You are on page 1of 173

A Calibrated Phase and Amplitude Control System

for Phased-Array Transmitters


Cameron T. Charles
A dissertation submitted in partial fulllment
of the requirements for the degree of
Doctor of Philosophy
University of Washington
2006
Program Authorized to Oer Degree: Electrical Engineering
University of Washington
Graduate School
This is to certify that I have examined this copy of a doctoral dissertation by
Cameron T. Charles
and have found that it is complete and satisfactory in all respects,
and that any and all revisions required by the nal
examining committee have been made.
Chair of the Supervisory Committee:
David J. Allstot
Reading Committee:
Hui Liu
Brian P. Otis
Jay Rajagopalan
Brian N. Bershad
Date:
In presenting this dissertation in partial fulllment of the requirements for the doctoral
degree at the University of Washington, I agree that the Library shall make its
copies freely available for inspection. I further agree that extensive copying of this
dissertation is allowable only for scholarly purposes, consistent with fair use as
prescribed in the U.S. Copyright Law. Requests for copying or reproduction of this
dissertation may be referred to Proquest Information and Learning, 300 North Zeeb
Road, Ann Arbor, MI 48106-1346, 1-800-521-0600, to whom the author has granted
the right to reproduce and sell (a) copies of the manuscript in microform and/or (b)
printed copies of the manuscript made from microform.
Signature
Date
University of Washington
Abstract
A Calibrated Phase and Amplitude Control System
for Phased-Array Transmitters
Cameron T. Charles
Chair of the Supervisory Committee:
Professor David J. Allstot
Electrical Engineering
A system is proposed that allows the phase and amplitude of a signal to be accu-
rately set and regulated over process and power supply variations. The intended
application is in a phased array transmitter, where the phase and amplitudes of the
array elements are electronically adjusted to shape and steer the beam of radiation
in the chosen direction. Phase and amplitude errors introduced in the array elements
lead to degradation of the array pattern through eects such as sidelobe growth and
reductions in directivity.
The conventional means of setting the phase in a branch of a phased array is
through look-up tables storing the control voltage - phase relationships for the phase
shifter. This method does not compensate for changes in these relationships intro-
duced by processing variations between fabricated devices or changes in operating
conditions such as power supply voltage and temperature. Additionally, if the phase
shifter has variable gain over the phase control range then the amplitude will vary
depending on the phase setting. The proposed system uses a variable gain amplier
(VGA) in conjunction with the phase shifter to compensate for the variable losses of
the phase shifter and simultaneously provide a means of adjusting the amplitude of
the signal. Dual feedback loops are employed to set the control voltages for the phase
shifter and VGA, allowing the phase and amplitude to be closely regulated across
process variations and adjusted to compensate for changes in operating conditions.
The phase shifter is implemented as a reective-type phase shifter which provides
up to 310

of phase shift, and the VGA is a narrowband implementation with a


resonant load that provides about 20 dB of gain and is intended to be used as a power
amplier driver. The phase feedback loop includes buers to remove the amplitude
dependancy of the signal, high-speed dividers, a charge pump, and an active loop
lter. The phase of the system is specied by digitally setting the charge pump
currents. The amplitude loop consists of an RF peak detector used in conjunction
with an active loop lter, and the amplitude is specied by setting an analog input
voltage to the loop lter.
The complete system has been fabricated in a 0.18 m CMOS process, and assem-
bled on a custom printed circuit board using chip-on-board bonding to allow testing
and characterization. The system operates at 1.9 GHz, and the phase can be set with
5 bits of control over a 240

range, and the amplitude can be varied over a 20 dB


range. The amplitude feedback loop reduces the variation in |S
21
| across the phase
control range from 12.1 dB for the open loop case to 0.4 dB. The phase feedback
loop reduces the variation in

S
21
across the amplitude control range from 32.1

to
7.4

. The standard deviation of the phase across the phase control range is reduced
from 9.4

to 1.8

for changing power supply voltages, and from 5.0

to 1.2

between
dierent test chips.
TABLE OF CONTENTS
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Chapter 1: Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Dissertation Overview . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Chapter 2: Phased Arrays . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1 Types of Multiple Antenna Systems . . . . . . . . . . . . . . . . . . . 7
2.2 Phased Array Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Phased Array Architectures . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Phased Array Applications . . . . . . . . . . . . . . . . . . . . . . . . 27
Chapter 3: Previously Published Work . . . . . . . . . . . . . . . . . . . . 29
3.1 Phased Array Calibration . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Phase and Amplitude Control Techniques . . . . . . . . . . . . . . . . 31
3.3 Proposed Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
Chapter 4: Proposed System . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.1 Qualitative Explanation . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Quantitative Explanation . . . . . . . . . . . . . . . . . . . . . . . . 43
4.3 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Chapter 5: Phase Loop Design . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.1 Phase Shifter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.2 Buer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.3 Divider . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.4 Pulse Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.5 Charge-Pump . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.6 Loop Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
i
Chapter 6: Amplitude Loop Design . . . . . . . . . . . . . . . . . . . . . . 100
6.1 Variable Gain Amplier . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.2 Peak Detector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.3 Loop Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
Chapter 7: Experimental Results . . . . . . . . . . . . . . . . . . . . . . . 121
7.1 Test and Measurement Circuits . . . . . . . . . . . . . . . . . . . . . 121
7.2 Test Conguration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
7.3 Measurement Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Chapter 8: Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
8.1 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
ii
LIST OF FIGURES
Figure Number Page
2.1 Worldwide smart antenna deployments (source: Visant Strategies) . . 6
2.2 Categories of multiple antenna techniques. . . . . . . . . . . . . . . . 7
2.3 A phased array transmitter. . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 A phased array receiver, showing the relationship between the angle of
the incident wave and the time delays at the antenna elements. . . . . 11
2.5 A cell that has been partitioned into three sectors, each using a direc-
tional antenna, to allow frequency reuse within the cell through space
division multiple access. . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.6 A switched beam phased array. Users are transferred from one beam
to another as they move through the cell. . . . . . . . . . . . . . . . . 14
2.7 An adaptive array diers from a switched beam array in that the main
lobe is continuosly variable, and the nulls can be placed to reject in-
terferers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.8 A plane wave of electromagnetic radiation impinging on a linear an-
tenna array. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.9 Radiation pattern for a 4-element broadside antenna array. . . . . . . 17
2.10 Radiation patterns for a 4-element antenna array at 0

, 30

, 60

, and
90

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.11 Radiation patterns for antenna arrays with M = 4 and M = 8 elements. 19
2.12 Radiation patterns for 8 element antenna arrays with d = /2 and
d = elements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.13 Radiation patterns for 8 element antenna arrays with equal amplitude
weighting, triangular amplitude taper, and binomial amplitude taper. 21
2.14 The eect of phase and amplitude errors on sidelobe amplitude. (After:
[1].) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.15 The eect of phase quantization on the radiation pattern for an 8 ele-
ment array steered to 21

from broadside. . . . . . . . . . . . . . . . 24
iii
2.16 Architecture choices for implementing a phase shift in a multiple an-
tenna receiver. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1 Points where phase and amplitude errors can be introduced in each
branch of a phased array transmitter. . . . . . . . . . . . . . . . . . . 30
3.2 Comparison of the control voltage to phase relationships of three dif-
ferent instances of the same phase shifter. . . . . . . . . . . . . . . . 33
3.3 Variation of the phase shifters from Fig. 3.2. . . . . . . . . . . . . . . 34
4.1 High level system diagram for phase/amplitude feedback system. . . . 37
4.2 System level block diagram. . . . . . . . . . . . . . . . . . . . . . . . 38
4.3 Transient behaviour of pulse generator block. . . . . . . . . . . . . . . 41
4.4 Transient signals for I
up
= 3 I
dn
. . . . . . . . . . . . . . . . . . . . 42
4.5 System diagram for quantitative analysis of phase loop . . . . . . . . 44
4.6 Transfer function for phase shifter model . . . . . . . . . . . . . . . . 45
4.7 Charge-pump output current for varying output phases . . . . . . . . 46
4.8 System diagram for quantitative analysis of the amplitude loop . . . . 47
4.9 Transfer function model for the VGA . . . . . . . . . . . . . . . . . . 48
4.10 Schematic for phase loop simulation. . . . . . . . . . . . . . . . . . . 49
4.11 Transient waveform for the control voltage, with a step in the charge-
pump current applied at t = 1 s. . . . . . . . . . . . . . . . . . . . . 50
4.12 Schematic for amplitude loop simulation. . . . . . . . . . . . . . . . . 51
4.13 Transient waveform for the amplitude loop control voltage, with a step
in the amplitude control voltage applied at t = 1 s. . . . . . . . . . 52
4.14 Schematic for complete system simulation. . . . . . . . . . . . . . . . 53
4.15 Transient waveforms of the control voltages for the complete system
simulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.16 Transient plot of control voltages for phase and amplitude controls
changing simultaneously . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.1 Illustration of the vector addition performed by a vector modulator
phase shifter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.2 Phase variation of all-pass phase shifter with frequency. . . . . . . . . 59
5.3 Distributed phase shifter implementations: (a) varactor loaded trans-
mission line, (b) lumped element implementation. . . . . . . . . . . . 60
5.4 Operation of a reective-type phase-shifter. . . . . . . . . . . . . . . . 62
iv
5.5 Lumped element hybrid 90
o
coupler. . . . . . . . . . . . . . . . . . . 64
5.6 Simulated S31 and S41 of the lumped element coupler (including par-
asitics). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.7 Reective loads: (a) Varactor, (b) Single resonated load (SRL), (c)
Transformed single resonated load (TSRL), (d) Dual resonated load
(DRL) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.8 RTPS phase shift superimposed on TSRL impedance over the varactor
range. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.9 Eect of zero location (value of C
N
) on phase shift characteristics. . . 68
5.10 Phase shift range vs. zero location (value of C
N
). . . . . . . . . . . . 69
5.11 TSRL with inclusion of parasitics: (a) Series resistance, (b) Parallel
capacitance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.12 The eect of parasitic series resistance on phase shift characteristics. . 71
5.13 The eect of parasitic parallel capacitance on the phase shift charac-
teristics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.14 Schematic of the TSRL of the implemented RTPS. . . . . . . . . . . 74
5.15 Die photo of the implemented RTPS. . . . . . . . . . . . . . . . . . . 76
5.16 Measured phase shift range of the implemented RTPS. . . . . . . . . 76
5.17 Measured loss of the implemented RTPS. . . . . . . . . . . . . . . . . 77
5.18 Measured noise gure of the implemented RTPS. . . . . . . . . . . . 77
5.19 Schematic of the buer block. . . . . . . . . . . . . . . . . . . . . . . 79
5.20 Delay introduced by the buer across the input amplitude range. . . . 81
5.21 Conventional frequency divider block diagram. . . . . . . . . . . . . . 82
5.22 Divider core schematic . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.23 Divider multiplexer schematic . . . . . . . . . . . . . . . . . . . . . . 84
5.24 Timing diagram for pulse generator. . . . . . . . . . . . . . . . . . . . 84
5.25 State diagram for pulse generator. . . . . . . . . . . . . . . . . . . . . 85
5.26 System diagram for pulse generator. . . . . . . . . . . . . . . . . . . . 85
5.27 Schematic for pulse generator. . . . . . . . . . . . . . . . . . . . . . . 86
5.28 Schematic for a standard charge-pump. . . . . . . . . . . . . . . . . . 87
5.29 Schematic for a charge-pump with current steering to reduce charge
injection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
5.30 Schematic for the new charge-pump with buers and no switching. . . 89
v
5.31 Simulated transient current waveforms for the three charge-pump topolo-
gies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.32 Toplogy of the phase loop lter. . . . . . . . . . . . . . . . . . . . . . 92
5.33 Schematic for the operational amplier used in the phase looplter. . 93
5.34 Schematic of the reective load as it appears to the looplter. . . . . 94
5.35 Approximate gain and phase relationships for the phase shifter and VGA. 97
6.1 Four VGA topologies: (a) variable feedback, (b) variable bias, (c) cur-
rent steering, and (d) simple cascode. . . . . . . . . . . . . . . . . . . 101
6.2 Transistor model used for noise gure analysis. . . . . . . . . . . . . . 103
6.3 Schematic of variable gain amplier. . . . . . . . . . . . . . . . . . . 106
6.4 Layout of the variable gain amplier. . . . . . . . . . . . . . . . . . . 108
6.5 Simulated S21 for the variable gain amplier. . . . . . . . . . . . . . 109
6.6 Simulated noise gure for the variable gain amplier. . . . . . . . . . 109
6.7 Schematic for peak detector design and analysis. . . . . . . . . . . . . 111
6.8 Transient behavior of peak detector output voltage from Matlab sim-
ulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.9 Transient behavior of peak detector output voltage from Cadence sim-
ulation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.10 Schematic for the amplitude loop lter. . . . . . . . . . . . . . . . . . 117
7.1 Schematic for the conguration register, showing 3 of the 20 bits. . . 122
7.2 Schematic of the Observe/Drive module used for the control voltages. 123
7.3 Photograph of the assembled printed circuit board. . . . . . . . . . . 125
7.4 Die photograph of the complete system. . . . . . . . . . . . . . . . . 127
7.5 Measured phase of S
21
plotted against steps in the digital control bits. 129
7.6 Layout of the charge pump unit current elements, with the Up and Dn
arrays denoted. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
7.7 Measured phase steps of S
21
for each increment in the digital control
bits. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
7.8 Measured phase of S
21
for individual control bits (0 corresponds to only
the xed current elements being active). . . . . . . . . . . . . . . . . 132
7.9 Measured magnitude of S
21
over the phase control range. . . . . . . . 134
7.10 Time domain waveforms of VGA and phase shifter control voltages for
a step in the digital phase control bits. . . . . . . . . . . . . . . . . . 135
vi
7.11 Phase of S
21
across the amplitude control range. . . . . . . . . . . . . 136
7.12 Time domain waveforms of VGA and phase shifter control voltages for
a step in the desired amplitude input. . . . . . . . . . . . . . . . . . . 137
7.13 Phase characteristics for changing supply voltage with open loop oper-
ation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
7.14 Phase characteristics for changing supply voltage with closed loop op-
eration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
7.15 Maximum deviation in phase characteristics across the control range
for changing power supply voltage. . . . . . . . . . . . . . . . . . . . 140
7.16 Phase characteristics for three dierent test chips with open loop op-
eration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
7.17 Phase characteristics for three dierent test chips with closed loop op-
eration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
7.18 Maximum deviation in phase characteristics across the control range
for three dierent test chips. . . . . . . . . . . . . . . . . . . . . . . . 144
vii
ACKNOWLEDGMENTS
viii
1
Chapter 1
INTRODUCTION
The era of wireless communications began in the 1860s when James Clerk Maxwell,
a Scottish Physicist, predicted the existence of radio waves. In 1886, a German
physicist names Heinrich Hertz was the rst to verify that electromagnetic waves exist,
and propagate with a nite velocity. An Italian inventor named Gugliemlo Marconi
sent and received the rst radio signals in 1895 using a spark gap transmitter, and
made headlines in 1899 by using his radio to provide up-to-the-minute coverage of the
Americas Cup yacht race. The vacuum tube was invented in 1907, and subsequently
the state of the art in radio advanced rapidly, with much of the early pioneering
work being done by Edwin Howard Armstrong, including the superheterodyne receiver
which he patented in 1917.
In recent years there has been tremendous growth in research into radio frequency
(RF) circuit design, driven largely by the advent of the cellular telephone and other
forms of wireless communications. The rst cell phone call was made in 1973 by Dr.
Martin Cooper, a former general manager for the systems division at Motorola. Nine
years later the rst American cellular service, AMPS (Advanced Mobile Phone Ser-
vice) was deployed in Chicago. This generation of cellular phone service was based on
analog modulation, and is known as 1G (for rst generation). The next generation of
cellular phone service (2G) was based on digital transmission, and included the stan-
dards D-AMPS (Digital AMPS), CDMA (Code Division Multiple Access), and GSM
(Global System for Mobile Communications) in Europe. The third generation (3G)
of cell phone service added data transmission capabilities, allowing users to transmit
2
e-mail, send text messages, and access the internet. Work is presently underway on
standards for the fourth generation (4G) of cell phone systems, and while the scope of
this generation has not been fully specied, it is clear that it will require increased data
rates as compared to 3G systems. Also driving the need for increased data rates in
wireless communications systems are wireless local area network (WLAN) standards.
The most prominent WLAN standards are IEEE 802.11a, b, and g (also known as
Wi-Fi). These standards accommodate data rates up to 54 Mbps, and many new
home and oce network installations use these for their benets of easy installation
and user portability. An emerging WLAN standard, IEEE 802.11n, will triple the
data rates of the existing standards. Personal wireless connectivity standards such as
Bluetooth which allow users to achieve high data rates over short distances have also
grown in popularity.
Bandwidth is a scarce commodity in modern communications systems, with the
Federal Communications Commission (FCC) deciding which parts of the spectrum can
be used for emerging communications standards. The easiest way to accommodate
the rising data rates required by new communications standards is to increase the
bandwidth. This is often not possible, so engineers are forced to make the most of
the limited bandwidth that is available.
In the past this has involved improving spectral eciency using time and fre-
quency domain methods, such as moving to more complex modulation schemes like
M-ary PSK, QAM, etc. [2]. These modulation schemes allow each transmitted sym-
bol to represent multiple bits, thus increasing data rates without aecting the the
bandwidth. The limiting factor with these methods is the requirement for higher
signal-to-noise (SNR) ratios. One way of improving the SNR ratio is to simply use
more power to transmit the signal, but this causes problems with battery life and
IIP3 limits of power ampliers, and is not eective in most modern wireless networks
(which are interference limited).
As it becomes increasingly dicult to attain further improvements in data rates
3
through time and frequency domain methods, communications engineers are turning
to spatial methods. The use of cells allowing frequency reuse in a modern cellular
phone system is one example of a spatial method. Currently the most active spatial
research area is multiple antenna systems [3]. While there are several ways in which
multiple antennas can be exploited in a wireless system (as will be discussed in the
next chapter), the focus of this work is on beamforming applications. The most basic
requirement for a beamforming (or phased array) system is the ability to accurately
set the phase and amplitude of the signal being transmitted (or received) on each an-
tenna. This work investigates a system which compensates for the variable amplitude
response of an RF phase shifter over its phase range while simultaneously allowing
both the phase and amplitude in each branch of a phased array transmitter to be
accurately set.
1.1 Dissertation Overview
This dissertation is organized in the following manner. Chapter 2 provides an intro-
duction to the dierent types of multiple antenna systems, with a focus on phased
arrays which are the target application of this work. Basic concepts pertaining to
phased arrays are reviewed, and discussion on the importance of calibrating the phase
and amplitude of each path in a phased array is provided. Chapter 3 reviews the state
of the art in phased array calibration methods, and presents details on recent pub-
lications addressing this topic. Chapter 4 describes the system being proposed in
this work. The system is initially presented from a qualitative standpoint to give an
overview of the operation of the system as a whole, and then a quantitative system-
level analysis is provided accompanied by simulation results. Chapter 5 describes the
circuit-level implementations of each of the components in the phase loop, starting
with the phase shifter. A more rigorous analysis of the phase loop dynamics is also
provided in this chapter to arrive at the design of the loop lter. Chapter 6 describes
the circuit-level implementations of the components in the amplitude loop, starting
4
with the variable gain amplier. Similar to the previous chapter, a more rigorous
examination of the amplitude loop dynamics is also presented. Chapter 7 describes
additional circuitry that was designed for test and measurement purposes, and then
presents the measurement results from the fabricated system. Finally, conclusions are
drawn in Chapter 8.
5
Chapter 2
PHASED ARRAYS
As mentioned in the previous chapter, recent years have seen growing interest in
the use of multiple antennas in communications systems as a means of increasing
data rates without requiring additional bandwidth. The basic denition of a multiple
antenna system is any conguration of several antenna elements at the transmitting
and/or receiving side of a communications link whose signals are processed adaptively
in order to exploit the spatial dimension of the wireless channel [4].
The logarithmic relationship between the capacity of a wireless link and the signal-
to-interference-and-noise ratio (SINR) has been recognized for many years [5], and
one obvious way to improve data rates over a wireless channel is simply to increase
transmit power levels. However, in the case of a mobile user this is extremely costly in
terms of battery life. In the context of interference-limited multi-user cellular systems,
increasing the transmit power provides no net gain as the interference received from
other users increases along with the signal received from the desired user. Most
mature cellular systems are interference limited [6], so this is not a viable option. As
mentioned previously, increasing the bandwidth is also an undesirable solution (and
may not even be possible), as bandwidth is an expensive and limited resource.
Multiple antenna systems achieve increases in channel capacity by exploiting the
spatial dimension of a wireless link. The type and degree of channel capacity im-
provement depends on the multiple-antenna technique being used. The concept of
exploiting the spatial dimension is not new, as cell sectorization allowing frequency
reuse (the fundamental concept upon which cellular communications networks are
built) can be regarded as a form of spatial processing [5]. Multiple antennas used
6
Figure 2.1: Worldwide smart antenna deployments (source: Visant Strategies)
for antenna diversity have been employed in cellular telephone base stations for a
number of years, where they mitigate the three major impairments caused by the
wireless channel: fading, delay spread, and co-channel interference. Recent studies
have reported that multiple antenna technology is now deployed in one of every 10
base stations worldwide, and deployment is expected to grow by 60 percent in the
next four years [4], as shown in Fig. 2.1. The same study showed that multiple an-
tenna systems have been implemented in base stations for as little as 30 percent more
cost than traditional base station antenna systems. As the push for higher data rates
intensies, moves are now being made to incorporate multiple antenna technology in
mobile elements of wireless communications systems.
Multiple-antenna techniques can be grouped into three main categories, as illus-
trated in Fig. 2.2. The following section will briey describe each of these techniques,
and the remainder of the chapter will be devoted to phased-array multiple antenna
systems, which are the target application for the work presented in this dissertation.
7
Figure 2.2: Categories of multiple antenna techniques.
2.1 Types of Multiple Antenna Systems
2.1.1 Diversity
Antenna diversity has been known for several decades [7], and has been employed in
cellular base stations for a number of years. The primary goal of diversity techniques
is to combat fading in the wireless channel. Fading results from the multiple paths
between the transmitter and the receiver over a wireless channel (caused by scatter-
ing). When the signals arriving from dierent paths cancel destructively, the received
signal undergoes what is known as a fade, where the received signal power is greatly
reduced.
Diversity systems employ multiple antennas that are uncorrelated, and rely on
the the fact that they will undergo fading independantly from one another. When
one antenna is experiencing a fade, it is likely that some of the other antennas are
not, and a strong signal can still be received. There are several means of ensuring
8
that the dierent antennas in the system will be uncorrelated. The simplest is space
diversity, where the antennas are placed far enough apart to have uncorrelated signals.
The required distance will depend on the degree of scattering in the wireless channel,
but a distance of about 10 wavelengths is usually sucient. Other approaches are
polarization diversity, where antennas with orthogonal polarizations are employed,
and angle diversity, where directional antennas are used. The benet of the latter
two approaches is that the antennas can be housed in the same location.
Diversity techniques can be employed at either the receiver or the transmitter.
When employed at the receiver, there are three main techniques: switch diversity,
equal gain combining, and maximal ratio combining [8]. Switch diversity is the sim-
plest of the three techniques, where the antenna with the highest signal to noise ratio
(SNR) is determined, and this antenna alone is used with the receive chain. For equal
gain combining, the signals are co-phased and added together, similar to what is done
in a RAKE receiver. The drawback of equal gain combining is that if one antenna
has a low SNR, it will degrade the SNR of the entire system. To address this prob-
lem, maximal ratio combining can be used. In this scheme, the signals are weighted
according to their SNRs before co-phasing and combining them. In the presence of
noise, this leads to the best SNR for the overall system, but also has the highest
degree of complexity.
The diversity methods employed at the transmitter depend on how much channel
state information (CSI) is available. If perfect channel state information is available,
then any of the techniques described above for the receiver can also be employed
at the transmitter. This is often not the case, and there are a group of techniques
known as space-time coding techniques that can be used at the transmitter without
any knowledge of the channel characteristics. The most well known was proposed by
Siavash Alamouti [9], and involves transmitting a dierent signal on each antenna,
with delayed and redundant versions of each symbol being transmitted. By employing
some signal processing at the receiver, the same diversity benets can be realized at
9
the transmitter as if the diversity were being implemented at the receiver.
2.1.2 Spatial Multiplexing
Spatial multiplexing evolved from seminal work done by Foschini and Gans [10] and
independantly by Telatar [11] in the late 1990s. Their work shed light on the pos-
sibility of using multiple transmit and receive antennas to achieve higher data rates
without any requiring extra bandwidth or transmission power. Telatar showed that
a wireless system using N
T
transmit antennas and N
R
receive antennas can theoret-
ically acheive data rates which are linearly proportional to the minimum of N
T
and
N
R
[11]. This result depends on the channel parameters being known at the receiver,
and on the path gains between the dierent antennas behaving independantly. In
practice these conditions will not be perfectly met and the theoretical linear increase
in data rates will not be realized, but signicant increases in channel capacity are still
possible. This is an impressive result, in light of the fact that increasing the power
(and thus the SNR) in a single user system only leads to logarithmic improvements
in channel capacity.
Spatial multiplexing multiple antenna techniques require multiple antennas at
both the transmitter and the receiver, and send independant data streams over the
dierent transmit antennas. In these schemes, multipath interference is exploited to
establish multiple parallel channels operating simultaneously in the same frequency
band. The data streams are recovered and seperated at the receiver using interfer-
ence cancellation algorithms. The number of antennas at the receiver must be greater
than or equal to the number of antennas at the transmitter. As mentioned, spatial
multiplexing techniques can theoretically acheive a linear growth in channel capacity
with min(N
T
, N
R
). One of the most well known processing algorithms for this tech-
nique is the Vertical Bell Laboratories lAyered Space-Time (V-BLAST) algorithm
[12]. In this scheme, each receive antenna receives all of the signals radiated from
the N
T
transmit antennas, but the interence between the bit streams is removed by
10
Figure 2.3: A phased array transmitter.
using both optimum combining and interference cancellation. The emerging wireless
LAN standard, IEEE 802.11n is expected to mandate the use of spatial multiplexing
techniques to achieve its target data rates.
2.1.3 Phased Arrays
The third multiple antenna technique is the beamforming, or phased array technique.
Multiple antenna systems using this technique are sometimes also referred to as smart
antennas, or adaptive antennas. For beamforming, the same signal is transmitted
or received on each of the antennas, but a time delay and amplitude adjustment
is introduced at each antenna to shape the overall radiation pattern and focus it
in a particular direction. An intuitive picture of a multiple antenna beamforming
transmitter is shown in Fig. 2.3. In the presence of interferers, nulls can be introduced
in the pattern to limit the interferers eect on the received signal. This process is
also referred to as spatial ltering. A beamforming receiver is illustrated in Fig. 2.4
where the relationship between the angle of the incident wave and the time delays at
the antenna elements can be seen.
Phased array techniques work by improving the eective SNR at the output of
11
Figure 2.4: A phased array receiver, showing the relationship between the angle of
the incident wave and the time delays at the antenna elements.
the receiver. If the antennas are spaced at a sucient distance from one another, the
radiation noise of each antenna is uncorrelated, and the receiver noise sources in each
signal path are independant. As a result, the signals from the antennas add in ampli-
tude (coherently), while the noise adds in power (incoherently), creating a 10log(N)
dB improvement in the SNR for an N element phased array receiver. Intuitively,
the phased array can be thought of as focusing the transmitted (or received) energy
into a directional beam, so that more signal power arrives at the target than would
be the case for the same power being transmitted by a single isotropic antenna. To
express the advantage in terms of channel capacity, the single-user data rate bound
for a single transmit antenna and single receiver antenna system can be expressed
using Shannons equation as
C = Blog
2
_
1 +
P
T
|h|
2

