 0, 1, and -1.
PT_I_Last
Sp_Q_Last
PT_Q_Last
Sp_I_Last
=
=
=
=
10;
10;
10;
10;
for i = 4:length(Trd_Ip)-3
[PC_I PT_I] = PeakCheck(Trd_Ip, i, 1);
% Check if there is a valid
% peak on I_p.
[TC_Q Sp_Q] = TrendCheck(Trd_Qp, i, 1); % Check if there is a
% consistent trend on Q_p
% If yes, a Type-I peak is
% identified on I_p.
if PC_I && TC_Q && (Sp_Q_Last ~= Sp_Q)
flip_I = ~flip_I;
% Switch flip to follow stage or follow
% to flip stage.
if flip_I == 1
I_p_fix(i:end) = I_p_fix(i:end) - 2*(I_p(i) - I_p(i1))*ones(length(I_p(i:end)),1); %Flipping I_p and store the data to I_p_fix.
end
PT_I_Last = PT_I;
%Store the current peak and sloep type.
Sp_Q_Last = Sp_Q;
else
if flip_I == 1
I_p_fix(i:end) = I_p_fix(i:end) - 2*(I_p(i) - I_p(i1))*ones(length(I_p(i:end)),1);
end
end
[PC_Q PT_Q] = PeakCheck(Trd_Qp, i, 1);
% Check if there is a valid
% peak on Q_p.
[TC_I Sp_I] = TrendCheck(Trd_Ip, i, 1); % Check if there is a
% consistent trend on I_p
% If yes, a Type-I peak is
% identified on Q_p
if PC_Q && TC_I && (Sp_I_Last ~= Sp_I)
flip_Q = ~flip_Q;
% Switch flip to follow stage or follow
% to flip stage.
if flip_Q == 1
Q_p_fix(i:end) = Q_p_fix(i:end) - 2*(Q_p(i) - Q_p(i1))*ones(length(Q_p(i:end)),1); %Flipping Q_p and store the data to Q_p_fix.
99
end
PT_Q_Last = PT_Q;
Sp_I_Last = Sp_I;
%Store the current peak and sloep type.
else
if flip_Q == 1
Q_p_fix(i:end) = Q_p_fix(i:end) - 2*(Q_p(i) - Q_p(i1))*ones(length(Q_p(i:end)),1);
end
end
end
plot(Data(:,1), I_p_fix, '-k'); %Plot the recovered waveforms.
plot(Data(:,1), Q_p_fix, '-g');
AX = legend('B_I(t)','B_Q(t)','B_I(t)','B_Q(t)');
hold off;
% ----------------------------------------------------------------------% Start complex signal demodulation [5]
CSD1 = I_p_fix + j*Q_p_fix; % Recovered data
CSD3 = I_p_raw + j*Q_p_raw; % Use raw data before the waveform smoothing.
L = 2^20;
%FFT number of point
[H1t, W1t] = dtft(CSD1, L);
H1 = H1t(L/2+1:L);
W1 = W1t(L/2+1:L);
[H3t, W3t] = dtft(CSD3, L);
H3 = H3t(L/2+1:L);
W3 = W3t(L/2+1:L);
%-----------------------------------------------------------------------figure;
plot (W3/pi*(fs/2)*60, abs(H3),'-b'); % Plot the baseband spectrum before
% recovery
h_axis=gca;
get(h_axis,'FontSize'); % displays the default Font size
set(h_axis,'FontSize',16); % sets the font size of axis
xlim([0 150]);
grid;
ylabel('CSD Spectrum');
h_ylabel = get(gca,'YLabel');
set(h_ylabel,'FontSize',16);
xlabel('Beat/Minute');
h_xlabel = get(gca,'XLabel');
set(h_xlabel,'FontSize',16);
title('Baseband Spectrum Before Recovery');
h = get(gca, 'title');
set(h, 'FontSize', 16);
figure;
plot (W1/pi*(fs/2)*60, abs(H1),'-r'); % Plot the baseband spectrum after
100
% recovery
h_axis=gca;
get(h_axis,'FontSize'); % displays the default Font size
set(h_axis,'FontSize',16); % sets the font size of axis
xlim([0 150]);
grid;
ylabel('CSD Spectrum');
h_ylabel = get(gca,'YLabel');
set(h_ylabel,'FontSize',16);
xlabel('Beat/Minute');
h_xlabel = get(gca,'XLabel');
set(h_xlabel,'FontSize',16);
title('Baseband Spectrum After Recovery');
h = get(gca, 'title');
set(h, 'FontSize', 16);
%-----------------------------------------------------------------------
function [PC PT] = PeakCheck(Trend, Idx, N)
% Idx: center index,
% Centered at Idx, check N indexes to the left and N-1 to the right to
% determine if it is a valid peak. N>=1.
