Nestor Lopez (Ph.D. 2008),
HIGH-EFFICIENCY POWER AMPLIFIERS FOR
LINEAR TRANSMITTERS
by
NÉSTOR DAVID LÓPEZ
B.S., University of Puerto Rico, 2001
M.S., University of Puerto Rico, 2003
M.S., University of Colorado, 2007
A thesis submitted to the
Faculty of the Graduate School of the
University of Colorado in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
Department of Electrical and Computer Engineering
2008
This thesis entitled:
High-Efficiency Power Amplifiers for Linear Transmitters
written by Néstor David López
has been approved for the
Department of Electrical and Computer Engineering
Prof. Zoya Popović
Prof. Dragan Maksimović
Date
The final copy of this thesis has been examined by the signatories;
and we find that both the content and the form meet acceptable presentation
standards of scholarly work in the above mentioned discipline
López, Néstor David (Ph.D., Electrical Engineering)
High-Efficiency Power Amplifiers for Linear Transmitters
Thesis directed by Prof. Zoya Popović
Digital modulation techniques used in wireless communications with radio frequency (RF) carriers can increase channel capacity, improve transmission quality,
enhance security, and provide services not possible with analog modulation. Improving spectral efficiency by allowing the envelope of the RF signal to vary with
time can enhance channel capacity. Envelope variations introduce RF power amplifier (RFPA) linearity requirements. The power conversion efficiency of RFPAs
operating in linear modes is limited to less than 25% for signals with high envelope
variations. Poor conversion efficiency leads to significant dissipated power that is
wasted as heat, and it also shortens the lifetime of battery operated equipment.
This thesis focuses on the design and implementation of high-efficiency but
non-linear RFPAs that can be linearized with dynamic biasing techniques, such
as polar modulation. In polar modulation the signal that is modulated in both
amplitude and phase is separated into two components; an envelope varying baseband signal that contains the amplitude modulation, and a phase modulated signal
with constant envelope. The non-linear PA can efficiently amplify the constant
envelope signal, while envelope variations can be injected by the supply. The
RFPA can reconstruct the amplitude and phase modulated signal by operating as
a time-domain multiplier. A 56% efficient linear polar transmitter with +20 dBm
of output power was designed and implemented for the EDGE modulation scheme.
This is the highest efficiency reported to date for a polar EDGE transmitter.
This thesis also studies the use of transistors such as GaN HEMTs, and SiC
MESFETs in the design of high-power high-efficiency RFPAs and how they com-
iii
pared to a high-efficiency RFPA implemented with standard Si LDMOS. Widebandgap semiconductors have better intrinsic material properties than silicon,
i.e. larger energy gap (support higher internal electric fields before breakdown),
lower relative permittivity (lower capacitive loading), higher thermal conductivity
(higher heat handling), and higher critical electric fields (higher RF power). A
45-W, 87% efficient UHF non-linear transmission line hybrid Class-E RFPA is
designed with a GaN HEMT on a SiC substrate.
iv
Dedication
A Mabel, Lourdes, Jackie y Esther.
Acknowledgments
I first would like to thank Prof. Zoya Popović for her invaluable guidance and
support during these five years, and for the opportunity of been a member of her
laboratory. I would also like to thank my wife Mabel Ramı́rez for literally been
there in every step of the way. I am also grateful to the members of my committee:
Prof. Dragan Maksimović, Prof. Dejan Filipović, Prof. James Curry and Prof.
Rafael Rodrı́guez Solı́s.
During this PhD expedition, I had the great opportunity to work closely with
Dr. Joseph Hagerty, Dr. Srdjan Pajić, Dr. Jim Vian, Dr. Patrick Bell, Dr.
Michael Forman, and Dr. Xufeng Jiang. I really appreciate the time invested in
discussing RF and its mysteries. I also, would like to acknowledge Milos Janković
and Quianli Mu, my partners in crime.
I am also grateful to the many members of Active Antenna Laboratory; Narisi
Wang, Charles Dietlein, Kenneth Vanhille, Sebastien Rondineau, Marcelo Perotoni, Alan Brannon, Jacques Loui, Jason Breitbarth, Milan Lukic, Michael Buck,
Christi Walsh, Negar Ehsan, Luke Sankey, Mike Elsbury, Nicola Kinzie, Evan
Cullens, Jonathan Chisum, Jason Shin, and Erez Falkenstein. Special thanks go
to John Hoversten for his help and many RFPA discussions. I would also like to
acknowledge Prof. Dana Anderson, Sid Gustafson, Bill McCalpin, Rachael Tearle,
and Jaroslava Hladisova.
I also would like to acknowledge Barbara Kraus, and Prof. Mark Hernández
vi
from the Colorado Diversity Initiative and to my friends Prof. Nelson Sepúlveda,
Dr. Fabio Sánchez, Luis Cuadra, Suinya Perez, Idalis Villanueva, and Neil Rodrı́guez for reminding me there is a life outside the lab. I would also like to acknowledge my family for their support, their prayers and understanding.
vii
Contents
1 Introduction & Thesis Outline
1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
1.2 Thesis Organization . . . . . . . . . . . . . . . . . . . . . . . . . .
5
2 Background
7
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7
2.2 Efficiency Definitions . . . . . . . . . . . . . . . . . . . . . . . . .
7
2.3 PA Sweeps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
2.4 Classic Amplifier Modes of Operation . . . . . . . . . . . . . . . .
10
2.5 RF Power Amplifiers with Harmonic
Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
12
2.6 Class-E Power Amplifiers . . . . . . . . . . . . . . . . . . . . . . .
14
2.6.1
Transmission-Line Class-E . . . . . . . . . . . . . . . . . .
16
2.7 High Efficiency and Linear Power Amplifiers . . . . . . . . . . . .
18
2.7.1
Enhancing Efficiency of Linear PAs . . . . . . . . . . . . .
18
2.7.2
Enhancing Linearity of High-Efficiency PAs . . . . . . . .
19
3 High-Efficiency Linear Polar Transmitters
23
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
3.2 Envelope Trackers . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
3.2.1
Switch-Mode Power Supplies . . . . . . . . . . . . . . . . .
viii
26
3.2.2
Linear Assisted Switch-Mode Power Supplies . . . . . . . .
27
3.3 Class-E Power Amplifiers in
Polar Transmitters . . . . . . . . . . . . . . . . . . . . . . . . . .
28
3.3.1
Bias-Tee Design . . . . . . . . . . . . . . . . . . . . . . . .
30
3.4 Example - Polar Two-Tone Architecture . . . . . . . . . . . . . .
32
3.4.1
Class-E PA Performance with a Single and
Two-Tone Input . . . . . . . . . . . . . . . . . . . . . . . .
32
3.4.2
Class-E Amplifier in a Polar Transmitter . . . . . . . . . .
35
3.4.3
Distortion and Predistortion . . . . . . . . . . . . . . . . .
39
3.4.4
Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . .
40
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
40
4 A High-Efficiency Linear Polar Transmitter for EDGE
43
4.1 EDGE Modulation Scheme . . . . . . . . . . . . . . . . . . . . . .
43
4.2 Polar EDGE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
46
4.2.1
Bandwidth Limited Envelope . . . . . . . . . . . . . . . .
48
4.2.2
EDGE Polar Transmitter . . . . . . . . . . . . . . . . . . .
48
4.3 High-Efficiency Envelope Tracker . . . . . . . . . . . . . . . . . .
49
4.4 High-Efficiency RFPAs . . . . . . . . . . . . . . . . . . . . . . . .
51
4.4.1
Transistor Selection . . . . . . . . . . . . . . . . . . . . . .
52
4.4.2
RF Amplifier Design . . . . . . . . . . . . . . . . . . . . .
53
4.4.3
RFPA Characterization
. . . . . . . . . . . . . . . . . . .
56
4.5 EDGE Polar Transmitter Implementation . . . . . . . . . . . . .
58
4.6 System Performance . . . . . . . . . . . . . . . . . . . . . . . . .
58
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
61
5 High-Efficiency UHF PA Design based on Load-pull
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ix
63
63
5.2 Load-pull Setup and Calibration . . . . . . . . . . . . . . . . . . .
65
5.2.1
Pre-matching Circuits . . . . . . . . . . . . . . . . . . . .
66
5.2.2
DUT Fixture Calibration . . . . . . . . . . . . . . . . . . .
68
5.2.3
Block Deembeding & Impedance Verification . . . . . . . .
69
5.2.4
Transistor Characterization . . . . . . . . . . . . . . . . .
71
5.3 Example with a RF3932 Prototype . . . . . . . . . . . . . . . . .
72
5.4 Optimum Amplifier Design . . . . . . . . . . . . . . . . . . . . . .
74
5.5 Class-E Amplifier Performance . . . . . . . . . . . . . . . . . . . .
77
5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
78
6 Technology Comparison of UHF Amplifiers
80
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
80
6.2 Transistor Technologies . . . . . . . . . . . . . . . . . . . . . . . .
81
6.3 Transistor Characterization in Class-E Mode . . . . . . . . . . . .
83
6.4 Class-E Amplifier Performance . . . . . . . . . . . . . . . . . . . .
85
6.4.1
Input Power Sweep . . . . . . . . . . . . . . . . . . . . . .
86
6.4.2
Supply Voltage Sweep . . . . . . . . . . . . . . . . . . . .
88
6.4.3
AM-to-AM and AM-to-PM Distortion . . . . . . . . . . .
90
6.5 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . . . .
92
6.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
94
7 Discussion & Future Work
95
7.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
7.2 Additional Optimization Examples . . . . . . . . . . . . . . . . .
98
7.3 Class-F Transistor Characterization . . . . . . . . . . . . . . . . . 100
7.4 Active Load-pull . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.5 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.5.1
Outphased Assisted Polar Transmitter . . . . . . . . . . . 106
x
7.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Bibliography
112
xi
Tables
3.1
IMD Level for Polar Two-Tone Test with and without Predistortion 39
4.1
Load-pull for HFET Transistor at 880-MHz. . . . . . . . . . . .
54
5.1
Measured Results for Class-E Amplifier . . . . . . . . . . . . . .
77
6.1
Summary of Transistor Properties . . . . . . . . . . . . . . . . .
82
6.2
Optimum POUT and ηD Impedances from Load-pull Contours. . .
85
6.3
Prototypes Power Sweep Performance . . . . . . . . . . . . . . .
86
7.1
Class-F Optimum POUT and ηD . . . . . . . . . . . . . . . . . . . 101
xii
Figures
1.1
Diagram of a polar transmitter . . . . . . . . . . . . . . . . . . .
3
1.2
Normalized dissipated power versus efficiency . . . . . . . . . . .
4
2.1
Power diagram of an RFPA . . . . . . . . . . . . . . . . . . . . .
8
2.2
Diagram of a bench setup for testing PAs . . . . . . . . . . . . .
10
2.3
Voltage waveform across the transistor versus RF cycle . . . . .
11
2.4
Diagram of transistor with input and output matching networks
13
2.5
Diagram of the Class-E power amplifier . . . . . . . . . . . . . .
16
2.6
Transmission line hybrid class-E power amplifier . . . . . . . . .
17
2.7
Diagram of an outphasing architecture . . . . . . . . . . . . . . .
20
2.8
Diagrams that illustrate dynamic biasing techniques . . . . . . .
22
3.1
Diagram of an IQ modulator . . . . . . . . . . . . . . . . . . . .
24
3.2
Diagram that illustrates polar modulation with an RFPA . . . .
24
3.3
Diagram of a polar transmitter . . . . . . . . . . . . . . . . . . .
25
3.4
Diagram of a polar transmitter with the envelope tracker implemented with a linear assisted switching-power supply
. . . . . .
27
3.5
Diagram of a bias-tee . . . . . . . . . . . . . . . . . . . . . . . .
30
3.6
Measured bias-tee performance . . . . . . . . . . . . . . . . . . .
31
3.7
Block diagram of a polar system with FPGA digital control . . .
31
xiii
3.8
Pout (solid line) and PAE (dashed line) load-pull contours for MESFET AFM04P2 (Alpha industries) . . . . . . . . . . . . . . . . .
3.9
33
Single tone power sweep and two-tone performance for 10 GHz
class-E PA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
3.10 Supply sweep for Class-E PA operated at 10-GHz . . . . . . . .
36
3.11 MESFET POUT and worst IMD3 polar load-pull contours . . . .
37
3.12 Measured performance of class-E PA with dynamic biasing . . .
38
4.1
EDGE frequency spectrum and constellation . . . . . . . . . . .
44
4.2
0.2 ms long sample of the normalized EDGE envelope and the
envelope PDF and CDF . . . . . . . . . . . . . . . . . . . . . . .
45
4.3
Envelope and phase baseband spectra of an EDGE modulated signal 47
4.4
Bandlimited envelope spectra and its effects on the combined
EDGE spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
4.5
Block diagram of the EDGE polar transmitter . . . . . . . . . .
49
4.6
Schematic of the closed-loop envelope tracker . . . . . . . . . . .
50
4.7
Photograph of the transmission-line hybrid 880-MHz class-E PA
53
4.8
Measurements of the EDGE 880-MHz class-E PA . . . . . . . . .
55
4.9
Block diagram of the implemented EDGE polar transmitter . . .
57
4.10 Photograph of the implemented setup. Baseband circuit design,
selection and FPGA programming was performed by Dr. Xufeng
Jiang.
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
4.11 Measured output spectrum of the polar transmitter when the envelope tracker is the linear amplifier (red) and fast switching buck
converter (blue). The dashed line indicates the EDGE spectral
5.1
mask requirement. . . . . . . . . . . . . . . . . . . . . . . . . . .
60
Setup for load-pull device characterization . . . . . . . . . . . . .
64
xiv
5.2
Typical constellation achieved by impedance tuners . . . . . . .
67
5.3
TRL fixture for transistor characterization in class-E mode . . .
68
5.4
Photograph of load-pull setup at dBm Engineering/Peak Devices
facilities in Boulder, CO . . . . . . . . . . . . . . . . . . . . . . .
5.5
POUT (W) and ηD (%) load-pull contours for the RF3932 (GaN
HEMT on a SiC substrate) biased at 28 V . . . . . . . . . . . . .
5.6
72
73
(a) POUT (W) and (b) ηD (%) contours when the transistor supply
voltage is 28 V, 36 V and 48 V . . . . . . . . . . . . . . . . . . .
73
5.7
Drain efficiency values for specified output power . . . . . . . . .
75
5.8
Load-pull contours in a normalized 20 Ω Smith Chart for different
5.9
values of the parameter α in Equation 5.12 . . . . . . . . . . . .
75
Photograph of the prototype amplifier . . . . . . . . . . . . . . .
77
5.10 Power sweep of the Class-E mode 370-MHz Power Amplifier for a
supply voltage of 28 V . . . . . . . . . . . . . . . . . . . . . . . .
6.1
78
POUT and ηD load-pull contours for each of the four different transistor technologies at a supply voltage of 28 V . . . . . . . . . . .
84
6.2
Input power sweep for each of the amplifier prototypes . . . . . .
87
6.3
Supply voltage sweep for each of the amplifier prototypes . . . .
89
6.4
(a) AM–to–AM and (b) AM–to–PM for each of the amplifier prototypes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.1
91
Load-pull contours for the parameter hα=0.75 of Equation 5.12 for
each of the four different technologies and a supply voltage of 28 V 98
7.2
Optimization of voltage dependent load-pull contours . . . . . .
99
7.3
Output matching network for Class-F operation . . . . . . . . . 101
7.4
POUT and ηD load-pull contours for three different transistor technologies for a supply voltage of 28 V and Class-F terminations . . 102
xv
7.5
Diagram of an active load-pull setup . . . . . . . . . . . . . . . . 103
7.6
Implemented active load . . . . . . . . . . . . . . . . . . . . . . 104
7.7
Active load measure contours for the NPTB00050 biased at 28 V. 105
7.8
Diagram of the outphased assisted polar transmitter . . . . . . . 108
xvi
Chapter 1
Introduction & Thesis Outline
1.1
Introduction
Digital modulation techniques used in wireless communications with radio frequency (RF) carriers increase channel capacity, improve transmission quality,
enhance security, and provide services not possible with analog modulation [1].
Improving spectral efficiency by allowing the envelope of the RF signal to vary
with time can enhance channel capacity. Envelope variations introduce radio frequency power amplifier (RFPA) linearity requirements. A metric used to quantify
required linearity is the crest factor or the peak-to-average ratio (PAR), which relates to waveform peak amplitude to its mean value:
·
PAR = 10 log10
max(|x(t)|2 )
mean(|x(t)|2 )
¸
(1.1)
where x(t) is the signal to be transmitted.
The envelope is constant for signals that have predominantly phase and frequency modulation such as Q-PSK or G-MSK (GSM) and the corresponding PAR
is 0 dB. This means that a non-linear high-efficiency RFPA can used to amplify
such signals. On the other hand, WCDMA signals have a large peak-to-average
ratio, between 9 dB and 11 dB, requiring a RFPA that is able to amplify both
small and large amplitude signals with equal fidelity. One way to achieve linearity is to back-off the RFPA input power from its 1 dB compression point by
an amount of approximately the PAR value to accommodate the signal envelope
variation [1]. Backing-off the amplifier input power significantly degrades power
consumption efficiency as well as average output power. For example, a 100 W
(+50 dBm) amplifier for WCDMA is needed for transmitting an average power of
10 W (+40 dBm). The corresponding mode of RFPA linear operation is typically
in class A or backed-off class AB modes. These linear operating modes limit the
efficiency of the RFPA to less than 25% where linearity is traded for efficiency [2].
The goal of this thesis is to make significant contributions towards obtaining
linear amplifiers that are simultaneously efficient. A number of researchers have
investigated dynamic biasing techniques that can simultaneously provide high
efficiency and good linearity, and most notable examples are Envelope Elimination and Restoration (EER), and polar modulation [3] - [11]. Dynamic biasing
techniques achieve linearity by allowing the RFPA supply voltage to follow the
signal envelope, since the output voltage to a constant load vary proportional to
the supply voltage. These types of transmitters are more complex, and the total transmitter efficiency depends on both the RFPA and the supply modulating
circuit efficiency, as shown if Figure 1.1.
Polar transmitters achieve high-efficiency with the use of both a high-efficiency
envelope tracker (ET) and a high-efficiency RFPA. Switch-mode power supplies
(SMPSs) are ideal for the role of envelope tracking, because they can achieve ultra
high-efficiencies, are light and can step up or down the DC voltage. High-efficiency
class-E non-linear power amplifiers are ideal for the role of the RFPA because they
can also achieve ultra high-efficiencies and more than that, the voltage delivered to
a constant load is linearly proportional to the supply voltage, achieving relatively
2
VDC
Envelope
Tracker
IQA(t)
~A(t)
IQ
RFPA
IQ
IQRF
RF
Figure 1.1: Diagram of a polar transmitter. The RFPA is fed with two signals;
f A . IQφRF is a phase modulated RF signal with constant envelope,
IQφRF and IQ
f A is a baseband signal that contains the signal envelope. The RFPA
while IQ
functions as a time-domain multiplier, reconstructing the desired signal IQRF .
Time alignment between the two signals is crucial to avoid distortion. The total
transmitter efficiency is the product of the efficiency of the envelope tracker and
the RFPA.
linear amplifiers without the need of pre-distortion techniques.
This thesis discusses the design of high-efficiency non-linear class-E power amplifiers that can be linearized with dynamic biasing techniques such as polar modulation. Results are presented for two polar transmitters; a 10-GHz polar transmitter implemented with a transmission line class-E RFPA and a linear assisted
switch-mode power supply as the ET [12] - [14], and a 880-MHz polar transmitter
implemented with a hybrid transmission line class-E RFPA and a Buck converter
as the ET [15] - [17]. Also, UHF Class-E high-efficiency high-power amplifiers are
design with wide-bandgap transistors and their performance is compared to an
amplifier implemented with standard Si LDMOS using the same techniques [18]
- [19].
The main motivation of this work is to save the prime resource that is power.
Figure 1.2 shows the normalized dissipated power as a function of supply efficiency. If a power amplifier is 25% efficient, it means that for every 1 W of RF
output power, the power supply needs to provide 4 W while the remaining 3 W
3
Normalized Dissipated Power
15
10
5
0
0
20
40
60
Supply Efficiency [%]
80
100
Figure 1.2: Normalized dissipated power versus efficiency. Typical RFPA efficiencies for linear amplifiers is below 25%. Due to the nonlinear relation between
dissipated power and efficiency, for efficiencies below 40%, increasing the efficiency
by a small amount reduces significantly the dissipated power enhancing the lifetime of battery operated equipment and reducing required heatsink mass.
are dissipated as heat. This ratio is maintained as the RF output power increases
to 100 W or 500 W of RF output power, where heat dissipation increases to 300 W
and 1.5 kW respectively. Increasing the amplifier efficiency from 25% to 40% reduces the amount of dissipated power by half, so that the dissipated power is
only 1.5 W for every 1 W of RF output power. This also means that the heatsink
mass can be halved or that a cellphone battery will last 1.6 times longer. Farther
increasing the efficiency to 75% means that only 1/3 W is dissipated for every 1 W
of RF output power. In this case the heatsink mass is farther reduced to approximately a tenth of the original mass at 25% efficiency and a cellphone battery will
last 3 times as long. Due to the nonlinear relation between dissipated power and
efficiency, for efficiencies below 40%, increasing the efficiency by a small amount
reduces significantly the dissipated power enhancing the lifetime of battery op-
4
erated equipment and reducing the heatsink mass. This work demonstrate that
efficiencies higher than 50% can be achieve in polar transmitters, while maintaining transmitter linearity.
1.2
Thesis Organization
Following this introductory chapter which briefly describes the motivation for the
work, six additional chapters provide details of the background and contributions
as follows:
• Chapter 2 - This chapter covers the basic definitions used throughout the
thesis and gives a background of different modes of RFPA operation. In
this context, the choice of switch-mode class-E operation used in the other
chapters and its transmission line version is explained. In addition to basics
of power amplifiers, this chapter also gives a brief discussion of different
methods that can be used to obtain high-efficiency and linear PAs.
• Chapter 3 - This chapter discusses in detail polar transmitters, which is
the particular choice of linearity adopted in this thesis. As an example, a
10-GHz polar transmitter is implemented with a transmission line class-E
RFPA and a linear assisted switched-mode power supply as the envelope
tracker. Load-pull data under Cartesian and polar two-tone excitation is
presented for amplifier output impedance matching selection.
• Chapter 4 - This chapter covers the design, implementation and performance of a high-efficiency linear polar transmitter for EDGE modulated
communication signals (one of the current standards for cellular communications). The 880-MHz transmitter uses a non-linear transmission line
hybrid class-E high-efficiency RFPA and an ultra-fast high-efficiency switch5
mode power supply as the envelope tracker. The polar transmitter meets the
EDGE envelope mask with +20 dBm of output power and 56% efficiency.
• Chapter 5 - This chapter covers the steps for the design of high-efficiency
UHF PAs with P OUT > 50 W. The steps are illustrated with the design of
a high-efficiency RFPA that uses a GaN HEMT on a SiC substrate prototype from RFMD as the active device. Load-pull characterization shows a
significant tradeoff between optimum output power and optimum efficiency
contours. An optimization procedure based on a weighted Euclidean distance is developed in order to take this tradeoff into account for the final
amplifier design.
• Chapter 6 - This chapter discusses the design and implementation of UHF
high-efficiency power amplifiers with four different transistor technologies;
GaN HEMTs on a Si substrate, GaN HEMTs on a SiC substrate, SiC MESFETs and Si LDMOS. The goal is to achieve more than 40 W of output
power with over 80% drain efficiency at 370-MHz with four different device
technologies, Si LDMOS being an old and established technology while the
others are emerging new semiconductor devices. The prototype amplifiers
are compared in terms of performance and static distortion.
• Chapter 7 - This chapter highlights the contributions of the presented work,
gives additional examples of the proposed optimization procedure introduced
in Chapter 5 and discusses future work.
6
Chapter 2
Background
2.1
Introduction
As discussed in Chapter 1, it is a challenge to design high-efficiency and linear
power amplifiers. The digital modulation schemes used to enhance data rates
in a limited bandwidth result in signals with PAR as high as 10 dB. The linear
RFPAs used to amplify these signals typically suffer from low power conversion
efficiency. There are two trends to conciliate linearity and efficiency; (1) enhance
the efficiency of linear amplifiers and (2) enhance the linearity of high-efficiency
non-linear power amplifiers. An overview of these techniques is presented in this
chapter, but first some basic definitions are given along with an overview of classes
of PA operation and PA testing procedures.
2.2
Efficiency Definitions
In the amplification process, the RF input signal (PIN RF ) is amplified by converting available DC power (PDC ) to RF output (POUT RF ). The remainder of
the power not converted to POUT RF is dissipated as heat. The closed system, as
shown in Figure 2.1, has two inputs, PIN RF and PDC and two outputs POUT RF and
PDC
PIN RF
Amplifier
POUT RF
PDISS
Figure 2.1: Power diagram of an RFPA. The inputs are PIN RF and PDC and the
outputs are POUT RF and PDISS . The PDC not converted to POUT RF is dissipated
as heat.
PDISS . Because this system has multiple inputs there are several ways to define
power conversion efficiency. The equations to be presented next assume power
levels to be given in Watts. The supply efficiency (ηS ), relates the consumed DC
power to the obtained POUT RF and it is given by
ηS =
POUT RF
PDC
(2.1)
This ratio is also known as collector efficiency (ηC ) or drain efficiency (ηD ) depending on whether a bipolar or a FET is used as the active device. This ratio
does not takes into account PIN RF , that might be significant for RFPAs. The
power-added efficiency (PAE) takes into account this input power and is given by
PAE =
=
POUT RF − PIN RF
PDC
RF
POUT RF − POUT
GRF
PDC
(2.2)
(2.3)
where GRF is the amplifier RF gain. Comparing ηS to PAE, it is evident that PAE
is always smaller than ηS and is a more conservative number to quote for efficiency.
The difference between these two power ratios increases when GRF decreases. A
third way to define efficiency of an amplifier is with the overall efficiency and is
8
defined by
ηoverall =
POUT RF
PDC + PIN RF
(2.4)
Although ηoverall gives the best indicator to the dissipated power PDISS , ηS and
PAE are the two most common methods to quote RFPA efficiency.
2.3
PA Sweeps
Figure 2.2 shows a diagram of a bench for testing PAs where all impedances