2
_
(2.1)
where B is the bandwidth, h is the channel gain, P
T
is the transmitted signal power,
and
2
is the noise power. For N transmit or recieve antennas using beamforming
(assuming that the beam is steered in the proper direction), the channel capacity can
12
be expressed as
C = Blog
2
_
1 +
NP
T
|h|
2

2
_
(2.2)
As N increases, the beam becomes more focused, with a corresponding increase in
channel capacity. In principle, all of the phased array techniques used for receivers
can be used at the transmitter.
Beamforming techniques can be implemented at either the receiver or the trans-
mitter, although to date most research eorts implementing beamforming systems for
communications have focused on the receiver, since the limited channel state informa-
tion (CSI) available at the transmitter can hamper the eectiveness of beamforming
techniques. However, in the case of a base station, the directional location of the
mobile user can be derived from the uplink [5], and CSI at the transmitter can also
be obtained through the use of feedback control sequences. Modern 3G wireless stan-
dards already employ feedback sequences for power control, adaptive modulation, and
closed loop diversity modes. Recent research has shown that even a few bits of feed-
back can provide performance close to that of having complete CSI at the transmitter
[13].
An elementary implementation of the space division multiple access (SDMA) prin-
ciples used in beamforming is the sectorization of cells in a modern cellular phone
network [14]. Sectorization (see Fig. 2.5) involves subdividing the cell into three or
six sectors with dedicated directional antennas and radio frequency (RF) paths. This
allows frequency channels to be reused within the same cell, increasing the capacity
of the cellular system. One limitation of the sectorization approach is that trac
can be distributed unevenly throughout the geographical regions of a cell, leading to
unbalanced distributions of users between the sectors, saturating the network while
some sectors have unused capacity. Eliminating this imbalance would require sectors
with exible orientations and beamwidths, but this is not possible with the directional
antennas used for conventional sectorization.
This leads to the use of antenna arrays which can be electrically congured through
13
Figure 2.5: A cell that has been partitioned into three sectors, each using a directional
antenna, to allow frequency reuse within the cell through space division multiple
access.
phased array techniques. Phased arrays are typically used in one of two congura-
tions: switched beam and adaptive array. A switched beam system is considered an
extension of the current sectorization scheme, and employs an antenna array where
the phases of the signals on each antenna are adjusted to create a number of xed
narrow beams (Fig. 2.6. As an individual user moves through the cell, the signal
strength is detected, and the user is passed from one beam to another.
Taking another step up in complexity and performance leads to adaptive array
systems. In these systems the beams are continuously steerable, and are adjusted to
track the user as they move through the cell. A primary advantage of these systems is
that in addition to adjusting the main beam to track the user, the nulls in the radiation
pattern can be placed to block interfering signals to yield the highest possible SNIR.
Another advantage of the adaptive array is that there is no need to hand o the user
between dierent beams, which leads to a degradation in throughput. The dierence
14
Figure 2.6: A switched beam phased array. Users are transferred from one beam to
another as they move through the cell.
15
Figure 2.7: An adaptive array diers from a switched beam array in that the main
lobe is continuosly variable, and the nulls can be placed to reject interferers.
between a switched beam system and an adaptive array is illustrated in Fig. 2.7.
All phased array systems shape the radiation pattern of the array by adjusting the
complex weights of the dierent antenna branches, which will be discussed in the next
section.
2.2 Phased Array Theory
Antenna arrays can be laid out in a number of dierent congurations, including
linear, planar, and circular. For the purposes of this discussion, we will assume a
linear array of M antennas. We will assume a plane wave of electromagnetic radiation
is impinging on the array at an angle of relative to the axis of the antenna array, as
depicted in Fig. 2.8. The complex signal received by the antenna can be expressed as
S() = S
e
()S
a
() (2.3)
where S
e
() is the element factor and S
a
() is the array factor. The element factor is
determined by the characteristics of the individual antennas in the array. In this dis-
cussion we will only be considering the eects of the array, and as such we will assume
16
Figure 2.8: A plane wave of electromagnetic radiation impinging on a linear antenna
array.
that S
e
() = 1. The array factor accounts for the eects of the array conguration,
and can be expanded as [15]
S
a
() =
M

i=1
a
i
e
j(k
0
(i1)d sin()+
i
)
(2.4)
where a
i
and
i
are the amplitude and phase adjustments performed at the i
th
antenna
element, d is the element spacing, and
k
0
=
2

0
(2.5)
where
0
is the wavelength of the radiation. Note that the angular frequency has been
omitted from (2.4), and it is assumed that the incoming radiation has a frequency

0
. Also note that for this analysis the array is assumed to be operating in only two
dimensions.
17
Figure 2.9: Radiation pattern for a 4-element broadside antenna array.
Initially we will assume that the amplitude weights are all equal, and consider only
the eects of adjusting the phases of the dierent elements by changing
i
. If
i
= 0
for all of the elements in the array, it is said to be operating in broadside mode, and
the maximum array factor will occur at an angle of incidence = 0

. A normalized
plot of the radiation pattern of an array with 4 elements operating in broadside mode
is shown in Fig. 2.9. The main lobe of radiation at 0

can be seen, with nulls and


sidelobes for angles of incidence moving towards 90

. The sidelobes are undesirable,


as they result in interference and wasted power.
If we set a linear phase taper for all of the elements, so that

i
= k
0
(i 1)d sin(
0
) (2.6)
where 90


0
90

, the array factor can be expressed as


S
a
() =
M

i=1
a
i
e
jk
0
(i1)d(sin()sin(
0
))
(2.7)
Now by choosing a beam-pointing direction
0
and by phasing the elements according
18
Figure 2.10: Radiation patterns for a 4-element antenna array at 0

, 30

, 60

, and
90

.
to (2.6) we can steer the beam in the desired direction. Fig. 2.10 demonstrates an
array with 4 elements being scanned from
0
= 0

to
0
= 90

. When
0
= 90

, the
array is said to be operating in endre mode. From Fig. 2.10 it can be seen that the
width of the main lobe increases for larger scanning angles, more detail on this can
be found in [15].
The two variables that can be chosen in conguring the array are the number
of elements, M, and the spacing between the elements, d. The eect of increasing
the number of elements in the array is shown in Fig. 2.11. It can be seen that by
increasing the number of elements, the width of the main lobe is decreased and the
number of sidelobes and nulls in the pattern increases, while the level of the sidelobes
relative to the main lobe increases.
Fig. 2.11 shows that by increasing the size of the array through increasing the
number of elements, we can improve its performance. Another way to increase the
size of the array is to increase the element spacing, d. This is illustrated in Fig. 2.12
19
Figure 2.11: Radiation patterns for antenna arrays with M = 4 and M = 8 elements.
where the radiation pattern for an 8 element broadside array has been plotted for
d = /2 and d = . We can see that the width of the mainlobe has been reduced,
but there are now large sidelobes at the same level as the main lobe at 90

and 90

.
These are referred to as grating lobes, and are undesirable as they result in wasted
power and interference. See [15] for a detailed discussion of grating lobes and the
conditions under which they appear.
Desirable characteristics for the radiation pattern of an antenna array are a narrow
main lobe and low sidelobe levels, which means that most of the energy is focused in
the main lobe. There are a number of metrics for quantifying these qualities, including
rst null beamwidth and half-power beamwidth [14]. Another important metric is the
directivity of the array, which is given by
D
o
=
4U
max
P
rad
(2.8)
where U
max
is the power radiated per unit solid angle in the direction of maximum
radiation intensity (the main lobe) and P
rad
is the total power radiated by the array.
20
Figure 2.12: Radiation patterns for 8 element antenna arrays with d = /2 and d =
elements.
Until this point we have only considered the case where the amplitude weights
for all of the elements are equal. By choosing the amplitude weights dierently, the
sidelobe levels and spacing can be altered. One approach is to taper the amplitude of
the elements moving out from the center of the array, which reduces the sidelobe levels.
Two variants of this approach are the triangular taper and the binomial taper [15],
illustrated in Fig. 2.13. It can be seen that in addition to reducing the sidelobe levels,
the width of the main lobe is also increased. There are numerous other methods for
setting the phase and amplitude weights to steer the main lobe and null out interferers,
more details on these can be found in [16].
2.3 Calibration
A phased-array transceiver can have errors present in the phase and amplitude settings
of the antenna elements. The errors can be introduced at any point from the RF chain
in the transceiver to the antenna, and can be caused by mismatches in the length of
21
Figure 2.13: Radiation patterns for 8 element antenna arrays with equal amplitude
weighting, triangular amplitude taper, and binomial amplitude taper.
the transmission lines feeding the dierent antennas, mismatches introduced by the
phase shifter, etc. These errors can be classied into two groups, static and dynamic
[17]. Static errors are caused by dierences in component tolerances and physical
construction, and in larger arrays attempts are typically made to calibrate these errors
out. Dynamic errors result from the fact that the parameters of electronic components
vary with temperature and can drift over time [18], and can be eliminated through
periodic recalibration of the array. Changes in the absolute values of the phase and
amplitude do not aect the beam pattern, only changes in the relative phase and
amplitudes between elements [19], so if the phases and amplitudes drift in unison
then the array performance is not aected. In this work we will consider phase and
amplitude errors that are introduced in the phase and amplitude control modules of
each element, and assume that the rest of the antenna array operates in an ideal
manner.
22
Prior work has examined in detail the eect of phase and amplitude errors on the
array pattern [20], here we will summarize the key results that pertain to this work.
Random phase and amplitude errors lead to three nondesirable eects in a phased
array: increased sidelobe levels, reduction in directivity, and beam pointing error. For
this discussion the errors are dened as a phase error
n
and an amplitude error
n
at array element n.
2.3.1 Sidelobes
As mentioned previously, the presence of sidelobes in the radiation pattern is un-
desirable, and a phased array is designed with the intent of minimizing them. The
most signicant impact of phase and amplitude errors is to increase the size of these
undesirable sidelobes. The residual sidelobe level due to the phase and amplitude
errors (normalized to the isotropic level) can be expressed as [16]

2
I
= g
e
(
2
+
2
) (2.9)
where g
e
is the element gain (directivity),
2
is the variance of the phase error and
2
is the variance of the amplitude error (both assumed to have a gaussian probability
distribution function with zero mean). From this it can be seen that as the variance
of the phase and amplitude errors increase, the sidelobe levels will also increase. A
plot of the residual sidelobe levels (relative to isotropic radiation) due to phase and
amplitude errors is shown in Fig. 2.14 [1].
From Fig. 2.14 it is evident that a phase (amplitude) error can be equated to an
equivalent amplitude (phase) error. It can be shown that [16] and amplitude error of
1 dB is equivalent to a phase error of 6.6

.
2.3.2 Directivity
As dened previously, the directivity of a phased array is the ratio of the power density
in the direction of the main lobe to the power density from an isotropic source. In
23
Figure 2.14: The eect of phase and amplitude errors on sidelobe amplitude. (After:
[1].)
the design of a phased-array system it is desirous to maximize the directivity. Phase
and amplitude errors cause a reduction in the directivity of the phased array, which
can be expressed as [20]
D
D
0
=
1
1 +
2
+
2
(2.10)
Where D is the directivity of the array with errors and D
0
is the directivity of the
error-free array. The reduction in directivity is usually insignicant compared to the
increase in sidelobes.
2.3.3 Beam Pointing Error
The third non-ideality introduced by phase errors is beam pointing error. For an array
of N elements with uniform amplitude, the variance of the beam pointing deviation
can be expressed as [16]

2
=
12
N
3

2
(2.11)
24
Figure 2.15: The eect of phase quantization on the radiation pattern for an 8 element
array steered to 21

from broadside.
2.3.4 Quantization Errors
Many phased-arrays make use of phase and amplitude control schemes with discrete
levels instead of a continuum of phases and amplitudes. This simplies the control
circuitry, but also results in periodic phase and amplitude errors across the array.
The quantization errors are highly correlated, and they result in large, well-dened
sidelobe pattern errors called quantization lobes [16]. Fig. 2.15 compares the radiation
patterns resulting from 2 bits and 5 bits of phase quantization to the ideal case. It
can be seen that 5 bits of quantization provides a reasonable approximation to the
ideal case, with less than 2 dB of error in the radiation pattern.
2.4 Phased Array Architectures
There are a number of dierent phased-array architectures, diering in where to the
phase shift is inserted in the receive/transmit chain [21]. Four dierent options are
25
Figure 2.16: Architecture choices for implementing a phase shift in a multiple antenna
receiver.
illustrated for the case of a receiver in Fig. 2.16, the same principles hold for the
transmitter.
The rst option is to insert the phase shift in the radio frequency (RF) path (Fig.
2.16(a)), which has the advantage of minimal power consumption and area, since only
the antennas and phase shifters (and variable gain ampliers if it is necessary to adjust
the amplitude of each path) must be duplicated, and the intermediate frequency (IF)
and baseband stages can be shared between all signal paths. Another advantage is
that since the interferers are nulled out at the RF stage, linearity requirements on the
IF/baseband stages can be relaxed.
The second option is to insert the phase shift at the IF stage (Fig. 2.16(b)),
which is desirable from the standpoint that it is easier to realize accurate phase
shifts at lower frequencies. Some disadvantages of this approach are that there will
be increased power and area consumption due to increased duplication (now each
26
antenna path requires a dedicated mixer for the downconversion to IF) and the phase
shifters themselves will be larger since the passive components required in a phase
shifter scale in inverse proportion to the frequency of operation, so inductors and
capacitors will be larger than they would be if the phase shifter was operating at RF.
The third option is to apply the phase shift to the local oscillator (LO) signal before
it is applied to the mixer (Fig. 2.16(c)). This scheme will have a power consumption
between that of the RF and IF phase shift architectures. It is attractive since it
relaxes the requirement for the phase shifter to have constant amplitude for varying
phase shifts. Since the mixers are typically hard driven, the LO stages are operated
in saturation, and variation in the phase shifter output will have a reduced impact. A
disadvantage of this scheme is that the time delay is not be applied to the modulated
signal, so for very wideband signals some distortion will be introduced. A detailed
derivation of this eect is given in [22]. However, for almost all communication systems
the bandwidths are such that the delay is negligible in the modulated signal.
The fourth option is to add an analog to digital (A/D) converter to each path
and perform the phase shift and any other required manipulation in the digital do-
main (Fig. 2.16(d)). This is an attractive option from a exibility standpoint, as
the receiver/transmitter could be recongured for any of the MIMO techniques that
have been discussed. However, the additional A/D converters (one for each antenna)
greatly increase the power consumption, and it is dicult to design A/D converters
with adequate performance as the linearity, dynamic range, and speed requirements
are very stringent.
The four architectures shown in Fig. 2.16 can be grouped into two categories.
The rst three architectures are analog implementations, and the fourth is a digital
implementation. The principle of analog beamforming has been known since the
1960s [23], while digital beamforming techniques have evolved more recently. The
advantage of the analog techniques is that they provide drastic reductions in power
dissipation and fabrication costs, since they eliminate the need to duplicate the entire
27
RF chains between the dierent branches [24]. Their limitation is that since the phase
shift is introduced after the baseband signals have been combined (for the case of a
transmitter), they are limited to single user or point to point communications. The
advantage of the digital approach lies in its exibility, and its capacity for multi-user
beamforming [3]. It allows multiple beams to be formed and directed to an arbitrary
number of users.
This work will focus on analog beamforming, using the architecture presented
in Fig. 2.16(a). This architecture holds the possibility of minimizing the power
consumption and area, which are the motivating factors for considering the analog
architecture. The remainder of the discussion in this chapter will focus on analog
beamforming applications.
2.5 Phased Array Applications
Phased arrays have traditionally been used primarily for radar applications [25], [26].
Most of these applications are for military purposes, although there are emerging
applications in the civilian sector. One example is the development of short-range
vehicular radar systems for collision prevention and driver assistance [21]. Most mil-
itary applications are less concerned with cost, and are implemented using discrete
components or high performance semiconductors (such as GaAs). With the emer-
gence of automotive radar, more eort is being put into reducing the cost of phased
array systems [26], and one avenue for doing this is by exploring the possibilities of
integrated CMOS implementations.
The increasing prevalence of lower cost MMIC phased-array implementations cou-
pled with the growing demand for high data rate wireless communications has led to
an increased interest in phased-array systems for communications applications. Digi-
tal beam steering implementations are more suited for use in cellular base stations, but
analog implementations have also been reported in recent years [27]. Another possible
use of analog beamsteering in a base station would be to enable adaptive sectorization
28
which would allow network loading to be optimized as compared to the current xed
sectorization schemes. There has also been interest in the use of phased arrays for
broadband wireless access (BWA) systems, which use wireless links to bridge the last
mile between the subscriber and the service provider [28]. This is attractive for the
speed of deployment and limited infrastructure requirements as compared with wired
solutions. Analog beamforming systems are also suitable for broadband terrestrial
LMCS/LMDS and satellite systems [3]. Another analog beamforming application is
wireless ad-hoc networks [29].
Other research has been done into combining beamforming with other MIMO tech-
niques to maximize spectral eciency [30]. This idea is incorporated into the soon
to be released IEEE 802.11n standard for wireless LANs, which uses MIMO spatial
multiplexing techniques with optional beamforming. The move of phased arrays into
the consumer electronics domain necessitates low cost and compact systems, which
point to fully integrated CMOS implementations. The work presented in this disser-
ation will focus on a CMOS implementation of phase and amplitude control circuits
for a phased-array transmitter.
29
Chapter 3
PREVIOUSLY PUBLISHED WORK
As discussed in the previous chapter, phase and amplitude errors in the branches
of a phased array have a negative eect on the systems performance. This chapter
will discuss how these errors are introduced, and dierent approaches for mitigating
them. The chapter will conclude with a brief introduction to the proposed technique,
which is expanded upon in the following chapter.
3.1 Phased Array Calibration
Phased-array calibration is widely recognized as an important topic, and numerous
papers have been published on research into methods and algorithms for phased array
calibration [17], [18], [19]. There are a number of points in a phased array transmitter
where phase and amplitude errors can be introduced, as pictured in Fig. 3.1. Errors
introduced in the RF phase and amplitude control sections are enumerated in numbers
1 to 3. These can result from improper control voltages being applied to the phase
shifter and VGA (1), as well as from unwanted amplitude variations across the control
range of the phase shifter (2) and unwanted phase variations across the amplitude
range of the VGA (3). Errors introduced in the physical antenna conguration are
enumerated in numbers 4 to 7. These include discontinuities at the interfaces between
lines and connectors (4), dierences in the length of transmission lines (5), the air
interface discontinuity (6), and mutual coupling between antennas (7).
Dierent types of phased array systems use dierent methods for dealing with
these phase and amplitude errors, as will be discussed in the following sections.
30
Figure 3.1: Points where phase and amplitude errors can be introduced in each branch
of a phased array transmitter.
3.1.1 High Precision Applications
In phased array applications that require high precision such as military radar, each
branch of each phased array system is calibrated. These systems are typically pro-
duced in low volumes, so the additional cost associated with the extra calibration is
not a signicant drawback. Additionally, these systems often operate at very high
frequencies where small dierences in feedline lengths introduce signicant phase er-
rors, so individual branch calibration is important. Even after extensive calibration,
errors can still be introduced by changes in system behaviour over operating condi-
tions such as temperature and power supply voltages (for the active RF components).
These errors can be addressed by including the capacity for on-line self calibration to
periodically adjust the operation according to the changes in operating conditions.
3.1.2 Low Cost Applications
In lower cost, high volume applications precision is often not as critical and the
cost associated with calibrating each individual branch of each phased array system
31
is prohibitively high. These systems often operate at lower frequencies, where the
phase errors caused by small length mismatches in antenna feed lines are insignicant
enough that they can be neglected. For these applications calibration is performed
on one unit, and this information is used across a large number of devices, trading
accuracy for cost savings. This method is susceptible to changes in operating con-
ditions (temperature and power supply voltages) as mentioned above, and it is also
susceptible to manufacturing variations, as the system behavior will change across
dierent fabrication runs.
In this work, we will focus on the errors introduced by the RF phase and amplitude
control chain of the transmitter (sources 1 to 3 in Fig. 3.1). Errors introduced by
the physical antenna conguration are either calibrated out (as described for high
precision applications) or are small enough to be neglected (as described for low cost
applications) and are less susceptible to changes in temperature and power supplies
than the RF chain. The following sections describe dierent methods for setting the
phase and amplitude in the RF chain and discuss the phase and amplitude errors that
can be introduced in each case.
3.2 Phase and Amplitude Control Techniques
For the case of a phased array transmitter using analog beamforming under con-
sideration here, there are several approaches for setting the phase and amplitude.
These will be reviewed in the following sections, citing examples of previous work.
The techniques used for a receiver are typically interchangeable with those used for a
transmitter, so several of the examples of prior work cited will be for receivers. The
advantages and disadvantages of each approach will be reviewed, and nally a brief
introduction to the approach proposed in this dissertation will be given.
32
3.2.1 Look-up Tables
The most common approach for setting the phase of the signal is through control
voltage to phase shift relationships stored in look-up tables [31]. For this approach,
the control voltage to phase shift behavior of the RF phase shifter is characterized
and stored in a ROM. When the control unit determines the phase shift required,
it consults the look-up table to determine the proper control voltage, and then uses
a D/A converter to apply the analog control voltage to the phase shifter. Many of
the designs that use this approach employ vector modulator phase shifters (to be
described in more detail in Chapter 5) which require more than one control voltage,
increasing the complexity of the look-up tables required [32], [33].
One advantage of this approach is that it allows arbitrary phase resolution, limited
only by the number of entries in the look-up table and the number of bits in the D/A
converter used to set the control voltage. As described in Chapter 2, quantized phase
and amplitude in a phased array lead to degradation of the radiation pattern through
the introduction of quantization lobes (similar to grating lobes). Another advantage
of this approach is that it compensates for non-linearities in the control voltage to
phase relationships of the phase shifter.
The primary disadvantage of this approach is that the phase shifter characteri-
zation that is stored in the look-up tables is performed under one set of operating
conditions, and changes in operating conditions during system operation or variations
between phase shifter instances (due to manufacturing variations) will cause errors
to be introduced. An example of this is shown in Fig. 3.2 where the control voltage
to phase relationships of three dierent instances of the same phase shifter have been
characterized. It can be seen that there is some variation between the dierent phase
shifters, which would result in errors being introduced if one of the curves is used to
represent all of the phase shifters. This variation is made more clear in Fig. 3.3 where
it can be seen that the phase error introduced can approach 40