% If it is a valid peak, PC==1. If it is not a valid peak, PC==0.
% If the peak type is rising first and then falling, PT == 1.
% If the peak type is falling first and then rising, PT == 0.
PC = 0;
PT = 0;
i = 1;
while i <= N
if (Trend(Idx) == Trend(Idx+i-1)) && (Trend(Idx) ~= Trend(Idx-i))
if i == N
PC = 1;
if Trend(Idx+i-1) == 0
PT = 1;
else
PT = 0;
end
end
i = i+1;
else
PC = 0; break
end
end
%-------------------------------------------------------------------------
function [TC Sp]= TrendCheck(Trend, Idx, N)
% Idx: center index,
% Centered at Idx, check N indexes to the left and N-1 indexes to the right
% to determine if there is a consistent trend. N>=1.
% If there is a consistent trend, TC == 1. If there is not a consistent
% trend, TC == 0.
101
% If the slope of the trend > 0, Sp == 1. If the slope of the trend < 0,
% then Sp == -1.
% The generate of trend functions Trd_Ip and Trd_Qp has eliminated the
% slope = 0 case.
TC = 1;
i = Idx-N+1;
while (TC == 1) && (i <= Idx+N-1)
if Trend(Idx-N) == Trend(i)
TC = 1;
if Trend(i) == 1
Sp = 1;
else
Sp = -1;
end
i = i+1;
else
TC = 0;
Sp = 0;
break
end
end
%------------------------------------------------------------------------
function [H, W] = dtft(h, N)
%DTFT
calculate DTFT at N equally spaced frequencies
%---%
Usage:
[H, W] = dtft(h, N)
%
%
h : finite-length input vector, whose length is L
%
N : number of frequencies for evaluation over [-pi,pi)
%
==> constraint: N >= L
%
H : DTFT values (complex)
%
W : (2nd output) vector of freqs where DTFT is computed
%--------------------------------------------------------------% copyright 1994, by C.S. Burrus, J.H. McClellan, A.V. Oppenheim,
% T.W. Parks, R.W. Schafer, & H.W. Schussler. For use with the book
% "Computer-Based Exercises for Signal Processing Using MATLAB"
% (Prentice-Hall, 1994).
%--------------------------------------------------------------N = fix(N);
L = length(h); h = h(:); %<-- for vectors ONLY !!!
if( N < L )
error('DTFT: # data samples cannot exceed # freq samples')
end
W = (2*pi/N) * [ 0:(N-1) ]';
mid = ceil(N/2) + 1;
W(mid:N) = W(mid:N) - 2*pi;
% <--- move [pi,2pi) to [-pi,0)
W = fftshift(W);
H = fftshift( fft( h, N ) ); %<--- move negative freq components
102
LIST OF REFERENCES
1.
C. Li, Y. Xiao, and J. Lin “Experiment and spectral analysis of a low-power Ka-band
heartbeat detector measuring from four sides of a human body ” IEEE rans. Microw.
Theory Tech., vol. 54, no. 12, pp. 4465-4471, Dec. 2006.
2.
A. D. Droitcour, V. M. Lubecke, J. Lin, and O. Boric-Lubecke “A microwave radio for
Doppler radar sensing of vital signs ” IEEE MTT-S Int. Microw. Symp. Digest, pp. 176178, May, 2001.
3.
C. Li X.Yu C. Lee L. Ran and J. Lin “High-sensitivity software configurable 5.8 GHz
radar sensor receiver chip in 0.13 μm CM S for non-contact vital sign detection ” IEEE
Trans. Microw. Theory Tech., vol. 58, no 5, pp. 1410-1419, May 2010.
4.