are matched to Z0 . All the parts are discussed in Chapter 5. The setup allows
calibration and measurement of PIN RF and POUT RF . Once a prototype of an
amplifier is designed, the characteristic impedance Z0 (usually 50 Ω) is presented
to the input and output of the RFPA. The setup allows for several testing methods
such as, input power sweep and supply voltage sweep that are the most useful for
polar PA design:
• Input Power Sweep - the amplifier is biased at the desired quiescent
point. PIN RF is increased within a defined range and parameters such as
PIN RF , POUT RF , gain, ηS and PAE are measured and recorded. This type of
sweep allows measurements such as 1 dB compression point and drive level
for optimum PAE.
• Supply Voltage Sweep - the input power is fixed, usually at the value
that optimizes PAE and the supply voltage is swept within a defined range.
Again, parameters such as PIN RF , POUT RF , gain, ηS and PAE are measured
and recorded as a function of supply voltage. In this type of sweep it is
also important to measure parameters such as AM-to-AM and AM-to-PM
distortion. AM-to-AM relates the supply voltage to the output voltage
9
PMETER
Input
Source
Variable
Attenuator
PMETER
Output
Input Block
High Power
Attenuator
3 dB
Attenuator
Pre Amp
Lowpass
Filter
Spectrum
Analyzer
DUT
Lowpass
Filter
Coupler
Output Block
Osc
PMETER
Reflect
Z0
Z0
Figure 2.2: Diagram of a bench setup for testing PAs where all impedances are
matched to Z0 . The setup allows to carefully calibrate and perform input power
and supply sweeps to measure parameters such as PIN RF , POUT RF , gain, ηS and
PAE.
across a constant load. In polar transmitters it is desirable for this relation
to be linear. The AM-to-PM relates how the delay through the amplifier
varies as a function of the supply voltage. In polar transmitters it is desired
for this response to be flat. Supply voltage sweeps are important when
implementing dynamic biasing techniques.
2.4
Classic Amplifier Modes of Operation
The classic amplifiers mode of operation, such as Class-A, Class-AB, Class-B,
and Class-C are defined in terms of conduction angle, 2θ [20]. In all the cases
matching conditions are such that maximum output power is achieved. Excellent
references about amplifier modes of operation are [2], [20] - [22]. The conduction
angle (Figure 2.3) is defined as the portion of the RF cycle during which the
device is in the active region [20] and it is a function of drive level, as well as the
quiescent bias point [2].
• Class-A - Class-A amplifiers are linear amplifiers and they are biased so
that the transistor remains in the active region during the entire RF cycle
and the conduction angle is 2θ=360◦ . A drawback of Class-A amplifiers is
10
Class−A
Class−AB
Saturated Class−AB
Voltage Across the Device, VDS
1
0.8
0.6
0.4
0.2
0
0
50
100
150
200
RF Cycle [deg]
250
300
350
Figure 2.3: Voltage waveform across the transistor versus RF cycle. In ClassA (blue) the transistor remains in the active region during the entire RF cycle
(360◦ ). In Class-AB (green) the transistor remains in the active region between
180◦ < 2θ < 360◦ and it is driven into cut-off for part of the RF cycle. For large
input signals the transistor can also be driven into saturation (red).
that significant DC power is dissipated even if no RF input is present. The
maximal theoretical supply efficiency is 50%.
• Class-AB - The conduction angle for Class-AB amplifiers is between 180◦ <
2θ <360◦ , so that the transistor remains in the active region for more than
50% but less than of the full RF cycle. They are not linear amplifiers and
signals with an amplitude-modulated envelope will be distorted significantly
at peak power levels. However, the change in conduction angle causes a
useful increase in efficiency at the expense of drive power and gain [2].
• Class-B - For transistor to operate in this mode the conduction angle is set
to 2θ = 180◦ so that the transistor remains in the active region during 50% of
the RF cycle. The theoretical optimum efficiency for this mode is 78.5% [2]
at the expense of gain reduction compared to Class-AB. [2] suggests the use
11
of high gain technologies, such as GaAs HBT and pHEMT to compensate
for the gain reduction in this mode of operation.
• Class-C - In this mode the conduction angle is small (2θ <180◦ ) and the
transistor remains in the active region for less than 50% of the RF cycle.
This mode of operation is highly nonlinear; the RF input signal turns “ON”
the device and the gain is highly dependent upon input power.
One of the main advantages of reducing the conduction angle is efficiency
enhancement, in the trivial case less power is dissipated with no input drive.
Class-B and Class-C are biased so that no power is consumed in the absence
of input drive. Modulation types such as FM or GMSK (GSM), do not require
linear amplification of the input signal allowing the use of these efficient modes of
operations.
2.5
RF Power Amplifiers with Harmonic
Terminations
Device nonlinearities are exhibited as generation of harmonics and intermodulation products in the output RF signal. These harmonics can be used to shape
the voltage and current waveforms across the device. For example, in Class-F
operation the harmonics are terminated so that the voltage waveform across the
transistor is a square wave and this leads to benefits in both power and efficiency [2]. The number of terminated harmonics depends on several parameters,
e.g. the ft of the device, its output capacitance (COUT ), the complexity of the
circuit, and passive circuit insertion loss.
At microwave frequencies, transmission line stubs can be used to implement
harmonic terminations [23]. This passive circuitry can be designed very precisely
12
+VDC
PA
RFC
f0 Input
Matching
Network
PIN RF
2f0, 3f0, …
Term
f0 Output
Matching
Network
POUT RF
Z0
Z0
Z0
Figure 2.4: Diagram of transistor with input and output matching networks.
The output matching network contains a pre-matching circuit that terminates
harmonics with desired reflective terminations to shape the voltage and current
waveforms across the device. Shaping the transistor waveforms can enhance, for
example, power conversion efficiency.
with full wave EM software to present the desired impedance conditions. A brief
description of modes of operation with terminated harmonics is given next.
• Class-F - In this mode the transistor operates as a saturated controlled
current source [2]. Terminating the even harmonics with a short circuit, and
the odd harmonics with an open circuit produce an approximate square-wave
voltage and a half sine-wave current waveform across the device.
• Class-F−1 - In the inverse class-F the even harmonics are now terminated
with open circuits, while the odd harmonics are terminated with short circuits, producing an approximate square-wave current and a half sine-wave
voltage across the device.
• Class-E - For Class-E operation all the harmonics are terminated with
an open circuits, while the fundamental is terminated with a particular
impedance that achieves soft-switching. The next section discusses the
13
Class-E mode in greater depth.
• Class-E−1 - In the inverse Class-E, again the fundamental impedance is
matched to a particular impedance that achieves soft-switching, but now
the harmonics are terminated with a short circuit.
• Class-J - This mode is similar to a Class-AB amplifier with a large shunt
capacitance following the device. This capacitance is of precise value so
that the harmonics are shorted, but the fundamental impedance can still be
matched [2].
Because power amplifiers are driven with large signals they usually display
strong nonlinear behavior. The active devices are driven into saturation or cut-off
for a certain portion of the RF cycle. Modeling these strong nonlinear effects
is a difficult task, even if CAD models and tools are available [20]. The loadpull technique, discussed in more detail in Chapter 5, can be used to obtain an
empirical model for the active device as a starting point for amplifier design.
2.6
Class-E Power Amplifiers
There are several sources of loss in power amplifiers: dissipated power in the
active device; loss due to power in generated harmonics; insertion loss of input
and output matching networks; loss in the bias network; losses in connections to
the outside world; and radiation loss. However, the dominant loss is the dissipated
power in the active device due to simultaneous presence of V and I. This loss
can be minimized by operating the transistor as a switch. Ideal switches have this
temporal orthogonality by instantaneous toggling between open and short circuit
conditions so that there is zero current if the switch is open and zero voltage
when the switch is close. In reality, when operating transistors as switches there
14
are two sources of loss; (1) the transistor ON resistance (RON ) responsible for
the presence of voltage under the short circuit condition, and (2) the output
capacitance (COUT ) that limits how fast can the switch toggle between states also
producing simultaneous presence of V and I.
In practice, RON can be made very small and depends on the transistor power
capabilities and technology. COUT also depends on these parameters and it is
responsible for limiting the maximal frequency the transistor can operate as a
switch. At low frequencies (hundred of kHz range) switch-mode power supplies can
achieve efficiencies in the order of 90% [24]. As the operating frequency increases,
resonant techniques such as zero-voltage-switching (ZVS) and soft-switching help
to decrease power lost in COUT by discharging it right before the switch is closed.
In soft-switching (Class-E) the voltage across the transistor is not only zero, but
its derivative is also zero.
Figure 2.5(a) shows a diagram of the classical Class-E amplifier. The transistor is modeled as a switch with an intrinsic output capacitance and it is followed
by a series resonator (LS and CS ). Shunt capacitance (Cp ) can be added externally at lower frequencies to shape the voltage and current waveforms across the
device. Figure 2.5(b) shows there is no overlap between ideal voltage and current
waveforms across the device leading to a 100% efficiency assuming no transient
effects and zero ON resistance. An important characteristic of Class-E is that the
voltage across the device can be higher than 3.56×VDC . VDC needs to be kept
below VDSS /3.56, where VDSS is the breakdown voltage of the transistor, to avoid
permanently damaging the device.
In Class-E operation the transistor behaves as a switch; the device is biased
close to cut-off and driven into compression. The transistor turns ON and OFF
with the RF drive which also provides the switching frequency. For the device to
operate in ideal Class-E mode, all the harmonics must be terminated in an open
15
Class-E Waveforms
+VDC
VDS/VDC and IDS/IDC
Voltage
Current
Series
Resonator
RFC
LS CS
COUT
Transistor
CP
RL
t/T
(a)
(b)
Figure 2.5: (a) Diagram of the Class-E power amplifier. The transistor operates as
a switch with its intrinsic output capacitance and it is followed by a series resonator
(LS and CS ). Additional shunt capacitance (CP ) can be added to shape the
waveforms across the device. (b) Voltage and current waveforms across the device.
Ideal maximum efficiency in Class-E mode is 100%. An important characteristic
of Class-E is that the voltage across the device can be higher than 3.56×VDC . VDC
needs to be kept below VDSS /3.56, where VDSS is the breakdown voltage of the
transistor, to avoid permanently damaging the device.
circuit, while the fundamental is matched to the Class-E impedance which depends
on the operating frequency and the device intrinsic output capacitance [23], [25]
- [30]; and is shown to be:
ZE =
0.28
◦
ej49
2πfs COUT
(2.5)
Class-E power amplifiers have been shown to achieve 95% in the low MHz range
and 70% drain efficiency at X-Band [25] - [31]. For excellent reviews on Class-E
operation the reader is refered to [20] - [22], [32], [34].
2.6.1
Transmission-Line Class-E
Lumped-component losses and parasitic effects are significant at microwave frequencies and the transmission line class-E PA is more appropriate in these frequen16
Fundamental
Tuning
2nd Harmonic
Termination
(a)
j10
j5
j20
2ND Harmonic
j2
0
2
5
10
¥
20
Fundamental
-j2
-j5
-j20
-j10
(b)
Figure 2.6: (a) Photograph of the output circuit of a transmission line hybrid
class-E PA. A λg /8 open stub at a λg /8 distance is used to terminate the 2nd harmonic, where λg is the guided wavelength at the fundamental frequency. Lumped
components are used to match the fundamental impedance. (b) Measured output
impedance as a function of frequency in a 10-Ω Smith Chart. A high-impedance
is presented at the 2nd harmonic, while the fundamental is matched to the class-E
impedance, ZE .
17
cies [32], [34]. In the transmission line class-E PA, harmonics are terminated with
open stubs. The number of harmonics that should be terminated depend on the
ft and COUT of the device. Significant efficiency enhancement is usually achieved
by terminating only the 2nd harmonic. Additional harmonics can be terminated
for small improvements in efficiency at the expense of circuit complexity. A λg /8
open stub at λg /8 distance from the transistor can be used to terminate the 2nd
harmonic, where λg is the guided wavelength at the fundamental frequency. The
fundamental impedance matching is also implemented with stubs, but depending
on the fundamental frequency the use of lumped components might be more appropriate, resulting in a transmission line hybrid class-E power amplifier as shown
in Figure 2.6(a). Figure 2.6(b) shows the measured output impedance in a 10-Ω
Smith Chart. A high impedance is presented to the transistor 2nd harmonic, while
the fundamental is matched to ZE .
2.7
High Efficiency and Linear Power Amplifiers
As was discussed in the introduction there are two trends when designing highefficiency linear power amplifiers: (1) enhancing the efficiency of linear power
amplifiers, and (2) enhancing the linearity of high-efficiency power amplifiers.
These two methods are discussed next.
2.7.1
Enhancing Efficiency of Linear PAs
As was mentioned in section 2.4 the easiest way to enhance the efficiency of a linear
amplifier is to reduce the conduction angle as in Class-AB amplifiers. However, it
is important to consider that there is a tradeoff between linearity and efficiency
and that Class-AB amplifiers are not linear since they exhibit strong nonlinearities at peak power levels [2]. At the expense of little additional degradation in
18
linearity the efficiency can be farther enhanced if the harmonics are terminated as
in Class-F or Class-J. It is important to point out that biasing conditions and fundamental matching networks are chosen to achieve amplifier linearity and output
power. More advanced techniques such as the Doherty architecture also enhance
the efficiency of linear amplifiers, this time by active load-pulling the RF load by
applying current from a second RFPA. An excellent review of the Doherty amplifier is in [2]. This architecture can be costly for high power PAs because it uses 2
devices, however the output power is not doubled.
2.7.2
Enhancing Linearity of High-Efficiency PAs
There are two techniques to enhance the linearity of non-linear high-efficiency
power amplifiers; (1) outphasing and (2) dynamic biasing. A brief description of
these techniques is presented next.
Outphasing
The idea behind the outphasing technique is that an amplitude and phase modulated signal can be resolved into two constant envelope signals that are out of
phase and that are applied to two highly efficient, highly nonlinear power amplifiers whose outputs are then summed appropriately [1]. The outphasing technique
is covered in detail in [1], a brief description is presented here for completeness.
The complex representation of a signal that is modulated in both amplitude and
phase can be represented as
s(t) = r(t)ejθ(t)
(2.6)
For outphasing, this signal is split into two signals with constant amplitudes with
modulated phases:
19
IQ Modulator 1
Iˆ1(t)
Mixer
RF Power
Combiner
+
90º
Q
ˆ1(t)
RF
Oscillator
IQ
RF1
RFPA
IQ
RF1
High-Efficiency
Class-E PA
Mixer
+
IQ Modulator 2
Iˆ2(t)
IQRF
Mixer
RF Power
Combiner
+
Q
ˆ2(t)
90º
RF
Oscillator
IQ
RF2
RFPA
IQ
RF2
High-Efficiency
Class-E PA
Mixer
Figure 2.7: Diagram of an outphasing architecture. The signal that is modulated
in both amplitude and phase is divided into two constant envelope phase modulated signals. This two signals when combined reconstruct the original signal.
Due to the envelope removal, nonlinear high-efficiency RF power amplifiers can
be used at each of the branches. Typical combiners will dissipate the out-of-phase
component, however advance techniques such as Chireix combiners reduce this
loss.
S1 (t) = s(t) − ξ(t)
(2.7)
S2 (t) = s(t) + ξ(t)
(2.8)
where ξ(t) is the quadrature signal and is defined by
s
ξ(t) = js(t)
2
rmax
−1
r2 (t)
(2.9)
When combining these two signals in a power combiner, the in-phase signal
components add together and the out-of-phase signal components cancel out. Because these two signals have constant envelopes, non-linear high-efficiency power
20
amplifiers can be used in each of the branches. Imperfections in the system such
as path imbalance (gain and phase), result in distortion. However, predistortion
techniques can be used to compensate for some of these nonidealities. Traditional
combining techniques are lossy because they dissipate the out-of-phase component. However, recycling techniques such as [37], [36] can be used to reduce this
loss. More advance combining techniques such as Chireix architecture [1], [2] can
be used to efficiently combine the two signals.
Dynamic Biasing
Dynamic biasing techniques, such as Envelope Elimination and Restoration (EER)
[3], [5] and polar transmitters [6] - [17], modulate the RFPA supply voltage to
proportionally scale the output voltage across a constant real load. Figure 2.8(a)
shows a diagram of an EER system. The amplitude and phase modulated RF
signal is converted to polar form. This can be accomplished with an envelope
detector to obtain the envelope signal that is used as a reference for an envelope
tracker, and a amplitude limiter to remove all the envelope to produce a constant
envelope phase modulated signal. The envelope tracker supplies the envelope
signal to the RFPA through the supply. The RFPA is used as a time domain
multiplier to amplify and reconstruct the RF signal. In polar transmitters, Figure
2.8(b), the signal conversion to polar form is done in the baseband digital domain
but otherwise the principle is the same as EER. Chapter 3 talks in detail about
polar transmitters and its components, which are the approach taken in this thesis.
21
Efficient
Envelope
Tracker
IQA(t)
IQRF
Complex
to Polar
Converter
IQA(t)
IQ
PA
RF
IQRF
Digital
Generator
(a)
IQA(t)
Efficient
Envelope
Tracker
IQA(t)
RF
Oscillator
Phase
Shifter
IQ
PA
RF
IQRF
(b)
Figure 2.8: Diagrams that illustrate dynamic biasing techniques. The RFPA is
f A . IQφRF is a phase modulated RF signal
fed with two signals; IQφRF and IQ
f A is a baseband signal that contains the signal
with constant envelope, while IQ
envelope. The RFPA functions as a time-domain multiplier, reconstructing the
desired signal IQRF . Time alignment between the two signals is crucial to avoid
distortion. The total transmitter efficiency is the product of the efficiency of the
envelope tracker and the RFPA. In EER (a) the RF signal is divided into polar
form, while in polar transmitters (b) the baseband signal is converted digitally to
polar form.
22
Chapter 3
High-Efficiency Linear Polar
Transmitters
3.1
Introduction
Polar modulation has become increasingly popular in RF transmitters due to its
potential to simultaneously achieve linearity and efficiency [6] - [17]. Increasing
the transmitter efficiency reduces heat dissipation and extends the lifetime of the
battery in portable equipment. Digital modulation uses symbols that are encoded
with in-phase (I) and quadrature (Q) baseband components. An IQ modulator is
used to combine and frequency shift these two signals to the RF carrier, creating
the modulated RF signal IQRF as shown in Figure 3.1.
Depending on the modulation scheme used, the IQRF signal might vary in
both amplitude and phase. A polar representation of this signal is
IQA = |IQRF |
IQΦRF =
IQRF
|IQRF |
(3.1)
(3.2)
IQ Modulator
Mixer
I(t)
RF Power
Combiner
IQRF(t)
+
RF
Oscillator
90º
Q(t)
Mixer
Figure 3.1: Diagram of an IQ modulator. The baseband I and Q signals modulate
the RF carrier. The output is the RF modulated signal, IQRF .
The resultant is the baseband signal IQA and the constant amplitude phase modulated RF signal IQΦRF . The original IQRF signal can be recovered from
IQRF = IQA · IQΦRF
(3.3)
A way to obtain the IQΦRF signal is with normalized Iˆ and Q̂ of the form,
I
Iˆ = p
I 2 + Q2
Q̂ = p
(3.4)
Q
I2
+ Q2
(3.5)
been the input to the IQ modulator of Figure 3.1.
IQA
IQ RF
RFPA
IQRF
Figure 3.2: Diagram that illustrates polar modulation with an RFPA. The IQΦRF
is the RF input to the PA while IQA is the supply voltage which changes with a
bandwidth dictated by the envelope signal bandwidth. The RFPA operates as a
time domain multiplier reconstructing the IQRF signal.
24
VDC
Envelope
Tracker
IQA(t)
~A(t)
IQ
IQ Modulator
Mixer
ˆI(t)
RF Power
Combiner
+
90º
RF
Oscillator
IQ
RFPA
RF
IQRF
Q(t)
ˆ
Mixer
Figure 3.3: Diagram of a polar transmitter with baseband IQA , Iˆ and Q̂ inputs.
The envelope tracker has two inputs (DC supply and the reference IQA signal)
f A ). IQ
f A is the tracked envelope signal that
and one output (low pass filtered IQ
feeds the RFPA. Because envelope and biasing is provided by the tracker, the
total efficiency depends on both the tracker and the RFPA.
As shown in Figure 3.2 supply modulating techniques, such as polar modulation and EER [3] implement the function of Equation 3.3 by operating the
RFPA as a time domain multiplier that reconstructs the IQRF signal by utilizing
the amplifier output power dependence to the supply voltage. Nothing is said in
this diagram about what sort of RFPA is needed to perform this function and in
practice people have used saturated Class-A, Class-AB or Class-E amplifiers for
this role [6] - [17], [39] - [42]. Later in this chapter we will show that Class-E amplifiers are ideal for polar transmitters because the output voltage to a constant
load is linearly proportional to the supply voltage.
Figure 3.3 shows a diagram of a polar transmitter. The envelope tracker
f A , a lowpass filtered version of IQA . Because IQ
f A contains biasing
supplies IQ
and envelope variations, it handles similar power levels compared to the RFPA.
The overall transmitter efficiency is given by,
25
POUT RF
PDC
PET POUT RF
=
·
PDC
PET
ηTransmitter =
=ηET · ηS
(3.6)
(3.7)
(3.8)
where PET and ηET is the output power and the efficiency of the ET, while POUT RF
and ηS is the output power and the supply efficiency of the RFPA. From this
equation we can conclude that it is not only important for the RFPA to be efficient,
but to obtain a high overall transmitter efficiency, it is crucial that the envelope
tracker is efficient and can provide the necessary envelope bandwidth.
3.2
Envelope Trackers
As was discussed in the previous section, the overall transmitter efficiency depends
on the efficiency of the RFPA and the efficiency of the envelope tracker. One
way to design ultra high-efficiency envelope trackers is with switch mode power
supplies (SMPSs). SMPSs can achieve high efficiencies, are small, light, economic
and have the capability to step-up or step-down voltages [43], [45]. However, for
these converters losses increase with switching frequency and it is a challenge to
design fast high-efficiency SMPSs. SMPSs can be assisted with linear amplifiers
to cover larger bandwidths [43].
3.2.1
Switch-Mode Power Supplies
SMPSs can be implemented with transistors and diodes to step up or step down a
supply voltage (VDC ). The output voltage depends on a reference signal converted
usually to a duty cycle. There are several SMPSs configurations, such as Buck,
boost, Buck-boost, and Ćuk converters. Each converter has its own properties;
26
Envelope Tracker
H1 (s)
IQA-LP(t)
Switching
Amplifier
fB
+
Linear Amplifier
H2 (s)
IQA-HP(t)
~A(t)
IQ
fB
Linear Assisted
Switching Power supply
IQ
RFPA
IQRF
RF
Figure 3.4: Diagram of a polar transmitter with the envelope tracker implemented
with a linear assisted switching-power supply [14], [43], and [44]. The envelope
signal IQA is band-separated into low (IQA−LP ) and high (IQA−HP ) frequency
components. The low frequency component is supplied by a slow but ultra-efficient
SMPSs, while the high frequency components are supplied by less efficient but fast
linear amplifier.
for example, Buck converters can only step down the supplied voltage. To reduce
switching harmonics, an SMPS is followed by a lowpass filter. The bandwidth
of envelope signals such as WCDMA is in the order of 4 MHz [6]. In addition,
the bandwidth provided by the tracker needs to take into account factors such as
oversampling ratio. Because losses increase with frequency it is not possible to
obtain high-efficiency with signal bandwidth exceeding the MHz range. However,
linear amplifiers can be used to take care of the high frequency components, while
the SMPS is responsible for supply the lower frequency envelope components.
3.2.2
Linear Assisted Switch-Mode Power Supplies
The envelope bandwidth of signals such as WCDMA is in the MHz range, but
85% of the envelope power is between DC and 300 kHz [6]. In such a system it
makes sense to have an ultra efficient SMPSs to supply the power contained in
the 300 kHz bandwidth while the remaining envelope spectrum is supplied by a
27
less efficient but high-bandwidth linear amplifier. In the case where the efficiency
of the SMPSs is 90% and the efficiency of the linear amplifier is 30% the total
envelope tracker efficiency is still on the order of 80%.
Figure 3.4 shows a diagram of a polar transmitter with the envelope tracker
implemented with a linear assisted switching power supply [6], [14], [43], and [44],
the envelope signal IQA , is assumed to have a bandwidth that might be on the
order of couple of MHz is band-separated into low (IQA−LP ) and high (IQA−LP )
frequency components. The low frequency components are supplied by a slow
ultra-efficient (9̃0%) SMPSs, while the high frequency components are supplied
by the linear amplifier. In [14] it is shown that depending on the signal statistics
and the efficiency of the SMPSs and the linear amplifier, there is a band-separation
frequency were efficiency is optimized.
When implementing this type of envelope tracker it is crucial to have a flat pass
band response. For example, ripple, delay or amplitude unbalance between low
and high frequency components will introduce distortion, which can be corrected
by pre-distortion [46] - [49].
3.3
Class-E Power Amplifiers in
Polar Transmitters
The class-E mode of operation lends itself well to the polar transmitter architecture. As shown in [30], for a 50% duty cycle of the switch drive, the voltage across
the transistor is given by:
vDS (t) =
IDS
[ωs t − 1.86(cos(ωs t − 32.5◦ ) − cos(32.5◦ ))],
ωs · COUT
(3.9)