. These errors can be


33
Figure 3.2: Comparison of the control voltage to phase relationships of three dierent
instances of the same phase shifter.
minimized by including a calibration mode to allow the look-up tables to be updated
[34], but this approach is more suited to phased array receivers than to transmitters.
Another possible source of error is the amplitude control mechanism. The look-
up tables typically only store the control voltage to phase relationships, and changes
in the amplitude (to shape the sidelobes of the radiation pattern) will alter these
relationships, as described in [34]. If a VGA is used to set the amplitude of the signal,
it will introduce some additional phase shift that will vary with the gain. Additionally,
phase shifters often have a varying gain across their phase shift range. These phase
shifter/VGA interactions (2 and 3 in Fig. 3.1) are not accounted for in the look-up
tables, and are additional sources of error.
34
Figure 3.3: Variation of the phase shifters from Fig. 3.2.
3.2.2 Direct Digital Setting
Another approach for setting the phase is to set it directly using digital control. This
approach is most often used with the local oscillator (LO) phase shifting architecture
(2.16(c)). In these designs, the phase shift is achieved by multiplexing between dier-
ent xed phase shifted versions of the LO signal. These can either be generated from
a higher frequency LO signal which is divided down to yield the dierent phases [35],
or by standard phase shifting techniques [22].
The primary advantage of this approach is its simplicity, since it eliminates the
need for ROM-based look-up tables and D/A converters to set analog control voltages.
In the case of [35] where there are no explicit phase shifters, the system can also be
more compact since the dividers can be implemented using only transistors, unlike RF
phase shifters which typically require large passive components. Another advantage
is that the variable losses of the phase shifter have less of an impact since the LO
signal can be hard-limited in driving the mixer.
35
The primary disadvantage of the direct digital approach is that the inherent quan-
tization places limits on the phase resolution that can be achieved, leading to unde-
sirable quantization lobes in the radiation pattern. The resolution can be increased
by using more bits, but this leads to very high LO frequencies in the divider scheme
[35] and to large numbers of xed phase shifters in the other case [22], and in both
cases the distribution of the dierent LO phases becomes increasing dicult as the
number of bits increases. Another disadvantage is that this method typically pro-
vides no control over the amplitude of the signals, or if it does then the varying
phase shift introduced by the VGA as its gain changes introduces errors in the phase
characteristics.
3.3 Proposed Technique
The approach for setting the phase and amplitude proposed in this dissertation seeks
to combine the strengths of the two approaches described above. It provides the
capability for continuous variation of both phase and amplitude, automatically com-
pensates for interactions between the phase shifter and VGA at dierent phase/gain
settings and varying operating conditions, and obviates the need for look up tables.
It achieves these goals by using feedback loops to generate on the y control voltages
(can be thought of as a dynamic single entry look-up table) whenever the phase or
amplitude needs to be adjusted. The proposed approach requires additional complex-
ity to implement the feedback loops, but the power and area requirements are small
in comparison to the phase shifter and VGA system components. Further discussion
of the proposed system is provided in the following chapter.
36
Chapter 4
PROPOSED SYSTEM
As described previously, a phased array must have a means of accurately setting
the phase and amplitude of each path in order to predictably steer the beam of radi-
ated (or received) power in the desired direction. This work develops and implements
a feedback system for accurately setting the gain and phase for each antenna path in
a phased-array transmitter. The aim of this work is to provide a means of controlling
the phase and amplitude which can compensate for changes in operating conditions
and manufacturing variations, and compensate for non-idealities in the phase shifter
and VGA characteristics.
The proposed system couples the phase shifter and VGA and allows them to
compensate for phase/amplitude non-idealities in each other, using methods similar
to those used in polar feedback linearization techniques for power ampliers (PAs)
[36]. A conceptual depiction of the phase/amplitude control system being proposed
is shown in Fig 4.1. The system is composed of two feedback loops, one for the
phase and one for the amplitude. The inputs to each feedback subsystem are the
phase/amplitude at the output, and the desired phase/amplitude. Since the output
is taken after both the phase shifter and the VGA, the phase feedback loop can
correct for phase errors introduced by the VGA and the amplitude feedback loop
can compensate for amplitude errors introduced by the phase shifter. Additionally,
with proper design of the phase and amplitude feedback loops, the proposed system
compensates for changes in operating conditions and provides an attractive alternative
to setting the phase and amplitude using look-up tables.
The rst section of this chapter provides a qualitative description of how the pro-
37
Figure 4.1: High level system diagram for phase/amplitude feedback system.
posed system operates, and the reasoning and design considerations used in choosing
the architecture. The second section provides a quantitative analysis of the system
loop dynamics, and describes how they aect the system design. The nal section
provides system simulation results that have been obtained from verilogA modeling.
4.1 Qualitative Explanation
A detailed block diagram of the proposed system is shown in Fig. 4.2. This diagram
shows the blocks that are present in each of the phase and amplitude feedback loops.
The system is designed to be used in two dierent modes of operation. In the transmit
mode, the single pole single throw (SPST) switches are in the o position, and the
feedback loops are disconnected. The single pole double throw (SPDT) switches
are in the right hand position, and the control voltages for the phase shifter and
VGA are supplied by the control logic Digital to Analog Converters (DACs). In this
conguration the feedback loops are not operating, and thus do not consume any
extra power. The upconverted, modulated signal is applied to the input, and the
38
Figure 4.2: System level block diagram.
phase shifter applies the desired phase shift to the RF signal. The VGA applies the
desired gain to the signal, allowing both the phase and amplitude to be controlled.
Since this structure is intended for transmit applications, the VGA in Fig. 4.2
could have been implemented as a power amplier (PA), and directly controlled the
power of the transmitted signal. However, several recent PA designs intended for
wireless standards requiring large dynamic ranges of transmission have implemented
the PA in two blocks to make it easier to provide the required dynamic range [37],
[38]. The rst block is a variable gain power amplier driver, and the second block
is a power amplier. We have chosen to design the system in this manner, and in
Fig. 4.2 the VGA is intended to be used as a power amplier driver. The inputs to
the phase and amplitude feedback loops in the nal system should still be taken after
the PA, since it will also introduce non-idealities to the phase and amplitude of the
output signal. The PA in Fig. 4.2 is shown in dotted lines because it will not be
implemented in this work.
The phase shift was implemented at RF (as compared with the other choices shown
39
in Fig. 2.16) for two reasons. First, as mentioned previously, this choice minimizes
duplication in the transmit path since none of the IF circuitry needs to be duplicated
for each antenna path, and thus can lead to area and power consumption improve-
ments. Secondly, it was desired to use the phase shifter and the VGA to compensate
for non-idealities in each other, this is most easily done if they are operating at the
same point in the signal chain.
The concepts being applied in this work are similar to those used in polar feedback
correction for power amplier linearization [36]. These systems make use of phase and
amplitude feedback loops to improve the linearity of a PA. If the proposed system
was used with a PA that employed polar feedback linearization, much of the phase
and amplitude detection circuitry for the feedback loops could be shared between the
proposed system and the linearization circuitry for the PA.
An ideal phase shifter would have a constant gain (or loss) over its entire phase
shift range, and an ideal VGA would have a constant phase shift over its entire
amplitude range. For real components this is not the case, the losses in the phase
shifter vary with the phase shift, and the phase shift introduced by the VGA varies
with the gain. Measurement results from a representative stand-alone phase shifter
indicate that the output amplitude will vary by 12 dB over the phase control range,
and simulations for the VGA indicate that the output phase will vary by 20

over
the amplitude control range. If a look-up table approach is used to independantly
set the control voltages for the phase shifter and VGA, the errors introduced to the
radiation pattern (in the form of directivity error, beam pointing error, and sidelobe
distortion as described in Chapter 2) will be signicant. In this work, the feedback
loops determine the required control voltages for the phase shifter and VGA, to allow
them to compensate for non-idealities in each other. The phase shifter adjusts its
phase shift to compensate for the undesired phase shift introduced by the VGA, and
the VGA adjusts its gain to compensate for undesired losses introduced by the phase
shifter.
40
When it is necessary to adjust the phase and amplitude of the antenna element
(to steer the main beam in another direction), the system is switched to the calibrate
mode. This is done by removing the modulation on the signal applied at the input
so that a single tone is being applied. The SPST switches in Fig. 4.2 are switched
to their on positions, and the SPDT switches are moved from the right to the left,
so that the control voltages for the phase shifter and VGA are being supplied by the
feedback loops.
The feedback for each of the phase and amplitude loops is taken from after both the
phase shifter and the VGA, allowing them to compensate for each other. Considering
rst the phase loop, the input tone is compared to the amplied and phase shifted
signal at the output. Both signals are amplied by the buer/limiters shown in the
diagram to remove any amplitude dependance, and are then applied to the divide by 2
circuits shown in the diagram. The purpose of the divide by 2 circuits is twofold. First,
they allow the pulse generator/charge-pump combination that is used as a variable
phase detector to operate at a lower frequency, relaxing the design constraints on
these blocks. Second, as will be explained subsequently, they allow the full 360

of
phase range to be specied at the phase control input without running into boundary
constraints. Next, the signals from the divide by 2 blocks are applied to the pulse
generator. The output of the pulse generator is shown in Fig. 4.3, it acts as a digital
circuit that sets its output high on the rising edge of one signal and resets its output
low on the rising edge of the second signal. The in1 and in2 signals in Fig. 4.3 are
shown as having a 25% duty cycle because of the preceding divider circuits which act
as pulse swallowers.
The output of the pulse generator is then applied to the charge-pump. For a high
input, the charge-pump block will source a current I
up
into the lter block, and for
a low input the charge-pump will sink a current of I
dn
out of the lter block. The
charge pump currents I
up
and I
dn
are set by the phase control input, and this is how
the desired phase shift is specied during the calibration mode.
41
Figure 4.3: Transient behaviour of pulse generator block.
The negative feedback action of the loop ensures that the control voltage at the
output of the lter block will settle to a steady state value, which means that in the
steady state condition, the charge being removed and added to the control node by the
charge-pump in each cycle must be equal. If I
up
and I
dn
are set equal to each other, we
can see from Fig. 4.3 that the two inputs to the pulse generator must be oset from
each other by 180

. Because of the divide by two circuits preceding the pulse generator


inputs, this means that the phase shifter/VGA combination will be applying 360

(or
equivalently, 0

) of phase shift. If I
up
is increased to 3 times the amplitude of I
dn
, we
can see from Fig. 4.4 that the inputs to the pulse generator will be at 90

oset from
each other, which means that the phase shifter/VGA combination will be applying
180

of phase shift. By symmetry it is clear that if we reverse the situation and set I
dn
to be 3 times that of I
up
, the phase shifter/VGA combination will be applying -180

.
From this analysis we can see that the inclusion of the divide by two circuit allows us
to cover the entire 360

range of phase shifts with a reduced range of charge-pump


currents, while maintaining the duty cycle of the pulse generator between 25% and
75%.
The lter can be designed in a similar manner to lters used for charge-pump
42
Figure 4.4: Transient signals for I
up
= 3 I
dn
phase-locked loops (PLLs), and its design will determine the loop dynamics and set-
tling time.
The amplitude loop operates in a similar manner to the phase loop in that it makes
use of negative feedback to set the VGA control voltage to the value that yields an
output with the desired amplitude. The rst component in the amplitude feedback
loop is the peak detector, which determines the amplitude of the output and converts
it to a dc signal. The output of the peak detector and the amplitude control signal
are the inputs to the integrator block. The integrator block is a dierential circuit
which integrates the dierence between the peak detector output and control input
(which can be positive or negative), and increases or decreases its output accordingly.
This allows the VGA control voltage to settle to the value that will yield an output
with the desired amplitude.
To resume the description of the calibration sequence, after a time period long
enough to permit the phase shifter and VGA control voltages to settle to their steady
43
state values for the desired control inputs, the control circuitry detects the steady
state values of these nodes using the Sense lines in Fig. 4.2, stores the values, and
then outputs them on the Control lines. The SPST switches are then turned o to
disconnect the feedback loops, and the SPDT switches are moved to the right hand
positions to allow the phase shifter and VGA control voltages to be set by the control
circuitry. The feedback loops can then be powered o, the modulated signal can be
resumed at the input, and transmission can continue as normal with the phase and
amplitude that was set during the calibrate phase.
4.2 Quantitative Explanation
This section will provide a quantitative analysis of the proposed system. The analysis
will assume linearized, frequency domain models for each of the blocks shown in Fig.
4.2, and will proceed in a manner similar to that used for the analysis of charge-pump
PLLs [39]. In order to simplify the analysis, the phase and amplitude loops will be
analyzed seperately. In reality they will not operate independantly, as changes within
the amplitude loop as it is settling will aect the settling of the phase loop, and vice
versa. They can be made to operate independantly (to a reasonable approximation)
by designing the loops with signicantly dierent time constants (so that one loop
changes by a negligible amount in the time it takes the other loop to settle). This
is not a desirable option as it increases the overall time for the calibration period,
which should be minimized. Interactions between the phase and amplitude loops will
be covered in more detail in Chapter 5 when the design of the phase loop lter is
discussed.
4.2.1 Phase Loop
The system diagram for the quantitative analysis of the phase loop is shown in Fig.
4.5. For analytical convenience, a xed 180

phase shift has been inserted after the


divide by 2 circuit for the input signal. The analysis assumes that the system is in
44
Figure 4.5: System diagram for quantitative analysis of phase loop
the locked state.
We rst make the assumption that the input phase is constant, and for convenience
we choose
in
(t) = 0

. We assume that the phase shifter transfer function is as shown


in Fig. 4.6. Since we have assumed that the input phase is zero, the output phase
can be expressed as

out
(t) = K
PS
V
cont
(t) (4.1)
where K
PS
is the phase shifter gain (the slope of the curve in Fig. 4.6) and V
cont
(t)
is the control voltage.
Next considering the pulse generator/charge-pump combination, the transient out-
put current waveform for dierent values of
out
is shown in Fig. 4.7. From this gure
we can see that the average current output over one cycle can be expressed as a sum
of the up and down pulses as
I
avg
(t) =
I
up
(t)( 0.5
out
(t))
2

I
dn
(t)( + 0.5
out
(t))
2
(4.2)
where I
up
(t) and I
dn
(t) are the magnitudes of the up and down current pulses, respec-
tively. They have been given time dependance because the output phase is adjusted
45
Figure 4.6: Transfer function for phase shifter model
by changing the amplitude of the current pulses with respect to each other. We can
factor this equation to express it as
I
avg
(t) =
1
2
(I
up
(t) I
dn
(t))

out
(t)
4
(I
up
(t) +I
dn
(t)) (4.3)
If we then express the sum and dierence of the up and down currents as I
sum
(t) =
I
up
(t) +I
dn
(t) and I
diff
= I
up
(t) I
dn
(t), we can express the average current as
I
avg
(t) =
I
diff
(t)
2


out
(t)I
sum
(t)
4
(4.4)
Since we will we working in the frequency domain, to avoid the convolution op-
erator when taking the Laplace transform of (4.4), we will make the simplifying as-
sumption that the sum of the up and down currents is a constant. This means that
I
sum
(t) = I
sum
, and when the up and down currents are adjusted to change the phase,
they are adjusted so that they change relative to each other but their sum remains
a constant (if one is increased the other is decreased). We can express the control
voltage as
V
cont
(t) = I
avg
(t) h(t) (4.5)
where h(t) is the impulse response of the loop lter. If we then take the Laplace
transform of these equations and substitute (4.4) into (4.5) and the result into (4.1),
46
Figure 4.7: Charge-pump output current for varying output phases
we can express the closed loop transfer function as

out
(s)
I
diff
(s)
=
2K
PS
H(s)
4 +K
PS
H(s)I
sum
(4.6)
To analyze the simplest case, we will assume that the lter in Fig. 4.5 is a simple
capacitor, so that H(s) = 1/sC and we can express the closed loop transfer function
as

out
(s)
I
diff
(s)
=
A
1 +s/p
(4.7)
where A = 2/I
sum
is the dc gain and p = K
PS
I
sum
/4C is the single pole. We can
see that this is a simple low pass lter with a bandwidth determined by the sum of
the up and down currents, the phase shifter gain, and C, the capacitor in the loop
lter. For this simplied analysis the loop will be unconditionally stable, however, to
conrm the stability it would be necessary to carry out an analysis which takes into
account the continuous time approximation used for the charge pump (similar to that
done by Gardner for charge-pump PLLs [40]). The result of this would be some limit
on the loop bandwidth to guarantee stability. From the analysis for this simple loop
lter choice we can see which variables aect the bandwidth (and thus settling time)
of the loop.
47
Figure 4.8: System diagram for quantitative analysis of the amplitude loop
4.2.2 Amplitude Loop
This section will provide a quantitative analysis of the amplitude loop, assuming it
is operating independantly from the phase loop. The analysis will be carried out in
the frequency domain using Laplace transforms, similar to the phase loop analysis
performed in the previous section. A system diagram of the amplitude loop for the
analysis is shown in Fig. 4.8. The quantity being operated on is the amplitude of
the sinusoidal signal being applied to the VGA, and as such the peak detector is not
necessary (it is assumed to be ideal) and has been omitted from Fig. 4.8.
It is assumed that the input amplitude A
in
is constant, and the output amplitude
is altered by adjusting V
amp
(t), the amplitude control input to the integrator. The
amplitude at the output of the VGA can be expressed as
A
out
(t) = A
in
K
V GA
V
cont
(t) (4.8)
where K
V GA
is the gain of the VGA. The VGA transfer function model is shown in
Fig. 4.9.
We can express the output of the integrator as
V
cont
(t) =
1