C. Li, and J. Lin, "Non-contact measurement of periodic movements by a 22-40 GHz
radar sensor using nonlinear phase modulation," IEEE MTT-S Int. Microw. Symp.
Digest, pp. 579-582, June 2007.
5.
C. Li, and J. Lin, "Complex signal demodulation and random body movement
cancellation techniques for non-contact vital sign detection," IEEE MTT-S International
Microwave Symposium, pp. 567-570, Atlanta, June, 2008.
6.
S. Nicolson, P. Chevalier, A. Chanter, B. Sautreuil and S. P. Voinigescu "A 77–79 GHz
Doppler radar transceiver in silicon", IEEE Compound Semiconduct. Integr. Circuits
Symp., pp.1, 2007
7.
S. Kim and C. Nguyen "On the development of a multifunction millimeter-wave sensor
for displacement sensing and low-velocity measurement", IEEE Trans. Microw. Theory
Tech., vol. 52, no. 52, pp.2503 -2512, Nov. 2004
8.
Y. Yan, L. Cattafesta C. Li J. Lin “Analysis of detection methods and realization of a
real-time monitoring RF vibrometer ” IEEE rans. on Microw. heory ech., vol. 59, no.
12, pp. 3556-3566, Dec 2011.
9.
Keyence LK-GSeries Laser Displacement Sensor User Manual, Keyence Corporation,
Woodcliff Lake, NJ.
10.
K. Abe, K. Otsuka, and J.-Y. Ko “Self-mixing laser Doppler vibrometry with high optical
sensitivity: Application to real-time sound reproduction,” New J. Phys. vol. 5 no. 1 pp.
8.1-8.9, 2003.
11.
K. Otsuka, K. Abe, J.-Y. Ko, and T.-S. Lim “Real-time nanometer-vibration
measurement with a self-mixing microchip solid-state laser ” pt. Letters, vol. 27, no. 15,
pp. 1339-1341, 2002.
12.
Y. Yan, C. Li, J.A. Rice J. Lin “Wavelength division sensing RF vibrometer ” IEEE
MTT-S Int. Microw. Symp., Baltimore, MD, June 5-11, 2011.
103
13.
A. D. Droitcour, O. Boric-Lubecke . M. Lubecke J. Lin and G. . A. Kovac “Range
correlation and I/Q performance benefits in single-chip silicon Doppler radars for
noncontact cardiopulmonary monitoring ” IEEE Trans. Microw. Theory Tech., vol. 52,
no.3, pp. 838-848, Mar. 2004.
14.
C. Balanis, Antenna Theory Analysis and Design. Hoboken, NJ: John Wiley & Sons,
Inc., 2005, ch. 2, pp. 110-112.
15.
E. Knott, Radar Handbook. New York, NY: McGraw Hill Company, 2008, ch. 14, pp.
10.
16.
C. Gu, C. Li, J. Lin, J. Long, J. Huangfu, and L. Ran, "Instrument-based noncontact
Doppler radar vital sign detection system using heterodyne digital quadrature
demodulation architecture, " IEEE Trans. Instrum. Meas., vol. 59, no. 6, pp. 1580-1588,
Jun. 2010.
17.
A. D. Droitcour, O. Boric-Lubecke, V. M. Lubecke, and J. Lin “0.25m CMOS and
BiCM S single chip direct conversion Doppler radars for remote sensing of vital signs ”
IEEE Int. Solid State Circuits Conference, Digest of Technical Papers, pp. 348-349, Feb.
2002.
18.
C. Li, Y. Xiao, and J. Lin, "A 5 GHz double-sideband radar sensor chip in 0.18 µm
CMOS for non-contact vital sign detection," IEEE Microw. and Wireless Components
Letters, vol. 18, issue 7, pp. 495-496, 2008.
19.
T. Yao, L. Tchoketch-Kebir, O. Yuryevich, M. Q. Gordon, and S. P. oinigescu “65
GHz Doppler sensor with on-chip antenna in 0.18 um SiGe BiCM S” IEEE M -S Int.
Microw. Symp. Dig., pp. 1493-1496, June 2006.
20.
W. Sansen “Analog IC Design in Nanometer CM S echnologies ” Proceedings of
VLSI Design 2009: Improving productivity through higher abstraction, The 22nd
International Conference on VLSI Design, New Delhi, India, Jan. 2009.