where IDS is the average drain current, COUT is the device nonlinear output capacitance, and fs is the input signal frequency. If the DC supply voltage is provided
28
through an ideal RF choke, the average value of the switch voltage has to be equal
to the DC drain supply voltage VDS :
VDS
1
=
Ts
Z
Ts
vDS (t)dt =
0
IDS
πωs COUT
(3.10)
By properly terminating the ideal class-E PA, the power delivered is
1
1
2
POU T = RE IOUT
= RE (1.86IDS )2
2
2
(3.11)
where RE is the real part of the optimal class-E load impedance, ZE = RE + jXE
and IOUT is the magnitude of the output current iOUT . From Equations 3.10 and
3.11 we obtain,
1
2
POUT = RE (1.86 · π · ωs · COUT )2 VDS
2
(3.12)
For a lossless output matching network, the output power is linearly proportional
2
to VDS
. As a result, the output voltage of a class-E PA across a constant load can
be linearly varied by varying the drain voltage:
³
´
p
VOUT = ± 26 · fs · COUT · RE · RL VDS
(3.13)
In addition, the optimal efficiency and optimal load impedance are not affected
by the bias variation, since the transistor current and voltage amplitudes only
change with bias, but not their time-domain waveform shapes. The power can
theoretically vary from zero to the maximal available power, but in practice the
lowest power is limited by feedthrough, and the maximal power is constrained by
the power handling of the device. In [13], and [12], the linearity of 10-GHz class-E
PAs are examined and it is demonstrated that these highly saturated PAs can be
linearized to some degree using polar modulation.
29
+VDC
Port 3
CBS
LRFC
Port 2
Port 1
CB
Figure 3.5: Diagram of a bias-tee with blocking capacitor (CB ), inductor (LRFC ),
and shunt capacitors (CBS ) in the DC path. The basic function of the bias-tee is
to diplex DC and RF signals.
3.3.1
Bias-Tee Design
Figure 3.5 shows the basic diagram of a bias-tee. The main purpose of the circuit
is to diplex RF and DC signals. It is desired that the DC signal only flows from
Port 3 to Port 1, while the RF signal only flows from Port 1 to Port 2. To deal with
unstabilities and to ensure little VDC variations it is a common practice to add
a number of capacitors (CBS ) shunted in the DC path and to wind the inductor
LRFC in ferrite cores.
In polar transmitters the envelope of the signal is been fed through Port 3.
Due to the large signal bandwidth, the bias-tee Port 3 to Port 1 path, needs to
be re-design taking into account the expected frequency content and a lowpass
filter approach is more appropriate. Figure 3.6 shows the measured S31 response
for a bias-tee using an air core inductor and a ferrite. The bandwidth of the biastee implemented with an inductor with a ferrite is limited and will distort large
bandwidth envelope signals when applying dynamic biasing.
30
Bias−Tee S31 vs. Frequency
10
Air Inductor
Inductor with Ferrite
0
Amplitude [dB]
−10
−20
−30
−40
−50
−60
−70
0
50
100
150
200
250
Frequency [MHz]
300
350
400
Figure 3.6: Measured S31 for bias-tee implemented with an air and ferrite core
wound inductors. In polar transmitters the supplied voltage is intentionally varied
to reproduce the signal envelope. Bias-tee bandwidth limitations will introduce
distortion in the envelope signal.
Envelope Tracker
H1 (s)
IQA-LP(t)
Switching
Amplifier
fB
FPGA
Xilinx
XC2V
1000
Vertix 2
+
Linear Amplifier
IQA-HP(t)
H2 (s)
fB
~A(t)
IQ
Linear Assisted
Switching Power supply
RFIN
Phase
Modulator
IQ
RFPA
IQRF
RF
Figure 3.7: Block diagram of a polar system with FPGA digital control. The
RFPA is a class-E 10-GHz MESFET amplifier [12]. The envelope signal is split
into a low frequency component which controls a DC-DC converter and a high frequency component which provides additional envelope AC variations [14]. Phase
variation is achieve with a digitally controlled phase shifter.
31
3.4
Example - Polar Two-Tone Architecture
In this section, the linearity of a polar transmitter implemented with a linear
assisted switch-mode power supply as the envelope tracker and an ultra-nonlinear
high-efficiency switched-mode class-E RFPA is examined [12]. The RF carrier is at
10 GHz, which is a higher frequency than most current commercial communication
systems, but is a common frequency in other applications, such as radar. At carrier
frequencies above 2 GHz, the circuit parasitics and device nonlinearities are more
pronounced and difficult to model. The class-E X-band PA used in this study
has been demonstrated in a two-stage efficient PA [30] and its linearity and EER
operation were investigated in [13].
A diagram of the implemented system is shown in Figure 3.7. The digitallycontrolled drain bias provides amplitude modulation of the output voltage through
a ultra-efficient slow DC-DC converter in combination with a fast less efficient
linear amplifier which provides the high frequency portion of the signal envelope.
The signal is generated digitally and converted from IQ to polar form. The digital
control of the bias allows allocation of the amplitude modulation between the slow
and fast circuits in order to optimize efficiency for different modulated signals [43].
IQΦRF is generated from an 10-GHz RF source and a phase shifter.
3.4.1
Class-E PA Performance with a Single and
Two-Tone Input
The class-E PA used throughout this work was designed using design equations
as in [29] and load-pull characterization of a GaAs MESFET at 10 GHz, shown
in Figure 3.8. The theoretical class-E impedance calculated from the output
capacitance of this device is 24.7 + j28.4 Ω. The maximum PAE of 59% at 27.2 +
j31.4 Ω is obtained from load-pull measurements. POUT , gain, and ηD for the
32
Figure 3.8: Pout (solid line) and PAE (dashed line) load-pull contours for MESFET
AFM04P2 (Alpha industries). The class-E impedance (27.2+j31.4 Ω) is indicated
with ‘×’. This amplifier was fabricated by Dr. Narisi Wang, formerly a student
at CU.
PA are +19.9 dBm, 8.1 dB, and 64% respectively when the PA is compressed by
2.2 dB. The input impedance is matched to 8.2 + j27.3 Ω.
Fig. 3.9(a) shows an input power sweep for the amplifier biased at 4.1 V
and 10 mA. The output spectrum for a two-tone test when the tones are offset
by 200 kHz and the input power is +11 dBm (optimum PAE) is shown in Figure
3.9(b). The 200 kHz bandwidth is chosen as an example since it corresponds to the
bandwidth of an EDGE communication signal. Class-E PAs are highly nonlinear,
this can be seen in the high upper and lower IMD3 levels of -19.5 dBc and 17.8 dBc, respectively. The PAE for the two-tone test with the same total input
power dropped to 45% and the output power dropped to +17.5 dBm indicating a
gain reduction.
To understand this degradation, load-pull measurements were taken for a twotone test with the same 200 kHz offset. Figure 3.9(c) shows POUT (solid line)
33
100
Pout
16
70
60
Gain
10
50
8
40
PAE
6
30
4
20
2
10
0
−10
−5
0
5
10
80
ηD
14
12
20
90
Output Spectrum (dBm)
18
Efficiency [%]
Output power [dBm] and Gain [dB]
20
−10
−20
−30
−40
−50
0
10
−17.8dBc
−19.5dBc
0
−1
−0.5
Input Power [dBm]
0
0.5
1
Offset Frequency (MHz)
(a)
(b)
Pout [dBm]
IMD3 [dBc]
j50
j100
0
10
15
−−26
28
3
−3 25
0
−18
16
−26
15
50−28 100
−28
1
1−6217
−2
j10
−30
17
−
−221
−26 3
j25
∞
−30
−j10
−j25
−j100
−j50
(c)
Figure 3.9: (a) Single tone power sweep for class-E PA at 10 GHz, (b) two-tone output spectrum for 200 kHz offset between tones (carrier frequency is 10 GHz), and
(c) MESFET POUT and worst IMD3 load-pull contours. The Class-E impedance
is indicated with an ‘×’; for this impedance PAE is 45% with +17.5 dBm of POUT .
The optimumal PAE of 48% is obtained for 22 + j38 Ω, ‘o’ with +17.5 dBm of
POUT . Worst IMD3 and maximum POUT contours coincide.
34
and worst IMD3 contours (dashed-line) for the MESFET with +11dBm of input
power. It is interesting to observe that there is an overlap between maximum
output power and worst IMDs. The POUT and PAE for the class-E impedance
are +17.5 dBm and 45% respectively. Optimum PAE (48%) impedance shifted to
22 + j38 Ω (indicated with a ‘o’) with same POUT .
3.4.2
Class-E Amplifier in a Polar Transmitter
Digital polar modulation was implemented using a Xilinx Virtex II FPGA to
provide baseband and control signals. Digital-to-Analog Converters (DAC) are
used to control a phase shifter responsible for modulating the RF carrier and
also to send control signals to the linear portion of the envelope amplifier. The
switching converter is also directly controlled by the FPGA. Time alignment of
these three components is critical in achieving PA linearity.
The output power level of the envelope amplifier is greater than that of the
RFPA, making it a significant factor in system efficiency. An efficient wideband
envelope amplifier is realized by taking advantage of the high efficiency of a lowswitching frequency converter and the broad bandwidth of a linear amplifier. The
envelope command is filtered into high and low frequency components which are
sent to the linear amplifier and switching converter respectively. This technique,
referred to as band separation, has the capability to improve efficiency for various
input signal types [14].
In a polar transmitter, the amplifier input power is held constant and output
power variations are achieved by varying the supply voltage. As previously mentioned, in class-E the output voltage is nearly proportional to the supply voltage.
Figure 3.10(a) shows how the amplifier parameters (POUT , Gain, ηD , and PAE)
change as a function of supply voltage. Since, the RF input power is held constant
throughout this test, it leads to negative gain and PAE, and drain efficiency higher
35
100
80
14
70
12
60
10
50
PAE
8
40
6
30
Gain
4
20
2
10
1
1.2
4
1
2
3
4
0.8
3
0.6
2
0.4
1
0.2
drain bias
phase control
0
0
5
−1
0
Supply Voltage [V]
20
40
60
80
Phase control signal [V]
Pout
16
0
0
5
90
ηD
Supply Voltage [V]
18
Efficiency [%]
Output power [dBm] and Gain [dB]
20
0
−0.2
100
Time [µs]
(a)
(b)
20
Output spectrum [dBm]
10
−23.6dBc
−24.4dBc
0
−10
−20
−30
−40
−50
−1
−0.5
0
0.5
1
Offset frequency [MHz]
(c)
Figure 3.10: (a) Supply sweep for Class-E PA operated at 10-GHz, (b) envelope
and phase control time domain signals for polar two-tone test, and (c) polar twotone output frequency spectrum for 200 kHz offset between tones for a carrier
frequency of 10 GHz.
than unity at low bias voltage. A maximal PAE of 56% is obtained for a supply
voltage of 3.8 V. The corresponding ηD and POUT are 68.5% and +18.5 dBm respectively. The gain under these conditions is 7.46 dB. As bias voltage varies from
0 V to 5 V output power varies from +1.4 dBm to +18.5 dBm.
Polar system linearity measurement requires the two-tone signal to be split
into amplitude and phase components, and fed to the envelope amplifier and
phase modulator. Figure 3.10(b) shows the time domain envelope and phase
control signals. For a polar two-tone test the envelope corresponds to a rectified
36
Pout [dBm]
IMD3 [dBc]
j50
j100
0
10
−30
−28
−26
−2 7
3 1
16
−
25 21 50
−2
3
16
15
16
15
j10
6
−2
17
j25
100
∞
15
−21
−j10
−j25
−j100
−j50
Figure 3.11: MESFET POUT and worst IMD3 polar load-pull contours. The classE impedance is indicated with an ‘×’, for this impedance, the ηD of the entire
polar transmitter is 29.5% with POUT of +17.5 dBm. The optimal ηD of 30.5%
is obtained for 31.3 + j22.7 Ω, ‘o’ with +17.1 dBm of POUT . Worst IMD3 and
maximum POUT contours do not coincide in this case.
sinewave and the phase control signal is a squarewave. The output spectrum
for a polar two-tone test with tone offset of 200 kHz is shown in Figure 3.10(c).
Upper and lower IMD measurements were -24.4 dBc and -23.6 dBc, significantly
improved over the standard two-tone conditions class-E PA.
A load-pull was performed under polar two-tone excitation. POUT and worst
IMD3 contours are shown in Fig. 3.11. Linearity improvements observed in the
output frequency spectrum of Figure 3.10(c) are consistent with IMD measurements. There is a dramatic difference in IMD3 impedance as compared to the
PA tested outside of the polar transmitter under standard two-tone conditions
from Figure 3.9(b). In this case, the optimum IMD region does not coincide with
maximum POUT . Therefore under polar modulation a different amplifier matching
circuit should be designed for optimal linearity.
The POUT and ηD for the entire polar transmitter are +17.5 dBm and 29.5%
37
−20
2
−30
1.5
−40
1
0.5
1
2
3
Supply Voltage [V]
output phase −50
output voltage
−60
4
5
0.12
0.1
3
0.08
2
0.06
0.04
1
0.02
drain bias
0
Phase control sigal [V]
2.5
0.14
4
Supply Voltage [V]
−10
Relative output phase [degree]
Output voltage [V]
3
0
0
5
0
3.5
0
phase control
−1
0
20
40
60
80
−0.02
100
Time [µs]
(a)
(b)
20
Output spectrum [dBm]
10
0
−29.8dBc
−28.8dBc
−10
−20
−30
−40
−50
−1
−0.5
0
0.5
1
Offset frequency [MHz]
(c)
Figure 3.12: (a) Measured static AM-to-AM and AM-to-PM distortion for Class-E
PA, (b) predistorted envelope and phase control time domain signals for polar twotone test, and (c) polar two-tone output frequency spectrum with predistortion
for 200 kHz offset between tones for a carrier frequency of 10 GHz [12].
respectively. Significant reduction in ηD was suffer because the spectral content
of the two-tone envelope. The 200 kHz tones are too high in frequency for the
high-efficiency DC-DC switching converter to track, limiting its contribution to
only the DC component. Therefore, significant portion of the total envelope power
is generated by the linear amplifier. Additional discussion is given in subsection
3.4.4. The maximal ηD of the entire polar loop is 30.5% with POUT of +17.1 dBm.
The real and imaginary parts of the best-efficiency load impedance of 31.2+j22.7Ω
differ by around 30% from the single-tone load pull value.
38
3.4.3
Distortion and Predistortion
Figure 3.12(a) shows measured AM-to-AM and AM-to-PM distortion for polar
class-E PA. The AM-to-AM distortion describes how the output voltage to a 50 Ω
load changes as the supply voltage is varied. The static measurements show that
there is significant phase distortion for voltages below 1 V. Other significant nonidealities are the feedthrough and the roll-off at high voltages.
Baseband predistortion was used to compensate for some of the polar system nonlinearities by distorting amplitude and phase information in a manner
complimentary to the static AM-to-AM and AM-to-PM characteristics. Baseband distortion was accomplished using a lookup table (LUT) implemented in
the FPGA. Figure 3.12(b) shows the predistorted baseband amplitude and phase
control signals.
Figure 3.12(c) shows the obtained output spectrum for the predistorted polar two-tone test with 200 kHz spacing between tones. With predistortion the
lower and upper IMD3 levels are reduced to -28.8 dBc and -29.8 dBc, for a 4.4 dB
and 6.2 dB improvement, respectively. Table 3.1 shows the results for a twotone test with tone spacing of 20 kHz, 200 kHz, 625 kHz and 1MHz offset between
tones. It is interesting the dramatic improvement in two tone performance for
20 kHz case, reducing lower and upper IMD3 levels to -33.2 dBc and -32.9 dBc
respectively. However, as the separation between tones increases effectiveness of
the predistortion is degraded, suggesting other types of non-linearities are taking
Table 3.1: IMD Level for Polar Two-Tone Test with and without Predistortion
∆f
kHz
IMD3L
dBc
IMD3U
dBc
IMD3Lpred
dBc
IMD3U pred
dBc
20
200
625
1000
-22.5
-24.4
-24.2
-22.2
-22.4
-23.6
-25.4
-25.5
-33.2
-28.8
-27.1
-27.6
-32.9
-29.8
-25.8
-23.2
39
place, such as memory effects. A more complex dynamic PA model which considers these effects is required to predistort wideband signals. Future work in this
area includes the implementation of adaptive predistortion, predistortion of memory effects, and the elimination of feedthrough using RF drive power modulation.
3.4.4
Discussion
A 10-GHz high-efficiency class-E PA was characterized and tested in polar modulation under two-tone conditions. Linearity in polar modulation is significantly
improved when compared to the linearity tested under standard two-tone conditions. Load-pull measurements suggest a region where linearity can be farther
enhanced. Therefore, under polar modulation a different matching circuit should
be designed for optimal linearity and efficiency.
The measured, not optimized, ηD for the entire polar class-E transmitter in
two-tone test with 200 kHz offset between tones is 29.5%. In [14], it is shown that
for an envelope signal with a known amplitude density distribution, an optimum
band-separation frequency fB can be found. Spectrally rich signals such as EDGE
can benefit from the band-separation optimization technique because significant
amount of power is concentrated at low frequency close to DC. For the case of
EDGE signals, optimal band separation (0.88 ratio of envelope sent to the DCDC converter and 0.12 to the linear amplifier) estimates efficiencies in the order of
72% for the linear-assisted switching converter and 50% for the entire transmitter.
3.5
Conclusion
It is shown in this chapter that polar modulation can be used to linearized highefficiency power amplifiers. However, the overall transmitter efficiency depends
on the efficiency, not only of the RFPA but the efficiency of the envelope tracker
40
as well. SMPSs can deliver ultra high-efficiencies, are small, light, economic and
have the capability to step-up or step-down voltages. However, for these converters losses increase with switching frequency and it is a challenge to design fast
high-efficiency SMPSs. With signal envelop bandwidths in the MHz region linear
amplifiers can be used to assist SMPSs to achieve larger bandwidths. In envelope
signals, most of the power is close to DC and even if linear amplifiers are not
efficient, they need to deal with a fraction of the total power maintaining a high
overall efficiency.
The measured data on the presented example of a 10-GHz polar transmitter
shows that for high-efficiency class-E power amplifiers linearity in polar form is
significantly improved when compared to the linearity under standard conditions.
Under polar modulation, linearity is not only enhanced, but there is an output
matching network were transmitters linearity is optimized.
Simple power managements techniques such as shutting down the PA when
it is not needed can significantly improve the transmitter efficiency. However,
power can be managed more efficiently by employing flexible or adaptive DC
bias approaches such as providing a power supply for the PA that is capable to
continually adapt the supplied voltage to achieve required output power variations
[14].
The main objective of the work presented in this chapter is to develop a
digitally-adaptive amplifier that can operate at high efficiency and linearity for
different signal types. A portion of this work was done by Dr. Narisi Wang [35]
& Dr. Vahid Yousefzadeh [43]. The specific contributions contained in this thesis
are as follows:
• The repeatability of the 10-GHz amplifier characterization, and load pull
measurements with single and two-tone (Cartesian and polar) excitation
was verified;
41
• Proper design of bias circuitry for RFPAs in polar transmitters was investigated.
42
Chapter 4
A High-Efficiency Linear Polar
Transmitter for EDGE
In the previous chapter, it was shown that high-efficiency power amplifiers can
be linearized to some degree with the use of dynamic biasing techniques such
as polar modulation. This chapter discusses how dynamic bias can be applied
to the design of a high-efficiency linear polar transmitter for EDGE modulated
signals [38] used in portable radios. Efficiency is especially relevant in portable
equipment handsets that depend on battery power.
4.1
EDGE Modulation Scheme
EDGE stands for E nhanced Data rates for the GSM E volution. It is considered
to be a 2.5 G or 2.5 Generation modulation scheme and it is the most important
transition from 2 G to 3 G for the GSM (Global S ystem for M obile communications) standard. Changing the modulation scheme from GMSK (GSM) to 8-PSK
improves data rates from 270.833 kbps to 384 kbps over the same 200 kHz spectral
bandwidth at one of the 3 career frequencies (900 MHz, 1800 MHz, and 1900 MHz).
The EDGE spectrum is shown in Figure 4.1(a). The dashed-line corresponds to
EDGE Spectrum
10
0
−10
Power Spectrum [dB]
−20
−30
−40
−50
−60
−70
−80
−90
−100
−1000
−500
0
Frequency [kHz]
500
1000
(a)
1
0.8
0.6
0.4
0.2
0
−0.2
−0.4
−0.6
−0.8
−1
−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
0.8
1
(b)
Figure 4.1: (a) EDGE frequency spectrum (solid-line) with the EDGE spectral
mask (dashed-line). (b) IQ constellation of the EDGE modulation scheme; 8-PSK
with 3π/8 phase rotations and filtering. The phase rotations prevent the signal
from passing through the origin.
44
EDGE Envelope Time−Domain Signal
0.9
0.9
0.8
0.08
0.8
0.07
0.7
0.06
0.6
0.05
0.5
0.04
0.4
0.03
0.3
0.02
0.2
0.01
0.1
Signal [V]
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
0.05
0.1
Time [ms]
0.15
0.2
(a)
0
0
1
0.2
0.4
0.6
Normalized Output Voltage
0.8
Cummulative Distribution Function
0.1
0.09
Probability Distribution Function
1
0
1
(b)
Figure 4.2: (a) 0.2 ms long sample of the normalized EDGE envelope. The envelope signal varies with time, but it does not have zero values. (b) Probability
distribution function (solid-line) and cumulative distribution function (dashedline) for the EDGE envelope.
the EDGE spectral mask. It is desired for the EDGE signal to remain within
the spectral mask limits to avoid adjacent channel interference. Distortions and
unbalances throughout the system are usually displayed as spectral regrowth that
commonly make the EDGE signal fall outside the spectral mask limits.
The EDGE constellation is shown in Figure 4.1(b). The 3π/8 phase rotations
prevent the signal from cross through the origin. Allowing envelope variations in
the EDGE modulation scheme achieves higher data rates within the same GSM
bandwidth. Figure 4.2(a) shows a normalized 0.2 ms long signal envelope sequence. The sequence does not cross through the origin, but varies with time.
These envelope variations corresponds to a peak-to-average ratio (PAR) of 3.2 dB
and they introduce linearity requirements. Figure 4.2(b) shows the calculated
EDGE envelope statistics (probability distribution function (PDF) and cumulative distribution function (CDF). From the PDF the mean normalized value of
the signal envelope can be calculated to be 0.65. The signal statistics can be used
to optimize transmitter performance in terms of power conversion efficiency.
45
4.2
Polar EDGE
There are several characteristics that make EDGE suitable for polar modulation.
For example, the 3π/8 phase rotations prevent the signal envelope from passing
through the origin. Also, its 3.2 dB PAR is low compared to 10 dB PA in WCDMA
signals. Due to these reasons there has been a lot of effort into developing polar
transmitters for EDGE. Companies such as Skyworks, TriQuint, Analog Devices,
and Anadigics are starting to release their versions of EDGE polar transmitters
giving higher emphasis towards linearity than efficiency [50] - [53]. In [50] is
quoted the best number for overall transmitter efficiency in the order of 35% with
+27 dBm of output power. This efficiency number is better than the one obtained
with a linear amplifier which is in the order of 22% for the same output power
level. Significant work has been done by the University of California, San Diego
with best numbers for overall transmitter efficiency in the order of 43% with POUT
= +20 dBm [54] - [56]. In these systems the envelope tracker is implemented with
a linear-assisted switch-mode power supply.
As described in the previous chapter, in polar transmitters the complex baseband signal is separated into two components; the envelope and a normalized
phase modulated signal. Figure 4.3 shows the baseband spectrum of these two
components corresponding to EDGE modulation. It is important to notice the
increase in spectral bandwidth of the amplitude and phase components compared
to the original EDGE spectrum, this implies that the circuits in these paths need
to support a bandwidth several times larger than the original signal. In the implemented system, the envelope tracker is implemented with solely a SMPS with a
switching frequency of 4.33 MHz and it is followed with a 1.3 MHz Bessel lowpass
filter. Details of the tracker are discussed in Section 4.3. The RFPA is a non-linear
high-efficiency transmission line hybrid class-E PA with details in Section 4.4.
46
EDGE Normalized Phase Spectrum
40
20
20
0
0
Power Spectrum [dB]
Power Spectrum [dB]
EDGE Envelope Spectrum
40
−20
−40
−20
−40
−60
−60
−80
−80
−100
−1000
−500
0
Frequency [kHz]
500
−100
−1000
1000
−500
(a)
0
Frequency [kHz]
500
1000
(b)
Figure 4.3: (a) Envelope and (b) phase baseband spectra of an EDGE modulated
signal compared to the total signal spectral mask (dashed-line). In polar transmitters the complex signal is divided into an envelope signal, and a normalized
phase modulated signal, which are fed through the bias and input, respectively.
It is important to consider the separate spectra of the envelope and phase when
designing the polar transmitter circuits.
EDGE Envelope Spectrum (Filtered)
40
EDGE Spectrum (Filtered Envelope)
fcutoff = 600kHz
20
cutoff
0
f
cutoff
Power Spectrum [dB]
original
−40
original
−40
−60
−80
−80
800
1000
(a)
= 1200kHz
fcutoff = 1500kHz
−20
−60
400
600
Frequency [kHz]
fcutoff = 900kHz
0
fcutoff = 1500kHz
200
fcutoff = 600kHz
= 1200kHz
−20
−100
0
fcutoff = 300kHz
20
fcutoff = 900kHz
f
Power Spectrum [dB]
40
fcutoff = 300kHz
−100
0
200
400
600
Frequency [kHz]
800
1000
(b)
Figure 4.4: (a) Envelope signal is lowpass filtered with different cutoff frequencies.
Band-limiting the envelope spectrum results in spectral regrowth (b). From here
we conclude that the cutoff frequency of the filter used in the envelope path should
be larger 1200 kHz.
47
4.2.1
Bandwidth Limited Envelope
The design of efficient envelope trackers with large bandwidths is a challenge.
Figure 4.4 shows the calculated effects of limiting the envelope bandwidth (Figure
4.4(a)) in the combined EDGE spectrum (Figure 4.4(b)). The envelope spectrum
is lowpass filtered with cutoff frequencies of 300 kHz, 600 kHz, 900 kHz, 1200 kHz,
and 1500 kHz. Figure 4.4(a) shows the envelope spectra for each of these cases
compared to the original spectrum. As can be seen in Figure 4.4(b), the main effect
of limiting the envelope spectral bandwidth is spectral regrowth in the combined
signal. This same figure shows that to meet the spectral mask, the cutoff frequency
of the filter should be higher than 1200 kHz. This figure illustrates the consequence
of band limiting the envelope signal, however it is important to point out that
the spectral response depends not only on the filter cutoff frequency, but also on
additional parameters such as the order and the type of filter used.
4.2.2
EDGE Polar Transmitter
Figure 4.5 shows a block diagram of the EDGE polar transmitter. The transmitter has 3 baseband input signals and the RF signal that determines the carrier
frequency. The 3 baseband input signals correspond to the envelope of the EDGE
ˆ and Q̂(t) signals which are quadrature
sequence IQA (t), and the normalized I(t)
phase parameters. Linearity is achieved when the RFPA performs the function
of multiplying the amplitude and normalized phase signal in time domain, thus