_
t
0

V
amp
(t) A
in
K
V GA
V
cont
(t)dt (4.9)
48
Figure 4.9: Transfer function model for the VGA
where is the time constant for the integrator. Here we have assumed that the
quantities being integrated are all zero at t = 0 to simplify the Laplace transforms.
Taking the Laplace transform of (4.9) yields
V
cont
(s) =
1
s
(V
amp
(s) A
in
K
V GA
V
cont
(s)) (4.10)
We can now take the Laplace transform of (4.8), sub in (4.10), and rearrange to
nd the closed loop transfer function as
A
out
(s)
V
amp
(s)
=
1
1 +s/p
(4.11)
where p is the single pole at p = K
V GA
A
in
/. We can see that the bandwidth of the
system depends on the VGA gain, the amplitude of the input signal, and the time
constant of the integrator. The dc gain of the closed loop system is unity, as we would
expect since the negative feedback and the integrator serve to drive the amplitude at
the output to match amplitude control input V
amp
.
4.3 Simulation Results
This section provides sytem level simulation results from verilogA modeling in Ca-
dence. The rst two subsections provide simulation results for the phase and ampli-
tude loops operating independantly to verify the analysis that has been carried out
49
Figure 4.10: Schematic for phase loop simulation.
in the preceding sections. The nal subsection provides simulation results for the
complete system.
4.3.1 Phase Loop
The schematic for the phase loop containing all of the verilogA blocks is shown in
Fig. 4.10. The loop lter capacitor has a value of 10 pF, and the gain of the phase
shifter is set at K
PD
= 2.09 rad/V. The bias currents for the charge pump are intially
both set at 100 A, and then at t = 1s one is increased to 110 A while the other
is decreased to 90 A.
The transient response of the control voltage is shown in Fig. 4.11. It can be seen
that it follows the familiar exponential form characteristic of a one-pole system. To
verify that the analysis and the simulation match, we can calculate the time constant
50
Figure 4.11: Transient waveform for the control voltage, with a step in the charge-
pump current applied at t = 1 s.
for the pole as

p
=
4C
I
sum
K
PS
= 300 nS (4.12)
After one time constant, the control voltage should have reached 63.2% of its nal
value. The control voltage starts at 900 mV and moves to 600 mV, so after one time
constant it should be at 900 (300 0.63) = 710 mV, and this is conrmed by the
markers in Fig. 4.11.
4.3.2 Amplitude Loop
The top level schematic for the amplitude loop containing the verilogA blocks is shown
in Fig. 4.12. The input is a sinusoidal source with A
in
= 100 mV. A peak detector
has been included with an exponential decay. The time constant for the decay of the
peak detector was set small enough so that it did not inuence the ideal peak detector
assumption in the analysis of the previous section. The gain of the VGA is set to
51
Figure 4.12: Schematic for amplitude loop simulation.
K
V GA
= 11.11, and the time constant of the integrator is set to
int
= 400 ns.
The transient response of the control voltage is shown in Fig. 4.13. It can be seen
that it follows the familiar exponental form characteristic of a one-pole system. To
verify that the analysis and the simulation match, the time constant of the loop can
be calculated as

p
=

int
K
V GA
A
in
(4.13)
After one time constant, the control voltage should have reached 63.2 % of its nal
value. The control voltage starts at 910 mV and moves to 610 mV, so after one time
constant it should be at 910 300 0.63 = 720 mV, and this is conrmed by the
markers in Fig. 4.13.
4.3.3 Complete System
This section provides simulation results for the complete system. The schematic for
the complete system is shown in Fig. 4.14. The verilogA components and loop param-
52
Figure 4.13: Transient waveform for the amplitude loop control voltage, with a step
in the amplitude control voltage applied at t = 1 s.
eters are the same as from the seperate loop simulations described in the preceding
sections, with the exception of the phase shifter and VGA blocks. These blocks have
been modied to include the non-ideal amplitude changes introduced by the phase
shifter, and the non-ideal phase shifts introduced by the VGA. These values were
chosen to be representative of transistor level simulations of the phase shifter and
VGA. The phase shifter gain was modeled as changing from -10 dB to 4 dB over the
phase shift range, and the VGA phase shift was modelled as changing from 0

to 22

over the amplitude range. Each non-ideality was modelled as having a linear relation-
ship with the control voltage, but more accurate relationships could be introduced
for more accurate modelling (at the cost of longer simulation times).
The transient waveforms of the two control voltages are shown in Fig. 4.15 where
the phase shifter control voltage is the thick line. From the plot it can be seen the
control voltages initially stabilize to their steady state values. At t = 2 s (marker A
on the plot) the amplitude control input is adjusted to change the amplitude of the
53
Figure 4.14: Schematic for complete system simulation.
54
Figure 4.15: Transient waveforms of the control voltages for the complete system
simulation.
output signal, and the VGA control voltage changes sharply to adjust for the new
amplitude. As the VGA changes its gain, its non-ideal phase shift also changes, and
a small change in the phase shifter control voltage can be seen, which compensates
for the VGA phase shift. At t = 4 s (marker B on the plot) the phase control input
is adjusted to change the phase of the output signal, and the phase shifter control
voltage changes sharply. As the phase shifter changes, the gain (or loss) of the phase
shifter changes, and the VGA control voltage changes to compensate for the change
in the phase shifter loss and maintain a constant amplitude at the output.
Simulations were also run with both the phase and amplitude control inputs chang-
ing at the same time. The transient waveforms of the control voltages for this case
are plotted in Fig. 4.16. Both control inputs were changed at t = 3 s, and it can
be seen that both the phase and amplitude control voltages settle quickly without
adverse eects from the feedback loops interacting with each other. The following
two chapters will give detailed descriptions of the implementations of the components
55
Figure 4.16: Transient plot of control voltages for phase and amplitude controls chang-
ing simultaneously
in the phase and amplitude loops.
56
Chapter 5
PHASE LOOP DESIGN
This chapter will cover the design of each of the components in the phase feedback
loop. The phase shifter is covered in the greatest detail, as its operation is most crucial
for the performance of the overall system. The general design strategy for each of
the blocks is to keep the topologies as simple as possible to minimize power and area
consumption of the system.
5.1 Phase Shifter
5.1.1 Requirements
There are a number of general requirements for a phase shifter to be used in a multiple
antenna beam steering system. The phase shifter should be compact and consume
minimal power (if an active topology is chosen), since the number of phase shifters
scales in proportion to the number of antennas in the system. Smaller phase shifters
allow more antennas to be incorporated into the system, and reduced power con-
sumption extends the battery life in portable applications. The phase shifter should
also have a low noise gure (NF) and minimal loss. In this work the phase shifter
will be used in conjunction with a VGA, so the gain of the VGA can compensate for
the loss of the phase shifter and this requirement is relaxed to some degree. Another
requirement is that the phase shift should be approximately constant over the signal
bandwidth, to minimize distortion eects which lead to increases in the bit error rate
(BER) in the case of communications applications. In most architectures it is impor-
tant for the phase shifter to have constant gain across the phase shift range, however
in this work the VGA will compensate for changes in the gain of the phase shifter.
57
Finally, the phase shifter must provide at least 180

of phase shift with high resolu-


tion. It would be preferable to have 360

of phase shift, but if only 180

is available
then the full range can still be covered through switching a fully dierential signal.
The switches required in this situation will introduce additional losses to the signal,
so this is less desirable than having a full 360

of phase shift. Phase shift ranges of


less than 180

and limited phase resolution both limit the eectiveness of the beam
steering.
With these requirements in mind, we will discuss dierent phase shifter topolo-
gies and select the one that comes the closest to meeting the requirements for this
application. The most well known architectures are switched high-pass/low-pass
(HP/LP) lters [41],[42], reective-type phase shifters [43],[44],[45],[46],[47], all-pass
lters [48],[49],[50], and vector modulators [51],[52]. For high phase resolution, switched
HP/LP phase shifters require large amounts of area, and are not suitable for low cost
designs [51]. For this reason, they will not be considered further here.
5.1.2 Vector Modulators
Vector modulator phase shifters work by weighting and combining several signal paths
with xed phase shifts in order to get the desired phase shift at the output. The
incoming signal is split into three paths, each with a xed 120
o
phase shift relative
to the next signal path. Weighting is performed by a variable gain amplier for each
path, and weights are chosen to obtain the desired total phase shift at the output
when the signal paths are recombined. This operation can be thought of in terms
of a vector addition between the dierent signal paths, as shown in Fig. 5.1. The
xed phase osets are typically generated by high-pass and low-pass passive networks.
Vector modulators are capable of phase shift ranges of 360

, can provide gain, and


are well suited for monolithic microwave integrated circuit (MMIC) integration [51].
The disadvantage of vector modulators is that they require multiple control voltages
to regulate the phase shift. Each control voltage requires a seperate digital to analog
58
Figure 5.1: Illustration of the vector addition performed by a vector modulator phase
shifter.
converter (DAC) to interface between the digital control circuitry and the vector
modulator, which adds to the complexity and power consumption of the design.
For the phase control architecture being proposed here, the phase shifter archi-
tecture is limited to one control voltage, since the control voltage is derived from a
feedback loop with only one output. A vector modulator architecture requiring only
one control voltage has been proposed [52], but this architecture is limited to 120
o
of
phase shift, which also excludes it from consideration for our purposes.
5.1.3 All-pass Phase Shifters
All-pass phase shifters are a class of active phase shifter which have the advantage of
requiring only one control voltage. An all-pass network has the characteristic of having
a constant amplitude response over frequency (the source of the all-pass designation),
and a phase response which varies with frequency. The transfer function of general
second-order all-pass network can be expressed as
H(s) =
s
2


0
Q
s +
2
0
s
2
+

0
Q
s +
2
0
(5.1)
The phase characteristics of this network can be expressed as
() = 2 tan
1
_

0

Q(
2
0

2
)
_
(5.2)
59
Figure 5.2: Phase variation of all-pass phase shifter with frequency.
The phase characteristics with changing frequency are plotted in Fig. 5.2. By
changing the value of
0
, the curve plotted in Fig. 5.2 is shifted to the right or
left, and the phase shift at the frequency of interest is altered. The value of
0
is usually changed using a varactor or active inductor. All-pass phase shifters are
usually active, and can provide gain in the signal path. This advantage is oset to
some extent by their high power consumption, 93 mW and 60 mW for the designs
reported in [48] and [49], respectively. They are also usually implemented in expensive
III-V semiconductors, with exceptions such as the SiGe design reported in [50].
Another drawback of the all-pass phase shifter is the variation in phase shift with
frequency. The variation of phase shift (as seen in Fig. 5.2) is determined by the
variable Q in (5.1). For varactors with limited tuning range, higher values of Q are
necessary to achieve the required phase shift over the varactor tuning range. This
leads to a higher variation in phase shift over the bandwidth of the signal, which
in turn will lead to distortion and increased bit error rates (BER). An advantage of
the all-pass architecture is that it provides a constant amplitude response over the
phase shift range, however, for this work that is not a requirement since the VGA will
60
Figure 5.3: Distributed phase shifter implementations: (a) varactor loaded transmis-
sion line, (b) lumped element implementation.
compensate for amplitude variations in the phase shifter output.
5.1.4 Distributed Phase Shifters
Another phase shifter architecture which requires only one control voltage is the
distributed phase shifter, which is typically implemented as a varactor loaded trans-
mission line. Two possible implementations are shown in Fig. 5.3. At higher frequen-
cies, coplanar waveguides (CPW) are periodically loaded with varactors (Fig. 5.3(a))
which allow the phase velocity of the line to be altered to achieve dierent phase shifts
[53]. This architecture has the advantage of a constant phase shift over a very wide
bandwidth. At lower frequencies, the transmission lines can be replaced with lumped
element equivalents [54], as shown in Fig. 5.3(b). This implementation has a reduced
bandwidth as compared with the previous case.
The disadvantage of these designs is that they typically require a large area, due
to the large number of sections required to achieve 360

of phase shift (16 sections in


[54]). The number of sections required can be reduced, but this increases the losses.
61
Distributed phase shifters are usually implemented as passive circuits, and for GaAs
implementations the losses can be as low as 4 dB. However, for a low cost CMOS
implementation the loss per section is much higher due to the conductive substrate
of CMOS technology. This has been addressed in [55] by adding active circuitry to
amplify the signal before and after the lumped element transmission line. This greatly
mitigates the loss, however, the drawback is a high power consumption of 170 mW.
5.1.5 Reective-Type Phase Shifters
A fourth phase shifter architecture is the reective-type phase shifter. A reective-
type phase shifter (RTPS) consists of a hybrid 90

coupler combined with two reec-


tive loads. The operation of an RTPS is illustrated in Fig. 5.4. The incoming signal
is split evenly into two parts, with one part of the signal experiencing a 90

phase
shift relative to the other. Each part of the signal is then reected by a load with a
dierent input impedance than the coupler, introducing an additional phase shift ().
This phase shift is equal to the phase of the reection coecient, which is expressed
as
r =
Z
L
Z
0
Z
L
+Z
0
(5.3)
The reected signals are then split and phase shifted again by the coupler, so
that the signal components emerging from the input are 180

out of phase and cancel


each other out. The signals emerging from the output port are in phase and combine
constructively, with a total phase shift of 90

+ . The phase shift is varied by


incorporating a varactor into the reective load, allowing the input impedance to be
changed, thereby changing the phase shift.
Many previously reported RTPS implemenations have shown favorable results,
with areas as small as 0.5 mm
2
, losses as low as 4.9 dB, and phase shift ranges of
over 360

. While none of the designs achieved all of these performance measures


simultaneously, it indicates that the RTPS is a promising architecture, and through
62
Figure 5.4: Operation of a reective-type phase-shifter.
appropriate tradeos it should be possible to realize an implementation which per-
forms adequately for the proposed work. Most of the previously reported work has
been implemented in GaAs [44],[46],[45],[43]. To minimize the cost of the phase shifter
it is desirable to implement it in a CMOS process. This presents a number of design
challenges, the foremost being the increased losses due to the conductive substrate
and the lower quality passives. In [47] a technique was presented for reducing the
losses in a CMOS RTPS by introducing active elements to compensate for losses in
the inductors. This design reported worst case losses of -11 dB, which is approaching
the losses of some of the GaAs designs. The total phase shift for this design was lim-
ited to 110

, so more work is needed, but the architecture shows promise for CMOS
implementations.
5.1.6 Reective-Type Phase Shifter Analysis
A reective-type phase shifter (RTPS) consists of a hybrid 90

coupler combined with


two reective loads. The operation of an RTPS was illustrated in Fig. 5.4.
63
Hybrid 90

Coupler
The hybrid 90

coupler is a microwave circuit that is typically implemented using


microstrip lines. It can also be implemented using lumped element equivalents, as
shown in Fig. 5.5. The component values are calculated as follows:
C
1
=
1

0
Z
0
(5.4)
L
1
=
Z
0

2
(5.5)
C
2
=
1

2
0
L
1
C
1
(5.6)
where Z
0
is the input impedance and
0
is the center frequency. Our design was
done for a center frequency of
0
= 2.0 GHz, which yielded values of C
1
= 1.77 pF,
C
2
= 323 pF, and L
1
= 2.25 nH (after adjustments to compensate for parasitics). The
simulated transmission magnitudes and phases for the coupler (including parasitics)
are shown in Fig. 5.6. Simulations were also run to determine the sensitivity of
the coupler performance to variations in component values. It was found that a
10% change in the value of C
1
or L
1
resulted in a dierence of 1.5 dB between the
magnitudes of the two paths, with a negligible eect on the phase. Altering the value
of C
2
by 10% had a negligible eect, and it was found that removing the four C
2
completely only shifted the crossover frequency (the frequency where the incoming
signal is evenly split) by 2.5%. The crossover frequency could then be restored to
the desired value by reducing the size of the C
1
capacitors, resulting in an increase in
loss of only 0.09 dB as compared to the coupler with the C
2
capacitors present, and
the same relative phase shifts. This is likely because the calculated value for the C
2
capacitor is small compared to C
1
, and the inductor has enough parasitic capacitance
to ground to adequately perform the function of this capacitor. In designs where the
omission of the C
2
capacitors will signicantly reduce layout and wiring overhead, it
may be a good choice to omit them, since the impact on perfomance is very small.
64
Figure 5.5: Lumped element hybrid 90
o
coupler.
Figure 5.6: Simulated S31 and S41 of the lumped element coupler (including para-
sitics).
65
Figure 5.7: Reective loads: (a) Varactor, (b) Single resonated load (SRL), (c) Trans-
formed single resonated load (TSRL), (d) Dual resonated load (DRL)
Reective Load
The amount of phase shift that can be obtained from an RTPS is determined by the
design of the reective load. A number of common reective loads are shown in Fig.
5.7. The simplest is a single varactor (Fig. 5.7(a)). The phase shift of the single
varactor load can be increased by adding a series inductor which resonates with the
varactor at the operating frequency, adding a zero to the impedance function (single
resonant load, Fig. 5.7(b)). The phase shift can be further increased by adding
a parallel capacitance, which adds a pole to the impedance function (transformed
single resonant load, Fig. 5.7(c)). Further increases in phase shift can be obtained by
using two parallel single resonant loads, each resonating and introducing a zero at a
dierent point in the varactor capacitance range (Fig. 5.7(d)).
The choice of reective load depends on how much phase shift is required, the
tuning range of available varactors, and the allowable loss. As the complexity of the
reective load increases, the total phase shift increases, as does the loss. For most
CMOS processes the passive components are relatively low quality, so it is best to
minimize the complexity of the load. For this work we have chosen the transformed
66
single resonated load (TSRL). The analysis presented here assumes that the varactor
has a tuning range of 2, which means that the maximum phase shift attainable with
the TSRL is 360

, and practical implementations will have less than this theoretical


limit. As mentioned previously, phase shifts of 180

are adequate provided that


dierential signaling is used, since 360

of phase shift can be obtained by switching


the polarity of the inputs to the phase shifter. The switching network necessary to
obtain a full 360

range will introduce loss, so in a complete implementation the loss


introduced by the switches would have to be compared with the loss introduced by
moving to a more complex reective load structure.
Ideal Components
This section will analyze the TSRL assuming ideal components, to obtain strategies
for sizing the components. The impedance of the circuit shown in Fig. 5.7(c) can be
expressed as
Z
L
=
1
2
LC
V
j (C
T
+C
V

2
LC
T
C
V
)
(5.7)
From this we can see that the TSRL impedance is purely imaginary (as expected),
and assuming that is constant, as C
V
changes, there are two singularities, one in the
numerator where the TSRL impedance will go to zero, and one in the denominator
where the TSRL impedance will go to innity. We will designate the values of C
V
where these singularities occur as C
N
and C
D
, respectively. If we then choose L =
1/
2
C
N
and C
T
= C
D
C
N
/(C
D
C
N
), we can express (5.7) as
Z
L
=
1 C
V
/C
N
jC
T
(1 C
V
/C
D
)
(5.8)
Since the change in the phase shift is determined by the change in the impedance,
the greatest change in the phase shift will occur when C
V
is in the range of C
D
, since
the impedance will transition from a nite value to innity to negative innity and
back to a nite value. This is shown in Fig. 5.8 where the phase shift and impedance
are plotted against C
V
. In this and the following plots we have assumed an operating
67
Figure 5.8: RTPS phase shift superimposed on TSRL impedance over the varactor
range.
frequency of 2 GHz and a tuning range of 2, since this is a good approximation for
what would be available in a CMOS process with standard supply voltages. From
gure 5.8 it can be seen that the phase changes most rapidly around C
D
, which has
been set to the middle (arithmetic mean) of the tuning range. From this we can
conclude that a good choice for C
D
is in the middle of the tuning range.
The rate of change of the phase will be determined by the rate of change of the
impedance, and this is set by the proximity of C
N
to C
D
, as well as by the magnitude
of C
V
. We have assumed that C
D
is chosen at the middle of the tuning range, and C
N
is chosen to be less than C
D
. As C
N
moves closer to C
D
, the impedance changes more
quickly in the neighborhood of C
D
, and thus the slope of the phase shift increases.
This is plotted in Fig. 5.9, where the phase shift is plotted for C
N
increasing from
C
V,min
to 1.5C
V,min
. As C
N
increases, the slope increases, as does the total phase
shift, approaching the limit of 360

(for a varactor tuning range of 2). The change in


the phase shift range for changing C
N
is plotted in Fig. 5.10. Increasing the nominal
68
Figure 5.9: Eect of zero location (value of C
N
) on phase shift characteristics.
value of C
V
has a similar eect to moving C
N
closer to C
D
. From these plots we can
see that the choice of C
N
is a compromise between the phase shift range and the slope
of the phase shift curve. For higher slopes the phase shift will be more sensitive to
noise on the control voltage of the varactor. In the presense of non-ideal components,
higher slope in the phase shift curve also leads to higher losses.
A previously reported design which implemented an RTPS in CMOS technology
[47] implemented the capacitor C
T
in Fig. 5.7(c) with a varactor of the same value as
C
V
. The result of this is that the slope of the phase shift in the neighborhood of C
D
is determined solely by the size of the varactor, which means that for a desired slope
at a specied operating frequency, the C
V
and L values are xed. By making C
T
a
xed capacitor (as we have done in this work), the varactor (and inductor) values are
decoupled from the slope of the phase shift characteristics.
69
Figure 5.10: Phase shift range vs. zero location (value of C
N
).
Impact of Parasitics
This section will analyze the rst-order eects of parasitics on the phase shift charac-
teristics. The two circuits which will be analyzed are shown in Fig. 5.11. The circuit
in (a) includes a series resistance that would result from parastic resistance in the
varactor or inductor, and the circuit in (b) includes a parasitic parallel capacitance
to ground that would be present between wiring/components and the substrate.
Series Resistance
The impedance of the circuit shown in Fig. 5.11(a) can be expressed as
Z
L
=
(1
2
LC
V
) +jR
P
C
V
j(C
V
+C
T
+
2
LC
V
C
T
)
2
C
V
C
T
R
P
(5.9)
Comparing (5.9) with (5.7), we can see that the parasitic resistor has created an
additional term in each of the numerator and denominator. As expected, as R
P
0,
the last terms in the numerator and denominator of (5.9) disappear, and it becomes
identical to (5.7). We can also observe that as C
V
and C
T
become smaller, the
70
Figure 5.11: TSRL with inclusion of parasitics: (a) Series resistance, (b) Parallel
capacitance
extra terms created by R
P
will be less signicant, and the behavior will more closely
approximate the ideal case.
To see the eect of the parasitic resistance on the phase shift, we can plot the
phase shift for increasing values of R
P
. This is shown in Fig. 5.12 for C
V,min
of 1
pF and R
P
values from 0 to 10 . It can be seen that as R
P
increases, the slope of
the phase shift in the vicinity of C
V
= C
N
increases. For R
P
> 4.5 the phase shift
characteristics change completely; the phase shift is greatly reduced and is no longer
monotonic.
As expected from (5.9), larger values of C
V
greatly increase the sensitivity of
the phase shifter to parasitic series resistance. For C
V,min
= 1 pF the phase shifter
switches to non-monotonic behavior (as shown in Fig. 5.12) at R
P
= 4.73 , while
for C
V,min
= 10 pF (and C
N
chosen to provide similar slope in the phase shift char-
acteristics) the phase shifter switches to non-monotonic behavior at R
P
= 0.41 .
The additional term in the denominator is also proportional to C
T
, so we can observe
that the sensitivity to R
P
will increase as C
N
moves closer to C
D
(which increases
the phase shift by steepening the phase shift characteristics in the vicinity of C
D
).
71
Figure 5.12: The eect of parasitic series resistance on phase shift characteristics.
From this analysis, it is evident that the value of the varactor in the TSRL should be
chosen as small as is practical to minimize sensitivity to parastic resistance.
Parallel Capacitance
The impedance of the circuit shown in Fig. 5.11(b) can be expressed as
Z
L
=
1
2
L(C
P
+C
V
)
j(C
T
+C
V