21.
E. Laskin, M. Khanpour, S. T. Nicolson, A. Tomkins, P. Garcia, A. Cathelin, D. Belot,
and S. P. Voinigescu “Nanoscale CM S transceiver design in the 90-170 GHz range ”
IEEE Trans. Microw. Theory Tech., vol. 57, pp. 3477-3490, Dec. 2009.
22.
J. Lee Y. Chen and Y. Huang “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.
23.
J. Lee Y. Li M. Hung and S. Huang “A fully-integrated 77-GHz FMCW radar
transceiver in 65-nm CM S technology ” IEEE J. Solid-State Circuits, vol. 45, no. 12,
pp.2746-2756, Dec. 2010.
24.
A. Tang, Q. J. Gu, and M.-C. F. Chang “CM S receivers for active and passive mmwave imaging ” invited paper to IEEE Communication Magazine vol. 49 no. 10 ct.
2011.
104
25.
B. Razavi “A 60-GHz CMOS receiver front-end ” IEEE J. Solid-State Circuits, vol. 41,
no.1, pp.17-22, Jan. 2006.
26.
B. Razavi “A millimeter-wave CMOS heterodyne receiver with on-chip LO and
divider ” IEEE J. Solid-State Circuits, vol. 43, no.2, pp.477-485, Feb. 2008.
27.
D. . Petkie C. Benton and E. Bryan “Millimeter wave radar for remote measurement
of vital signs ” IEEE Radar Conference pp.1-3, May 2009
28.
Y. Xiao, J. Lin, O. Boric-Lubecke and . M. Lubecke “Frequency tuning technique for
remote detection of heartbeat and respiration using low-power double-sideband
transmission in Ka-band ” IEEE rans. Microw. heory Tech., vol. 54, pp. 2023-2032,
May 2006.
29.
B. Razavi, "Design of millimeter-wave CMOS radios: A tutorial," IEEE Transactions on
Circuits and Systems I: Regular Papers, vol.56, no.1, pp.4-16, Jan. 2009
30.
A. Parsa et al. “A 60 GHz CM S receiver using a 30 GHz L ” in ISSCC Dig. ech.
Feb. 2008, pp. 190-191.
31.
H. Chuang H. Kuo F. Lin . Huang C. Kuo and Y. u “60-GHz millimeter-wave life
detection system (MLDS) for noncontact human vital-signal monitoring ” IEEE Sensors
Journal, vol. 12, issue 3, pp. 602-609, March 2012.
32.
C. Li and J. Lin “ ptimal carrier frequency of non-contact vital sign detectors ”
Proceedings of IEEE Radio and Wireless Symp., pp. 281-284, Long Beach, Jan. 2007.
33.
Y. P. Zhang D. Liu “Antenna-on-chip and antenna-in-package solutions to highlyintegrated millimeter-wave devices for wireless communications ” IEEE rans. Antennas
Propagat., vol. 57, no. 10, pp. Oct. 2008.
34.
K. Kang et al. “A 60-GHz OOK receiver with an on-chip antenna in 90 nm CM S ”
IEEE J. Solid-State Circuits, vol. 45, no. 9, pp. 1720-1731, Sept. 2010.
35.
S. Reynolds, B. Floyd, U. Pfeiffer, T. Beukema, J. Grzyb, C. Haymes, B. Gaucher, and
M. Soyuer “A sillicon 60 GHz receiver and transmitter chipset for broadband
communications ” IEEE J. Solid State Circuits vol. 41 no. 12 pp. 2 20-2831, Dec.
2006.
36.
A. Jentzsch and W. Heinrich “ heory and measurements of flip-chip interconnects for
frequencies up to 100 GHz ” IEEE rans. Microw. Theory Tech., vol. 49, pp. 871-878,
May 2001.
37.
W. Heinrich “ he flip-chip approach for millimeter wave packaging ” IEEE Microw.
Mag., vol. 6 IEEE Microw. Mag., vol. 6, pp. 36-45, Sept. 2005.
105
38.