reconstructing the original signal. Time alignment between envelope and phase
signals is therefore crucial for output signal fidelity.
To make the transmitter efficient, a high-efficiency envelope tracker and a highefficiency RFPA are needed. As was discussed in the previous chapter, SMPS are
suitable for implementing high-efficiency envelope trackers, while Class-E ampli-
48
Signal Envelope
Envelope
Tracker
|I(t)|2+|Q(t)|2
IQA(t)
~A(t)
IQ
IQ Modulator
I (t )
A(t )
Mixer
Normalized
ˆI andQˆ
Q(t )
A(t )
100
fc 1000
ˆI(t)
IQA(t)
Q(t)
ˆ
fc 500
fc
fc+500
fc+1000
RF Power
Combiner
+
90º
RF
Oscillator
IQ
RFPA
RF
IQRF(t)
High-Efficiency
RFPA
Mixer
Figure 4.5: Block diagram of the EDGE polar transmitter. Three baseband signals
ˆ and normalized Q̂(t). The
are generated; the envelope IQA (t), normalized I(t)
RF output is the reconstructed signal IQRF (t).
fiers are ideal for the RFPA because the output voltage across a constant load
is theoretically linearly proportional to the supply voltage. The next section discusses the designed envelope tracker and the RFPA.
4.3
High-Efficiency Envelope Tracker
The envelope tracker used in this work is implemented with a synchronous Buck
SMPS (Figure 4.6) and the envelope variations are achieved with pulse width modulation. Deadtime (time when both transistors in Figure 4.6 are ON) control of
the synchronous switches in SMPS is essential to enhance the tracker efficiency by
eliminating the short circuit current and it is implemented with a digital counter49
Switcher IC
Vout
L
AM
C
PM RFin
Vg
PA
RFout
Deadtime
Control
LPF
Clk_in
Ts
+
PWM
d
Compensator
Gc (s)
Vref
EDGE
Envelope
signal
-
Figure 4.6: Schematic of the closed-loop envelope tracker. The tracker is implemented with a synchronous Buck switching converter. Envelope variations are
achieved with pulse width modulation and deadtime control is used to enhance
the tracker efficiency [16].
based adaptive scheme. Control is also required to produce a well regulated output
voltage in the presence of variations in the input voltage, the load and/or element
tolerances. The feedback loop around the power stage also achieves disturbance
rejection and sensitivity reduction. The feedback loop also improves the tracking performance when compared to open loop operation. To minimize switching
harmonics, the envelope tracker is followed by a 4th order low pass Bessel filter
with a cutoff frequency of 1300 kHz that keeps the 2nd switching harmonic below
-65 dBc, with a constant group delay of 259 ns [16]. This bandwidth is larger than
the 1200 kHz found in the previous section.
Two factors were considered when choosing the switching frequency: signal
bandwidth and switcher efficiency. The bandwidth of EDGE signals is 200 kHz,
but the spectral requirements increase when the signal is converted to polar form
as discussed in the previous section. The switching frequency was chosen to be
16 times higher than the EDGE standard transmission rate of 270.833 kHz. This
corresponds to a switching frequency of 4.33 MHz. The tracker supply voltage is
3.6 V, which corresponds to the nominal voltage of a single-cell Li-Ion battery. The
maximum output voltage of the tracker is 3.3 V. The switcher measured efficiency
50
is 79% [45]. For additional information about the switch-mode envelope tracker,
the reader is referred to [16], [45].
In section 4.6, the performance of the transmitter with the switching supply
from Figure 4.6 is compared to that with a linear amplifier envelope tracker such
as the one presented in [17]. Although the power efficiency of a linear amplifier
is not suitable, it can provide higher bandwidths and better dynamics than other
analogous tracking systems. The linear amplifier is followed by a 4th order lowpass
Bessel filter with a cutoff frequency of 2.5 MHz and constant group delay of 140 ns.
4.4
High-Efficiency RFPAs
The transmitter efficiency depends not only on the efficiency of the envelope
tracker, but on the efficiency of the RFPA as well. The steps to follow for the
design of a high-efficiency RF power amplifier at a specified operating frequency
and output power are:
(1) Choose mode of operation, suitable for the operating frequency. As was
discussed in Chapter 2, there are several high-efficiency modes of operation,
among them Class-E. The main advantage of the Class-E mode is that the
output RF voltage to a constant load is linearly proportional to the supply,
³
´
p
VOUT = ± 26 · fS · COUT RE · RL · VDS
(4.1)
were RE is the real part of the optimal impedance presented to the device
for class-E switched mode operation, fS is the operating frequency which
is also the switching frequency, and COUT is the output capacitance of the
device.
(2) Select the transistor. The main parameters of interest for class-E operation
are the device internal output capacitance, the break down voltage and the
51
maximum output power capability of the device.
(3) In class-E the transistor is biased at cutoff and the RF input signal turns
ON and OFF the device. It is important to determine the voltage range in
which the transistor is going to operate. In class-E mode the voltage across
the device might be higher than 3.56×VDS . This mean that the maximum
supply voltage needs to be limited to a maximum of VDSS /3.56, where VDSS is
the breakdown voltage of the device. Reducing the maximum supply voltage
will limit the maximum RF output power of the device.
(4) Design input and output matching networks. For Class-E operation the
harmonics should be terminated with a high impedance, in the transmission
line version this is accomplished with stubs. The fundamental impedance
should be terminated in a particular impedance that ensures soft switching.
This impedance can be theoretically from the transistor output capacitance
and the operating frequency [23],
ZE =
0.28
◦
ej49
2πfS COUT
(4.2)
The input needs to be matched to maximize gain, otherwise the ηS will be
much larger than PAE.
4.4.1
Transistor Selection
Class-E is a switched mode of operation. Switching at RF frequencies is approximated by choosing a device with a low output capacitance relative to the operating
frequency and a high fT . The EDGE standard has 3 operating bands; 900 MHz,
1800 MHz, and 1900 MHz. For the 900 MHz band mobile transmitters operate
around 880 MHz with output power of approximately +30 dBm. The transistor
selected is the TGF-4240 from TriQuint. It is 2.4-mm HFET with and output
52
Input HFET Transistor
Matching
& Baising
Output
Matching &
Lowpass
Bias-Tee
2nd Harmonic
Stub
Figure 4.7: Photograph of the transmission line hybrid 880-MHz class-E power
amplifier with integrated bias lines and DC blocking capacitors which also serve
for impedance matching. A 10-Ω chip resistor in series with the gate terminal
ensures stability of the PA.
capacitance of approximately 1 pF. The transistor is intended for operation at
8 GHz with over +30 dBm of output power and 10 dB of gain when operated from
a 8 V supply [57]. Thus, at 880-MHz, it should be able to operate in Class-E.
4.4.2
RF Amplifier Design
This section details the design of the high-efficiency class-E PA for the polar
EDGE transmitter, Figure 4.7. The design method combines transmission lines
and lumped components for the implementation of the matching networks in order
to keep loss low, but also reduce size. Transmission lines are used at the output to
present a high impedance to the transistor second harmonic. The fundamentalfrequency input and output matching circuits are implemented with lumped inductors and capacitors. The amplifier is build on a Rogers TMM6 0.635-mm thick
substrate (²r = 6). A 10-Ω series resistor is included in the gate of the transistor
for stabilizing the PA at the operating frequency.
Using Equation 4.2 where the output capacitance for this particular device
is estimated to be 1 pF based on measured small-signal S-parameters. Because
the exact large-signal value of the output capacitance is not known, a source and
load-pull measurement is performed around the initial estimate of ZE to determine
53
Table 4.1: Load-pull for HFET Transistor at 880-MHz.
COUT
(pF)
0.5
0.6
0.7
0.8
0.9
1.0
1.1
66
55
47
42
37
33
30
ZE
(Ω)
ηD
(%)
+
+
+
+
+
+
+
49
58
55
67
62
62
62
j77
j64
j55
j48
j43
j38
j35
PAE Gain POUT
(%) (dB) (dBm)
41
48
47
58
55
56
56
8
8.2
8.7
9.7
9.6
9.8
10
+18
+18.2
+18.7
+19.7
+19.6
+19.8
+20
the optimum impedance. The only unknown in Equation 4.2 is COUT . Because
there is uncertainty in the exact value of the output capacitance a load-pull was
performed presenting to the device the calculated Class-E impedance from COUT
in the range between 0.5 pF and 1.1 pF as shown in Table 4.1. Measurements are
taken for VGS = -2 V, VDS = 3.5 V and +10 dBm of input power. A tradeoff analysis
needs to be made in terms of gain, output power and efficiency for choosing the
desired impedances. Since the main goal of the project is to improve linearity
and efficiency, the output impedance of 42 + j48 Ω is chosen, corresponding to an
output capacitance of 0.8 pF. The resulting PAE is 58% with a drain efficiency of
67%, a gain of 9.7 dB and an output power of +19.7 dBm.
The amplifier final design is biased at VDS = 2.16 V taking into account the
EDGE envelope mean of 0.65 and the maximum supply voltage from the SMPS
of 3.3 V (0.65×3.3 V) and VGS = -1.7 V, for IDS = 72 mA. It was found that
varying VGS from -2 V to -1.7 V increased the gain and the output power with no
degradation in efficiency. The input is matched to 15 + j35Ω for an input return
loss of -10 dB. Figure 4.8(a) shows the measured power sweep, which resulted in
a choice for the input power of +7.6 dBm. If the transistor is compressed further,
the drain efficiency will continue to increase but at the expense of gain and PAE.
For this particular bias point, with +7.6 dBm of input power, the output power
54
100
25
100
PAE
20
80
Pout
70
15
60
50
10
40
30
5
20
Output power [dBm] and Gain [dB]
ηd
ηd
90
Efficiency [%]
Output power [dBm] and Gain [dB]
Gain
90
20
80
PAE
70
15
60
50
Pout
10
40
30
Gain
5
20
10
0
−10
−5
0
Input power [dBm]
10
0
10
5
Efficiency [%]
25
0
0.5
1
1.5
2
2.5
Supply Voltage [V]
0
(b)
4.5
60
4
40
3.5
20
3
0
∆ φ [deg]
Output Voltage [V]
(a)
3
2.5
−20
2
−40
1.5
−60
1
−80
0.5
0
0
−100
0.5
1
1.5
2
Supply Voltage [V]
2.5
3
−120
(c)
Figure 4.8: Measurements for EDGE 880-MHz class-E PA with VGS = −1.7 V. (a)
Power sweep for VDS = 2.16 V, the maximal PAE is 70% with a drain efficiency of
75%, gain of 11.46 dB and output power of +19 dBm for input power of +7.6dBm.
(b) Measured output power, gain, drain efficiency and PAE for supply sweep with
input power of +7.6 dBm. (c) Measured AM-to-AM (solid line) and AM-to-PM
conversion (dashed line).
is +19 dBm with a gain of 11.5 dB, drain efficiency of 75% and PAE of 70%.
Careful attention needs to be given to drain line design to avoid distortion and
linear memory effects [58]. The drain line lowpass performance has to be designed
to allow for supply modulation at the signal envelope bandwidth. The measured
performance of the PA is shown in Figure 4.8: at 880 MHz, the PA gives +19 dBm
output power at 3 V supply voltage with 70% power added efficiency and 11.5 dB
gain.
55
4.4.3
RFPA Characterization
Figure 4.8(b) shows the results for the measured output power, gain and efficiencies for the class-E PA when VDS is swept from 0.5 V to 3.3 V with VGS = -1.7V
and an input power of +7.6dBm. As can be observed from the figure, ηD is high
throughout the entire supply voltage range, while the PAE is above 50% during
a substantial part of the sweep, ensuring high-efficiency performance for this particular applications. Considering the EDGE statistics the mean values ηD , PAE
and POUT are 76%, 65% and +19 dBm, respectively.
Figure 4.8(c) shows the AM-to-AM (solid line) and the AM-to-PM (dashed
line) conversion. The AM-to-AM measurements show the output voltage variations produced to a 50-Ω load by sweeping the supply voltage for fixed input
power. For a supply voltage of 0 V the output voltage is non-zero due to feedthrough. A linear relationship is observed in the range of interest (0.5 V to 3.3 V)
and can be described by
Vout = 1.2VDS + 0.091V.
(4.3)
The AM-to-PM conversion curve shows the phase offset produced by the amplifier as the supply is swept. As desired this value is almost constant throughout
the range of interest. The measurement shows significant variations in phase for
supply voltages below 0.5 V.
A summary of the measured PA properties is as follows:
• When the drain bias is swept from 0.5 V to 3.3 V, the output power varies
from +7.01 dBm to +22.45 dBm giving an output power range of 15.44 dB.
• The drain efficiency, PAE and gain for maximal output power at VDS = 3.3 V
are 71.6%, 69% and 14.9 dB, respectively.
56
D/A Converters
I(t)
I2(t)° + ° Q2(t)
Q(t)
DAC
DAC
Board
t
I (t )
A(t )
DAC
Envelope
Tracker
IQA(t)
~A(t)
IQ
IQ Modulator
Mixer
I(t)
ˆ
RF Power
Combiner
+
Q(t )
A(t )
DAC
90º
Q(t)
ˆ
RF
Oscillator
IQ
RF
RFPA
IQRF
High-Efficiency
Class-E PA
Mixer
FPGA board
Digital Baseband FPGA
Figure 4.9: Block diagram of a polar transmitter. A random string of symbols is
stored in the FPGA board as a look up table and transmitted at 2.17 MHz which
corresponds to an over-sampling ratio of 8. The FPGA has three outputs; the amplitude of the complex signal, IQA (t) which is the envelope tracker reference and
ˆ and Q̂(t) that are inputs to an are the input to an IQ modulator;
normalized I(t)
the output is a constant amplitude phase modulated RF signal. Digital-to-analog
converters are used at each output.Time alignment between amplitude and modulated phase path is crucial since both values determine each symbol.
• The maximal PAE is 70.2% obtained for a drain voltage of 2.6 V. At this
bias point, the output power is +20.6 dBm, with a gain of 13 dB and a ηD
of 74%.
• At a drain voltage of 0 V the output power is -4 dBm due to feed-through.
• Due to feed-through, drain efficiencies higher than 100% can be achieved
since the input power is not considered in drain efficiency calculations.
• At high drain voltages, the drain efficiency and PAE exceed 70%. Signals
with statistics that are weighted towards higher output powers are the ones
that benefit from this technique.
57
4.5
EDGE Polar Transmitter Implementation
The diagram shown in Figure 4.9 is implemented with a Xilinx Virtex II FPGA
and commercial oscillator, IQ modulator and DACs as shown in Figure 4.10.
The oscillator is an Analog Devices ADF4360-7 and the IQ modulator is Texas
Instruments TRF3701. For testing purposes an EDGE signal segment consisting
of 256 pseudo random symbols is stored as a look up table in the FPGA and
streamed out repeatedly. Given the standard EDGE transmission rate of 270.833
kHz and an oversample ratio of 8, the signals are streamed out of the FPGA at
the rate of 2.17 MHz. The FPGA and baseband circuits were programmed and
design by Dr. Xufeng Jiang.
Time synchronization between envelope and phase signals is crucial to minimize distortion, as discussed in [12]. For time alignment, it was found to be
sufficient to compensate for the group delay τ produced by the lowpass Bessel filter. It was also found to be important to compensate for the DC offset introduced
by the DACs in the I and Q channels. The implementation shown in Figure 4.10
allows all the required adjustments to be performed digitally.
4.6
System Performance
The system described in section 4.5 was implemented as shown in Figure 4.10.
Figure 4.11 shows the measured output spectrum of the polar transmitter, with the
switcher (blue) and the linear amplifier (red) as the envelope trackers. Because
the linear amplifier bandwidth is larger than the SMPS it shows less spectral
regrowth, however, the transmitter with the 79% SMPS also meets the EDGE
spectral mask. It is important to emphasis that the high-efficiency RFPA was
linearized with dynamic biasing without any pre-distortion. Adding pre-distortion
to the system will farther enhance its linearity.
58
Figure 4.10: Photograph of the implemented setup. Baseband circuit design,
selection and FPGA programming was performed by Dr. Xufeng Jiang.
59
Figure 4.11: Measured output spectrum of the polar transmitter when the envelope tracker is the linear amplifier (red) and fast switching buck converter (blue).
The dashed line indicates the EDGE spectral mask requirement.
60
The total transmitter efficiency taking into account the 79% efficiency SMPS
and the 71% RFPA is 56% with an average RF output power of +20 dBm. The
RFPA is able to deliver more than +30 dBm of output power at 8 V drain voltage,
however it is desired to use a 3.6 V Li-Ion power battery as the DC power source
of the system. This power source in combination with a Buck converter limits the
maximum PA supply voltage to 3.3 V thus limiting the output power capabilities
of the RFPA.
4.7
Summary
In conclusion, it is possible to linearize high-efficiency switched mode RFPAs
to meet the EDGE spectral mask using a polar architecture, while maintaining
high overall transmitter power consumption efficiency. This work demonstrates
a 880-MHz 56% efficient polar transmitter for EDGE with +20 dBm of output
power. The envelope tracker is a 79% efficient switch-mode buck converter and
the RFPA is a 71% non-linear transmission line Class-E power amplifier. The
transmitter meets the EDGE mask with two different envelope trackers, a less
efficient linear amplifier with a 2.5 MHz bandwidth and a high-efficiency switchmode power supply with a 1.3 MHz bandwidth.
The main objective of the work presented in this chapter is to described the
details of the EDGE polar transmitter. The specific contributions contained in
this thesis are as follows:
• Design and implementation of a transmission line hybrid Class-E power
amplifier for the 880-MHz polar transmitter. The maximal PAE is 70%
with ηD of 75% with POUT = +20 dBm. The maximal POUT is +22.4 dBm
with little degradation in PAE.
• The amplifier is linearized with polar modulation without the need of pre61
distortion and meets the EDGE spectral mask.
• The total transmitter efficiency is 56% with POU T = +20 dBm for EDGE
signals. This is the highest efficiency reported for a EDGE polar transmitter to date. The closest work [56] is a parallel effort at the University of
California, San Diego with a 44% efficient polar transmitter with the same
output power.
62
Chapter 5
High-Efficiency UHF PA Design
based on Load-pull
5.1
Introduction
RF amplifier design consists of biasing and impedance matching. These parameters are selected so that the PA performs a desired function at the desire frequency
range, i.e. low-noise, maximum output power, linearity, or efficiency. Linear simulators are great tools for designing linear amplifiers. However, power amplifiers
are driven with large signals and particularly in switch-mode operation display
strong nonlinear behavior. The devices are driven into saturation and cutoff for
a certain portion of the RF cycle and modeling these strong nonlinear effects is a
difficult task, even if CAD models and tools are available [20]. Nonlinear device
models often fail to accurately represent devices operating in class-E mode where
the dynamic load line is close to the I-V axes.
Load-pull techniques can be modified for effective amplifier characterization
in this mode of operation as shown in [34]. The load-pull technique consists of
empirically presenting calibrated input and output impedances, RF excitation
and biasing conditions to the DUT and recording desired responses such as, input
power, output power, gain, and power consumption. As long as the responses of
interest can be measured, impedance contours that map the PA performance as
a function of these controllable parameters can be created.
As an example, this method is used for designing a high-efficiency power amplifier at 370-MHz with a GaN HEMT on a SiC substrate prototype from RFMD,
the RF3932. This transistor can deliver POUT = 60 W in Class-AB at 2 GHz and
it is rated for 48 V operation with 14 dB of gain.
The steps for characterizing and designing a high-efficiency power amplifier
are as follows:
(1) Select biasing conditions and characterize the transistor via load-pull;
(2) Design input and output matching networks and perform impedance verification;
(3) Repeat load pull to verify amplifier optimum operation; and
(4) Fine tune input and output matching networks until optimum PA performance is achieved.
PMETER
Input
Source
Variable
Attenuator
PMETER
Output
Input Block
3 dB
Attenuator
Pre Amp
Input
Tuner
Output
Tuner
High Power
Attenuator
Lowpass
Filter
Spectrum
Analyzer
DUT
Lowpass
Filter
Coupler
Output Block
Osc
PMETER
Reflect
ZIN
Z OUT
Figure 5.1: Setup for load-pull device characterization. The setup consists of a
source, input and output blocks, the device under test (DUT), and impedance
tuners.
64
5.2
Load-pull Setup and Calibration
Figure 5.1 shows a schematic of a load-pull setup. The setup consists of a source,
input and output blocks, impedance tuners, and the device under test [59]. Loadpull characterization depends on carefully calibrating the input and output blocks,
and the impedance tuners usually with a network analyzer. For device characterization the input and output blocks, as well as the tuners in the setup of Figure 5.1
are modeled with S-parameter matrices. Impedance tuners are calibrated creating
look-up tables that map tuner position to S-parameters [59].
In this case, S-parameters do not take into account power dependent variations, such as thermal drift. The circulators/isolators and the attenuators might
be sensitive to power or temperature. To reduce uncertainty it is important to use
robust components that do not vary with power. A description of each of these
elements is given next:
• Source - The source determines the operating frequency and the system
input power.
• Pre-Amplifier - Usually, the source’s output power is low compared to
input power needed for amplifier testing, i.e. measuring compression. A
linear pre-amplifier can be used to increase the available input power. Preamplifiers might generate harmonics and these should be lowpass filtered.
• Input Block - The input block consists of a coupler, a circulator/isolator
and a 3 dB attenuator. The circulator/isolator ensures the coupler is terminated in 50 Ω. Also redirects the input reflected power to a load. A circulator
or an isolator is used, depending on whether or not it is desired to measure
reflected power. The 3 dB attenuator reduce out-of-band reflections produced by the the circulator/isolator. Out-of-band reflections might present
impedances that drive the DUT into unstable regions causing oscillations.
65
• Output Block - The output block consists of a high-power attenuator, a
coupler and a lowpass filter. The attenuator reduces POUT to a suitable
level for RF instruments. For example, the maximum power handling of RF
power sensors is usually +20 dBm, while the power handling of spectrum
analyzers is usually +30 dBm. The attenuator is followed by a coupler to
sample the output power. This signal is lowpass filtered to farther reduce
harmonic levels. If it is desired to view the output spectral content, it should
be through the direct path, because the coupled path is frequency dependent
and will affect spectral measurements.
• Impedance Tuners - The impedance tuners transform a known impedance,
usually a 50 Ω load, into a range of impedances. Some of the tuner important parameters are the maximum reflection coefficient (|Γmax |), impedance
resolution and insertion loss. Impedance tuners can be manually or electrically controlled mechanical [59]- [61], based on MEMS, semiconductors [68]
or active [62]. Active tuners inject a signal in a controllable manner mimicking a load. For additional information on tuners the reader is referred
to [59] - [68].
5.2.1
Pre-matching Circuits
Impedance tuners can cover a set of points on a Smith Chart, referenced to an
impedance constellation as shown in Figure 5.2(a). The constellation is centered
around the tuner characteristic impedance, usually 50 Ω with points equally spaced
radially around this value. The tuners are able to present impedances within a
range in |Γ| (from 0 to |Γmax |) with a particular density (number of constant |Γ|
and ∆ phase).
Small impedances are needed to match high-power devices. In this case, pre66
j50
j50
j25
|max|
j10
0
j25
j100
10
−j10
25
50
j10
∞ 0
100
10
−j10
Region of
Interest
25
50
∞
100
Region of
Interest
−j25
−j100
−j25
j100
−j100
−j50
(b)
−j50
(a)
Figure 5.2: Typical constellation achieved by impedance tuners. (a) The points
enclose the system characteristic impedance (usually 50 Ω). (b) Small impedances
are necessary to match high-power devices. Pre-matching circuits can be used to
shift the constellation to the region of interest, in this example 5 Ω.
matching circuits can be used to shift the constellation to a new impedance range
as shown in Figure 5.2(b). In this example, the constellation is centered around
5 Ω.
For class-E mode characterization the pre-matching circuit not only re-centralizes the impedance constellation closer to the region of interest, but also terminates
harmonics in an open circuit. For ideal Class-E operation, it is necessary to
terminate all the transistor harmonics. However, reasonable performance can be
achieved by terminating only the 2nd harmonic. Terminating additional harmonics
adds complexity and insertion loss for little enhancement in efficiency. Harmonic
stubs are resonating structures and it is difficult to present the correct amplitude
and phase for more than two harmonics simultaneously. Harmonic stubs also
affect the fundamental frequency making multiple harmonic termination circuit
design complex.
67
(a)
(b)
(c)
(d)
Figure 5.3: TRL fixture for transistor characterization in class-E mode. The input
and output blocks transform the load-pull contours to a low impedance suitable
for characterization. The output block includes an open stub for terminating the
second harmonic with a high impedance. (a) TRL thru connection, (b) TRL
reflect connection implemented with a short circuit, and (c) TRL line, and (d)
input and output block fixtures with the DUT.
5.2.2
DUT Fixture Calibration
A modular mechanical fixture was designed to allow for pre-match circuit characterization and impedance verification. The fixture is shown in Figure 5.3. Sparameters of the pre-matching circuits are found using a TRL calibration algorithm requiring that the input and output pre-matches be:
• Connected together directly at the DUT plane (Figure 5.3(a));
• Terminated with high reflection (implemented in Figure 5.3(b) with a short
circuit); and
• Connected together via a short length of transmission line with a known
characteristic impedance (Figure 5.3(c)).
68
The modular nature of the mechanical fixture allows these configurations to be
used easily and repeatably without fabricating additional calibration standards.
Copper tape or solder is used to make microstrip connections between blocks. The
modular fixture allows the use of passive lumped elements such as inductors and
capacitors. Using the same fixture for calibration and testing eliminates element
tolerance variations, reducing uncertainties.
5.2.3
Block Deembeding & Impedance Verification
The TRL algorithm returns two S-parameter matrices corresponding to the fixture
input (SA ) and output blocks (SB ):