2
L(C
T
C
P
+C
V
C
T
+C
V
C
P
))
(5.10)
If we make the same substitutions that were made in moving from (5.7) to (5.8), we
can express this as
Z
L
=
_
C
N
C
N
C
P
_
1
C
P
+C
V
C
N
1
C
V
C
D
_
1+
C
P
C
D
C
N
(C
N
C
P
)
1
_
(5.11)
From comparing (5.8) to (5.11) we can see that the parastic capacitance has re-
duced the zero from C
N
to C
N
C
P
, and the pole has been reduced from C
D
to
C
D
_
1
1+C
P
C
D
/C
N
(C
N
C
P
)
_
. Shifting the zero will reduce the slope of the phase shift
characteristics, and shifting the pole will move the phase shift characteristics to the
72
Figure 5.13: The eect of parasitic parallel capacitance on the phase shift character-
istics.
left. If we make the assumption that C
P
C
D
, C
N
, we can see that the zero will be
reduced by C
P
and the pole will be reduced by C
P
(C
D
/C
N
)
2
. Since C
D
> C
N
, the
pole will move more than the zero, and the phase shift characteristics will steepen in
addition to moving to the left. This can be seen in Fig. 5.13, where C
V,min
= 1 pF and
C
P
increases from 0 to 300 fF. As C
P
increases, the phase shift curve steepens and
shifts to the left. When the varactor is larger, the phase shift will be less susceptible
to the parasitic capactitance, but if an accurate estimate of C
P
is known, the values
of C
N
and C
D
can be adjusted to restore the pole and zero to their desired locations.
Design Strategy
This section will consolidate the ndings of the previous sections into a general pro-
cedure which can be followed in the design of an RTPS with a TSRL. To simplify the
design, the initial design should be simulated using ideal components, then simulated
with models that include component parasitics, and nally simulated with extracted
models which include layout parasitics, with adjustments to component values being
73
made at each step as necessary.
1. Size the hybrid 90
o
coupler using the design equations provided.
2. Choose a value for L which will yield the highest quality factor at the frequency
of interest.
3. Calculate the value of C
N
to resonate with L at the frequency of interest using
C
N
= 1/L
2
.
4. As a starting point, choose the varactor size to satisfy C
V,min
= C
N
.
5. Choose C
D
to be in the middle of the varactor tuning range, using C
D
= C
V,min
+
0.5(C
V,max
C
V,min
).
6. Calculate the value of C
T
using C
T
= C
D
C
N
/(C
D
C
N
).
7. Simulate the phase shift characteristics and adjust the slope of the phase shift
curve by adjusting the value of C
N
and recalculating C
T
. Increasing C
N
(moving
it closer to C
D
) will increase the total phase shift at the expense of making the
phase shift curve steeper in the vicinity of C
D
.
At this point real component models should be introduced for the varactor, capac-
itor, and inductor. After doing this, the greatest eect on the phase shift characteris-
tics will be due to the parasitic resistances in the non-ideal components. Resimulate
the circuit, and if the phase shift is greatly reduced and non-monotonic it will be
necessary to return to the rst step and redo the sizing procedure with a larger value
of L (and thus a smaller C
V
) to reduce the sensitivity to the parasitic resistance. The
loss of the RTPS should also be simulated at this point, the loss will be greatest in
the vicinity of C
D
. If the loss is too large, it can be reduced by moving C
N
away from
C
D
and reducing the slope of the phase shift characteristics.
74
Figure 5.14: Schematic of the TSRL of the implemented RTPS.
Finally, extract and simulate the circuit from the layout. The largest eect on the
phase shift characteristics after this step will be due to the parasitic capacitance to
ground. This can be compensated for by adjusting C
N
and C
D
and recalculating C
T
to restore the pole and zero to their desired locations, and by adjusting the varactor
range if necessary.
RTPS Design
In this section we will report the performance of the RTPS that was designed using
the methods and analysis outlined above. The hybrid 90

coupler is shown in Fig. 5.5


and was designed using (5.4) - (5.6) for
0
= 2.0 GHz. The schematic of the TSRL
is shown in Fig. 5.14. Capacitors C
B1
- C
B3
are blocking capacitors to allow proper
biasing of the varactor and the cross-coupled pairs. The NMOS and PMOS cross-
coupled pairs are included to reduce the loss of the RTPS by introducing a negative
resistance that compensates for the losses of the inductor, as described in [47].
The sizing for the dierent components is given in Table 5.1. The component
sizes were chosen using the design procedure outlined in the previous section, and
then adjusted after simulations to minimize loss and maximize the phase shift range.
75
Component Size
C
B1
43.8 pF
C
B2
7.0 pF
C
B3
42.9 pF
C
T
3.08 pF
C
V,min
0.73 pF
C
V,max
1.62 pF
L 4.5 nH
M
1
M
4
4 m/180 nm
Table 5.1: Sizing of components in the implemented TSRL
Blocking capacitor C
B3
was sized considerably smaller than the other blocking ca-
pacitors in order to minimize the parasitic capacitance to ground that is introduced
between the bottom plate and the substrate. The varactor was implemented with a
hyper-abrupt junction varactor (HAV), with a tuning range t = 2.23.
A die photo of the RTPS is shown in Fig. 5.15. The RTPS has been implemented
in a 0.18 m CMOS process, and occupies an area of 0.75 mm
2
. The measured and
simulated phase shifts are compared in Fig. 5.16. It can be seen that the measured
phase shift range is 308

for a control voltage from 0 - 1.8 V (the standard power


supply range for this process. The measured results are very similar to the simulated
results, with the exception of a shift with control voltage which is likely due to a
mismodelling of parastic capacitances. The measured loss over the nominal phase
shift range is shown in Fig. 5.17, and it can be seen that the loss varies from -4 dB to
-16.5 dB. The measured noise gure is plotted in Fig. 5.18, and it can be seen that
the noise gure varies from 12 dB to 27 dB.
The performance of the implemented RTPS is summarized in table 5.2, where it
is compared to the performance of previously reported CMOS phase shifters. It can
76
Figure 5.15: Die photo of the implemented RTPS.
Figure 5.16: Measured phase shift range of the implemented RTPS.
77
Figure 5.17: Measured loss of the implemented RTPS.
Figure 5.18: Measured noise gure of the implemented RTPS.
78
Specication [55] [47] This work
Process 0.18 m 0.18 m 0.18 m
Frequency 8 GHz 2.4 GHz 2 GHz
Control voltage -1 to 1 V 0 to 1.8 V 0 to 1.8 V
Phase shift 180
o
105
o
308
o
Max. loss NA -11 dB -16 dB
Max. NF NA 17 dB 27 dB
Area 0.23 mm
2
1.08 mm
2
0.75 mm
2
Power 170 mW 1.8 mW 5.4 mW
Table 5.2: Comparison of this work with previously reported CMOS phase shifters
be seen that the phase shift range is the greatest of reported phase shifters, and the
loss and noise gure are comparable to previously reported work.
5.2 Buer
The next block to be discussed is the buer, which appears in the phase feedback
loop before and after the phase shifter/VGA combination (shown in Fig. 4.2). The
purpose of the buer stage is to remove the amplitude dependancy of the input and
output signals before they are fed into the phase feedback loop, by converting the
sine wave signals to full scale digital waveforms where the information is contained
in the edge crossing locations. The digital output of the buers is then used to drive
the pulse generator.
The most important requirement for the buer is that it be able to produce a
digital output signal for a range of sine wave amplitudes at its input. This is necessary
since the buer at the output of the phase shifter/VGA chain will have a varying
input amplitude depending on what amplitude has been specied at the input to the
amplitude feedback loop. Additionally, the delay introduced by the buer should
79
Figure 5.19: Schematic of the buer block.
be minimized for stability considerations, and should not vary across dierent input
amplitudes, as this will introduce error in the phase of the output signal. The buer
should also contribute minimal phase noise (jitter), and consume minimal power and
area.
The buer has been implemented with three dierent cascaded gain stages, as
shown in Fig. 5.19. Several stages were necessary to provide sucient gain. Stages
1 and 2 provide the gain, and stage 3 produces the digital output that drives the
divider.
The buer was designed for operation with input amplitudes ranging from 10 mV
to 100 mV. This range was chosen to allow operation over a 20 dB range of output
amplitudes. The minimum input amplitude of 10 mV determined the number of gain
stages required in the buer.
80
5.2.1 Stage 1
The common source amplier of stage 1 was chosen as the rst stage since it provides
the most gain, at the cost of higher power consumption and increased area (primarily
due to the blocking capacitors C
B
required for biasing). It was biased with 150 A
of current, and with a load resistor R
1
of 5 k it provides about 12 dB of gain. The
blocking capacitors were sized at 1.3 pF, which was chosen as a tradeo between
signal attenuation and area consumption.
5.2.2 Stage 2
The second stage was implemented as a resistor-biased inverter, to allow signicant
gain with minimal power and area consumption. To provide sucient gain for input
amplitudes down to 10 mV it was necessary to use three cascaded instances of stage
2 (not shown in Fig. 5.19). Each stage draws 50 A of current and provides 8 dB
of gain. The biasing resistor R
2
was sized at 28 k. This choice was a tradeo
between higher gain (for larger resistor values) and faster settling times to the correct
bias point for a change in input amplitude. This settling time is not relevant for the
phase loop design, but is important for the dynamics of the amplitude loop, as will
be discussed in the design of the peak detector in the following chapter.
5.2.3 Stage 3
The nal stage is a standard inverter that is used to drive the divider stage. Dier-
ent combinations of stages were simulated (i.e., more instances of stage 1 and fewer
instances of stage 2) to try and minimize the number of stages required while mini-
mizing the variation in delay over the input amplitude range, and it was found that
one instance of stage 1 followed by three instances of stage 2 and one instance of stage
3 was optimal.
The change in the delay introduced by the buer across the range of input am-
81
Figure 5.20: Delay introduced by the buer across the input amplitude range.
plitudes is plotted in Fig. 5.20. The delay is plotted in units of degrees so that it
is clear how much phase error will be introduced. It can be seen that, while the
buer operates properly (outputs a full scale digital signal to drive the divider) for
input amplitudes down to 10 mV, for inputs below 30 mV the change in the buer
delay introduces signicant phase error. Thus, the buer input amplitude should be
maintained above this level to insure that the phase error introduced by the buer
remains below a few degrees.
5.3 Divider
As mentioned in the previous chapter, the divider circuits are included to allow the
full 360

of phase to be specied for the duty cycle of the pulse generator output
ranging from 25% to 75%, with a nominal value of 50% (as shown in Fig. 4.3). This
exposes the primary requirement for the divider: since the phase shifter has a limited
phase shift range, the divider must start up in the correct state so that the pulse
82
Figure 5.21: Conventional frequency divider block diagram.
generator has a nominal duty cycle of 50% and can span the range from 25% to 75%.
As will be described, this requirement is met by using multiplexing circuitry to choose
the correct phase of the divider upon start up.
The other requirements for the divider are that it contribute minimal phase noise,
consume minimal power, and be able to operate at 2 GHz (which is not dicult in a
modern CMOS process). The divider block can be partitioned into the divider core
and multiplexing circuitry, each of which is described in the following sections.
5.3.1 Divider Core
Most high speed dividers are fully dierential structures that make use of two D ip-
ops connected in a master/slave conguration, as shown in Fig. 5.21 [56]. The D
ip-ops are typically implemented using source-coupled logic (SCL). Since the ip-
ops will not be used in a general digital circuit, design techniques can be used to
optimize them for speed that would pose timing problems in some applications.
The topology chosen for the divider core is a well-known design that was reported
by Razavi in 1997 [57]. The schematic is shown in Fig. 5.22. The divider uses two
identical latches in a master-slave conguration, driven by complementary clocks.
Since the input to the divider in our application is single-ended, the complementary
clock inputs are generated using inverters and transmission gates to approximate an
inverter delay. The dierential structure of the divider naturally generates 4 phases of
83
Figure 5.22: Divider core schematic
the output clock (Out
0
Out
270
in Fig. 5.22), which are the input to the multiplexer
(described in the following section) which selects the proper phase to yield a nominal
50% duty cycle for the pulse generator.
5.3.2 Multiplexer
The schematic for the multiplexer is shown in Fig. 5.23. It is a standard design
using transmission gates, and the transistors are minimum size to minimize RC time
constants and allow high speed operation. Since there are four clock phases to be
multiplexed, three of the 2 input blocks shown in Fig. 5.23 are used together to
realize a 4:1 multiplexer.
In this design the control of the multiplexer is done manually through o-chip
switches, but in a real implementation it would be straightforward to design a start-
84
Figure 5.23: Divider multiplexer schematic
Figure 5.24: Timing diagram for pulse generator.
up circuit which automatically congures the multiplexer into the proper state upon
circuit initialization.
5.4 Pulse Generator
The pulse generator generates pulses to drive the charge-pump, with a duty cycle that
varies according to the phases of the two input signals. The primary requirements for
the pulse generator are that it be insensitive to the duty cycle of the inputs, and have
symmetry for reliable operation at high frequencies. Fig. 5.24 shows the required
input/output waveforms and Fig. 5.25 shows the state diagram that is required for
the pulse generator.
85
Figure 5.25: State diagram for pulse generator.
Figure 5.26: System diagram for pulse generator.
The insensitivity to the duty cycle of the inputs is necessary because the pulse
swallowing divider outputs have a duty cycle of 25%. This requirement points to an
edge triggered implementation, and a simple circuit that meets the requirements of
the state diagram of Fig. 5.25 is shown in Fig. 5.26. For a typical D ip-op imple-
mentation the clock and reset inputs are not symmetrical, which is a key requirement
for our design. With this in mind, a proposed D ip-op implementation is shown
in Fig. 5.27. The cross coupled latches hold the internal state, and transistors M1
and M2 set and reset the output. The edge sensitivity is accomplished through the
delayed and inverted input signals arriving at transistors M3 and M4, which disable
the set/reset transistors M1 and M2.
Due to its inherently dierential operation, this design has the additional advan-
tage of generating dierential outputs for use by the charge-pump. Care was taken
in the design of the circuit to size the latch transistors to be small enough so that
the set/reset transistors (M1M4) could change the state of the circuit without any
86
Figure 5.27: Schematic for pulse generator.
glitches.
5.5 Charge-Pump
The charge-pump is driven by the pulse generator, and is the key component for
setting the phase of the feedback loop. When the pulse generator output is high,
the charge-pump pushes the specied current onto the loop lter node (referred to as
the Up current), and when the pulse generator output is low, the charge-pump pulls
the specied current o of the loop lter node (referred to as the Down current). As
shown in Fig. 4.4, the phase is adjusted by changing the values of the Up and Down
currents relative to each other.
With this in mind, the most important requirement for the charge-pump is that it
be able to generate accurate, programmable, Up and Down currents. Inaccuracies in
the charge-pump current will result in phase errors between the intended and actual
phase, and the resolution with which the Up and Down currents can be specied will
determine the phase resolution of the phased-array element. Additional requirements
for the charge-pump are that it should consume minimal power and area.
A schematic for a standard charge-pump is shown in Fig. 5.28. The current in
the branches is set by the current source and the diode connected transistor MN0.
87
Figure 5.28: Schematic for a standard charge-pump.
It is assumed that the charge-pump is employed in conjunction with an active loop
lter which holds the output node Out at voltage V ref. Operational amplier 1
incorporates feedback to insure that the Up and Down currents are equal. To illustrate
how this works, if the Up current is higher than the Down current, the voltage on
node A will increase, causing the opamp to increase the voltage on the gates of MP1
and MP3, reducing their gate to source voltages, and reducing the Up current.
The primary weakness of the charge-pump in Fig. 5.28 is the charge-injection from
transistors MP3 and MN3 during switching. When the switch transistor MP6 is in
the o state, current continues to ow through transistor MP3, charging the parastic
capacitances on node B. When MP6 is switched on, this excess charge ows onto
the output node. Since this charge is not related to the duration of time for which
the Up current is switched on, it adds a non-linear component to the charge-pump
output, which is highly undesirable. A parallel situation holds for the Down current
and node C.
The most common modication to ameliorate this problem is to add a second set
of switch transistors so that the current in transistors MP3 and MN3 is steered back
88
Figure 5.29: Schematic for a charge-pump with current steering to reduce charge
injection.
and forth between two branches, instead of being switched on and o. This topology
is shown in Fig. 5.29, where transistors MP5 and MN5 have been added to form
the second branch. These transistors are driven with the inverse of the output switch
transistors, so that when MP6 is switched o, MP5 is switched on, and the current
from MP3 is steered to this branch instead of accumulating on node B. The voltage
on this dummy output node is set to the same voltage as the output node.
This modication provides a signicant improvement over the initial design in Fig.
5.28, but there is still residual charge injection due to mismatches in the switching
times of transistors MP5, MN5 and transistors MP6, MN6. Since the accuracy of
the charge-pump currents translates into phase accuracy in our application, an alter-
native charge-pump topology has been developed which almost completely eliminates
charge injection. The new charge-pump architecure is shown in Fig. 5.30. The central
features are the addition of operational ampliers 2 and 3, and transistors MP4 and
MN4.
The charge injection problem arises from switching the current from transistors
MP3 and MN3, so in this topology the current from these transistors ows perma-
89
Figure 5.30: Schematic for the new charge-pump with buers and no switching.
nently through the dummy switches MP4 and MN4, which are always on. In the
previously described topologies, charge accumulation at nodes B and C resulted in a
change in the voltage at those nodes. In this topology, there is no switching taking
place so these nodes are held at steady state values. Operational amplers 2 and 3
serve as buers, and hold nodes D and E at the required voltages. Since all of the
switch transistors have identical sizing, the currents in the output branch will match
those in the dummy branch of MP4 and MN4. Although the switching in the out-
put branch will provide small disturbances at the output node, as long as operational
ampliers 2 and 3 have sucient gain and bandwidth, the charge injection will be
negligible.
A comparison of the charge-injection performance of the three topologies is shown
in Fig. 5.31. It can be seen that the current steering topology provides signicant
improvement over the standard topology, but the newly proposed topology goes a step
further and completely eliminates charge injection eects, resulting in highly linear
operation.
In designing the circuit shown in Fig. 5.30 there are several important issues to be
aware of. To maintain stable operation, extra care has to be taken in designing the
90
Figure 5.31: Simulated transient current waveforms for the three charge-pump topolo-
gies.
feedback loop containing operational amplier 1 which maintains equality between
the unit Up and Down currents. A standard operational amplier is usually designed
to have about 60 dB of gain and about 75

of phase margin, but in this feedback loop,


transistors MP1 and MP2 act as an additional cascoded common source amplier,
adding about 30 dB to the gain of the feedback loop, making it potentially unstable.
With this in mind, operational amplier 1 should have much less gain than operational
ampliers 2 and 3 (which can be designed as standard opamps) to maintain stability.
Another consideration is that the charge-pump must provide means of setting the
Up and Down currents. This was accomplished by implementing transistors MP3
and MN3 as switchable arrays of unit transistors. The unit transistors are switched
in and out to allow precise digital control over the Up and Dn currents, which will
be multiples of the unit current set by the current source driving transistor MN0.
In this implementation, 5 bits of digital control were used for each of the Up and
91
Dn currents. If more resolution is required it is straightforward to add more bits to
the charge-pump transistor arrays. Since these arrays are isolated from the switching
nodes by the operational ampliers 2 and 3, the additional parasitic capacitance does
not cause any problems. In addition to the switchable unit currents, a xed current
equal to the most signicant control bit was added to each of the Up and Dn branches
to avoid boundary conditions with zero current.
One nal consideration relates to charge injection onto the output node from the
switches MP6 and MN6. The preceding discussion on charge injection has focused
on charge buildup at nodes D and E which is the dominant source of charge injection,
but injection of the channel charge of the transistors MP6 and MN6 during switching
can also have an eect. To ameliorate this eect, half-sized dummy transistors driven
by the inverse of the input with their drain and sources connected were added to the
output node.
5.6 Loop Filter
A preliminary analysis of the loop dynamics of the phase feedback was given in the
previous chapter, this section will build upon that analysis to account for second
order eects and to enable the design of the loop lter. The primary requirement
for the looplter is that it combine with the other loop components to yield the
desired dynamics for the feedback loop. The two most important elements of the
loop dynamics are the stability and the loop bandwidth, which determines how fast
the control voltage will settle to its nal value.
Since we are using a charge-pump in our feedback loop, it was mentioned in the
previous chapter that the simplest choice of looplter that would provide an integra-
tion is a capacitor. However, from the design of the charge-pump in the previous
section, it is clear that we need an active loop lter which holds the charge-pump
output at a stable voltage. The looplter topology is shown in Fig. 5.32. Here I
in
is
the charge-pump output current, and V
ref
is a reference voltage. Assuming an ideal
92
Figure 5.32: Toplogy of the phase loop lter.
opamp, the transfer function can be shown to be
H(s) =
1
sC
(5.12)
where C is the capacitor value. This is the same as for the case of the passive capacitor
looplter, with the exception of the sign inversion.
5.6.1 Operational Amplier
The most important requirement for the operational amplier used in the looplter
is that it be able to drive the output from rail to rail, to allow the full control range
of the phase shifter to be used. It is not necessary to have a rail-to-rail input stage,
as the feedback within the looplter will hold both inputs at V
ref
.
Initial designs were done using a 1.8 V opamp with a class AB rail-to-rail output
stage. However, it was found that the performance of the opamp was not consistent
across the process corners, so a decision was made to use a standard opamp topology
with a higher power supply to allow a 0 1.8 V output range. The schematic for the
nal opamp is shown in Fig. 5.33. It has a common source dierential pair for the
rst gain stage, a common source second gain stage, and a source follower buer for
the third stage to provide a low output impedance. The opamp was designed using
93
Figure 5.33: Schematic for the operational amplier used in the phase looplter.
thick oxide transistors to allow a 3.3 V power supply, and it provides 60 dB of gain.
A Miller compensation network (not shown in Fig. 5.33) is used to provide a phase
margin of 78