C.-C. Kuo, P.-A. Lin, H.-C. Lu, Y.-S. Jian, C.-M. Liu, Y.-M. Hsin and H. Wang “Wband flip chip assembled CMOS amplifier with transition compensation network for SiP
integration ” in IEEE M -S Int. Microw. Symp. Dig., Jun. 2010, pp. 465-468.
39.
C.-C. Kuo, P.-A. Lin, C.-F. Tai, Y.-L. Chang, Y.-S. Jiang, J.-H. Tsai, H.-C. Lu, Y.-M.
Hsin, and H. Wang “Flip chip assembled W-band CMOS chip modules on ceramic
substrate with transition compensation for millimeter-wave system-in-package
integration ” in IEEE Trans. Microw. Theory Tech., vol. 60, no. 3, pp. 766-777, Mar.
2012
40.
M. C. Budge, Jr. and M. P. Burt, “Range correlation effects on phase and amplitude
noise,” Proc. IEEE Southeastcon, Charlotte, NC, p. 5, 1993.
41.
T. Yao, M. Q. Gordon, K. K. W. Tang, Kenneth H. K. Yau, M. Yang, P. Schvan, and S.
P. oinigescu “Algorithmic design of CM S LNAs and PAs for 60 GHz Radio ” IEEE
J. Solid-State Circuits, vol. 42, no. 5, pp. 1044-1057, May 2007.
42.
S. Pellerano Y. Palaskas and K. Soumyanath “A 64 GHz LNA with 15.5 gain and 6.5
dB NF in 90 nm CM S ” IEEE J. Solid-State Circuits, vol. 43, no.7, pp. 1542-1552, Jul.
2008.
43.
E. Cohen, S. Ravid and D. Ritter “An ultra low power LNA with 15 dB gain and 4.4 dB
NF in 90 nm CM S process for 60 GHz phase array radio ” in IEEE Radio Freq.
Integrated Circuits Symp., pp. 61-64, Jun. 2008.
44.
Clayton R. Paul, Inductance: Loop and Partial. Hoboken, NJ: John Wiley & Sons, Inc.,
2009, ch. 5, pp. 201-205.
45.
J. L. Kuo Z. M. sai K. Y. Lin and H. Wang “A 50 to 70 GHz power amplifier using
90 nm CM S technology ” in IEEE Microw. Wireless Compon. Lett., vol. 19, no.1, pp.
45-47, Jan 2009.
46.
B. Razavi, Design of Analog CMOS Integrated Circuits. New York, NY: McGraw Hill
Company, 2001, ch. 7, pp. 215.
47.
T. O. Diskson, M. A. LaCroix, S. Boret, D. Gloria, R. Beekens, and S. P. Voinigescu,
“30-100 GHz inductors and transformers for millimeter-wave (Bi)CMOS integrated
circuits ” IEEE rans. Microw. heory ech. vol. 53 no. 1 pp.123-134, Jan. 2005.
48.
Y. Cao, R. A. Groves, X. Huang, N. D. Zamdmer, J. O. Plouchart, R. A. Wachnik, T. J.
King and C. Hu “Frequency-independent equivalent circuit model for no-chip spiral
inductors ” IEEE J. Solid-State Circuits, vol. 38, no. 3, pp. 419-426, Mar. 2003.
49.
M. Kraemer D. Dragomirescu and R. Plana “Accurate electromagnetic simulation and
measurement of millimeter-wave inductors in bulk CM S technology ” 10th Topical
Meeting on Silicon Monolithic Integrated Circuits in RF Systems, Jan. 2010.
106
50.
Ali M. Niknejad and Robert G. Meyer, Design, simulation, and applications of inductors
and transformers for Si RF ICs. New York, NY: Kluwer Academic Publishers, 2000, ch.
5, pp. 67-70.
51.
C. K. Lian and B. Razavi “Systematic transistor and inductor modeling for millimeterwave design ” IEEE J. Solid-State Circuits, pp. 450-457, Feb. 2009.
52.
C. Cao and K. K. “Millimeter-wave voltage-controlled oscillators in 0.13-m CMOS
technology ” IEEE J. Solid State Circuits, vol. 41, no. 6, pp. 1297-1304, June 2006.
53.