A
s11
SA = 
sA
21
sA
12 
sA
22

(5.1)
and


B
s11
SB = 
sB
12 
B
sB
21 s22

(5.2)
These matrices can be used to calculate parameters such as pre-matching
impedances and insertion loss. For example, sA
22 corresponds to the central reflection coefficient presented to the DUT input port, while sB
11 corresponds to the
central reflection coefficient presented to the DUT output port. The insertion loss
(IL) quantifies how much power is dissipated in the matching block and is given
by
ILdB = 10 log10
|s21 |2
1 − |s11 |2
(5.3)
The S-parameters obtained from the TRL algorithm can also be used to perform impedance verification. Once the optimum impedances have been identified
69
from load-pull measurements we would like to synthesize matching networks that
present the targeted input (sA-Amp
) and output (sB-Amp
) reflection coefficients to
22
11
the DUT, where SA-Amp and SB-Amp correspond to the S-parameter matrices of
the input and output synthesized matching networks. For example, to perform
impedance verification of the synthesized output matching network SB-Amp , the
block can be connected directly following the calibrated block SA . The total
S-parameters of these two cascaded blocks SM can be measured using a network
analyzer. Cascade matrices such as ABCD can be used to deembed the unknown
output block SB-Amp from the measured SM . The ABCD matrix can be obtain
from the S matrix by [69]


A B 
ABCD = 

C D

(5.4)

(1+s11 )(1−s22 )+s12 s21
2s21

=
22 )−s12 s21
Z0 (1−s11 )(1−s
2s21
1 (1+s11 )(1+s22 )−s12 s21

Z0
2s21
(1−s11 )(1+s22 )+s12 s21
2s21

(5.5)
where Z0 is the system impedance, usually 50-Ω. After transforming to ABCD
matrices we obtain:
ABCDM = ABCDA · ABCDB-Amp
(5.6)
Then the ABCD parameters of the unknown block ABCDB-Amp can be easily
obtained from,
ABCDB-Amp = [ABCDA ]−1 · ABCDM
ABCD matrices can be converted back to S with [69]
70
(5.7)


s11 s12 
S =

s21 s22

(5.8)