.
5.6.2 Sizing
The most important decision in the looplter design is the sizing of the capacitor,
as this will determine the loop dynamics. This can be done by starting with Eq.
(4.6) from the previous chapter. Two non-idealities that were not taken into account
in this expression are the eect of the biasing network of the phase shifter, and the
eects of the delays introduced by the other components (buers, dividers, etc.). The
reective load of the phase shifter was shown in Fig. 5.14. There it was implied that
the control voltage directly sets the voltage at one terminal of the varactor, but in the
actual design it is necessary to include a large blocking resistor between the output
of the loop lter and the varactor terminal. This blocking resistor combined with the
capacitance of the reective load has a signicant impact on the loop dynamics and
must be included in the analysis.
A representation of the load that must be driven by the looplter is given in Fig.
94
Figure 5.34: Schematic of the reective load as it appears to the looplter.
5.34. Here V
in
is the output of the looplter opamp and V
out
is the voltage that is
applied to the varactor. The resistors labeled R
B
are the large blocking resistors and
R
C
is the coupler input impedance. The capacitors labeled C
B
are the large blocking
capacitors, and the other components comprise the reective load. After making a
number of approximations based on the operating frequency and component values, it
can be shown that the transfer function of the network in Fig. 5.34 can be expressed
as
V
out
(s)
V
in
(s)
=
1
1 +sR
B
C
eq
(5.13)
where R
B
= 9 k and C
eq
= C
B
/2 +C
V
= 7.5 pF.
After incorporating (5.12) and (5.13) into (4.6), we arrive at a nal expression for
the closed loop gain of the phase loop:

out
(s)
I
diff
(s)
=
K
PS
CR
B
C
eq
_
_
1
s
2
+s
1
R
B
Ceq
+
K
PS
Isum
4CR
B
Ceq
_
_
(5.14)
where K
PS
is the phase shifter gain, C is the looplter capacitor, R
B
and C
eq
result
from the phase shifter bias network (as described above), and I
sum
is the sum of the
Up and Down charge-pump currents.
The denominator is second order, and can be put in the form of the characteristic
equation for a second order system. After doing so, we can express the natural
95
frequency and damping factor as:

n
=

K
PS
I
sum
4CR
B
C
eq
(5.15)
=

C
R
B
C
eq
K
PS
I
sum
(5.16)
Using these equations, the design procedure is to set the damping factor () to 0.707
(for critical damping), and solve for the relation between the congurable parameters,
which are the loop lter capacitor and the charge-pump currents. These can then be
chosen to yield a reasonably sized capacitor. Since we have only one parameter
to adjust in the looplter, we cannot specify both the damping factor and natural
frequency, so if it was necessary to set these quantities independantly we would be
forced to move to a higher order loop lter.
The phase shifter gain varies across its control range (as seen in Fig. 5.16), so
to ensure that the system remains stable across the entire operating range the worst
case (for stability) of the maximum phase shifter gain was used (K
PS
= 8.2 rad/V).
Using this procedure, we arrive at a loop lter capacitor value of 35 pF for nominal
Up and Down currents of 126 A (I
sum
= 252 A). This yields a critically damped
system with a natural frequency of 250 KHz, so the settling time will be on the order
of several s which is acceptable for our purposes.
Since the absolute value of integrated components is not well controlled, the ca-
pacitor in the looplter was implemented as a switchable array, allowing 2 bits of
control over the capacitor value. Additional control over the loop dynamics can be
realized by adjusting the charge-pump unit current, but this method is limited by the
need to maintain stability in the replica bias feedback network of the charge-pump.
One other non-ideality that must be considered is the additional delay introduced
by the buer, divider, pulse generator, and charge-pump blocks in the feedback loop.
Since added delay reduces the phase margin at the unity gain frequency, it is necessary
to verify that the system is stable in the presence of these additional delays which were
not accounted for in the analysis presented above. Simulations were run to determine
96
the magnitude of these delays, which were incorporated into a Matlab script which
veried that they have a negligible impact on the phase margin at the unity gain
frequency.
The nal non-ideality to be considered is the eect of nite gain and output
resistance in the op-amp, since the analysis presented above assumes an ideal opamp.
These eects were incorporated into a Matlab script which conrmed that they have
a negligible eect at the frequencies of interest.
5.6.3 Phase/Amplitude Loop Interaction
The nal eect to be considered is the interaction between the phase and amplitude
loops, as this has the potential to degrade the stability of the loops. If the phase
shifter had constant gain across its phase range or if the VGA had constant phase
across its gain range then this issue would not arise, as the loops would not interact
with each other. For a real phase shifter and VGA this condition does not hold, and
part of the motivation for this work was to correct for the errors introduced by this
non-ideal behavior in the phase shifter and VGA.
The eect of the variable gain in the phase shifter and the variable phase shift in
the VGA is to change the eective gains of each of the blocks. As we have seen in the
previous chapter, the phase shifter and VGA gains play a role in determining the loop
dynamics of the system, so changes in these quantities caused by loop interactions
can either make the system more or less stable, depending on the gain and phase
relationships. This section will determine the changes in the phase shifter and VGA
gains introduced by the loop interactions, and the eect on the stability.
The approximate gain and phase relationships versus control voltage for each of
the phase shifter and VGA are shown in Fig. 5.35. It can be seen that the gain
relationship for the phase shifter is non-monotonic. As we will see, the characteristic
on the left part of the curve actually increases the stability of the loop, so we will
only consider the right portion of the curve.
97
Figure 5.35: Approximate gain and phase relationships for the phase shifter and VGA.
98
To examine the eect of the amplitude loop on the stability of the phase loop,
rst consider the solid arrow on the phase relation plot for the phase shifter (the top
plot in Fig. 5.35). For an increase in phase, as shown by the solid arrow, there is
an increase in the control voltage, and from the next plot down we can see that this
corresponds to an increase in the amplitude of the signal at the output (assuming
we are operating on the right part of the phase shifter gain curve). This increase in
amplitude will then be detected by the amplitude loop, which will then decrease its
control voltage to restore the amplitude to its original value (solid arrow in the third
plot). From the solid arrow in the fourth plot showing the VGA phase relation, we
can see that this decrease in control voltage will lead to an increase in the output
phase, in addition to the increase in phase brought about by the original change in
the phase loop. So, we can see that the eect of the amplitude loop is to increase the
eective gain of the phase shifter, which makes the system less stable. The increase
in eective phase shifter gain can be calculated as
K
PS
=
K
G,PS
K
P,V GA
K
V GA
=
1.7 0.9
7.3
= 0.22 (5.17)
where K
G,PS
is the amplitude gain of the phase shifter and K
P,V GA
is the phase gain
of the VGA. By following this procedure while assuming operation on the left portion
of the phase shifter amplitude plot, we can see that the eect of the amplitude loop
is to reduce the eect phase shifter gain, making the phase loop more stable.
A similar procedure can be followed to determine the eect of the phase loop on
the amplitude loop stability, by examining the dashed arrows in the order denoted
by their accompanying numbers. From the VGA amplitude plot we can see that
an increase in amplitude corresponds to an increase in control voltage, which from
the VGA phase plot, results in a decrease in phase. From the phase shifter phase
relation, we can see that the phase shifter will increase its control voltage to increase
its phase to compensate for the decrease of the VGA, and from the phase shifter
amplitude plot it is seen that this corresponds to an increase in amplitude, which
99
adds to the original intended increase in amplitude of the VGA. In a similar manner
to above, this increases the eective VGA gain, making the amplitude loop less stable.
The increase in eective VGA gain can be calculated as
K
V GA
=
K
P,V GA
K
G,PS
K
PS
=
0.9 1.7
1
= 1.6 (5.18)
It should be noted that in this expression the minimum phase shifter gain of 1 rad/s
was employed, as this represents the worst case for this situation.
The calculated increases in the phase shifter and VGA eective gains can now
be factored into the design equations to recalculate the loop lter component values.
Since the above analysis assumes that the phase shifter and VGA operate in a linear
manner, there may be further loop interactions resulting from the nonlinear nature of
the real components that have been ignored. With this in mind, some further margin
of error is designed into the gains for the loop lter calculations, in addition to the
exibility provided by the programmable nature of the loop lter components.
100
Chapter 6
AMPLITUDE LOOP DESIGN
This chapter will cover the design of the components in the amplitude feedback
loop. The most detail is provided for the variable gain amplier (VGA), as its oper-
ation is most crucial for the performance of the overall system. The general design
strategy for each of the blocks is to keep the designs as simple as possible to minimize
power and area consumption of the system.
6.1 Variable Gain Amplier
As mentioned in Chapter 4, a decision was made to implement the variable gain
for the system in a variable gain amplier (VGA) that will drive a power amplier
(PA), rather than implementing the variable gain in the power amplier itself. This
decision was made both to avoid the linearity problems that can be encountered when
adjusting the bias of the power amplier, and to limit the scope of the project. The
approach of achieving variable gain by dividing the power amplier into a VGA and a
PA has been used in several PA implementations that require a large dynamic range
[37], [38]. The complexity of the amplier block was limited by using a narrowband
design, making the design of a VGA with sucient gain an easier proposition.
6.1.1 Requirements
The main requirement for the VGA is imposed by the losses of the phase shifter
in conjunction with the necessary amplitude range for providing sucient sidelobe
reduction [58]. The phase shifter measurements indicate that the loss varies over a
range of 12 dB. Since it is desirable to implement the VGA in a single stage, it has
101
Figure 6.1: Four VGA topologies: (a) variable feedback, (b) variable bias, (c) current
steering, and (d) simple cascode.
been decided to aim for 20 dB of gain with a control range of over 30 dB. This will
allow the VGA to fully compensate for the loss variation of the phase shifter while
providing 20 dB of gain control for setting the gain of each branch to reduce the
sidelobes of the array pattern.
It is also desirable to minimize the power consumption and noise gure of the VGA
since it will be a part of the transmit path, along with the phase shifter. Another
requirement and motivation for limiting the VGA to a single stage design is the need
to have a single control voltage, since the feedback loop can only generate one control
voltage.
6.1.2 Topology
There are a number of dierent techniques for varying the gain of an amplier, four
of which are shown in Fig. 6.1. The rst is the variable feedback approach, where
the gain is varied by altering the feedback resistance. The variable resistor can be
implemented as an FET device. While this approach has the potential for a high IP3,
its disadvantages are possible stability problems and a limited gain control range [59].
102
The second approach, shown in Fig. 6.1(b), is to incorporate a variable bias
network to alter the gain by varying the bias current of the drive transistor (to change
its transconductance). This approach has the advantage of a low noise gure, but
the disadvantage that the linearity of the amplier is strongly dependant on the bias
current.
The third approach is to use dierential cascode transistors to steer current to
and from the load. This approach has been reported frequently in the literature, with
implementations in both Si-BJT and GaAs HBT [37], [38], [60]. The advantage of
this approach is that it provides a large gain control range, and the disadvantage is
that it tends to suer more from noise than the other approaches [59].
The fourth approach can be thought of as a variant of the third, where the ad-
ditional transistor for steering current away from the load has been omitted and the
gain is controlled by reducing the current in both the cascode transistor M
2
and the
main drive transistor M
1
. This method has also been reported in the literature, and
it is attractive for its simplicity [61].
The architecture employed in this work is a combination of the third and fourth
approaches. These topologies are attractive for their large gain control ranges, and
an optimization technique is used to minimize the noise gure, which is their primary
disadvantage. If we assume that the current steering transistors M
2
and M
3
in Fig.
6.1(c) are sized with equal W/L ratios and we consider the topology shown in Fig.
6.1(d) to be a variant of this where the current steering transistor has been sized with
W
3
/L
3
= 0, we can see that it is possible to size the current steering transistor M
3
with
a continuum of sizes between 0 and W
2
/L
2
. In the following Section we will analyze
the general case, and use a combination of simulations and analytical techniques to
show that optimal noise performance can be achieved by sizing 0 W
3
/L
3
W
2
/L
2
.
It will be shown that this minimizes the combined noise contributions of all of the
transistors in the amplier.
103
Figure 6.2: Transistor model used for noise gure analysis.
6.1.3 Noise Analysis
To obtain a sizing strategy for minimizing the noise gure of the topology shown in
Fig. 6.1(c) we will use combination of simulations and analytical techniques. First
we will derive an expression for the noise gure of the VGA. For conciseness, we will
only consider the noise contributions of the transistors (which are the dominant noise
sources in the circuit). The transistor model used is shown in Fig. 6.2. The drain
current noise is expressed as [36]
i
2
nd
= 4kTg
d0
f (6.1)
where is a constant that is approximately equal to 4/3 in short channel devices, and
g
d0
is the drain source conductance at V
DS
= 0. In this analysis we will neglect gate
noise and 1/f noise, since simulations have shown that thermal noise in the channel
dominates at the 2 GHz operating frequency of interest.
Considering the transistor noise sources individually and then superimposing their
eect at the output, we can derive the noise gure for the VGA shown in Fig. 6.1(c)
as
NF = 1 +
(i
2
nd,1
+i
2
nd,3
)
i
2
ns
s
2
C
2
gs,1
g
2
m,1
+
i
2
nd,2
i
2
ns
s
2
C
2
gs,1
[g
ds,1
+s(C
gs,2
+C
gs,3
) +g
m,3
]
2
g
2
m,1
g
2
m,2
(6.2)
where i
2
ns
is the noise contribution of the source and we have omitted terms that were
shown to be negligible by simulation results. It can be seen that the second term is
104
due to noise contributions from M
1
and M
3
and the third term is due to noise from
M
2
. We can obtain the noise gure for the VGA topology shown in Fig. 6.1(d) by
omitting all of the terms in (6.2) due to M
3
. We then obtain
NF = 1 +
i
2
nd,1
i
2
ns
s
2
C
2
gs,1
g
2
m,1
+
i
2
nd,2
i
2
ns
s
2
C
2
gs,1
[g
ds,1
+sC
gs,2
]
2
g
2
m,1
g
2
m,2
(6.3)
First we will look at the noise performance of the simple cascode topology com-
pared with that of the current steering topology. We will focus on their noise per-
formance when the gain is reduced from its maxiumum value, as this is the region
where the VGA is most susceptible to noise. As the gain is reduced, the signal power
at the output is reduced while most of the noise sources remain constant, resulting
in a degraded signal to noise ratio (SNR). From (6.1), the drain noise contributions
depend on g
d0
, which in turn depends on the V
GS
bias condition. Simulations have
shown that V
GS
stays approximately constant for the topologies as the gain is varied,
so in this analysis we will focus on the coecients of the drain noise contributions,
and assume that the i
2
nd
terms remain relatively constant as the gain is varied.
Intuition might indicate that the simple cascode topology will have superior noise
performance since by adding M
3
for the current steering topology we have added an
additional noise source, but to arrive at a nal result we must examine the coecients
of the noise contributions as the gain is varied. The two architectures operate in a
similar manner when the gain is high, since negligible current ows through M
3
in
the current steering topology. As the gain is reduced, in the current mirror topology
the current through M
1
stays approximately constant while in the simple cascode
topology the current in M
1
is greatly reduced.
For the current steering topology, simulations reveal that the noise from M
2
dom-
inates at low gain settings. In the simple cascode topology, as the current in M
1
is
reduced to very low levels, g
m,1
decreases signicantly causing the the second term
in (6.3) to increase sharply so that the noise from M
1
begins to dominate, increasing
the overall noise gure. The result of this is that as the gain decreases, the noise
105
gure of the simple cascode topology degrades considerably faster than that of the
current steering topology. As a quantitative example, in preliminary simulations it
was found that 20 dB down from the maximum gain the noise gure for the simple
cascode topology was 10.1 dB, while for an identically sized current steering VGA
with W
3
/L
3
= W
2
/L
2
it was 8.4 dB.
As mentioned previously, simulations have shown that at lower gain settings the
third term in (6.2) (the contribution from M
2
) dominates the noise gure of the
current steering topology. Since this term depends partially on the characteristics of
M
3
, we can infer that it is possible to reduce this term by changing the size of M
3
so
that it is not necessarily equal in size to M
2
. By reducing the size of M
3
, both g
m,3
and C
gs,3
are reduced. Since both of these quantities appear in the numerator of the
M
2
drain noise coecient, by scaling down M
3
we can minimize the eect of the M
2
drain noise. Since this is the dominant noise source at reduced gain settings for this
topology, this will reduce the overall noise gure.
As we scale down M
3
we also reduce the current that it can carry, thus reducing the
current through M
1
for low gain settings. If M
3
is scaled down too much, the current
in M
1
will be reduced enough so that it becomes the dominant noise source at low
gain settings, as in the cascode topology. From this discussion, it is evident that an
optimum size for M
3
can be found to minimize the noise, with W
3
/L
3
being between
W
3
/L
3
= 0 (simple cascode topology) and W
3
/L
3
= W
2
/L
2
(standard current steering
topology). As a quantitative example, preliminary simulations showed that a VGA
with W
3
/L
3
= 0.2W
2
/L
2
had a noise gure of 8.0 dB, as compared with 8.4 dB for
the same VGA with W
3
/L
3
= W
2
/L
2
.
While the analytical noise gure equations are useful for gaining intuition into
which parameters we should adjust to minimize the noise gure, they neglect higher
order eects that are important for accurately determining the overall noise gure.
For this reason, the nal optimization must be carried out using simulations. In the
following section the VGA is designed making use of the principles discussed in this
106
Figure 6.3: Schematic of variable gain amplier.
section.
6.1.4 Design and Sizing
The schematic of the implemented VGA is shown in Fig. 6.3. It has been designed to
operate at 2.0 GHz, and has been implemented in 0.18 m CMOS technology. The
C
B
capacitors are large blocking capacitors to allow proper biasing, and R
B
is a large
resistor included to allow biasing of M
1
. Transistor M
3
is sized to be 1/5 the size of
M
2
, a ratio that was arrived at through simulator optimization. L
L
and C
L
form a
resonant load, and transistors M
4
and M
5
act as a buer to minimize loading eects
on the VGA output. Due to the gain of the rst stage, the noise contribution of these
transistors is negligible.
Improvements in the gain could have been obtained by including matching net-
works at the input and output, however this would have required additional inductors.
One of the goals for this VGA was to create a compact implementation, so matching
networks were not included. A summary of the sizing for the VGA components is
107
Component Size
C
B
10 pF
R
B
12 K
L
L
3.2 nH
C
L
1.5 pF
M
1
250 m
M
2
100 m
M
3
20 m
M
4
100 m
M
5
60 m
Table 6.1: Sizing of components in the implemented VGA
given in Table 6.1, where all transistors are sized with the minimum length of 180
nm. The layout for the standalone VGA is shown in Fig. 6.4. It can be seen that the
area of the VGA is dominated by the size of the inductor. The rectangles around the
perimeter are pads to allow wafer probing to measure the VGA in isolation.
6.1.5 Simulation Results
It had been planned to fabricate the VGA in a standalone conguration prior to its
inclusion in the nal system (as was done for the phase shifter), but the targeted
fabrication run was cancelled so the post-layout extracted simulations are presented
here. These simulations include the eects of parastic resistances and capacitances,
and are expected to be close to the measured results. The power supply voltage
for this process is 1.8 V, and the bias currents were 2 mA and 3 mA for M
1
and
M
4
respectively, leading to a total power consumption of 8 mW (not including bias
circuitry). The simulated S21 characteristics over the control voltage range are shown
108
Figure 6.4: Layout of the variable gain amplier.
in Fig. 6.5. It can be seen that the maximum gain is 20.6 dB. As noted above, this
could be improved by adding matching networks at the input and output of the VGA,
at the cost of increased area due to the additional inductors. The bandwidth of the
VGA is 340 MHz, centered around 2 GHz.
The simulated noise gure is plotted in Fig. 6.6. Both the actual noise gure
and the minimum achievable noise gure are plotted, and it is evident that further
improvements in the noise gure are possible with the addition of matching networks.
The sharp increase in the noise gure as the gain is reduced is evident in Fig. 6.6. At
maximum gain, the actual noise gure is 1.47 dB, and the minimum achievable noise
gure is 0.76 dB. When the gain is reduced to 20 dB below the maximum, the actual
noise gure is 8.8 dB, and the minimum achievable noise gure is 5.1 dB.
The simulated IP3 for the bias conditions cited above was -14 dBm. This is quite
low, and if necessary it could be improved by increasing the bias current levels.
109
Figure 6.5: Simulated S21 for the variable gain amplier.
Figure 6.6: Simulated noise gure for the variable gain amplier.
110
6.2 Peak Detector
The primary requirements for the peak detector are that it be capable of operating
at 2 GHz, and that it have sucient dynamic range. The dynamic range must be at
least 10 dB, as that is the chosen amplitude range for the output to provide sucient
sidelobe reduction. As with the other components, it is preferable that the peak
detector have low complexity and power consumption.
The simplest peak detector consists of a diode for rectication followed by a low-
pass RC lter for averaging. Monolithic peak detectors are typically implemented
using schottky diodes or bipolar transistors [62]. A number of low-power peak de-
tectors with large dynamic ranges have been reported, but they are limited to low
frequency operation [63]. Recently, several CMOS implementations of power detec-
tors have been reported [64], but many of them have high complexity and power
consumption, and are not appropriate for this work.
In this work we employ a CMOS implementation of the bipolar RF peak detector
introduced in [62], similar to one that was described in [65]. The peak detector in its
simplest form is shown in Fig. 6.7. The actual circuit employed has some additional
stages, but the core circuitry in Fig. 6.7 is sucient for the analysis and design. The
following sections will present an analysis of the DC behavior of the peak detector
(using a similar approach as in [62]) as well as the transient behavior, which will be
used in the design of the system loop dynamics.
6.2.1 DC Analysis
In the circuit shown in Fig. 6.7, the transistor acts as a non-linear rectifying element.
The capacitor acts as a hold capacitor, and its value is determined (in conjunction
with the current source) by the allowable droop on the output voltage over a half
period of the input frequency. When the transistor is o, the current I
1
comes from
111
Figure 6.7: Schematic for peak detector design and analysis.
the capacitor, and can be expressed as
I
1
= C
1
dV
out
dt
(6.4)
The output voltage droop can then be expressed as
V
out
= t
I
1
C
1
(6.5)
where t is one half of the period of the input frequency. As will be shown in the
section on transient analysis, the choice of the capacitor also inuences the settling
behavior of the peak detector.
To determine the relationship between the amplitude of the input signal and the
output voltage of the peak detector, we assume that the input voltage is a sinusoidal
signal specied as
V
in
= V
B
+V
1
sin t (6.6)
where V
B
is the bias voltage, V
1
is the amplitude of interest, and is the angular
frequency of the input signal. If we assume that the droop is small enough to be
neglected, then V
out
can be assumed to be constant. The average current through
transistor M1 must be equal to the current through the current source, so if we
assume that the transistor is biased so that it remains in the saturation region for the
112
majority of the time, then we can write
I
1
=

n
C
ox
2
W
L
(V
in
V
out
V
T
)
2
(6.7)
Now if we represent the constant as =
nCox
2
W
L
and sub in (6.6), we get
I
1
= (V
B
+V
1
sin t V
out
V
T
)
2
(6.8)
Combining the three constant voltages as V
C
= V
B
V
out
V
T
, expanding the squared
quantity and using a trignometric identity yields
I
1
=
_
V
2
C
+ 2V
C
V
1
sin t +
V
2
1
2
(1 cos 2t)
_
(6.9)
Since we are interested in the average current, we can set the two sinusoidal terms
on the right hand side of (6.9) to zero, leaving
I
1
=
_
V
2
C
+
V
2
1
2
_
(6.10)
Rearranging to express in terms of V
C
yields
V
C
=