L. Li P. Reynaert and M Steyaert “Design and analysis of a 90 nm mm-wave oscillator
using inductive-division LC tank ” IEEE J. Solid-State Circuits, vol. 44, no. 7, pp.19501958, Jul. 2009.
54.
I.C.H. Lai and M. Fujishima, "Pseudo-millimeter-wave up-conversion mixer with onchip balun for vehicular radar systems," The 12th Asia and South Pacific Design
Automation Conference, pp. 106-107, 2007
55.
J. R. Long “Monolithic transformers for silicon RF IC design” IEEE J. of Solid-State
Circuits., vol. 35, no.9, pp. 1368-1382, Sept. 2000.
56.
C. Li and J. Lin “A 1-9 GHz linear-wide-tuning-range quadrature ring oscillator in 130
nm CMOS for non-contact vital sign radar application ” IEEE Microw. Wireless
Compon. Lett., vol. 20, no.1, pp. 34-36, Jan. 2010
57.
W. Yan and H. Luong “A 900-MHz CMOS low-phase-noise voltage-controlled ring
oscillator ” IEEE rans. Circuits Syst. II: Analog Digital Signal Process. vol. 4 no. 2
pp.216-221, Feb. 2001.
58.
T. Soorapanth, C. P. Yue D. K. Shaeffer . I. Lee . I. and S. S. Wong “Analysis and
optimization of accumulation-mode varactor for RF ICs ” in Symp. LSI Circuits Dig.
Tech. Papers, Jun. 1998, pp. 32-33.
59.
D. M. Pozar “Considerations for millimeter wave printed antennas ” IEEE rans.
Antennas Propagat., vol AP-31, pp. 740-747, Sept. 1983.
60.
D. Pozar “Microstrip antennas ” Proc. IEEE vol. 0 no.1 pp. 79-81, Jan 1992.
61.
Y.-F. Lu and Y.-C. Lin “A 60-GHz rectangular slot antenna with backed cavity in
L CC ” in Asia-Pacific Microw. Conf. Yokohama, Japan, pp. 1677-1680, Dec. 2010.
62.
W. Stutzman, and G. Thiele, Antenna Theory and Design. Hoboken, NJ: John Wiley &
Sons, Inc., 2005, ch. 1, pp. 30.
63.
X. Yu, C. Li, and J. Lin, "Two-dimensional noncontact vital sign detection using Doppler
radar array approach," IEEE MTT-S Int. Microw. Symp. Dig., pp. 1-4, June, 2011.
107
64.
C.-C. Tsai, Y.-S. Cheng, T.-Y. Huang, Y.-P. A. Hsu, and R.-B. Wu “Design of
microstrip-to-microstrip via transition in multilayered LTCC for frequencies up to 67
GHz ” IEEE Trans. Compon., Packag. Manuf. Technol., vol. 1, no. 4, pp. 595–601, Apr.
2011.
65.
S.-K. Lin and Y.-C. Lin “A compact sequential-phase feed using uniform transmission
lines for circularly polarized sequential-rotation arrays ” IEEE rans. Antennas Propag.,
vol. 59, no. 7, pp. 2721-2724, July 2011
66.
B. Y. oh R. Cahill and . F. Fusco “Understanding and measuring circular
polarization ” IEEE rans. Educat. vol. 46 pp. 313–318, Aug. 2003.
108
BIOGRAPHICAL SKETCH
Mr. Te-Yu Jason Kao received the B.S. degree in electrical and control engineering from
National Chiao Tung University, Hsin-Chu, Taiwan R.O.C. in 2004 and M.S. degree in electrical
engineering from the University of Washington, Seattle, WA in 2008. He received his Ph.D.
degree in electrical and computer engineering at the University of Florida in the spring of 2013.
From January to August 2011, he worked as a graduate intern technical at Intel Corporation in
Chandler, AZ. He worked on signal integrity for high speed IO circuits including PCI Express
Gen2 interface. In the Ph.D. study, his research interests include RF and millimeter-wave CMOS
circuit and system, passive component design and EM modeling, millimeter-wave packaging,
Doppler radar sensors, antennas, and bio-medical applications of RF systems. Mr. Kao is
currently a student member of IEEE and is the Student Paper Competition finalist in IEEE 2012
Radio Frequency Integrated Circuits (RFIC) Symposium, Montreal, Canada.
109