A+B/Z0 −CZ0 −D
 A+B/Z0 +CZ0 +D
=
2
A+B/Z0 +CZ0 +D
2(AD−BC)
A+B/Z0 +CZ0 +D
−A+B/Z0 −CZ0 +D
A+B/Z0 +CZ0 +D


(5.9)
From the S-parameters the insertion loss of the output block can be calculated,
as well as the output impedance presented to the device that is given by
ZOUT = Z0
1 + sB
11
1 − sB
11
(5.10)
The same procedure can used with the synthesized input block by connecting
the unknown block SA-Amp followed by the calibrated block SB and measuring
the combined S-parameters. After following a procedure similar to the one just
described the input impedance presented to the device is calculated to be,
ZIN
1 + sA
22
= Z0
1 − sA
22
(5.11)
This procedure can be iterated till the synthesized input and output matching
networks match the identified optimum impedances.
5.2.4
Transistor Characterization
Transistor characterization begins with low supply voltage to avoid device damage
due to voltage breakdown, power density, and instability. One convenient and
unique property of class-E is that the optimal efficiency impedance does not change
with changing supply voltage. High-efficiency impedance regions can be more
safely identified at low voltage, and will not change considerably with increasing
supply level.
71
Figure 5.4: Photograph of load-pull setup at dBm Engineering/Peak Devices facilities in Boulder, CO.
The input impedance is found using a low RF input power source pull. The
objective of this test is to find a region of input impedances where the device is
stable and has large gain. If the device remains unstable over the useful impedance
region the input pre–match can be adjusted to stabilize the device. Parallel or
series components such as capacitors or resistors can serve to suppress oscillations.
Since the load-pull technique relies on careful calibration, it is prone to errors
in the absolute sense. The data should be used to determine optimum impedance
regions and to observe trends in performance. A PA prototype design is required
to determine the absolute performance of characterized devices.
5.3
Example with a RF3932 Prototype
The procedure described above was used for the characterization and design of a
high-efficiency UHF power amplifier with a GaN HEMT on a SiC substrate. The
transistor is a RF3932 prototype from RFMD capable of delivering 60 W of output
power at 2 GHz in Class-AB mode. The transistor was characterized via Focus
electro-mechanical tuners at the facilities of dBm Engineering/Peak Devices in
Boulder, CO. A photograph of the setup is shown in Figure 5.4.
72
427
4
j4
j40
84
*
j20
POUT [W]
D [%]
j10
ZE []
89
84
4742
0
4
10
20
∞
40
Figure 5.5: POUT (W) and ηD (%) load-pull contours for the RF3932 (GaN HEMT
on a SiC substrate) biased at 28 V. Optimum POUT and ηD regions do not overlap. The figure also shows the optimal Class-E impedance (9 + j10 Ω) estimated
from the device 9 pF output capacitance marked with ‘*’. The Smith charts is
normalized to 20 Ω.
j10
j20
28V
36V
48V
j40
j10
j20
28V
36V
48V
j40
51
56
6763
42
4
84
38
61
47
47
0
89
j4
68
42 43
j4
10
20
40
∞ 0
(a)
4
10
20
40
8743
68
∞
(b)
Figure 5.6: (a) POUT (W) and (b) ηD (%) contours when the transistor supply
voltage is 28 V, 36 V and 48 V. The optimal output power impedance varies as a
function of the supply voltage, while the high efficiency region remains constant.
The Smith Charts are normalized to 20 Ω.
73
For this transistor at a frequency of 370-MHz the optimum POUT and ηD regions
were close to 20-Ω and the Smith Chart is normalized to this value. Figure 5.5
shows measured POUT and ηD contours for a supply voltage of 28 V. The figure
also shows the Class-E impedance predicted by the simple theoretical formulae,
e.g. [23], with a 9 pF device output capacitance. From the figure it can be observed
that the optimal output power and drain efficiency contours do not overlap. This
means that a tradeoff analysis is necessary in order to choose the amplifier output
impedance. Ultimately, the PA designer needs to deal with this analysis depending
on the particular application. The next section discusses a proposed systematic
optimization procedure based on a weighted Euclidean distance.
Similar measurements were taken for the device biased at 36 V and 48 V. The
contours corresponding to optimal POUT are shown in Figure 5.6(a), while optimum ηD contours are shown in Figure 5.6(b). A trend can be observe in terms of
optimum POUT as a function of supply voltage. As the supply voltage increases
the imaginary part of the optimal impedance increases toward higher inductance
values, while the real part remains approximately constant.
Figure 5.6(b) shows that the optimum ηD remains constant as the supply
voltage is varied. This is consistent with Class-E theory. Assuming no varactor
effect in the output capacitance the Class-E impedance should remain constant
as a function of supply voltage, since the supply voltage only scales the amplitude
of the voltage and current waveforms across the device but does not change their
shape.
5.4
Optimum Amplifier Design
It can be seen from the examples in Figure 5.5 that a load impedance that optimizes power is not the same as the impedance that optimizes efficiency. For a
74
100
90
Drain Efficiency [%]
80
70
60
50
40
Pout
2
2
N D
h= 1
P out ,max 30
20
10
0
0
10
20
30
40
Output Power [W]
50
60
Figure 5.7: Drain efficiency values for specific output power are plotted as obtained
by the load-pull measurement (Figure 5.5). The distance of each point to the origin
is used a as metric for optimization.
4
10
20
(a)
40
∞ 0
4
5
10 6
(b)
j4
j40
70
66
α = 0.8
74
72
40
68 70
58
62
20
j10
56 60
62
64
α = 0.7
6508
90
80 70 0
6
50
0
j4
50
60
70
80
90
j40
56
58
60
64
j10
66
68
j40
α=0
j4
j20
j20
j20
j10
∞ 0
66
4
10
20
40
∞
(c)
Figure 5.8: Load-pull contours in a normalized 20 Ω Smith Chart for different
values of the parameter α in Equation 5.12. High values of h mean that the
specific requirements for a specific tradeoff are matched for all the impedances on
this contour; (a) show impedance contours for α=0, when power is optimize and
efficiency is sacrificed, (b) α=0.7 gives more weighting to efficiency and (c) α=0.8
further maximizes efficiency.
75
particular PA design it would be useful to have a guideline of how much output
power needs to be sacrificed to meet a particular efficiency specification. In this
section we propose a simple method which allows a systematic approach to this
tradeoff.
The method is outlined as follows:
• For each impedance the measured output power versus measured efficiency
is plotted from load-pull data. An example is given in Figure 5.7;
• Given that the data deviates significantly from a straight line we defined the
metric “h” as the weighted Euclidean distance from the origin;
• If defined in terms of output power and efficiency the metric is as follows;
s
h=
µ
(1 − α) ·
POUT
POUT, max
¶2
2
+ α · ηD
(5.12)
• In the case when the output power is maximized without concern for efficiency, the parameter α = 0;
• In the case when the efficiency is maximized without concern for output
power sacrifice, the parameter α = 1;
• Typically, however, there would be a tradeoff between these two parameters
corresponding to different values of α between 0 and 1;
• Once α is chosen for a given design the load-pull data can be replotted to
target the optimal impedance for a given tradeoff;
Examples are shown in Figure 5.8 for the load-pull data corresponding to
Figure 5.5. Load-pull contours for different values of the parameter α in Equation
5.12 are chosen. High values of h mean that the specific requirements for a specific
tradeoff are matched for all the impedances on this contour. For example α=0.7
76
Figure 5.9: Photograph of the prototype amplifier. It is a high-efficiency power
amplifier that uses a GaN HEMT on a SiC as the active device.
(Figure 5.8(b)) gives more weighting to efficiency than power. It seems that in this
case the solution is not unique as there are two distinct contours that maximize
the parameter h. This method can be extended to combine additional parameters
in a slightly more complicated metric. For example, IMD level would be relevant
for linearized Class-E PAs with envelope tracking [12]. In the presented design we
choose α = 0.7 and the resulting performance is presented in the next section.
Table 5.1: Measured Results for Class-E Amplifier
VDS
Pin
Pout
Gain (dB)
ηD
PAE
14 dB
15 dB
15 dB
87%
85%
71%
84%
82%
70%
28 V +32 dBm 45 W
35 V +33 dBm 65 W
48 V +34 dBm 87 W
5.5
Class-E Amplifier Performance
The measured power sweep for the α = 0.7 amplifier is shown in Figure 5.10 for
the 28 V supply voltage. The amplifier was designed using the measured loadpull data as a starting point for the output match and a similar source pull for
the input match. The matching network were verified separately using the three
block approach from Figure 5.3. The final impedances at the operating frequency
are ZIN = 10.1 + j1 Ω, and ZOUT = 12 + j6 Ω. Table 5.1 shows a summary of
77
RFMD060 28V − IMS 2008
50
100
45
90
D
40
80
PAE
35
70
30
60
25
50
Gain
20
40
15
30
10
20
5
10
0
10
15
20
25
Input Power [dBm]
30
35
Efficiency [%]
Output power [dBm] and Gain [dB]
Pout
0
Figure 5.10: Power sweep of the Class-E mode 370-MHz Power Amplifier for a
supply voltage of 28 V. The maximum PAE is 84% with a ηD of 87% with 45 W
of POUT .
the measured PA performance at three different supply voltages (28 V, 35 V, and
48 V). As a trend it is observed that by increasing the supply voltage the amplifier
is capable of producing additional output power at the expense of efficiency. For
example, the output power increased from 45 W to 87 W by increasing the supply
voltage from 28 V to 48 V. However, in this same range the ηD reduced from 87% to
71%. The supply voltage can be used as an additional dimension for optimization.
It is important to remain within transistor limits to avoid permanently damaging
the transistor.
5.6
Summary
The main objective of the work presented in this chapter is to layout the steps for
designing high-efficiency transmission line hybrid class-E power amplifiers. Due
to the highly nonlinear behavior of Class-E amplifiers, transistor characterization via load-pull techniques are more reliable than using nonlinear models. A
RF3932 GaN HEMT on a SiC substrate prototype from RFMD is used to illus78
trate the steps for amplifier design. Load-pull data showed little overlap between
optimum output power contours and optimum drain efficiency contours. An optimization procedure based on weighted Euclidean distance is developed as an
objective method for optimum impedance selection. The implemented amplifier
achieves POUT = 45 W with ηD = 87% with VDS = 28 V at 370 MHz. Additional
output power can be achieved by increasing the supply voltage at the expense
of efficiency. The supply voltage can be an additional parameter for amplifier
optimization.
79
Chapter 6
Technology Comparison of UHF
Amplifiers
6.1
Introduction
State-of-the-art wide-bandgap semiconductor RF transistors are delivering recordbreaking power levels in the UHF and lower microwave frequency range, with
demonstrations of excellent amplifiers as reported in [6], [18], [70] - [73]. Widebandgap semiconductors have better intrinsic material properties than silicon,
i.e. larger energy gap (support higher internal electric fields before breakdown),
lower relative permittivity (lower capacitive loading), higher thermal conductivity
(higher heat handling), and higher critical electric fields (higher RF power) [74].
High voltage operation and high power density with low parasitic reactance translate into robust devices that can withstand high-stress conditions typically associated with switch-mode operation. For example, in Class-E mode the peak voltage
across the device can be more than 3.56 times higher than the supply voltage [26].
The supply voltage must then be limited by this factor (VDSS /3.56 where VDSS
is the absolute maximum drain-to-source voltage). Therefore, devices with high
breakdown voltage are ideal for this mode of operation.
In this work, the performance of wide-bandgap transistors (GaN HEMTs on a
Si substrate, GaN HEMTs on a SiC substrate, and SiC MESFETs) is compared
to standard Si LDMOS by designing UHF high-power high-efficiency transmission
line Class-E power amplifiers. Load-pull techniques under Class-E conditions are
used for device characterization. Finally, the power amplifiers are compared in
terms of output power, gain, PAE and static distortion.
6.2
Transistor Technologies
The four different transistor technologies compared in this work are GaN HEMT
on a Si substrate, GaN HEMT on a SiC substrate, SiC MESFET, and Si LDMOS.
Today, the two main substrates used to grow GaN HEMTs are either Si or SiC.
The main advantages of Si as a substrate for GaN is its low cost, and devices are
commercially available in a variety of power levels ranging from 4 W to 180 W,
with a promise of 200 V breakdown voltage [75]. SiC substrates are expensive, but
the thermal conductivity is about 2.6 times higher than that of Si (4 W/◦ k-cm
for SiC vs. 1.5 W/◦ k-cm for Si) [74]. Thermal conductivity is critical for highpower PAs because device performance degrades with temperature. For additional
information on wide-bandgap transistors the reader is refered to [76]. The devices
used in this work have no internal prematching and they are:
• GaN HEMT on a Si substrate
Transistor used is the NPTB00050 from Nitronex. This transistor is able
to deliver 50 W of output power with 11.5 dB of gain and 60% ηD at 3 GHz
with a supply voltage of 28 V. The breakdown voltage is 100 V, the output
capacitance is 9 pF and the threshold voltage is about -2.1 V.
81
• GaN HEMT on a SiC substrate
Transistor used is a RF3932 prototype from RFMD. The transistor can
deliver 60 W of peak power and it is rated for 48 V operation with 14 dB of
gain in Class-AB. It has 9 pF of output capacitance and a threshold voltage
of -4.2 V.
• SiC MESFET
Transistor used is the CRF24060 from Cree. It is rated for operation up
to 2.4 GHz with 13 dB of gain and 50 W of output power in Class-AB. It
has a typical performance of 60 W of output power with 45% ηD at 1.5 GHz
with a supply voltage of 48 V. The threshold voltage is -10 V and the output
capacitance is approximately 11 pF.
• Si LDMOS
The transistor used is the AGR09045E from Peak Devices/Agere. Its recommended operation is from 865 MHz to 895 MHz with 45 W of output power,
and ηD of 59%. The breakdown voltage is 65 V and the threshold voltage is
3.5 V. The output capacitance is 23 pF.
Table 6.1 summarizes device parameters, i.e. drain-to-source breakdown voltage, output capacitance, RON DC , junction temperature and turn on threshold
voltage.
Table 6.1: Summary of Transistor Properties
Part Number
VDSS
COUT
RON-DC
NPTB00050
RF3932
CRF24060
AGR09045E
100 V
150 V
120 V
65 V
9 pF
9 pF
11 pF
23 pF
0.23 Ω
0.32 Ω
0.46 Ω
0.35 Ω
82
TJ
VTH
200◦ C -2.1 V
250◦ C -4.2 V
250◦ C -10 V
200◦ C 3.5 V
The breakdown voltage of wide-bandgap transistors is over 100 V, while standard Si LDMOS in only 65 V. Higher voltages lead to higher matching impedances.
Also, in Class-E operation the drain-to-source voltage should be limited to at least
1/3.56 the breakdown voltage of the device to avoid permanently damaging the
transistor. This might limit the maximum output power attainable in Class-E
mode.
Matching impedances are not only affected by the supply voltage, but by the
output capacitance as well [2]. The higher dielectric value of Si and its lower critical field (a larger number of fingers is needed to achieve desired POUT compared
to GaN) results in an output capacitance more than twice that of wide-bandgap
transistors. This parameter also limits the maximum frequency for Class-E operation.
Another important parameter for Class-E operation is the transistor RON . The
RON limits the maximum efficiency attainable with these devices in Class-E mode.
Table 6.1 shows the measured RDC ON for each of the transistor technologies.
The lower value was obtained with the GaN HEMT on a Si substrate (0.23 Ω)
followed by the GaN HEMT on a SiC substrate (0.32 Ω), while the largest value
was obtained with the SiC MESFET (0.46 Ω). The table also shows the maximum
junction temperature and the threshold voltage.
6.3
Transistor Characterization in Class-E Mode
A transistor of each of the four different technologies was characterized with the
procedure described in the previous chapter at an operating frequency of 370 MHz.
The POUT and ηD load-pull contours for a supply voltage of 28 V when the device
is in compression are shown in Figure 6.1. Closed contours for the Si LDMOS
were not obtained due to voltage breakdown problems. The supply voltage of
83
j10
j10
10
(a)
36
48
∞0
20
2
5
j10
j20
40
2
5
52
48
44
48
0
73 69
323
64
0
44
j2
77
81
10
(c)
89
85
77
69
20
j5
90
j20
POUT
D
78
7
704
32
36
∞
2024
POUT
D
78
j10
j2
73 81
j5
82
10 74 20
(b)
36
∞0
32
5
52
32
2
48
7478
82
4440
0
j2
32
69 1
6
459 7
441 3
33
73 5
6
65
j20
POUT
D
36
40 44
j2
33
41
61
37 65
45 73 77
81 77
j5
86
j20
POUT
D
61 69
j5
62
28
2
5
66
10
20
∞
(d)
Figure 6.1: POUT and ηD load-pull contours for each of the four different transistor
technologies at a supply voltage of 28 V in a 10-Ω Smith Chart. (a) GaN HEMT
on a Si substrate, (b) GaN HEMT on a SiC substrate, (c) SiC MESFET, and (d)
Si LDMOS. Notice that optimum POUT contours do not overlap with optimum ηD
contours.
28 V exceeds the maximum allowed for Class-E mode which is 18 V (65 V/3.56).
In this set of data the main difference between wide-bandgap transistors and
Si LDMOS is the value of the output capacitance. The output capacitance for
the wide-bandgap transistors is between 9 pF to 11 pF, while for the Si LDMOS
is 23 pF. It is interesting to observe the rotation of the optimal efficiency region
as a function of output capacitance. If the intrinsic output capacitance keeps
increasing, the transistor will no longer be operating in Class-E mode, but in
Class-AB.
From Figure 6.1 we can observe that the optimal output power regions do
not overlap with the optimal drain efficiency regions.
Table 6.2 summarizes
performance for each of the transistors in the optimum POUT and optimum ηD
84
Table 6.2: Optimum POUT and ηD Impedances from Load-pull Contours.
Part Number
NPTB00050
RF3932
CRF24060
AGR09045E
Optimum POUT
Optimum ηD
63 W, 68% @ 5+j7.5Ω
30 W, 84% @ 14+j22Ω
53 W, 75% @ 8+j4.4Ω 14 W, 93% @ 11.8+j2.9Ω
53 W, 70% @ 6+j3.6Ω
20 W, 89% @ 13.5+j16Ω
40 W, 74% @ 5.2+j5.6Ω
24 W, 82% @ 4+j9.3Ω
impedances. A tradeoff analysis as the one proposed in the previous chapter is
necessary in order to choose the amplifier optimum output impedance.
6.4
Class-E Amplifier Performance
Amplifier prototypes were implemented with each of the transistor technologies.
It is important to point out that in the remainder of this chapter, measured results
for the PA prototypes are compared, whereas it might not be a direct comparison
of transistor technologies.
The goals for these prototypes are to achieve POUT > 40 W with ηD > 80%.
The efficiency goal is much harder to meet. Increasing the supply voltage usually
increases POUT at the expense of ηD . The supply voltages used for each of the
prototypes are 28 V for the GaN HEMTs, 35 V for the SiC MESFET and 18 V
for the Si LDMOS. The supply voltage for the Si LDMOS is limited to only 18 V
because of transistor failure at higher levels. However, even for this low voltage
the Si LDMOS was able to meet the specifications of 40 W and 80%. As for the
SiC MESFET, the output matching gave significantly more weight to ηD with less
POUT and the supply voltage had to be increased to 35 V to meet the power goal.
The amplifiers were tested while sweeping the input power, and sweeping the
supply voltage. For the input power sweep, the supply voltage is held constant
while PIN increases until the amplifier starts to saturate and a peak is observed in
the PAE. This results in an optimum PIN for a given PAE. In the supply sweep,
85
Table 6.3: Prototypes Power Sweep Performance
Part Number
VDS
PIN
POUT
Gain
ηD
NPTB00050
RF3932
CRF24060
AGR09045E
28 V +29 dBm 43 W 17.5 dB 83%
28 V +32 dBm 45 W 14.6 dB 85%
35 V +31 dBm 40.5 W 15 dB 80%
18 V +34 dBm 42.6 W 12.5 dB 80%
PAE
81%
82%
78%
75%
the amplifier PIN is fixed to this optimum value while the supply voltage varies
from 0 V up to a maximum supply voltage. For each of these sweeps parameters
such as POUT , Gain, and PAE are recorded. Supply sweeps are important because
these amplifiers are intended for polar transmitters. Also, for this application it
is important to measure the amplifiers’ AM-to-AM and AM-to-PM properties.
Results for these sweeps are discussed next.
6.4.1
Input Power Sweep
Figure 6.2 shows measured POUT , Gain and PAE as a function of PIN for each of
the amplifier prototypes. In terms of POUT , Figure 6.2(a) shows that each of the
amplifiers is able to reach over 40 W (+46 dBm). This same figure shows that for
these prototypes the saturated gain follows a similar path. It is important to point
out the differences in supply voltage for each of the prototypes as discussed above
(28 V for the GaN HEMTs, 35 V for the SiC MESFET and 18 V for Si LDMOS).
Table 6.3 summarizes the performance of the four prototypes.
Figure 6.2(b) shows the measured PAE for each of the amplifiers. Some interesting plot properties are:
• PAE is above 80% for the GaN HEMTs amplifiers.
• The GaN HEMT on a Si substrate amplifier compresses with 3 dB less PIN
compared to the GaN HEMT on a SiC substrate amplifier, requiring less
drive.
86
50
Output Power [dBm] and Gain [dB]
45
P
OUT
40
GaN HEMT on Si @ 28V
35
GaN HEMT on SiC @ 28V
30
SiC MESFET @ 35V
25
Si LDMOS @ 18V
20
15
Gain
10
5
0
10
15
20
25
Input Power [dBm]
30
35
30
35
(a)
100
90
Power Added Efficiency [%]
80
GaN HEMT on Si @ 28V
GaN HEMT on SiC @ 28V
SiC MESFET @ 35V
Si LDMOS @ 18V
70
60
50
40
30
20
10
0
10
15
20
25