_
I
1


V
2
1
2
(6.11)
We can now sub in the expression for V
C
to solve for the output voltage, giving
V
out
= V
B
V
T

_
I
1


V
2
1
2
(6.12)
It can be seen that (6.12) only provides a valid result when
V
2
1
2
<
I
1

(6.13)
If the input amplitude increases beyond this threshold then on the positive swings
the current through M1 is so large that to maintain an average current of I
1
, on the
negative swings the transistor would have to pull current o of the capacitor. This
is clearly not possible, so the result would be that C
1
would charge up so that M1 is
113
no longer in the saturation region, invalidating the assumptions used to derive (6.12).
This case could also be analyzed, but for predictable operation according to (6.12)
the range of input voltages should be determined and then the W/L ratio for M1 can
be adjusted so that (6.13) is always satised.
To more clearly observe the relation between V
in
and V
out
in (6.12), we can expand
the last term using its MacLaurin series. If we truncate the series after the rst three
terms, we get
V
out
= V
B
V
T
+

I
1

_
V
2
1

4I
1
1
_
(6.14)
From this we can see that V
out
has a square law relationship with the input amplitude
V
1
.
6.2.2 Transient Analysis
The peak detector is a non-linear device, thus Laplace analysis does not strictly apply.
However, since we would like to include the eects of the peak detector in the Laplace
domain analysis of the overall amplitude loop behavior, it is desirable to develop a
linear approximation for the transient behavior of the peak detector.
To aid in the analysis of the peak detector, Matlab code has been written to plot
the output voltage for changes in the input amplitude. As in the previous section, it
is assumed that when the transistor is on, it is in the saturation region. A plot of the
output voltage (as produced by the Matlab code) for a positive and negative voltage
step is shown in Fig. 6.8. It can be seen that each of these transitions has settling
behavior which can be approximated by an exponential settling, so in this section we
will model the peak detector transient behavior as a one-pole system.
Step Up Behavior
The conceptual behavior of the peak detector is simplied if the input amplitude is
equated to an equivalent DC input voltage V
eff
. Then when the input amplitude
114
Figure 6.8: Transient behavior of peak detector output voltage from Matlab simula-
tion.
increases, V
eff
increases so that the output voltage increases so that the V
GS
for the
transistor settles to the proper value (where I
DS
of the transistor is equal to I
1
, as
must be the case in the steady state). The relation between the input amplitude and
the newly dened V
eff
can be determined from the DC analysis performed in the
previous section.
With this representation, we can express the current owing onto the hold capac-
itor C
1
as the dierence between the current of M1 and the current source I
1
:
I
C
= (V
eff
V
out
)
2
I
1
(6.15)
If we assume that the current source I
1
is small relative to the current owing through
the transistor, and use the relation for the current on a capacitor, we get
dV
out
dt
=

C
1
(V
eff
V
out
)
2


C
1
(6.16)
From this we see that the slope is proportional to /C
1
.
115
For an exponential settling characteristic with time constant , the slope is pro-
portional to 1/. From this, we can conclude that for the exponential approximation,

C
1

(6.17)
Thus, for the transient behavior for an positive step in the input amplitude, we can
use curve tting to determine the time constant for the exponential approximation,
and model the peak detector using a one pole transfer function
V
out
(s)
V
in
(s)
=
K
PD
1 +s
(6.18)
where the peak detector gain K
PD
incorporates the DC relationship between the input
and output voltages which was derived in the previous section. From the analysis
above, we can see that if it is necessary to adjust the time constant of the peak
detector, we can do so by changing the the W/L of M1 and/or the size of C
1
.
Step Down Behavior
From Fig. 6.8 it is evident that the peak detector behaves in a dierent manner for
a step down in the input amplitude than it does for a step up. For a step down,
the eective voltage V
eff
described in the previous section is reduced so that the
transistor M1 is initially turned o. During this period, V
out
declines linearly, with a
slope of I
1
/C
1
. After V
out
is reduced enough so that the transistor starts to turn on,
the behavior can then be approximated with an exponential.
The initial slope of the exponential is equal to the slope of the linear region, so
in a similar manner as for the step up behavior, we can relate the exponential time
constant as

C
1
I
1
(6.19)
Again we see that the settling time can be reduced by reducing C
1
, and in this case
it can also be reduced by increasing I
1
. From this expression and (6.5) we can see
that achieving a faster settling time is in direct conict with minimizing the droop
voltage, so these two factors must be traded o.
116
6.2.3 Design Procedure
The peak detector can now be designed with the following procedure. First, the
allowable droop is used to determine the relative values of C
1
and I
1
using (6.5). The
transistor W/L ratio is then chosen to ensure that (6.13) is satised. Next, the peak
detector is simulated for positive and negative voltage steps, curve tting is performed
to determine the exponential time constants for each case, and the worst case (slowest)
time constant is assumed for the design of the loop dynamics. If necessary, the time
constants can be altered using the relations given in (6.17) and (6.19). Finally, using
the largest allowable step in the input amplitude, the maximum delay due to the
linear portion of the curve for a step down in input amplitude is determined. This
can then be inserted as a xed delay in the loop dynamics to ensure that the system
remains stable in this situation.
The nal peak detector design contains additional buering, level-shifting, and
biasing circuitry not shown in Fig. 6.7. A transient simulation of the peak detector
output voltage for a changing input voltage is shown in Fig. 6.9, which conrms the
behavior shown in the Matlab simulations.
6.3 Loop Filter
As discussed in Chapter 4, the loop lter for the amplitude loop was chosen as an
integrator to force the dierence between the desired output amplitude and the actual
output amplitude (as detected by the peak detector) to zero. There are a number
of dierent choices for implementing an integrator, including a simple operational
amplier design making use of a resistor and a capacitor to set the time constant, and
the well-known g
m
C integrator. A comprehensive summary of integrator structures
and their characteristics can be found in [66]. The main requirement for the integrator
is that the output voltage be able to cover the required range for the VGA control
voltage, which from Fig. 6.5 is from 0.8 V to 1.8 V.
117
Figure 6.9: Transient behavior of peak detector output voltage from Cadence simu-
lation.
The topology selected for the loop lter is the RC opamp integrator, as shown in
Fig. 6.10, where V
des
is the desired output amplitude and V
pd
is the output of the
peak detector. This topology was selected for its low complexity and ease of design.
Figure 6.10: Schematic for the amplitude loop lter.
118
6.3.1 Operational Amplier
Unlike the operational amplier for the phase loop lter, the operational amplier for
the amplitude loop lter does not require a rail to rail output stage. This enabled the
use of a standard opamp topology as shown in Fig. 5.33, with the standard power
supply voltage of 1.8 V. In contrast to the opamp shown in Fig. 5.33, an NMOS
dierential input stage was used, to permit the use of a PMOS source follower output
stage which can drive the output voltage to the positive power supply rail. The opamp
was designed for 60 dB of gain and a phase margin of 75

.
6.3.2 Sizing
The most important decision in the looplter design is the sizing of the resistor-
capacitor combination, as this will determine the time constant of the integrator,
which will in turn determine the loop dynamics. This can be done by starting with
Eq. (4.11) from Chapter 4. In deriving this equation it was assumed that the peak
detector behaved in an ideal manner. As seen in the previous section, a real peak
detector introduces additional delay in the feedback loop, and the behavior can be
approximated with the one pole behavior.
After incorporating (6.18) into (4.6), we arrive at the nal expression for the closed
loop gain of the amplitude loop:
A
out
(s)
V
des
(s)
=
A
in
K
V GA
RC
_
(1 +sRC)(1 +s)
s
2
+s
1

+
A
in
K
V GA
K
BUF
K
PD
RC
_
(6.20)
where A
out
is the output amplitude, V
des
is the desired output amplitude, A
in
is the
input amplitude, K
V GA
, K
BUF
, and K
PD
are the gains of the VGA, buer, and peak
detector, is the time constant of the peak detector, and R and C are the looplter
components. The closed loop gain has been expressed with the assumption that the
desired voltage input to the feedback loop has been adjusted while the input amplitude
stays constant. It can also be expressed with the assumption that the desired voltage
input to the feedback loop stays constant and the input amplitude changes (as would
119
occur for a change in the phase shifter losses as the phase loop adjusts the phase
shifter control voltage), which yields the same expression multiplied by a dierent
constant.
The denominator is second order, and can be put in the form of the characteristic
equation for a second order system. After doing so, we can express the natural
frequency and damping factor as:

n
=

K
V GA
K
BUF
K
PD
RC
(6.21)
=
1
2

RC
K
V GA
K
BUF
K
PD

(6.22)
Using these equations, the design procedure is to set the damping factor () to 0.707
(for critical damping), and solve for the relation between the congurable parameters,
which are the loop lter components. These can then be chosen to yield a reasonably
sized capacitor. Since we have only one parameter to adjust in the looplter (the
resistor and capacitor are inversely proportional to one another), we cannot specify
both the damping factor and natural frequency, so if it was necessary to set these
quantities independantly we would be forced to move to a higher order loop lter. To
conrm the values obtained using this procedure, Matlab code was written to check
the phase margin and bandwidth of the complete loop including the two zeros in
(6.20).
As in the previous chapter, the adjusted VGA gain was employed in the design
equations to account for the interactions with the phase loop due to the variable losses
of the phase shifter. The other gain values were determined through simulation, and
after the design calculations, we arrive at a loop lter capacitor value of 5 pF for
a resistor value of 24 k. This yields a critically damped system with a natural
frequency of 4.7 MHz, so the settling time will be on the order of several hundreds
of ns which is acceptable for our purposes. The natural frequency of the amplitude
loop is over an order of magnitude greater then that of the phase loop, which will
minimize interaction between the two loops.
120
Since the absolute value of integrated components is not well controlled, the ca-
pacitor in the looplter was implemented as a switchable array, allowing 2 bits of
control over the capacitor value. Additional control over the loop dynamics of the
nal system can be realized by adjusting of the peak detector by changing the bias
current. The nal non-ideality to be considered is the eect of nite gain and output
resistance in the opamp, since the analysis presented above assumes an ideal opamp.
These eects were incorporated into a Matlab script which conrmed that they have
a negligible eect at the frequencies of interest.
121
Chapter 7
EXPERIMENTAL RESULTS
This chapter presents the the measured results from the fabricated system. The
system was fabricated in the IBM 7RF process. This is a CMOS mixed-mode process,
with 6 metal layers and a minimum feature size of 0.18 m. In addition to the standard
CMOS devices, the process also oers hyper-abrupt junction varactors (which were
used in the phase shifter), characterized inductor models, and high K metal-insulator-
metal capacitors. The power supply voltage is 1.8 V for the core, with thick oxide
devices available with a power supply voltage of 3.3 V.
The rst section of this chapter describes the design of the additional circuitry
that was included on the die for test and measurement purposes, such as conguration
registers and buers. The next section describes the test setup, including the printed
circuit board design and measurement equipment used. The nal section presents
and discusses the measured results that were obtained for the complete system.
7.1 Test and Measurement Circuits
This section presents the additional circuitry that was included with the system to
allow its functionality to be fully veried.
7.1.1 Conguration Registers
There are a number of digital control signals needed to congure the system, such
as the digital control for the charge-pump currents and the conguration bits for the
looplter components. Using I/O pins for each of these bits would require too many
pins, so a conguration register was designed to allow these bits to be loaded onto
122
Figure 7.1: Schematic for the conguration register, showing 3 of the 20 bits.
the chip in a serial manner. The schematic for the conguration register is shown in
Fig. 7.1.
The register consists of two chains of D ip ops. The Data in and Clock inputs
are used to load the top chain of registers in a serial manner, with the data being
clocked in on the rising edge of the Clock signal. Once the top chain of registers has
been loaded, the Load signal is clocked to load the data into the bottom chain of
registers, where it is available on the Q outputs of the ip ops to drive the digital
control bits. The Reset signal is an active low signal which clears all of the registers.
To allow the system to start in a default state after a reset of the conguration register,
some of the outputs are taken from the Q outputs and the others are taken from the
Q outputs, depending on whether the bit should be 0 or 1 in the default setting.
7.1.2 Control Voltage Observe/Drive Modules
As described in Chapter 4, the phase shifter and VGA control voltages driven by
either the feedback loops or by an external source, depending on whether the system
is in the calibrate or transmit mode. In a nal implementation this would all take
123
Figure 7.2: Schematic of the Observe/Drive module used for the control voltages.
place on the die with analog-to-digital and digital-to-analog converters, but in the
prototype implemented here it is necessary to have a means of observing the control
voltage and driving it from an external source.
To enable this, Observe/Drive modules were implemented for each of the two
control voltages. The schematic for the Observe/Drive module is shown in Figure
7.2. It has two modes of operation, determined by the observe and drive control
bits. When both control bits are low, the transmission gates are open and the control
voltages are disconnected from the external circuitry. When observe is driven high,
the voltage follower opamp buers the control voltage so it can be observed through
o-chip measurement equipment. When drive is driven high, the upper transmission
gate allows the control voltage to be driven by an external source (as would be done
after the feedback loop has been shut o and the system is in the transmit mode).
The additionaly nMOS transistor is included to ensure that the input to the opamp
is not oating when the observe control bit is low.
124
7.2 Test Conguration
This section describes the printed circuit board that was used to test the system, as
well as the congurations of measurement equipment that were used in the dierent
test phases.
7.2.1 Printed Circuit Board Design
A four-layer printed circuit board (PCB) was designed to enable the testing of the
system. The nal system has 44 I/O pins, which eliminated wafer probing as a
test option. The die was bonded directly to the PCB using Chip-On-Board (COB)
bonding to minimize parasitics due to longer than necessary bondwires (as would
have been the case if the die had been packaged). The AppCad software package was
used to design 50 transmission lines for routing the RF input and output signals
on the PCB, and SMA connectors were used to interface between the PCB and the
measurement equipment.
A photograph of the nal assembled PCB is shown in Figure 7.3. The main fea-
tures of the PCB are power connections with decoupling capacitors, potentiometers
for setting bias voltages and currents, buering circuitry for interfacing the congu-
ration registers to the parallel port of a PC, and switches for conguring the divider
multiplexors in the phase feedback loop.
7.2.2 Measurement Equipment
The test board was powered using two HPXXXX adjustable power supplies, and bias
currents were supplied and adjusted using on board potentiometers. The congura-
tion registers were loaded using the parallel port on a laptop running custom C++
code, and control voltages were measured and recorded using an HPXXXX multime-
ter. Phase and gain measurements were made using an Agilent XXXX vector network
analyzer (VNA). For the open loop measurements the VNA was calibrated and op-
125
Figure 7.3: Photograph of the assembled printed circuit board.
126
erated in the frequency domain, and for the closed loop measurements the VNA was
operated in the time domain in Continuous Wave mode. The gain measurements were
also conrmed using an Agilent XXXX signal generator to drive the RF input and
an Agilent XXXX spectrum analyzer to observe the output power levels.
7.3 Measurement Results
A die photo of the fabricated chip is shown in Figure 7.4. In the meaurement results
to follow, the system was operated with a 1.9 GHz input signal at -8.0 dBm. Sim-
ulations were performed for a 2.0 GHz signal, however, the phase characteristics of
the fabricated phase shifter at this frequency were signicantly steeper which reduced
the stability of the phase feedback loop. Reducing the operating frequency to 1.9
GHz reduced the total phase control range from 310

to 275

, but also improved the


stability of the phase feedback loop and reduced the losses of the system.
7.3.1 Phase Characteristics
As mentioned in Chapter 5, the charge pump has been designed with the Up and
Dn branches each having a xed current of 16 I
CP
(where I
CP
is the unit current
set by the charge pump bias current) as well as a digitally switchable current which
varies from 0 to 31 I
CP
. As described in Chapter 4, the phase loop dynamics remain
constant if the sum of the Up and Dn currents remains constant. This provides 5 bits
of control over the phase, by varying (Up, Dn) from (0, 31) I
CP
to (31, 0) I
CP
, with
a constant sum of I
sum
= 63 I
CP
(taking into account the xed components of the
charge pump currents in addition to the variable components). From (4.4), by setting
the average current over a cycle (I
avg
) to zero the output phase can be expressed as
in radians as

out
=
I
diff
I
sum
2 (7.1)
127
Figure 7.4: Die photograph of the complete system.
128
Since I
diff
varies by 2 I
CP
with each step, the theoretical phase change for each step
can be calculated as 11.4

. This allows the entire 360

of phase to be covered with


the 5 control bits if an ideal phase shifter is available. With a real phase shifter, the
system control range may be limited by either the range of the phase shifter or by
the middle of the system control range (which corresponds to a phase shift of 180

)
not coinciding with the middle of the phase shifter range.
The measured results for the phase characteristics of the system are shown in Fig.
7.5. Due to the limitations described above, only 240

of the range is accessible. Since


the middle of the phase shifter control range is not aligned with the 180

center of the
charge pump control range, the low end of the charge pump control range (Up, Dn)
= (31, 0) coincided with a control voltage of 0.12 V, meaning that the full range of
the phase shifter (275

) could not be exploited. The 1.8 V limit on the phase shifter


control voltage was reached for (Up, Dn) = (8, 23), yielding 24 steps of phase control
over the 240

range.
The average phase step across the phase control range is 10.3

, which is less than


the ideal value of 11.4

calculated above. From (7.1) it can be seen that this is due


to the xed components of the charge pump currents (present in the denominator)
being too large relative to the variable components (present in the numerator and the
denominator). The layout of the charge pump current elements is shown in Fig. 7.6,
with the arrays of Up and Dn unit current elements denoted. The transistors were not
arranged in a completely symmetrical array, which has resulted in dierences between
the unit currents in the xed and variable current components. A revised version of
the charge pump should have the unit current elements arranged in a completely
regular array, with elements for the dierent control bits interdigitated with each
other and dummy devices surrounding the entire array.
A line with the same average slope as the measured data has been plotted alongside
the data points (labeled as Ideal), and it can be seen that the data points exhibit
some deviation from the constant phase step that is expected. This is more clearly
129
Figure 7.5: Measured phase of S
21
plotted against steps in the digital control bits.
illustrated in Fig. 7.7 where the phase dierence between successive settings of the
data points is plotted along with the average phase dierence over all of the data
points. This deviation can also be attributed to the charge pump layout causing (or
rather, failing to prevent) dierences between the unit currents among the dierent
control bits.
To conrm this, the system phase shift was measured with individual pairs of Up
and Dn control bits activated, i.e. (Up, Dn) = (0, 0), (1, 1), (2, 2), (4, 4), etc. Ideally,
the phase shift for each case should be 180

since the currents should be equal, or if


the Up and Dn currents are dierent but the unit currents for each are the same, then
all of the bits should display a phase oset from the ideal case of 180

proportional to
the bit value. As seen in Fig. 7.8 this is not the case, indicating that the unit currents
of the charge pump elements vary between the dierent control bits within the Up and
Dn currents. The supposition that this is due to the charge pump layout is conrmed
130
Figure 7.6: Layout of the charge pump unit current elements, with the Up and Dn
arrays denoted.
131
Figure 7.7: Measured phase steps of S
21
for each increment in the digital control bits.
by the fact that the pattern of deviations from the ideal step size was repeated for the
dierent chips that were tested (as will be seen in Section 7.3.5) rather than varying
randomly. This indicates that the linearity of the phase characteristics in Fig. 7.5
could be greatly improved through the use of more rigorous layout techniques for the
charge pump.
One other point worth mentioning is that while the system has only been demon-
strated with the equivalent of 5 bits of resolution over a control range of 360

(assum-
ing that a phase shifter with sucient range were used), if the condition of having
constant loop dynamics across the phase control range were relaxed much greater
phase resolution could be achieved through using dierent combinations of Up and
Dn currents. For each of the 2
5
settings of the Up current, 2
5
unique settings for the
Dn current are possible, meaning that close to 2
10
unique phase settings can be real-
ized (the actual number is less than 2
10
since combinations such as (1,1) and (4,4) yield
132
Figure 7.8: Measured phase of S
21
for individual control bits (0 corresponds to only
the xed current elements being active).
133
the same phase oset). These are not linearly distributed through the phase range
(they are concentrated around the nominal phase shift of 180

), but it still opens the


door for increased phase resolution without any change in the circuitry (other than
insuring that the phase loop remains stable across the range of I
sum
values).
7.3.2 Amplitude Regulation for Changing Phase
As mentioned in Chapter 4, one of the advantages of the proposed system is that it
allows the VGA to compensate for the variable losses of the phase shifter across its
phase control range. The measured gain of the system is plotted in Fig. 7.9, where
the magnitude of S
21
has been plotted for the open loop case (with the VGA control
voltage xed at the maximum of 1.8 V) and the closed loop case (with the amplitude
feedback loop setting the VGA control voltage). For the closed loop case, the desired
amplitude input has been set lower than the minimum amplitude for the open loop
case to ensure that the amplitude requirement can be met across the entire phase
control range. If some variation in the magnitude of S
21
can be tolerated, then the
desired amplitude input can be increased to provide a higher average gain over the
phase control range in exchange for some variation in output amplitude. From Fig.
7.9 it can be seen that the system gain for the open loop case varies by 12.1 dB (which
is what would be expected based on the measurement results for the standalone phase
shifter described in Chapter 5), while the feedback of the closed loop case reduces the
gain variation to just 0.4 dB.
The settling behavior of the phase shifter and VGA control voltages is shown in
Fig. 7.10 for a change in the phase control inputs from (Up, Dn) = (31, 0) to (12,
19), which corresponds to a phase step of 204

. The third waveform in the plot is the


Load signal for the conguration register, which applies the new digital control values
to the charge pump. It can be seen that the phase shifter control voltage settles in
about 6 s, whereas the amplitude control loop is faster with the VGA control voltage
settling in about 3 s. The loop lters were designed to err on the conservative side
134
Figure 7.9: Measured magnitude of S
21
over the phase control range.
of stability, so it is likely that the settling speed could be increased signicantly from
the behavior shown here. This plot also conrms the supposition made in Chapter 5
that there will be limited interaction between the settling of the phase and amplitude
loops.
7.3.3 Phase Regulation for Changing Amplitude
Another benet of the proposed system is that it provides the ability to vary the
amplitude of the output signal (which can be useful for reducing the sidelobes of the
array pattern, as described in Chapter 3) while maintaining a constant phase shift.
The measured phase shift of the system as the gain is varied is plotted in Fig. 7.11.
For the open loop case the phase shifter control voltage is held steady at 0.75 V (which
yields the same phase shift as the closed loop case at the maximum gain setting), and
for the closed loop case the phase feedback loop adjusts the phase shifter control
135
Figure 7.10: Time domain waveforms of VGA and phase shifter control voltages for
a step in the digital phase control bits.
136
Figure 7.11: Phase of S
21
across the amplitude control range.
voltage to compensate for the changing phase shift introduced by the VGA as its gain
is reduced. From the plot it can be seen that the feedback loop reduces the variation
in phase from 32.1

(for the open loop case) to 7.4

. The phase regulation across


the amplitude range is not as eective as the amplitude regulation across the phase
range, and this is because of the buer block used in the phase loop (and described
in Chapter 5). As the amplitude of the input signal to the buer changes, the buer
undergoes variation in the phase shift that it introduces, and this variation appears
as phase error at the output. However, the phase variation introduced by the buer
is small enough that the phase loop still provides considerable improvement over the
open loop case.
The settling behavior of the phase shifter and VGA control voltages is shown in
Fig. 7.12 for a step change in the desired amplitude input to the amplitude feedback
loop. The step change was achieved by using an analog multiplexor to switch abruptly
137
Figure 7.12: Time domain waveforms of VGA and phase shifter control voltages for
a step in the desired amplitude input.
between two input voltages, with the switching initiated by the changing of a bit in
the conguration register. The third waveform in the plot is the Load signal for the
conguration registers. It can be seen that the settling behavior is similar to that for
a phase step in Fig. 7.10, only now the changes in control voltages are being initiated
by a change in the VGA control voltage, with the phase loop altering the phase shifter
control voltage to compensate for the altered phase shift of the VGA at its new gain
setting.
7.3.4 Phase Regulation Across Power Supply Voltages
As mentioned in Chapter 4, one of the intentions in implementing the proposed system
was to provide stable operation across variations in process, voltage, and temperature.
The measured phase characteristics for changing power supply voltages are plotted in
138
Figure 7.13: Phase characteristics for changing supply voltage with open loop opera-
tion.
Figs. 7.13 and 7.14. The power supply voltage was varied from 1.7 V to 1.9 V, and
Fig. 7.13 plots the system behavior for xed phase shifter and VGA control voltages
while Fig. 7.14 plots the system behavior when the phase shifter and VGA control
voltages are set by the feedback loops. It can be seen that the closed loop case does
a better job of regulating the phase to a constant value.
To better illustrate this point, the maximum deviation between the three power
supply voltages at each phase control point is plotted in Fig. 7.15. The maximum
deviation for the open loop case is 28.1

, which is reduced to 7.3

in the closed loop


case. The standard deviation in phase for the open loop case is 9.4

, which is reduced
to just 1.8

when the feedback loops are active in the closed loop case.
The variation in phase with changing power supply voltage is primarily due to the
active loss reduction circuitry included in the phase shifter architecture (described
139
Figure 7.14: Phase characteristics for changing supply voltage with closed loop oper-
ation.
140
Figure 7.15: Maximum deviation in phase characteristics across the control range for
changing power supply voltage.
141
in Chapter 5). A passive phase shifter implementation would have greatly reduced
phase variation for a changing power supply voltage, but completely active phase
shifter implementations (such as All-Pass phase shifters) would have greater phase
variation than shown here, and the use of the feedback loops in the system would
have an even greater benet. The deviation in the phase for the closed loop case that
is shown in Fig. 7.15 is due to the changing power supply voltage altering the delays
in the feedback loop and introducing phase error at the output.
7.3.5 Phase Regulation Across Test Boards
The measured phase characteristics for three dierent test chips are plotted in Figs.
7.16 and 7.17. Fig. 7.13 plots the system behavior for xed phase shifter and VGA
control voltages while Fig. 7.17 plots the system behavior when the phase shifter and
VGA control voltages are set by the feedback loops. It can be seen that the closed
loop case does a better job of regulating the phase to a constant value, and both cases
have less variation than what was seen in the previous section for a changing power
supply voltage.
To better illustrate this point, the maximum deviation between the three test
chips at each phase control point is plotted in Fig. 7.18. The maximum deviation for
the open loop case is 15.4

, and this is reduced to 4.4

in the closed loop case. The


standard deviation in phase for the open loop case is 5.0

, and this is reduced to just


1.2

when the feedback loops are active in the closed loop case.
The variation in phase characteristics between dierent test chips is due to process
variations across the wafer. Since all of the test chips were fabricated on the same
processing run, there is fairly minimal variation between the dierent test chips, as
shown in Fig. 7.16. For a system that was manufactured in higher volume across
dierent fabricaton runs there would be a more signicant variation in the phase
characteristics between dierent chips, and the benets of the feedback loops would
be more pronounced.
142
Figure 7.16: Phase characteristics for three dierent test chips with open loop oper-
ation.
143
Figure 7.17: Phase characteristics for three dierent test chips with closed loop op-
eration.
144
Figure 7.18: Maximum deviation in phase characteristics across the control range for
three dierent test chips.
145
7.3.6 Noise
Since this work is intended for use in a transmitter the noise performance is of less
signicance than if it were to be used in a receiver. Due to the position of the
phase shifter (with its comparatively high losses) preceding the VGA, the noise gure
was relatively poor, varying from 25 dB at the low end of the phase control range
(where phase shifter losses are at a maximum) down to 11 dB at the high end of the
phase control range (where phase shifter losses are at a minimum). This variation
in noise gure is roughly the same as the variation in the losses of the phase shifter.
In a receiver implementation the positions of the phase shifter and VGA would be
reversed, putting the high gain, low noise VGA before the phase shifter so that the
noise and losses of the phase shifter would have a greatly reduced impact on the
overall noise gure.
7.3.7 Power Consumption
The power consumption for each of the various blocks in the system is shown in Table
7.1. The total current draw (excluding pads and test and observation circuitry) is
12.6 mA, and the total power consumption is 23.6 mW. Since the feedback circuitry
is only used during periods when the phase and amplitude are being calibrated or
adjusted, the power consumption during normal operation is just that of the phase
shifter and VGA, which is 13 mW.
146
Block Current (mA) V
dd
(V) Power (mW)
Phase Shifter 4 1.8 7.2
VGA 3.2 1.8 5.76
Buer 20.5 1.8 1.8
Divider 20.5 1.8 1.8
Pulse Generator 0.2 1.8 0.36
Charge Pump 1.7 1.8 3.06
Phase LPF 0.6 3.3 1.98
Peak Detector 0.3 1.8 0.54
Amplitude LPF 0.6 1.8 1.08
Table 7.1: Power consumption for each of the system blocks.
147
Chapter 8
CONCLUSIONS
A system was proposed that allows the phase and amplitude of a signal to be
accurately set and regulated over process and power supply variations. The intended
application is in a phased array transmitter, where the phase and amplitudes of the
array elements are electronically adjusted to shape and steer the beam of radiation
in the chosen direction. Phase and amplitude errors introduced in the array elements
lead to degradation of the array pattern through eects such as sidelobe growth and
reductions in directivity.
The conventional means of setting the phase in a branch of a phased array is
through look-up tables storing the control voltage - phase relationships for the phase
shifter. This method does not compensate for changes in these relationships intro-
duced by processing variations between fabricated devices or changes in operating
conditions such as power supply voltage and temperature. Additionally, if the phase
shifter has variable gain over the phase control range then the amplitude will vary
depending on the phase setting. The proposed system uses a variable gain amplier
(VGA) in conjunction with the phase shifter to compensate for the variable losses of
the phase shifter and simultaneously provide a means of adjusting the amplitude of the
signal. Dual feedback loops were employed to set the control voltages for the phase
shifter and VGA, allowing the phase and amplitude to be closely regulated across
process variations and adjusted to compensate for changes in operating conditions.
The phase shifter was implemented as a reective-type phase shifter which provides
up to 310

of phase shift, and the VGA is a narrowband implementation with a


resonant load that provides about 20 dB of gain and is intended to be used as a power
148
amplier driver. The phase feedback loop includes buers to remove the amplitude
dependancy of the signal, high-speed dividers, a charge pump, and an active loop
lter. The phase of the system is specied by digitally controlling the charge pump
currents. The amplitude loop consists of an RF peak detector used in conjunction
with an active loop lter, and the amplitude is specied by setting an analog input
voltage to the loop lter.
The complete system was fabricated in a 0.18 m CMOS process, and assembled
on a custom printed circuit board using chip-on-board bonding to allow testing and
characterization. The system operates at 1.9 GHz, and the phase can be set with
5 bits of control over a 240

range, and the amplitude can be varied over a 20 dB


range. The amplitude feedback loop reduced the variation in |S
21
| across the phase
control range from 12.1 dB for the open loop case to 0.4 dB. The phase feedback
loop reduced the variation in

S
21
across the amplitude control range from 32.1

to
7.4

. The standard deviation of the phase across the phase control range was reduced
from 9.4

to 1.8

for changing power supply voltages, and from 5.0

to 1.2

between
dierent test chips.
8.1 Future Research
The implemented system performed well, and successfully demonstrated the benets
of the proposed architecture. There were a few limitations that could be corrected in
a revised version of the implemented system, such as the phase control range being
limited to 240

. A future revision could provide the full control range of 360

by
employing a phase shifter with over 360

of phase control range. Another limitation


was the nonlinearity in the phase characteristics introduced by variations between
dierent charge pump unit current elements. This could be improved through the use
of better layout techniques, such as laying out all of the current elements in a regular
array with interdigitated devices and dummy devices surrounding the array.
There are also a number of extensions that could be made to the work described in
149
this dissertation, such as employing a similar architecture for a phased array receiver.
The biggest change for this case would be to reverse the positions of the phase shifter
and VGA to achieve a better noise gure. A practical implementation of this system
in a phased array would have more than one branch integrated on the same die,
so another step towards a complete system would be to implement a number of
phase/amplitude control paths on the same die and investigate the impact of the
substrate noise and interactions on the accuracy of each branch.
150
BIBLIOGRAPHY
[1] J. Ruze, Pattern degradation of space fed phased arrays, MIT Lincoln Labo-
ratory, Project Rept. SBR-1, Dec. 1979.
[2] S. Haykin, Digital Communications. New York: John Wiley & Sons, 1988.
[3] S. Blostein and H. Leib, Multiple antenna systems: Their role and impact in
future wireless access, IEEE Commun. Mag., vol. 41, no. 7, pp. 94101, 2003.
[4] A. Alexiou and M. Haardt, Smart antenna technologies for future wireless sys-
tems: Trends and challenges, IEEE Commun. Mag., vol. 42, no. 9, pp. 9097,
2004.
[5] A. Lozano, F. Farrokhi, and R. Valenzuela, Lifting the limits on high-speed
wireless data access using antenna arrays, IEEE Commun. Mag., vol. 39, no. 9,
pp. 156162, 2001.
[6] D. Cox, Universal digital portable radio communications, Proc. of the IEEE,
vol. 75, no. 4, pp. 436477, 1987.
[7] D. Brennan, Linear diversity combining, Proc. IRE, vol. 47, pp. 10751102,
June 1959.
[8] R. Murch and K. Letaief, Antenna systems for broadband wireless access,
IEEE Commun. Mag., vol. 40, no. 4, pp. 7683, 2002.
[9] S. Alamouti, A simple transmit diversity technique for wireless communica-
tions, IEEE J. Select. Areas Commun., vol. 16, no. 8, pp. 14511458, 1998.
[10] G. Foschini and M. Gans, On limits of wirelss communications in a fading
environment when using multiple antennas, Wireless Pers. Commun., vol. 6,
no. 3, pp. 311335, 1998.
[11] E. Telatar, Capacity of multi-antenna gaussian channels, European Tran. on
Commun., vol. 10, no. 6, pp. 585595, 1999.
151
[12] G. Golden et al., Detection algorithm and initial laboratory results using
V-BLAST space-time communication architecture, IEE Electronics Letters,
vol. 35, no. 1, pp. 1416, 1999.
[13] D. Love, R. Heath, W. Santipach, and M. Honig, What is the value of limitied
feedback for mimo channels? IEEE Commun. Mag., vol. 42, no. 10, pp. 5459,
2004.
[14] A. E. Zooghby, Smart Antenna Engineering. Boston: Artech House, 2005.
[15] H. J. Visser, Array and Phased Array Antenna Basics. West Sussex, England:
John Wiley and Sons, 2005.
[16] R. J. Mailloux, Phased Array Antenna Handbook. Boston: Artech House, 2005.
[17] N. Tyler, B. Allen, and H. Aghvami, Adaptive antennas: The calibration prob-
lem, IEEE Commun. Mag., vol. 42, no. 12, pp. 114122, 2004.
[18] R. Sorace, Phased array calibration, IEEE Trans. Antennas Propagat., vol. 49,
no. 4, pp. 517525, 2001.
[19] G. Tsoulos and M. Beach, Calibration and linearity issues for and adaptive
antenna system, in IEEE Vehicular Tech. Conf., vol. 3, May 1997, pp. 1596
1600.
[20] M. Skolnik, Antenna Theory. New York: McGraw-Hill, 1969, pp. 227234.
[21] A. Hajimiri et al., Phased array systems in silicon, IEEE Commun. Mag.,
vol. 42, no. 8, pp. 122130, 2004.
[22] T. Yamaji et al., A four-input beam-forming downconverter for adaptive anten-
nas, IEEE J. Solid-State Circuits, vol. 38, no. 10, pp. 16191625, 2003.
[23] J. Butler et al., Beamforming matrix simplies design of electronically scanned
antennas, Electronics Design, vol. 9, pp. 170173, Apr. 1961.
[24] T. Ohira, Adaptive array antenna beamforming architectures as viewed by a
microwave circuit designer, in Asia-Pacic Microwave Conf., Sydney, Dec. 2000,
pp. 828833.
[25] D. Parker and D. Zimmermann, Phased arrays-part ii: Implementations, ap-
plications, and future trends, IEEE Trans. Microwave Theory Tech., vol. 50,
no. 3, pp. 688698, 2002.
152
[26] E. Brookner, Phased arrays around the world - progress and future trends,
in IEEE Int. Symp. on Phased Array Systems and Tech, vol. 3, May 1997, pp.
15961600.
[27] C. du Toit et al., Smart multibeam phased array antenna for GSM, GPRS and
EDGE, in IEEE Int. Symp. on Phased Array Systems and Tech., Oct. 2003, pp.
146151.
[28] K. Sheikh, D. Gesbert, D. Gore, and A. Paulraj, Smart antennas for broadband
wireless access networks, IEEE Commun. Mag., vol. 37, no. 11, pp. 100105,
1999.
[29] T. Ohira, Analog smart antennas: an overview, in IEEE Int. Symp. on Per-
sonal, Indoor, and Mobile Radio Com., vol. 4, Sept. 2002, pp. 15021506.
[30] P. Sudarshan, N. Mehta, A. Molisch, and J. Zhang, Antenna selection with RF
pre-processing: robustness to RF and selection non-idealities, in IEEE Radio
and Wireless Conf., Sept. 2004, pp. 391394.
[31] D. Banbury, N. Fayyaz, S. Safavi-Naeini, and S. Nikneshan, A CMOS 5.5/2.4
GHz dual-band smart-antenna transceiver with a novel RF dual-band phase
shifter for WLAN 802.11a/b/g, in IEEE RFIC Symp., June 2004, pp. 157160.
[32] T. Ohira et al., Megalithic microwave signal processing for phased-aray beam-
forming and steering, IEEE Trans. Microwave Theory Tech., vol. 45, no. 12, pp.
23242332, 1997.
[33] P. Wu et al., New miniature 15-20-GHz continuous-phase/amplitude control
MMICs using 0.18-/mum CMOS technology, IEEE Trans. Microwave Theory
Tech., vol. 54, no. 1, pp. 1019, 2006.
[34] F. Ellinger, R. Vogt, and W. Bachtold, Calibratable adaptive antenna combiner
at 5.2 ghz with high yield for laptop interface card, IEEE Trans. Microwave
Theory Tech., vol. 48, no. 12, pp. 27142720, 2000.
[35] A. Natarajan, A. Komijani, and A. Hajimiri, A 24ghz phased-array transmitter
in 0.18 m cmos, in Proc. Int. Solid-State Circuits Conf., San Francisco, USA,
Feb. 2005.
[36] T. H. Lee, The Design of CMOS Radio-Frequency Integrated Circuits. New
York: Cambridge University Press, 2004.
153
[37] V. Vintola et al., Variable-gain power amplier for mobile wcdma applications,
IEEE Trans. Microwave Theory Tech., vol. 49, no. 12, pp. 24642471, 2001.
[38] D. Cong et al., A variable gain amplier with 50-db control range for 900-mhz
applications, IEEE J. Solid-State Circuits, vol. 45, no. 12, pp. 23242332, 1997.
[39] B. Razavi, RF Microelectronics. New Jersey: Prentice Hall, 1998.
[40] F. Gardner, Charge-pump phase-locked loops, IEEE Trans. Commun., vol. 28,
no. 11, pp. 18491858, 1980.
[41] J. Wallace, H. Redd, and R. Furlow, Low cost MMIC DBS chip sets for phased
array applications, in IEEE MTT-S Int. Microwave Symp. Dig., June 1999, pp.
677680.
[42] K. Fujii, A sophisticated analysis procedure for an MMIC phase shifter, IEEE
J. Solid-State Circuits, vol. 33, no. 4, pp. 666668, 1998.
[43] H. Hayashi et al., A high-Q broad-band active inductor and its application to
a low-loss analog phase shifter, IEEE Trans. Microwave Theory Tech., vol. 44,
no. 12, pp. 23692374, 1996.
[44] F. Ellinger, R. Vogt, and W. Bachtold, Compact reective-type phase-shifter
MMIC for C-band using a lumped-element coupler, IEEE Trans. Microwave
Theory Tech., vol. 49, no. 5, pp. 913917, 2001.
[45] K. Miyaguchi et al., An ultra-broad-band reection-type phase-shifter MMIC
with series and parallel LC circuits, IEEE Trans. Microwave Theory Tech.,
vol. 49, no. 12, pp. 24462452, 2001.
[46] F. Ellinger, R. Vogt, and W. Bachtold, Ultracompact reective-type phase
shifter MMIC at C-band with 360
o
phase-control range for smart antenna com-
bining, IEEE J. Solid-State Circuits, vol. 37, no. 4, pp. 481486, 2002.
[47] H. Zarei and D. Allstot, A low-loss phase shifter in 180 nm cmos for multiple-
antenna receivers, in Proc. Int. Solid-State Circuits Conf., Feb. 2004, pp. 392
393.
[48] D. Viveiros, D. Consonni, and A. Jastrzebski, A tunable all-pass MMIC active
phase shifter, IEEE Trans. Microwave Theory Tech., vol. 50, no. 8, pp. 1885
1889, 2002.
154
[49] H. Hayashi and M. Muraguchi, An MMIC active phase shifter using a variable
resonant circuit, IEEE Trans. Microwave Theory Tech., vol. 47, no. 10, pp.
20212026, 1999.
[50] H. Zarei et al., A full-range all-pass variable phase shifter for multiple antenna
receivers, in Proc. Int. Symp. on Circuits and Systems, Kobe, Japan, May 2005.
[51] F. Ellinger, U. Lott, and W. Bachtold, An antenna diversity MMIC vector mod-
ulator for HIPERLAN with low power consumption and calibration capability,
IEEE Trans. Microwave Theory Tech., vol. 49, no. 5, pp. 964969, 2001.
[52] F. Ellinger and W. Bachtold, Novel principle for vector modulator-based phase
shifters operating with only one control voltage, IEEE J. Solid-State Circuits,
vol. 37, no. 10, pp. 12561259, 2002.
[53] A. Nagra and R. York, Distributed analog phase shifters with low insertion
loss, IEEE Trans. Microwave Theory Tech., vol. 47, no. 9, pp. 17051711, 1999.
[54] F. Ellinger, H. Jackel, and W. Bachtold, Varactor-loaded transmission-line
phase shifter at C-band using lumped elements, IEEE Trans. Microwave Theory
Tech., vol. 51, no. 4, pp. 11351140, 2003.
[55] S. Hamedi-Hagh and C. Salama, A novel C-band CMOS phase shifter for com-
munication systems, in Proc. Int. Symp. on Circuits and Systems, vol. 2, May
2003, pp. 316319.
[56] J. Wong, V. Cheung, and H. Luoung, A 1-v 2.5-mw 5.2-ghz frequency divider
in a 0.35-m CMOS process, IEEE J. Solid-State Circuits, vol. 38, no. 10, pp.
16431648, 2003.
[57] B. Razavi, K. Lee, and R. Yan, Design of high-speed, low-power frequency
dividers and phase-locked loops in deep submicron CMOS, IEEE J. Solid-State
Circuits, vol. 30, no. 2, pp. 101109, 1995.
[58] D. Parker and D. Zimmermann, Phased arrays-part i: Theory and architec-
tures, IEEE Trans. Microwave Theory Tech., vol. 50, no. 3, pp. 678687, 2002.
[59] K. Kobayashi et al., Gaas HBT 0.75-5 ghz multifunctional microwave-analog
variable gain amplier, IEEE J. Solid-State Circuits, vol. 29, no. 10, pp. 1257
1261, 1994.
155
[60] R. Meyer and W. Mack, A DC to 1-ghz dierential monolithic variable-gain
amplier, IEEE J. Solid-State Circuits, vol. 26, no. 11, pp. 16731680, 1991.
[61] K. Koh, Y. Youn, and H. Yu, A gain boosting method at RF frequency using
active feedback and its application to RF variable gain amplier (VGA), in
Proc. Int. Symp. on Circuits and Systems, Arizona, USA, May 2002.
[62] R. Meyer, Low-power monolithic rf peak detector analysis, IEEE J. Solid-State
Circuits, vol. 30, no. 1, pp. 6567, 1995.
[63] S. Zhak, M. Baker, and R. Sarpeshkar, A low-power wide dynamic range en-
velope detector, IEEE J. Solid-State Circuits, vol. 38, no. 10, pp. 17501753,
2003.
[64] Y. Zhou and M. Wah, A wide band CMOS RF power detector, in Proc. Int.
Symp. on Circuits and Systems, Kos, Greece, May 2006.
[65] D. Su and W. McFarland, An IC for linearizing RF power ampliers using enve-
lope elimination and restoration, IEEE J. Solid-State Circuits, vol. 33, no. 12,
pp. 22522258, 1998.
[66] T. Georgantas, Y. Papananos, and Y. Tsividis, A comparative study of ve
integrator structures of monolithic continuous-time lters, in Proc. Int. Symp.
on Circuits and Systems, Chicago, USA, May 1993.

You might also like