Input Power [dBm]
(b)
Figure 6.2: Input power sweep for each of the amplifier prototypes: GaN HEMT
on a Si substrate biased at 28 V, GaN HEMT on a SiC substrate biased at 28 V,
SiC MESFET biased at 35 V and Si LDMOS biased at 18 V. (a) POUT and Gain,
and (b) PAE versus PIN .
87
• All the amplifiers achieve more than 75% PAE.
• PAE achieved by the SiC MESFET is between the PAE acheived by the
GaN HEMTs and the Si LDMOS.
6.4.2
Supply Voltage Sweep
Figure 6.3 shows POUT , gain and PAE as a function of the supply voltage for
each of the four amplifiers with dynamic biasing for linearization in mind. In this
measurement, the input power is fixed to the one that gives optimum PAE in the
input power sweep, while the supply voltage is swept from 0 V up to a maximum
supply voltage. The maximum voltage is transistor dependent (VDSS /3.56).
Figure 6.3(a) shows POUT and gain. All the amplifiers reach POUT higher than
the specified but not at the same supply voltage. The Si LDMOS prototype amplifier reaches the specified output power for lower supply volage than the other ones.
The performance of the GaN HEMTs amplifiers is identical, while higher voltage
is needed for the SiC MESFET amplifier. Additional POUT can be achieved at
higher supply voltages, however it is important to consider the transistor breakdown voltage. The figure also shows the amplifier gain as a function of supply
voltage. Below an amplifier-dependent supply voltage threshold, the gain is negative and the amplifier attenuates the input signal. The amplifiers with higher
gain are the GaN HEMT on a Si substrate and the Si LDMOS. The amplifier
implemented with the SiC MESFET has the lowest gain.
Figure 6.3(b) shows PAE as a function of supply voltage. Due to the PAE
dependence upon the amplifier gain, the PAE is negative for voltages where POUT
is lower than PIN . In this plot we can observe the advantages of high gain and
high breakdown voltage. These two conditions ensure high overall PAE. If the
amplifier is used in a polar transmitter with a specified modulation scheme, the
88
50
Output Power [dBm] and Gain [dB]
45
40
GaN
GaN HEMT
HEMT on
on Si
Si @
@ +29dBm
+29dBm
35
GaN
GaN HEMT
HEMT on
on SiC
SiC @
@ +33dBm
+33dBm
30
SiC
SiC MESFET
MESFET @
@ +33dBm
+33dBm
POUT
Si
Si LDMOS
LDMOS @
@ +34dBm
+34dBm
25
20
15
Gain
10
5
0
0
5
10
15
20
25
Supply Voltage [V]
30
35
40
(a)
100
90
Power Added Efficiency [%]
80
70
60
50
40
30
20
GaN HEMT on Si @ +29dBm
GaN HEMT on SiC @ +33dBm
SiC MESFET @ +33dBm
Si LDMOS @ +34dBm
10
0
0
5
10
15
20
25
Supply Voltage [V]
30
35
40
(b)
Figure 6.3: Supply voltage sweep for each of the amplifier prototypes: GaN
HEMT on a Si substrate with +29 dBm of PIN , GaN HEMT on a SiC substrate
with +33 dBm of PIN , SiC MESFET with +33 dBm of PIN , and Si LDMOS with
+34 dBm of PIN . (a) POUT and Gain, and (b) PAE versus supply voltage.
89
input power and the supply voltage range can be optimized in terms of power
consumption efficiency.
6.4.3
AM-to-AM and AM-to-PM Distortion
In polar transmitters the two main sources of distortion are AM-to-AM and AMto-PM. The first describes how the supply voltage affects the RF output voltage
across a constant load, and the second relates the supply voltage and the phase
difference between the PA input and output RF signals. Figure 6.4 shows these
two sources of distortion for each of the amplifier prototypes.
Figure 6.4(a) shows AM-to-AM curves and a linear relation can be observed.
At low voltages AM–to–AM curves are limited due to feedthrough, related to the
leakage from input to output due to the transistor internal capacitances at a supply
voltage of 0 V. At the other end, the amplifier is limited by either saturation or
breakdown voltage. A third interesting parameter is the slope, which is higher for
Si LDMOS, followed by GaN HEMTs, while the smallest slope corresponds to the
SiC MESFET. The slope of the AM-to-AM line depends on impedance matching.
We can observe that in order to achieve the output power requirements with the
Si LDMOS transistor, the impedance matching was agressive towards achieving
the power at a low voltage (18 V). For the SiC MESFET, significant weight was
given to efficiency. As a consequence, a higher voltage was needed to achieve
the desired POUT specification, therefore a smaller slope. The GaN HEMTs were
matched at some point in between.
The AM-to-PM characteristics for each of the amplifiers is plotted in Figure
6.4(b). The AM-to-PM distortion is characterized by an abrupt knee at lower
voltages. This distortion should be corrected with predistortion because it has
significant effects on ACPR and EVM in polar transmitters. GaN HEMT on a Si
substrate shows the smallest knee voltage, followed by the Si LDMOS. The SiC
90
100
90
80
AM−AM [V]
70
60
50
40
30
20
GaN HEMT on Si @ +29dBm
GaN HEMT on SiC @ +33dBm
SiC MESFET @ +33dBm
Si LDMOS @ +34dBm
10
0
0
5
10
15
20
25
Supply Voltage [V]
30
35
40
(a)
0
−20
−40
AM−PM [V]
−60
−80
−100
−120
−140
GaN HEMT on Si @ +29dBm
GaN HEMT on SiC @ +33dBm
SiC MESFET @ +33dBm
Si LDMOS @ +34dBm
−160
−180
0
5
10
15
20
25
Supply Voltage [V]
30
35
40
(b)
Figure 6.4: (a) AM–to–AM and (b) AM–to–PM for each of the amplifier prototypes: GaN HEMT on Si with +29 dBm of PIN , GaN HEMT on SiC with +33 dBm
of PIN , SiC MESFET with +33 dBm of PIN , and Si LDMOS with +34 dBm of PIN .
The feedthrough values for the prototypes (GaN HEMT on a Si substrate, GaN
HEMT on a SiC substrate, SiC MESFET and Si LDMOS) are 0.75 V, 2.4 V, 4.28 V
and 1.39 V (respectively).
91
MESFET amplifier has the largest knee. Ideally, it is desired that this response
is constant. We can observe from the figure that the overall shape is amplifier
dependent.
6.5
Discussion of Results
In this work, high-efficiency UHF power amplifiers were designed with four different transistor technologies; GaN HEMT on a Si substrate (NPTB00050 from
Nitronex), GaN HEMT on a SiC substrate (RF3932 prototype from RFMD),
SiC MESFET (CRF24060 for CREE), and Si LDMOS (AGR09045E from Peak
Devices/Agere). These transistors have similar output power capabilities. Specifications such as breakdown voltage, output capacitance, and RON DC were given
for each of these transistors.
A modular fixture that allows TRL calibration and impedance verification was
designed and the transistors were characterized with a load-pull measurement
system under Class-E conditions. The POUT and ηD load-pull contours showed
that for wide–bandgap transistors used at UHF there is a significant tradeoff
between POUT and ηD since the contours do not overlap. A simple optimization
procedure, based on a weighted Euclidean distance was applied in order to deal
with this tradeoff.
Transmission line Class-E amplifier prototypes were implemented for each of
the transistor technologies. These prototypes had the specification of achieving
more than 40 W of POUT with over 80% ηD . Input power and supply sweeps
characterized performance of each amplifier prototype. The prototypes were also
characterized in terms AM-to-AM and AM-to-PM, due to potential use in polar
transmitters. Following is a brief summary of the overall technology performance.
• GaN HEMT on a Si substrate
92
The particular Nitronex transistor used in this work, has lower RON DC
compared to the other technologies and a low output capacitance of approximately 9 pF. The prototype built with this transistor seems to have
good overall performance. It had the largest gain and best properties for
polar transmitters; regarding AM-to-AM characteristics it had the the lowest feedthrough and regarding AM-to-PM the lowest knee voltage, and a
relatively flat response.
• GaN HEMT on a SiC substrate
The overall performance of the RFMD GaN HEMT on a SiC substrate was
comparable to that of the GaN HEMT on a Si substrate, with the advantage
of a higher breakdown voltage. The main advantages of a higher breakdown
voltage is that the implemented matching network can be weighted more
towards optimum efficiency, because more output power can be obtained at
a larger supply voltage.
• SiC MESFET
Due to the larger RON resistance of the CREE SiC MESFET transistor
compared to the tested GaN HEMTs, the amplifier prototype impedance
matching network gave a lot of weight towards optimizing efficiency. The
supply voltage needed to be increased to 35 V in order to achieve performance comparable to the other prototypes. Tests showed high output power
with little degradation in efficiency for voltages above 42 V (70 W of POUT
with 77% ηD with a supply voltage of 47 V).
• Si LDMOS
The AGR09045E Peak Devices/Agere LDMOS has the largest output capacitance (23 pF) and lowest breakdown voltage (65 V) compared to the
93
other technologies. Its overall performance was comparable to the other
technologies, although the low breakdown voltage played a significant role
in its limitation.
6.6
Summary
The main objective of the work presented in this chapter is to assess the benefits of wide-bandgap transistors compared to standard Si LDMOS at UHF by
designing high-efficiency power amplifiers. Si LDMOS has been the technology of
choice at UHF, and wide-bandgap transistors have been designed to target the
GHz range. The low cost of Si LDMOS compared to new wide-bandgap transistor
and its reliable performance are dominant factors to keep Si LDMOS as the technology of choice. However, wide-bandgap transistor also demonstrate excellent
performance at UHF with the advantage of higher breakdown voltage, and lower
output capacitance leading to higher output matching impedances and additional
flexibility in amplifier design. For example, the output matching design can give
higher weight towards improving efficiency, since higher output power can usually
be achieved by increasing the supply voltage. The specific contributions contained
in this thesis are as follows:
• Design and implementation of high-efficiency power amplifiers at UHF with
new wide-bandgap technologies; GaN HEMT on a Si substrate, GaN HEMT
on a SiC substrate, and SiC MESFET.
• Performance comparison of amplifiers implemented with wide-bandgap transistors to an amplifier implemented with standard Si LDMOS.
94
Chapter 7
Discussion & Future Work
7.1
Discussion
The topic of this thesis is efficiency optimization in UHF and microwave power
amplifiers used in communications systems. In order to eventually obtain an amplifier architecture in which efficiency and linearity can be optimized for a given
signal, some choices have to be made. For example, a specific class of singleended transistor operation for the RFPA can be chosen to enable straightforward
dynamic biasing. The choice of class-E with its soft-switching properties is also
appropriate for hybrid circuit implementation for which some tolerance for component values is advantageous. At the beginning of this work, it was an open
question as to what type of transistor works best in this switched-mode of operation. In addition to well known devices such as MESFETs and Si LDMOS,
new emerging technologies such as GaN HEMTs and SiC MESFETs needed to be
considered.
After choosing the type of PA, which is not commonly used in commercial
products, amplifiers over a wide frequency range were implemented. 60 W single
device amplifiers with over 80% efficiency at 370 MHz were designed in pure class-E
mode, while it was shown that at higher frequencies the devices operate in a suboptimal, slightly degraded class-E mode. Nevertheless, the results demonstrated
for EDGE signals at 880-MHz and the results at 10 GHz are highest reported in
the literature.
In order to linearize the PAs with minimal added complexity and minimal
sacrifice in efficiency, a choice was made for the linearization technique which
matches the class-E operation. This polar architecture was then demonstrated at
880-MHz and 10-GHz with some of the best reported simultaneous linearity and
efficiency as demonstrated by spectral mask measurements and reduction in IMD
level.
The results of this work open up numerous possibilities for future research as it
will be discussed in the next section. On the power amplifier side, modified modes
of operation, with different harmonic terminations are likely to give improved
performance especially for higher powers and at lower microwave frequencies. As
semiconductor technologies improve, more power will be available from single GaN
transistors with higher supply voltages and the technique developed in this thesis
can be directly applied to this future devices. For even higher power levels, low
loss combining techniques need to be investigated.
In terms of linearization for communication transmitters the trend of increasing
bandwidth and spectral efficiency will demand new integrated solutions such as
possibly combinations of polar, outphasing and predistortion.
The specific contributions contained in this thesis are as follows:
• The repeatability of the 10-GHz amplifier characterization, and load-pull
measurements with single and two-tone (Cartesian and polar) excitation
was verified, as well as an investigation of proper design of bias circuitry for
RFPAs in polar transmitters [12], [17].
96
• Design and implementation of a transmission line hybrid Class-E power
amplifier for the 880-MHz polar transmitter. The maximal PAE is 70%
with ηD of 75% with POUT = +20 dBm. The maximal POUT is +22.4 dBm
with little degradation in PAE [15].
• The amplifier is linearized with an efficient envelope tracker [16].
• The total transmitter efficiency is 56% with POUT = +20 dBm for EDGE
signals. This is the highest efficiency reported for a EDGE polar transmitter
that meets the EDGE mask [17]. The closest work [56] is a parallel effort at
the University of California, San Diego with a 44% efficient polar transmitter
with the same output power.
• Development of an optimization procedure based on weighted Euclidean
distance an objective method for optimum impedance selection [18]. This
contribution will be very useful for future designs, and thus more examples
are given in Section 7.2.
• Implementation of a UHF high-efficiency amplifier implemented with a GaN
HEMT on a SiC substrate that achieves POUT = 45 W with ηD = 87% with
VDS = 28 V at 370 MHz [18]. These PAs were design using a commercial
Focus load-pull system. Some of the active devices are in the experimental
stage and the manufacturer provided no specs. Another in-house developed
active load-pull technique could prove useful in the future, and some results
developed for the work in this thesis is described in Section 7.4.
• Design and implementation of high-efficiency power amplifiers at UHF with
new wide-bandgap technologies; GaN HEMT on a Si substrate, GaN HEMT
on a SiC substrate, and SiC MESFET [19].
97
j10
j10
64
66
69 71
=0.75
j2
71
j20
5
10
(a)
84
89
∞ 0
20
2
5
10
(b)
j5
7274
79
77 7742
0
2
j5
j20
5
82
10
(c)
j2
7747
=0.75
79
∞
j10
j10
j2
20
87
2
87
2
789
64 66
0
79
82
=0.75
69
j20
=0.75
71
79 72
20
∞ 0
2
5
78
76
73
j2
j5
j20
84
j5
10
(d)
20
∞
Figure 7.1: Load-pull contours for the parameter hα=0.75 of Equation 5.12 for each
of the four different technologies and a supply voltage of 28 V. (a) GaN HEMT
on a Si substrate, (b) GaN HEMT on a SiC substrate, (c) SiC MESFET, and (d)
Si LDMOS. Notice that optimum h depends significantly upon technology.
• Performance comparison of amplifiers implemented with wide-bandgap transistors to an amplifier implemented with standard Si LDMOS [19].
7.2
Additional Optimization Examples
Section 5.4 introduces a proposed method for the design of optimal power amplifiers. Amplifier design consists of biasing and impedance matching. Load-pull
techniques allows transistors to be characterized under different input and output
impedance terminations. In cases as the one discussed in Chapter 5 impedance
selection is non-trivial as different impedances provide desired functionality such
as maximal output power or maximal efficiency. Additional contours can be obtained from load-pull in terms of measurable quantities such as additive phase
noise or IMD level. The amount of data increases if parameters such as supply
98
j10
j10
∞ 0
20
(a)
2
5
78
10
20
(b)
10
(c)
20
84
87
j2
VDS = 32V
=0.75
79
∞ 0
2
5
10
(e)
79
8687
84
890
83
86879
7 80
84
j2
80
20
j20
=0.5
88
89
83
5
∞
∞ 0
2
5
10
(f)
87
2
20
(d)
j5
j20
80 9
7
0
10
85
8380
j10
j5
28V=0.75
32V=0.75
0
883
85
83
j10
j2
∞
j20
90
80
5
79
82
87
84
89
VDS = 28V
=0.75
2
j5
j20
2
789
0
88
j10
j10
j5
j2
883
9
7883
44
10
58
5
83
88
2
j2
58
49
44
0
86 91
96
91
6
8
81 76
18
27
35
81
18
27
76
j20
41
35
j20
32 1
49
4
j2
VDS =j5
32V
POUT [W]
D [%]
32
VDS =j5
28V
POUT [W]
D [%]
87
20
∞
Figure 7.2: POUT and ηD load-pull contours for the RF3932 for a supply voltage
of (a) 28 V, and (b) 32 V. (c) h for an α = 0.75 and a supply voltage of 28 V, and
(d) 32 V. (d) h for an α = 0.75 and supply voltages of 28 V and 32 V, and (e)
farther combined h contours (hω ) for an ω = 0.5.
voltage are also taken into account. Due to the large quantity of information it
is a difficult task to objectively select optimal impedances even if a very specific
application is in mind. It is in this sense that a method just as the one described
in Section 5.4 is important.
Figure 7.1 shows additional examples of optimized contours. In this case for
an α = 0.75 in Equation 5.12 is chosen for each of the transistors characterized
in Chapter 6. As can be seen from the figure the optimal impedance depends
significantly upon technology.
99
This technique also allows for additional optimization by combining for example voltage dependence. By defining:
q
hω =
(1 − ω) · h2V 1 + ω · h2V 2
(7.1)
an overall optimal result that combines hα contours for specified voltages is obtained. Figure 7.2 shows a sequence of contours that combine POUT and ηD for
two different voltages, 28 V and 32 V for the RF3932. It is interesting to observe
how the data from Figures 7.2 (a) and (b) is simplified to the one shown in Figure 7.2(e). This optimization procedure allows for the consideration of voltage
dependent effects when designing amplifiers with dynamic biasing. The weighting
coefficients can potentially be the envelope PDF of the modulation scheme to be
used.
7.3
Class-F Transistor Characterization
The procedure described in Section 5.2 was used for transistor characterization
in Class-F mode. In Class-F the 2nd harmonic is terminated with a short circuit,
while the 3rd harmonic is terminated with an open stub as shown in Figure 7.3.
The wide-bandgap transistors were characterized under these conditions for a supply voltage of 28 V. The POUT and ηD load-pull contours are shown in Figure 7.4.
It is important to notice the difference in impedances compared to Class-E. This
means that harmonic terminations affect the necessary impedances for optimal
output power or efficiency at the fundamental frequency. Table 7.1 summarizes
the measured data. It was observed that the tradeoff between POUT and ηD is
not as severe as in Class-E. Due to time constraints Class-F prototypes were not
implemented.
100
Figure 7.3: Output matching network for Class-F operation. For Class-F transistor characterization the 2nd harmonic is terminated with a short circuit, while the
3rd harmonic is terminated with an open stub.
Table 7.1: Class-F Optimum POUT and ηD .
Part Number
NPTB00050
RF3932
CRF24060
7.4
Optimum POUT
Optimum ηD
50 W, 77% @ 11.2-j3.6Ω 40 W, 86% @ 13+j0.4Ω
42 W, 67% @ 8.6-j4.7Ω
30 W, 75% @ 19-j5Ω
54 W, 66% @ 6.8-j5.9Ω 37 W, 76% @ 15-j1.9Ω
Active Load-pull
The procedure for amplifier design via load-pull is as followed:
(1) Select transistor to be used.
(2) Select mode of operation, i.e. Class-A, Class-AB, Class-E, Class-E, etc...
(3) Characterize transistor in the desired mode of operation at the correct biasing conditions.
(4) Synthesize and verify input and output matching networks.
(5) Test prototype.
(6) Load-pull the prototype to ensure it is delivering optimum performance.
In this work, the transistor were characterized in dBm Engineering/Peak Devices/TriQuint facilities in Boulder, CO and it was impractical to repeat load-pull
for every designed prototype. An active load-pull system was implemented to
101
j10
j10
j5
2
5
10
24
20
74
∞
42
49
64
∞ 0
20
33
10
69
40
59
54
5
j20
81
66371
762 32
−j2
2
32
0
j5
RF3932
POUT [W]
D [%]
j2
7166
76
j2
j20
NPTB00050
POUT [W]
D [%]
−j2
(a)
(b)
j10
j5
j2
j20
CRF24060
POUT [W]
D [%]
36
2
1045
5
20
∞
71
0
27
66 71
61
56
−j2
61
66
(c)
Figure 7.4: POUT and ηD load-pull contours for three different transistor technologies for a supply voltage of 28 V and Class-F terminations.(a) GaN HEMT on a
Si substrate, (b) GaN HEMT on a SiC substrate, and (c) SiC MESFET.
102
POUT
Load
Circulator
DUT
G =
VV+
Load
Linear
Amplifier
Vector
Modulator
Figure 7.5: Diagram of an active load-pull setup. The DUT delivers power to the
circulator. A sample of the power is coupled and controlled via a vector modulator
before been amplified and injected back to the DUT via the circulator. Control
of the amplitude and the phase of the injected wave mimics different Γ presented
to the DUT.
perform this last task. A diagram of the active load-pull setup is shown in Figure
7.5. The DUT delivers power to the circulator. A sample of the power is coupled
and controlled via a vector modulator before been amplified and injected back to
the DUT, again via the circulator. Control of the amplitude and the phase of
the injected wave mimics different Γ presented to the DUT. This particular setup
can be calibrated because it only depends on the input wave. Some of the system
details and assumptions are as follows:
• Circulator can handle the power level; output of the DUT plus the injected
power.
• Circulator insertion loss does not vary with power or temperature.
• Infinite circulator isolation.
• The amplifier is linear with a constant phase as a function of input power.
• For a Γ=1, the linear amplifier is injecting the same amount of power than
the one been generated by the DUT. This can be a problem when characterizing high-power devices.
103
j50
j25
j100
j10
0
10
25
50
∞
100
−j10
−j25
−j100
−j50
(b)
(a)
Figure 7.6: (a) Photograph of an implemented active load. The vector modulator
is implemented with the RVA-2500 attenuator and two JSPHS-446 phase shifters
(180◦ ) from Minicircuits. (b) Measure active load contours.
• Γ can be larger than 1.
• For Γ=0.7 the active load injects half of the DUT output power.
• The system is calibrated with small-signal S-parameters in a network analyzer. The accuracy of the calibration is compromised as the DUT output
power increases and as Γ increases.
Several active loads were implemented and one of them is shown in Figure
7.6(a). The vector modulator is implemented with the RVA-2500 attenuator and
two JSPHS-446 phase shifters (180◦ ) from Minicircuits. The active load contours
are shown in Figure 7.6(b). The procedure for characterizing amplifiers via the
active load-pull is as follows:
• Active load is characterized with small-signal in a network analyzer. The
calibration is saved as a look-up-table to present desired impedance to the
DUT.
104
j50
P [W]
j25 OUT
[%]
D
h
j100
25
87
8150
100
47
10
46
0
82
79
76
84
j10
∞
87
−j10
Figure 7.7: POUT , ηD , and hα=0.7 contours for the NPTB00050 measured with the
active load for a supply voltage of 28 V.
• The calibrated impedances are presented to the DUT and performance is
recorded.
Figure 7.7 shows POUT , ηD , and hα=0.7 contours in a 50 Ω for the NPTB00050
measured with the active load for a supply voltage of 28 V. Some of the problems
encountered in the system are measurement of efficiencies higher than 100%, no
linear amplifier with over 100 W output power was available at the time, and the
large-signal behavior is approximated with small-signal S parameters. The system
accuracy can improve with measurement of the forward and reverse waves. This
can be accomplished via a coupler placed between the DUT and the circulator. A
circuit that can compare amplitude and phase of these waves is also necessary. A
circuit that can achieve this is the AD8302 from Analog Devices [77].
105
7.5
Future Work
Some of the topics that need to be address in the design of high-efficiency power
amplifiers for linear transmitters are:
• How to deal with larger bandwidths in systems that implement dynamic
biasing for signals such as WCDMA or signals that have not yet been developed but that the envelope exceeds the MHz range?
• How does predistortion can be used to improve linearity in polar systems?
• How does wide-bandgap transistors scale with power and frequency?
• How to efficiently combine several high-efficiency power amplifiers without
significantly degrading performance?
• How to minimize package parasitic effects that limit higher frequency performance?
• How to maintain high-efficiency over broad bandwidths?
7.5.1
Outphased Assisted Polar Transmitter
A way to deal with the bandwidth and high PAR of signals such as WCDMA is
with a system that combines outphasing and dynamic biasing. We are calling this
system the Outphased Assisted Polar Transmitter. What is intended with this
system is to combine the strength of polar transmitters with the strength of the
outphasing architecture.
A summary of the properties of the outphasing architecture:
• Can deal with feedthrough;
• Sensitive to phase and amplitude unbalances;
106
• Needs 2 synchronized IQ channels;
• Needs 2 RFPAs;
• In constant load combiners the outphased power is dissipated. Can it be
harvested?
• “Chireix” combiners can be used to efficiently combined the signals;
• Combiners losses cannot be avoided.
A summary of dynamic biasing is as follows:
• Scales the supply voltage to the desired output power;
• Problems due to feedthrough;
• Sensitive to time-alignment between envelope and phase;
• Only one IQ channel that needs to be synchronized with the envelope;
• Only one RFPA is needed;
• Performance depends on the bandwidth and the efficiency of the envelope
tracker.
Outphasing can assist a polar transmitter by providing the bandwidth that the
envelope tracker cannot supply; similar to the linear assisted envelope tracker.
However, an advantage of combining techniques is that the system is redundant;
only outphasing or only polar. The hybrid system can deal with feedthrough, and
techniques such as optimal band-separation can be applied making the transmitter
highly adaptable. The system implementation can be described as:
(1) Decomposed the amplitude and phase modulated signal into two components; amplitude and phase;
107
Envelope
Tracker
IQA(t)
~
IQA(t)
IQ Modulator 1
Iˆ1(t)
Mixer
RF Power
Combiner
+
90º
Q
ˆ1(t)
RF
Oscillator
IQ
RF1
RFPA
IQ
RF1
High-Efficiency
Class-E PA
Mixer
+
IQ Modulator 2
Iˆ2(t)
IQRF
Mixer
RF Power
Combiner
+
Qˆ 2(t)
90º
RF
Oscillator
IQ
RF2
RFPA
IQ
RF2
High-Efficiency
Class-E PA
Mixer
Figure 7.8: Diagram of the outphased assisted polar transmitter. The system
combines the advantages of outphasing with the advantages of a polar transmitter
making it highly adaptable.
(2) Lowpass filter the envelope with the bandwidth that the envelope tracker
can supply;
(3) Recombine highpass envelope to the phase signal;
(4) Apply outphasing techniques to the remainder of the signal.
A diagram of such as system is shown in Figure 7.8. The system combines the
advantages of outphasing architecture with the advantages of a polar transmitter
making it highly adaptable. Mathematically the system is describes as follows,
108
IQRF = IQA · IQφ
(7.2)
IQφ = ej∠IQRF
(7.3)
IQA = |IQRF |
(7.4)
where
The envelope signal is decomposed into lowpass and highpass frequency components,
IQA = IQA LP + IQA HP
(7.5)
The original signal can now be expressed as,
IQRF =(IQA LP + IQA HP ) · IQφ
(7.6)
=IQA LP · IQφ + IQA HP · IQφ
(7.7)
Because dynamic biasing can only multiply,
µ
IQRF = IQA LP
IQA HP
IQφ + IQφ ·
IQA LP
¶
(7.8)
and the outphasing component is then,
µ
IQoutphasing = IQφ
IQA HP
1+
IQA LP
¶
(7.9)
Because dynamic biasing can only multiply, there is a low frequency component
(IQA LP ) in Equation 7.9. However as the tracking bandwidth of the envelope
tracker increases the IQA HP /IQA LP component goes to zero.
109
7.6
Summary
Digital modulation techniques used in wireless communications with radio frequency (RF) carriers can increase channel capacity, improve transmission quality,
enhance security, and provide services not possible with analog modulation [1].
Improving spectral efficiency by allowing the envelope of the RF signal to vary
with time can enhance channel capacity. Envelope variations require RFPA linearity. The power conversion efficiency of RFPAs operating in linear modes is
limited to less than 25% for signals with high envelope variations. Poor conversion efficiency leads to significant dissipated power, and shortens the lifetime of
battery operated equipment.
Polar transmitters can be used to linearize RFPAs to meet the spectral mask
for signals such as EDGE without the use of pre-distortion. In these systems the
baseband signal that contains, both amplitude and phase modulation is divided
into two signals; one that contains just the amplitude modulation and one that
contains the phase modulation with constant amplitude. The phase modulated
signal is the RF input to an RFPA, while the envelope is supplied by an envelope
tracker. The RFPA is able to reconstruct the signal due to its output voltage
dependence to the supply voltage.
To maintain a high-overall efficiency it is necessary to use a high-efficiency
envelope tracker and a high-efficiency RFPA. Switch-mode power supplies are
suitable for envelope trackers, because they can achieve high-efficiencies, are small,
light, economic and have the capability to step-up or step-down voltages. However,
for these converters losses increase with switching frequency and it is a challenge
to design fast high-efficiency SMPSs. SMPSs can be assisted with linear amplifiers
to cover larger bandwidths. The class-E mode of operation lends itself well to the
polar transmitter architecture because the output voltage across a constant load
can be linearly varied by varying the supply voltage. A 56% efficient linear polar
110
transmitter with +20dBm of output power was designed and implemented for the
EDGE modulation scheme. This is the highest efficiency reported to date for a
polar EDGE transmitter.
The work presented in this thesis illustrates the design, implementation and
performance of transmission line class-E power amplifiers at UHF and microwave
frequencies. 60-W single device amplifiers with over 80% efficiency at 370 MHz
were designed with transistor technologies such as GaN HEMTs, and SiC MESFETs. As these transistor technologies continue to mature with more power at
higher frequencies, the techniques illustrated in this thesis can be directly applied to these future devices. For even higher power levels, low loss combining
techniques need to be investigated.
111
Bibliography
[1] X. Zhang, L.E. Larson, P.M. Asbeck, Design of Linear RF Outphasing Power
Amplifiers, Norwood, MA: Artech House, Inc. 2003.
[2] S.C. Cripps, RF Power Amplifiers for Wireless Communications, 2nd Edition,
Norwood, MA: Artech House, Inc. 2006.
[3] L.R. Kahn, “Single sideband transmission by envelope elimination and restoration,” Proc. IRE, vol. 40, pp. 803-806, July 1952.
[4] M. Kazimierczuk, “Collector amplitude modulation of the class E tuned power
amplifier,” IEEE Trans. Circuits and Systems, Vol. 31, No. 6, pp. 543-549, Jun
1984.
[5] F.H. Raab and D.J. Rupp, “High-efficiency single-sideband HF/VHF transmitter based upon envelope elimintation and restoration,” in Proc. 6th Int.
Radio Systems Techniques Conf., Jul. 1994, pp. 21-25.
[6] D.F. Kimball, J. Jeong, C. Hsia, P. Draxler, S. Lanfranco, W. Nagy, K.
Linthicum, L.E. Larson, and P.M. Asbeck, “High-Efficiency Envelope-Tracking
W-CDMA Base-Station Amplifier using GaN HFETs,” IEEE Transactions on
Microwave Theory and Techniques, vol. 54, No. 11, pp. 3848-3856, November
2006.
[7] http://www.tropian.com/polarImpact.php
112
[8] J. Groe, “Polar Transmitters for Wireless Communications Groe,” IEEE Communications Magazine, vol. 45, No. 9, pp. 58-63, September 2007.
[9] F. Wang, A.H. Yang, D.F. Kimball, L.E. Larson, P.M. Asbeck, “Design of
Wide-Bandwidth Envelope-Tracking Power Amplifiers for OFDM Applications,” IEEE Transactions on Microwave Theory and Techniques, vol. 53, No.
4, pp. 1244-1255, April 2005.
[10] S. Mann, M. Beach, P. Warr and J. McGeehan, “Increasing the talk-time
of mobile radios with efficient linear transmitter architectures,” Electronics &
Communication Engineering Journal, pp. 65-76, April 2001.
[11] M.R. Elliott, T. Montalvo, B. Jeffries, F. Murden, J. Strange, A. Hill,
S. Nandipaku, and J. Harrebek, “A Polar Modulator Transmitter for
GSM/EDGE,” IEEE Journal of Solid-State Circuits, vol. 39, No. 12, pp. 21902199, December 2004.
[12] N. Wang, N.D. Lopez, V. Yousefzadeh, J. Hoversten, D. Maksimovic and Z.
Popovic, “Linearity of X-band Class-E Power Amplifiers in a Digital Polar
Transmitter,” IEEE International Microwave Symposium Digest, June 2007.
[13] N. Wang, X. Peng, V. Yousefzadeh, D. Maksimovic, S. Pajic, Z. Popovic,
“Linearity of X-Band Class-E Power Amplifiers in EER Operation,” IEEE
Trans. on Microwave Theory and Techn., vol. 53, No. 3, Mar. 2005.
[14] V. Yousefzadeh, E. Alarcon, and D. Maksimovic, “Band Separation and Efficiency Optimization in Linear-Assisted Switching Power Amplifiers,” 37th
IEEE Power Electr. Specialists Conf., 2006, pp. 1-7, June 2006.
[15] N.D. Lopez, X. Jiang, D. Maksimovic, and Z. Popovic, “Class-E Power Amplifier in a Polar EDGE Transmitter,” IEEE International Microwave Symposium Digest, pp. 785-788, June 2006.
[16] X. Jiang, N.D. Lopez, and D. Maksimovic, “A Switched Mode Envelope
113
Tracker for Polar EDGE Transmitter,” IEEE 37th Annual Power Electronics
Specialists Conference, pp. 1-7, June 2006.
[17] N.D. Lopez, X. Jiang, D. Maksimovic, and Z. Popovic, “A High-Efficiency
Linear Polar Transmitter for EDGE,” IEEE Radio and Wireless Symposium
Digest, pp. 199-202, January 2008.
[18] N.D. Lopez, J. Hoversten, Matthew Poulton, and Z. Popovic, “A 65-W HighEfficiency UHF GaN Power Amplifier,” submitted to IEEE International Microwave Symposium Digest, June 2008.
[19] N.D. Lopez, J. Hoversten, and Z. Popovic, “Wide-Bandgap High-Efficiency
UHF Power Amplifiers,” to be submitted to IEEE Trans. on Microwave Theory and Techniques.
[20] M. Albulet, RF Power Amplifiers, Atlanta, GA: Noble Publishing Corporation 2001.
[21] H.L. Krauss, C.W. Bostonian, and F.H. Raab, Solid State Radio Engineering,
New York, NY: John Wiley & Sons 1980.
[22] M. Albulet, RF and Microwave Power Amplifiers Design, New York, NY:
McGraw-Hill 2005.
[23] T. Mader and Z. Popovic, “The Transmission-Line High-Efficiency Class-E
Amplifier,” IEEE Microwave Guided Wave Letters, vol. 5, No. 9, pp. 290-292,
September 1995.
[24] R.W. Erickson, and D. Maksimovic, Fundamentals of Power Electronics, 2nd
Edition, Norwell, MA: Kluewer Academic Pulishers 2001.
[25] N.O. Sokal and A.D. Sokal, “Class-E a new class of high-efficiency tuned
114
single-ended switching power amplifiers,” IEEE Journal of Solid-State Circuits, Vol. SC 10, pp. 168-176, June 1975.
[26] F.H. Raab, “Idealized operation of class-E tuned power amplifier,” IEEE
Trans. Circuits Syst. vol. CAS-24, No. 12, pp. 725-735, Dec. 1977.
[27] F. H. Raab and N. O. Sokal, “Transistor power losses in the class E tuned
power amplifer,” IEEE Journal of Solid-State Circuits, Vol. SC- 13, NO. 6,
pp. 912-914, Dec. 1978.
[28] M. Kazimierczuk, and K. Puczko, “Exact analysis of class E tuned power
amplifier at any Q and switch duty cycle,” IEEE Trans. Circuits and Systems,
Vol. 34, No. 2, pp. 149-159, Feb 1987.
[29] T.B. Mader, E.W Bryerton, M. Markovic, M. Forman, M., and Z. Popovic,
“Switched-mode high-efficiency microwave power amplifiers in a free-space
power-combiner array,” IEEE Trans. on Microwave Theory and Techn., Vol.
46, No. 10, Part 1, Oct. 1998, pp. 1391 - 1398.
[30] S. Pajic, N. Wang, P. Watson, T. Quach, and Z. Popovic, “X-Band TwoStage High-Efficiency Switched-Mode Power Amplifiers,” IEEE Trans. on Microwave Theory and Techn., vol. 53., No. 9., September 2005.
[31] H. Zirath, D.B. Rutledge, “LDMOS VHF class-E power amplifier using
a high-Q novel variable inductor,” IEEE Trans. on Microwave Theory and
Techn., Vol. 47, No. 12, pp. 2534-2538, Dec. 1999.
[32] T.B. Mader, Quasi-Optical Class-E Power Amplifiers, PhD Dissertation,
Univ. of Colorado, Boulder, CO 1995.
[33] M.D. Weiss, Switched-Mode Microwave Circuits for High-Efficiency Transmitters, PhD Dissertation, Univ. of Colorado, Boulder, CO 2001.
115
[34] S.A. Pajic, Robust Design Methodology for Class-E Amplifiers for Microwave
Applications, PhD Dissertation, Univ. of Colorado, Boulder, CO 2005.
[35] N. Wang, X-Band Class-E Power Amplifiers with Dynamic Bias Control,
PhD Dissertation, Univ. of Colorado, Boulder, CO 2006.
[36] J.A. Hagerty, F. Helmbrecht, W. McCalpin, R. Zane, and Z. Popovic, “Recycling ambient microwave energy with broadband antenna arrays,” IEEE
Trans. Microwave Theory and Techn., pp. 1014-1024, March 2004.
[37] J.A. Hagerty, Nonlinear Circuits and Antennas for Microwave Energy Conversion, PhD Dissertation, Univ. of Colorado, Boulder, CO 2003.
[38] S.V. Schell, “Implementation Effects on GSM’s EDGE Modulation,” Tropian
Inc., Technology White Paper.
[39] J.N. Kitchen, I. Deligoz, S. Kiaei, and B. Bakkaloglu, “Polar SiGe Class E
and F Amplifiers Using Switch-Mode Supply Modulation,” IEEE Trans. on
Microwave Theory and Techniques, Vol. 55, No. 5, pp. 845-856, May 2007.
[40] S. Hong, Y.Y. Woo, I. Kim, J. Kim, J. Moon, H.S. Kim, J.S. Lee, and B.
Kim, “High Efficiency GaN HEMT Power Amplifier optimized for OFDM EER
Transmitter,” International Microwave Symposium, 2007.
[41] A. Diet, C. Berland, M. Villegas, and G. Baudoin, “EER architecture specifications for OFDM transmitter using a class E amplifier,” IEEE Microwave
and Wireless Components Letters, Vol. 14, No. 8, pp. 389-391, Aug. 2004.
[42] F.H. Raab, B.E. Sigmon, R.G. Myers, and R.M. Jackson, “L-band transmitter using Kahn EER technique,” IEEE Trans. on Microwave Theory and
Techniques, Vol. 46, No. 12, Part 2, pp. 2220-2225, Dec. 1998.
116
[43] V. Yousefzadeh, Digitally Controlled Power Converters for RF Power Amplifiers, PhD Dissertation, Univ. of Colorado, Boulder, CO 2006.
[44] V. Yousefzadeh, E. Alarcon, D. Maksimovic, D., “Efficiency optimization in
linear-assisted switching power converters for envelope tracking in RF power
amplifiers,” IEEE International Symposium on Circuits and Systems, May
2005 Pages:1302 - 1305 Vol. 2
[45] X. Jiang, Envelope-tracking Power Supplies for RF Power Amplifiers in
Portable Applications, PhD Dissertation, Univ. of Colorado, Boulder, CO
2006.
[46] J. Cha, J. Yi, J. Kim, B. Kim, “Optimum design of a predistortion RF power
amplifier for multicarrier WCDMA applications,” IEEE Trans. on Microwave
Theory and Techniques, Vol. 52, No. 2, pp. 655-663, Feb. 2004.
[47] J. Kim, Y.Y. Woo, J. Moon, and B.Kim, “A New Wideband Adaptive Digital
Predistortion Technique Employing Feedback Linearization,” IEEE Trans. on
Microwave Theory and Techniques, Vol. 56, No. 2, pp. 385-392, Feb. 2008.
[48] J.C. Pedro, J.A. Garcia, P.M. Cabral, “Nonlinear Distortion Analysis of Polar
Transmitters,” IEEE Trans. on Microwave Theory and Techniques Vol. 55, No.
12, Part 2, pp. 2757-2765, Dec. 2007.
[49] L. Ding, G.T. Zhou, D.R. Morgan, M. Zhengxiang, J.S. Kenney, J. Kim,
and C.R. Giardina, “A robust digital baseband predistorter constructed using
memory polynomials,” IEEE Trans. on Communications, Vol. 52, No. 1, pp.
159-165, Jan 2004.
[50] http://www.skyworksinc.com/rfsbu helios II edge.asp
[51] http://www.triquint.com/prodserv/more info/?prod id=1015&page id=154H
117
[52] http://www.analog.com/en/prod/0,2877,AD6546,00.html
[53] http://www.anadigics.com
[54] J. Popp, D.Y.C. Lie, F. Wang,D. Kimball, and L. Larson, “Fully-integrated
highly-efficient RF Class E SiGe power amplifier with an envelope-tracking
technique for EDGE applications,” IEEE Radio and Wireless Symposium, pp.
231-234, Jan. 2006.
[55] D.Y.C. Lie, J.D. Popp,J.F. Rowland,A.H. Yang, D. Kimball, “Highly Efficient
and Linear Class E SiGe Power Amplifier Design,” 8th International Conference on Solid-State and Integrated Circuit Technology, pp1526-1529, 2006.
[56] D.Y.C Lie, J.D. Popp, F. Wang, D. Kimball, and L.E. Larson, “Linearization of Highly-Efficient Monolithic Class E SiGe Power Amplifiers with
Envelope-Tracking (ET) and Envelope-Elimination-and-Restoration (EER) at
900MHz,” 6th IEEE Dallas Circuits and Systems Workshop on System-onChip, pp. 1-4, Nov. 2007.
[57] http://www.triquint.com/docs/t/TGF4240/4240.pdf
[58] K.A. Remley, D.F. Williams, D.M. Schreurs, J. Wood, “Simplifying and interpreting two-tone measurements,” IEEE Trans. on Microwave Theory and
Techn., vol. 52., No. 11., September 2004.
[59] Measurement System Operation Manual, Focus Microwaves, 1999.
[60] Product Note 52, Prematching Tuners for Very High VSWR and Power Load
Pull Measurements, Focus Microwaves, 1999.
[61] http://www.focus-microwaves.com/
118
[62] D. D. Poulin, J. R. Mahon, and J. P. Lanteri, “A high power on- wafer
pulsed active load pull system,” IEEE Transactions on Microwave Theory
and Techniques, Vol. 40, pp. 2412-2417, Dec. 1992.
[63] W. Bischof, “Variable Impedance Tuner for MMICs,” IEEE Microwave
Guided Wave Lett., Vol. 4, pp. 172-174, June 1994.
[64] C. E. Collins, R. D. Pollard, and R. E. Miles, “A Novel MMIC Source
Impedance Tuner for On-Wafer Microwave Noise Parameter Measurements,”
IEEE Microwave and Millimeter-Wave Monolithic Circuits Symp., June 1996,
pp. 123-126.
[65] C. E. McIntosh, R. D. Pollard, and R. E. Miles, “Novel MMIC SourceImpedance Tuners for On-Wafer Microwave Noise-Parameter Measurements,”
IEEE Trans. Microwave Theory Tech., Vol. 47, pp. 125-131, Feb. 1999.
[66] J. de Mingo, A. Valdovinos, A. Crespo, D. Navarro, and P. Garcia, “An RF
Electronically Controlled Impedance Tuning Network Design and Its Application to an Antenna Input Impedance Automatic Matching System,” IEEE
Trans. Microwave Theory Tech., Vol. 52, pp. 489-497, Feb. 2004.
[67] H. T. Kim, S. Jung, K. Kang, J. H. Park, Y. K. Kim, and Y. Kwon, “LowLoss Analog and Digital Micromachined Impedance Tuners at the Ka-Band,”
IEEE Trans. Microwave Theory Tech., Vol. 49, pp. 2394-2400, Dec. 2001.
[68] P. Bell, MEMS-Reconfigurable Microwave Power Amplifiers, PhD Dissertation, Department of Electrical and Computer Engineering, University of Colorado, Boulder 2006.
[69] G. Gonzalez, Microwave Transistor Amplifiers, Upper Saddle River, NJ:
Prentice Hall 1997.
[70] Y-S Lee, and Y-H Jeong, “A High-Efficiency Class-E GaN HEMT Power
119
Amplifier for WCDMA Applications,” IEEE Microw. and Wireless Letters,
Vol. 17, No. 8, Augutst 2007, pp. 622-624.
[71] Y. Xu, S. Gao, S. Heikman, S. Long, U. Mishra, and R. York, “A HighEfficiency Class-E GaN HEMT Power Amplifier at 1.9 GHz,” IEEE Microw.
and Wireless Letters, Vol. 16, No. 1, January 2006, pp. 22-24.
[72] T. Kikkawa, T. Maniwa, H. Hayashi, M. Kanamura, S. Yokokawa, M. Nishi,
N. Adachi, M. Yokoyama, Y. Tateno, and K. Joshin, “An Over 200-W Output
Power GaN HEMT Push-Pull Amplifier with High Reliability,” IEEE MTT-S
Int. Microw. Symp. Dig., June 2004, pp. 1347-1350.
[73] M. Franco, and A. Katz, “Class-E Silicon Carbide VHF Power Amplifier,”
IEEE MTT-S Int. Microw. Symp. Dig., June 2007, pp. 19-22.
[74] R. Trew, “SiC and GaN Transistors - Is There One Winner for Microwave
Power Applications?,” Proceedings of the IEEE, vol. 90., No. 6., June 2002,
pp. 1032-1047.
[75] http://www.nitronex.com/products.html
[76] F. Schwierz, and J. Liou, Modern Microwave Transistors, Hoboken, NJ: John
Wiley & Sons, 2003.
[77] http://www.analog.com/en/
120
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising