5a 059 PDF

Mixed-Signal Instrumentation
for Large-Signal Device

Characterization and
Modelling
Mauro Marchetti
Mixed-Signal Instrumentation for
Large-Signal Device Characterization and
Modelling
PROEFSCHRIFT
ter verkrijging van de graad van doctor

aan de Technische Universiteit Delft,
op gezag van de Rector Magnificus Prof. ir. K. C. A. M. Luyben,
voorzitter van het College voor Promoties,
in het openbaar te verdedigen
op maandag 25 november 2013 om 15.00uur
door
Mauro MARCHETTI
Dottore in Ingegneria Elettronica
van Università degli Studi di Napoli “Federico II”, Italië
geboren te Napoli, Italië
Dit proefschrift is goedgekeurd door de promotor:
Prof. dr. J. R. Long
Samenstelling promotiecommissie:
Rector Magnificus, voorzitter

Prof. dr. J.R. Long, Technische Universiteit Delft, promotor
Dr. ing. L.C.N. de Vreede, Technische Universiteit Delft, toegevoegd promotor
Prof. dr. L.K. Nanver, Technische Universiteit Delft
Prof. dr. V. Teppati, Eidgenössische Technische Hochschule Zürich
Prof. dr. H.F.F. Jos, Chalmers University of Technology
Prof. dr. D. Schreurs, Katholieke Universiteit Leuven
Dr. M. Vanden Bossche, National Instruments Belgium NV/SA
Prof. dr. A. Neto, Technische Universiteit Delft, reservelid
Mauro Marchetti,
Mixed-Signal Instrumentation for Large-Signal Device Characterization and
Modelling,
Ph.D. Thesis Delft University of Technology,
with summary in Dutch.
Keywords: load-pull, isothermal, large-signal, complex modulated signals, dy-

namic range, linearity, device, power amplifier, radio frequency (RF), base
station, handset, high power.
ISBN: 978-94-6203-471-6
Copyright c 2013 by Mauro Marchetti

All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted in any form or by any means without the
prior written permission of the copyright owner.
Cover design by twoeight studio, www.twoeight.it.

Printed by CPI, Wohrmann Print Service, Zutphen, the Netherlands.
To my family
Contents
1 Introduction 1
1.1 Trends in wireless communication . . . . . . . . . . . . . . . . . 1
1.2 Requirements on the power amplifier . . . . . . . . . . . . . . . 4
1.3 The need for advanced measurement tools . . . . . . . . . . . . 6
1.4 Thesis objectives . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2 Isothermal Measurement Systems 11

2.1 Pulsed measurements fundamentals . . . . . . . . . . . . . . . . 12
2.2 System configuration . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2.1 Pulsed VNA architecture . . . . . . . . . . . . . . . . . 17
2.2.2 Pulsed-DC measurements . . . . . . . . . . . . . . . . . 22
2.3 System performance . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.1 Pulsed VNA dynamic range . . . . . . . . . . . . . . . . 24
2.3.2 DC measurement accuracy . . . . . . . . . . . . . . . . 25
2.4 Measurements examples . . . . . . . . . . . . . . . . . . . . . . 27
2.4.1 LDMOS devices . . . . . . . . . . . . . . . . . . . . . . 27
2.4.2 III-V HBT devices . . . . . . . . . . . . . . . . . . . . . 29
2.4.3 SiGe HBT devices . . . . . . . . . . . . . . . . . . . . . 29
2.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3 Source and Load-pull Architectures 33

3.1 Passive load-pull . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2 Active load-pull . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.2.1 Closed-loop active load-pull . . . . . . . . . . . . . . . . 38
3.2.2 Open-loop active load-pull . . . . . . . . . . . . . . . . . 39
3.3 Electrical delay in load-pull systems . . . . . . . . . . . . . . . 40
3.4 Injection power and load amplifier linearity . . . . . . . . . . . 42
3.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
i
ii CONTENTS
4 Mixed-Signal Active Load-Pull with Realistic Wideband Mod-

ulated Signals 49
4.1 The wideband, open-loop load-pull approach . . . . . . . . . . 49
4.2 System configuration . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2.1 Signal generation . . . . . . . . . . . . . . . . . . . . . . 54
4.2.2 Data acquisition . . . . . . . . . . . . . . . . . . . . . . 55
4.2.3 System operating frequency and bandwidth . . . . . . . 55
4.2.4 System calibration . . . . . . . . . . . . . . . . . . . . . 56
4.3 Signal processing for modulated signal measurements . . . . . . 57
4.3.1 Signal generation . . . . . . . . . . . . . . . . . . . . . . 57
4.3.2 Data acquisition . . . . . . . . . . . . . . . . . . . . . . 58
4.3.3 Working example . . . . . . . . . . . . . . . . . . . . . . 59
4.4 System performance . . . . . . . . . . . . . . . . . . . . . . . . 59
4.4.1 Measurement linearity and dynamic range . . . . . . . . 61
4.4.2 Active load dynamic range . . . . . . . . . . . . . . . . 61
4.5 Measurement results . . . . . . . . . . . . . . . . . . . . . . . . 62
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5 High-Speed, High-Power, Fully-Controlled, Multi-dimensional

Load-Pull Parameter Sweeps 67
5.1 Generation and measurement of load and source terminations . 68
5.1.1 Measurement examples and results . . . . . . . . . . . . 70
5.2 High-power, real-time pulsed-RF measurements . . . . . . . . . 71
5.2.1 High-power measurement examples . . . . . . . . . . . . 74
5.3 High-power measurements with modulated signals . . . . . . . 75
5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6 Measurement of Time-Domain Waveforms 79

6.1 Time-domain waveform measurement fundamentals . . . . . . . 79
6.1.1 Waveform reconstruction with VNAs . . . . . . . . . . . 82
6.1.2 Waveform reconstruction with sub-sampling techniques 83
6.2 Waveform reconstruction on the mixed-signal load-pull system 84
6.2.1 System phase measurement repeatability . . . . . . . . . 85
6.2.2 Phase reference specifications . . . . . . . . . . . . . . . 85
6.2.3 Measurement example . . . . . . . . . . . . . . . . . . . 96
6.3 Waveform reconstruction for closely–spaced multi-tone signals . 99
6.3.1 System description . . . . . . . . . . . . . . . . . . . . . 99
6.3.2 Phase calibration . . . . . . . . . . . . . . . . . . . . . . 102
6.3.3 Phase stability results . . . . . . . . . . . . . . . . . . . 104
6.3.4 Waveform reconstruction . . . . . . . . . . . . . . . . . 108
6.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
CONTENTS iii
7 Application Examples 109

7.1 Out-of-band linearity optimization . . . . . . . . . . . . . . . . 109
7.1.1 HBT device linearity optimization . . . . . . . . . . . . 110
7.1.2 Device characterization for LTE applications with base-
band, fundamental and harmonic
wideband impedance control . . . . . . . . . . . . . . . 113
7.2 High-power device measurements for base-station applications . 119
7.3 Device characterization for high efficiency power amplifier design122
7.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8 Conclusions and Future Work 127

8.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
8.2.1 Supporting new generation signals and systems . . . . . 128
8.2.2 Supporting high-efficiency and high-linearity advanced
PA design . . . . . . . . . . . . . . . . . . . . . . . . . . 130
8.2.3 Supporting next generation device modelling . . . . . . 130
8.3 Future trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
Bibliography 133
Summary 145
Samenvatting 149
List of Publications 153
Acknowledgments 157
About the Author 161

Chapter 1
Introduction
The appearance of modern smart phones and their quick penetration to con-
sumer markets has drastically changed the way we communicate as a society.
The manner in which social networks, for example, are influencing every day
life has been magnified by the capabilities of truly portable, always connected
devices. While just a few years ago the cellular phone was only used for voice
calls and SMS, the newer generation smart phones are reaching more and more
the capabilities of pocket size personal computers which are always connected
to the network. Consequently, applications ranging from internet browsing
to video chatting, from high-quality video streaming to internet calls and so-
cial networks, which were previously tied to personal computers alone, have
become accessible anytime and from almost anywhere to the user.
As a result, it is becoming increasingly more difficult to imagine our life
without a mobile internet connection. This trend is reflected in the 2012 Cisco
World Technology Report [1], which indicates that more than 60 % of “Gen
Y” (18 to 30 year olds) do not even wait to get out of bed in the morning
before checking their smart phones, while more than 40 % of them (including
the author of this thesis) would feel anxious, almost lost, if they couldn’t check
their smart phones constantly.
1.1 Trends in wireless communication

As more and more people and devices get connected, the amount of data traffic
increases exponentially. Consequently, while according to the United Nations
expectations there will be 7.6 billion people in 2017, in that same year, based
on the scenario depicted by the Cisco visual networking index (VNI) global
mobile data traffic forecast 2012-2017 [2], there will be more internet access
1
2 Introduction
Figure 1.1: Mobile data traffic forecast vs. time in exabytes per month.
Source Cisco VNI Mobile Forecast [2].
points than people on this planet. Moreover, driven by applications such as

high definition video streaming or cloud computing, the mobile data traffic
will grow 13 times in the coming five years, increasing from the 0.9 exabytes
per month of 2012 to the 11.2 exabytes per month of 2017, as shown in Fig.
1.1.
The exponential increase in the amount of mobile data transferred and
the consequent request for higher data rates in communication systems has
led, over the past decade, to the introduction of 3G and 4G communication
standards, such as UMTS and LTE, which employ larger bandwidths and
more complex modulation schemes to achieve higher data rate capabilities.
The development of data rates over time in wireless networks is depicted in
Fig. 1.2. It can be seen that the transmission speed has been increasing by
approximately a factor of 10 every 5 years [3].
Another important aspect in the development of mobile communications
and the related infrastructure is the electricity consumption of mobile net-
works. From an economic point of view this is a primary issue, since the
biggest portion of the operational expenditures (OPEX) of mobile networks
resides into energy costs. In fact, it is estimated that the worldwide mobile
network OPEX for electricity is more than 10 billion dollars [4]. Furthermore,
due to the growing awareness of the environmental impact of wireless com-
munication systems, reducing their energy consumption is becoming a more
crucial point. According to some figures [5], information and communication
technology (ICT) is already responsible for about 2 to 4 % of the world-wide
1.1 Trends in wireless communication 3
10000
802.11ac
LTE
1000 Advanced
802.11n
Peak Data Rate [Mb/s]
LTE
100 WiMAX
HSPA+
802.11ag
10 802.11b HSPA Mobile
HSDPA WLAN
802.11
1
UMTS R99
EDGE
0.1
GPRS
0.01 GSM
1990 1995 2000 2005 2010 2015
Year
Figure 1.2: Development of data rates over time in wireless networks [3].
CO2 emissions and its contribution is growing by 16 − 20 % yearly, doubling

every 4 to 5 years. To put things into perspective, the CO2 emissions caused
by ICT correspond already to those caused by all international air traffic [3],
and it is equivalent to the annual greenhouse gas emissions from about 8
million cars [4]. While in 2002, the contribution of mobile communications
corresponded only to about 10 % of the total amount for ICT, this value is
expected to grow three times by 2020 [6].
However, when considering a wireless network, it is estimated that 60 to
80 % of the total energy is consumed by the base-station [4], the fixed radio
station which relays the information to and from the mobile devices. As a
consequence, cellular network operators have been placing particular emphasis
into trying to minimize the energy consumption of base-stations, by constantly
evaluating different strategies, ranging from hardware improvements, to better
cooling approaches, to advanced operation and deployment schemes. A few
potential opportunities which are currently under study, and that can allow a
significant reduction of the energy expenditure of wireless networks involve:
• Operational strategies such as dynamic shut down of some base-stations

during off-peak hours [4].
• Network architectures based on low power microcells, instead of high

power macrocells [3].
4 Introduction
• The increase of the transmission range of base-stations by using smart

antenna technology [7], such as multiple receiving and transmitting an-
tennas (Multiple-Input and Multiple-Output).
• The design and implementation of novel and more energy efficient hard-
ware.
Nevertheless, when considering the architecture of any base-station, the

most “power-hungry” component in these networks is the power amplifier
(PA) which is boosting the signal power before this is transmitted over the
air [8]. This is true not only because of the power consumption of the amplifier
itself, but also due to the energy which is necessary to cool it. Therefore, it is
not surprising that, despite all the different available strategies for minimizing
energy consumption, there is currently, worldwide, an enormous interest in
designing more energy efficient PAs, while at the same time improving their
signal handling capabilities in terms of bandwidth and linearity [8].
1.2 Requirements on the power amplifier

Society’s need for higher data rates and reduced energy consumption translates
into very stringent technology requirements for the underlying hardware.
First of all, modern advanced modulation schemes employ both phase and
amplitude modulation to increase the transmission data-rate. As a conse-
quence, the signal peak-to-average ratio, which is the ratio between the peak
power and the average power of the signal, is high (e.g., ≈ 9.8 dB for a typical
W-CDMA signal [9] for downlink transmission). This requires that the PA is
as efficient as possible not only close to compression, but also in power back-
off where statistically the PA is transmitting most of the time, and therefore
most of the energy is consumed. This can be easily explained by looking, as
an example, at the probability distribution function of a W-CDMA signal in
Fig. 1.3, which shows how the signal instantaneous power has the highest
probability to be about 9.8 dB in back-off with respect to the peak power.
Secondly, to operate correctly, the wireless transmitter should be able to
broadcast the desired signals without generating significant in-band distortion
and without interfering with the other channels. This means that the signal
should not leak considerably into the other transmission channels and there-
fore its spectrum should not exceed the levels given by the so called spectral
mask (see Fig. 1.4), which is defined by the communication standard. This
is achieved either by designing PAs which are extremely linear over a wide
frequency range, or by linearizing the final transmitted signal by digitally pre-
distorting the data at the source to compensate for the PA distortion.
1.2 Requirements on the power amplifier 5
0.08
0.07
0.06
0.05
Density
0.04
0.03
0.02
0.01
0
−30 −25 −20 −15 −10 −5 0
Normalized Signal Power [dB]
Figure 1.3: Probability Density Function (PDF) for a downlink W-CDMA

signal [9].
−50
Spectral Power Density [dBm/Hz]
−60
−70
−80
−90
−100
−110
2392 2402 2412 2422 2432 2442 2452 2462 2472
Frequency [MHz]
Figure 1.4: Example of a spectral mask of a 802.11n WiFi signal. The

transmitted signal should not exceed the levels specified by the spectral
mask to avoid interfering with the other channels [10].
6 Introduction
To complicate matters, the demands for high efficiency and high linearity
are normally in conflict with each other. The linearity requirement, in fact,
could very simplistically be met by operating the PA well below its saturation
point, in its linear region of operation. However, since the peak efficiency of a
PA is normally achieved at peak output power, this would drastically reduce
PA efficiency [11]. On the other hand, driving the PA closer to its saturation
point, in its nonlinear region of operation, would certainly increase efficiency,
but would result in nonlinear distortion effects.
For this reason, higher complexity amplifier concepts like Doherty [12],
outphasing [13] and envelope tracking [14], which make use of dynamic varia-
tions of the load and voltage conditions for the output stage transistor(s), are
becoming increasingly more popular, since they can relax the compromise be-
tween efficiency and linearity. It is clear that all these constraints on efficiency
and linearity, in combination with these dynamic changes to the operating
conditions for the active devices, make designing a power amplifier capable of
meeting today’s communication industry standards a challenging task, which
requires accurate, fast and versatile measurement tools.
1.3 The need for advanced measurement

tools
Looking in detail at the hardware design of state-of-the-art PAs for telecom-
munications, it can be stated that, for a given transistor technology, the per-
formance parameters of a power amplifier depend entirely on how well the
active devices, embedded in these circuits, are matched (i.e., electrically ter-
minated), or embedded in their surrounding circuitry. In other words, active
devices employed in these applications can only perform well if the proper
electrical terminations are provided around the fundamental frequency band,
as well as at all of the other frequencies (i.e., harmonics and baseband), be-
cause they strongly influence the device operation in terms of linearity and
efficiency [15]. Therefore, a PA designer has to find the optimum trade-off
among the different performance parameters, such as efficiency, linearity and
output power, by “searching” the multi-dimensional design space across all of
the different tunable parameters, such as bias, source and load impedance at
baseband, fundamental and harmonic frequencies.
To accomplish this difficult task, the PA designer has to rely on either
a model of the active device to perform the design in a circuit simulator,
or on measurement data coming from a load-pull measurement system (see
Chapter 3), which can measure the transistor performance parameters, such
1.3 The need for advanced measurement tools 7
as output power, efficiency, linearity, and many others, while tuning the various
impedances offered to the active device.
Due to the aforementioned need to operate the transistor close to its non-
linear region, the models used for PA design need to correctly represent the
large-signal and distortion behaviors of the active device. Moreover, to prop-
erly predict the transistor performance when employing wideband complex
modulated signals, effects such as bias or temperature-induced memory [16]
must be included in the model. Finally, the transistor model needs to be
verified.
Therefore, to enable the extraction and the verification of accurate large-
signal transistor models, there is a great need for advanced large-signal RF
and microwave measurement tools. In fact, while linear device characterization
(e.g., small-signal S-parameters) is mainly useful to extract the small-signal
behavior of the transistor, pulsed DC, RF and large-signal measurements are
mandatory to extract the nonlinear behavior of the active device and to verify
the model. As an example, pulsed-DC and S-parameters measurements are re-
quired to characterize the active device under isothermal conditions (maintain-
ing the core of the transistor at a constant temperature), in order to develop
device models which include self-heating or trapping effects [17–20]. Moreover,
load-pull measurements are necessary to obtain realistic large-signal data (e.g.,
gain and phase distortion data), for the extraction of large-signal behavioral
models [21], or for model validation.
Additionally, since the model creation process can be a lengthy and cum-
bersome task, it is in many cases common practice to directly use load-pull
measurement data for all activities related to PA design, from the technology
development to the actual power amplifier design. Therefore it is of fundamen-
tal importance that the measurement data extracted from these characteriza-
tion systems is able to predict with the highest level of accuracy the active
device behavior in the final application environment (i.e., circuit termination
and signal stimulus).
However, in spite of the significant progress made over the last decade in
the field of RF measurements, there is still a great need for more accurate and
advanced isothermal and large-signal characterization systems. In particular,
pulsed-DC and S-parameter measurement systems still suffer from limitations
in the minimum pulse widths that can be used and in measurement accu-
racy [22–26]. These impairments limit the usability of these systems for the
characterization of new device technologies under truly isothermal conditions.
A comparable situation exists in load-pull measurement techniques. Al-
though significant advances were made during past years [27], bringing conven-
tional load-pull setups employing passive mechanical tuners and power meters
8 Introduction
to advanced load-pull setups with large-signal vector measurements, there is

still a big discrepancy between these traditional single-tone or two-tones char-
acterization methods, and the actual signal conditions present in telecommu-
nication systems operating with standard compliant signals like WCDMA or
LTE.
As a consequence, the development and testing of new power amplifier
technologies (e.g., LDMOS, GaN, SiGe, etc.) and their application in PAs
are significantly hampered, yielding costly and lengthy technology and design
iterations.
1.4 Thesis objectives

Having reviewed in Section 1.1 and 1.2 the current trends in the telecommu-
nication industry and how these specifically reflect on the PA development,
two major areas of improvements for PAs can be identified: the increase in
available bandwidth, and the improvement of the overall average efficiency for
modern communication signals, while at the same time satisfying the linearity
specifications of the communication standard under consideration.
This thesis concentrates on the development of advanced large-signal mea-
surement and characterization tools that support transistor technology de-
velopment, model extraction and validation, and design of PAs that address
these new communication standards. In particular, the measurement systems
which will be described throughout this thesis work aim to extend the limits of
state-of-the-art, large-signal characterization systems in terms of bandwidth,
peak output power, speed and multi-functional capabilities.
For this purpose, an innovative mixed-signal approach, which replaces tra-
ditional analog techniques with digital data acquisition and digital signal gen-
eration and advanced digital signal processing, will be followed. This method
provides, compared to traditional techniques, a much higher flexibility, func-
tionality, performance and speed in many different applications.
In particular, to support compact model developments, and to facilitate the
characterization of RF / microwave devices under truly isothermal conditions,
the first part of this thesis work focuses on the development of an innovative
pulsed-DC and pulsed-RF system, capable of providing very short DC and
RF pulses (down to 200 ns), with arbitrarily low duty-cycle and an improved
dynamic range. The setup will be used to characterize several different PA
device technologies to gain insight into their electrical / thermal behavior.
The second, and dominant part of this thesis work introduces a revolution-
ary active harmonic load-pull approach to enable for the first time the charac-
terization of RF and microwave transistors under realistic (circuit-like) condi-
1.5 Thesis outline 9
tions with wideband modulated signals. By further exploiting this technique,

a system will be presented to allow extremely high-speed PA device character-
ization of up to 1000 measurement points per minute, with impedance control
of up to three harmonics, and the capability to measure RF voltage and cur-
rent waveforms at the device under test. The usefulness of the realized system
will be evaluated by applying the newly developed load-pull characterization
system to several relevant application examples.
1.5 Thesis outline

This thesis is organized as follows.
In Chapter 2, the theory and the requirements for pulsed-RF and pulsed-
DC measurements are discussed. Moreover, a new isothermal measurement
approach with its hardware implementation is introduced, providing the ability
to measure with DC and RF pulses as low as 200 ns, while featuring a high
dynamic range under pulsed-RF conditions. The realized system performance
is discussed in detail through a set of benchmarks, and some measurement
examples on active devices are provided.
In Chapter 3, conventional passive and active source and load-pull archi-
tectures are reviewed. The basic limitations of conventional load-pull mea-
surement systems are examined, with particular attention to the problems
arising when characterizing devices with wideband complex modulated sig-
nals. Moreover, the requirements of active load-pull for performing high power
measurements with complex modulated signals are also explained.
In Chapter 4, a new active harmonic load-pull system based on a mixed-
signal approach is developed and described in detail. The system and the un-
derlying measurement concepts were developed during this thesis work specif-
ically to solve the problems of conventional load-pull setups when dealing
with wideband modulated signals. In its current implementation, it enables
the measurement of active devices up to 120 MHz of modulation bandwidth.
Measurement data highlighting the system performance, and measurement
results on active devices are presented.
In Chapter 5, a new approach for enabling high-speed multidimensional
source and load-pull parameter sweeps is introduced. The method described
allows any combination of multiple parameters (e.g., input power and/or
fundamental and harmonic load impedance) to be swept, at a very high
speed, while maintaining all other parameters (e.g., second harmonic source
impedance) accurately controlled to a user-defined constant value. Moreover,
several measurements are reported, with particular emphasis on the high-
power capabilities of the system, both in CW as well as under modulated
10 Introduction
signal excitations.
In Chapter 6, the basic theory behind the measurement of the high fre-
quency time-domain voltage and current waveforms at the device reference
planes is discussed. An extension of the mixed-signal load-pull system de-
scribed in the previous chapters is presented, with particular attention on the
requirements of the calibration device used for the system calibration. Finally,
an approach to time-domain waveform analysis of multi-tone signals which are
closely spaced in frequency is introduced.
In Chapter 7, several examples of applications that highlight the most
unique capabilities of the system described are reported. In particular an out-
of-band linearity optimization of an HBT device, the characterization of a GaN
device for high efficiency PA design, and some high power device measurements
for base-station applications are described.
Finally, in Chapter 8 some conclusions are drawn, with recommendations
for future work.
Chapter 2
Isothermal Measurement
Systems
Isothermal measurement systems are typically used to evaluate the perfor-

mance of microwave devices under non-continuous (pulsed-RF and pulsed-DC)
operation. These measurements are essential when dealing with pulsed-RF
applications like radar or burst-mode transmitters, or in semiconductor de-
vice characterization when dissipation problems need to be avoided. From a
modelling point of view, isothermal characterization is indispensable in the
development of device models which include self-heating and trapping effects.
Currently, these advanced device models [17–20] receive high interest in litera-
ture since they provide a more accurate prediction for the linearity degradation
of PA’s due to memory effects. Typically, the extraction of the parameters for
these models involves the characterization of the bias dependent small-signal
parameters under pulsed-bias and pulsed-RF conditions. In general, activities
such as database-oriented modelling strongly benefit from improvements in
measurement accuracy and speed, when aiming for high-quality device data
which is densely spaced versus bias. Over time, several implementations of
isothermal setups have been reported [22–26], however, previously reported
isothermal measurement setups only partly fulfill the above mentioned re-
quirements since their core hardware was based on equipment, which was not
fully optimized for the pulsed conditions used in isothermal measurements.
This resulted in restrictions in pulse width, duty cycle, dynamic range and
measurement speed.
In this chapter, the theory and the requirements for pulsed-RF and pulsed-
DC measurements are discussed. Moreover, a new isothermal measurement
approach with its hardware implementation is introduced, providing the ability
to measure with DC and RF pulses as low as 200 ns, while featuring a high
11
12 Isothermal Measurement Systems
Pulsed width (PW)
Carrier frequency fC
Pulse repetition interval (PRI)
Figure 2.1: Pulsed-RF signal.
dynamic range under pulsed-RF conditions. The realized system performance

is discussed in detail through a set of benchmarks, and some measurement
examples on active devices are provided.
2.1 Pulsed measurements fundamentals

A pulsed-RF signal is a continuous wave (CW) signal with a certain frequency
fC , that is switched on and off at regular intervals. The pulse width (PW)
is the time interval during which the signal is switched on, while the pulse
repetition interval (PRI) is the time period at which the pulses are repeated
(Fig. 2.1). The inverse of the PRI is the pulse repetition frequency (PRF)
and the duty cycle is the amount of time the pulse is on, compared to the
period of the pulses (DutyCycle = P W/P RI). To see what the frequency-
domain spectrum of a pulsed signal looks like it is necessary to analyze the
time-domain response mathematically [28]. A pulsed signal is generated by
first creating a rectangular-windowed version of the signal with pulse width
PW. A shah function, which is defined as a train of impulses equally spaced
1
in time, is then realized with a period of P RF . The windowed version of the
signal is then convolved with the shah function to generate a periodic pulse
train in time corresponding to the pulsed signal (Fig. 2.2).

1
y(t) = [rectP W · x(t)] ∗ shah (t) (2.1)
P RF
where, X
1 1
shah (t) = δ(t − n( )) . (2.2)
P RF n
P RF
The Fourier transform of the signal y(t) is,
Y (s) = [P W · sinc(P W · s) ∗ X(s)] · [P RF · shah(P RF · s)] (2.3)

2.1 Pulsed measurements fundamentals 13
Y (s) = [P W · sinc(P W · s)] · [P RF · shah(P RF · s)] (2.4)
Y (s) = DutyCycle · sinc(P W · s) · shah(P RF · s) . (2.5)

As shown in equation 2.5 the signal’s spectrum in the frequency domain has a
sin(x)
x envelope (Fig. 2.3). The width of the lobes are inversely related to the
pulse width. This means that as the pulses get shorter in duration, the spectral
energy is spread across a wider bandwidth. Furthermore, the spacing between
the various spectral components is equal to the pulse repetition frequency.
In general two types of measurements can be distinguished when consider-
ing pulsed-RF characterization, based on the bandwidth of the IF filter and the
timing circuits available in the network analyzer: wideband and narrowband
detection [29].
Wideband detection can be used when the receiver is synchronized with
the pulse stream, and only the “on state” of the pulse is acquired. In this
case the majority of the pulsed-RF spectrum falls within the bandwidth of the
receiver. This requires a pulse trigger that is synchronized to the pulse repe-
tition frequency of the signal. The advantage of the wideband mode is that
there is no loss in dynamic range when the pulses have a low duty cycle. The
disadvantage of this technique is that when the pulse duration gets shorter,
the bandwidth of the signal becomes larger and will eventually fall outside the
bandwidth of the receiver. Consequently, to measure smaller pulse widths the
bandwidth of the receiver, along with the bandwidth of the intermediate fre-
quency (IF) filter, must increase, yielding a higher noise power, which reduces
the dynamic range.
Narrowband detection is based on the continuous asynchronous sampling
of the RF pulses, yielding a discrete, sinc-like frequency spectrum. With this
technique, all of the pulse spectrum is removed by digital filtering except
for the central frequency component, which represents the RF carrier. After
filtering, the pulsed-RF signal appears as a sinusoid or CW signal. With
PRF
-1/2PW 0 1/2PW -2/PRF -1/PRF 0 1/PRF 2/PRF -1/PRF 0 1/PRF
Figure 2.2: Mathematical representation of a pulsed-RF signal. Source

Agilent Technologies [28].
Pulse repetition frequency

(PRF = 1/PRI)
1/PW
fC
Figure 2.3: Frequency spectrum of the pulsed-RF signal of Fig. 2.2.
narrowband detection, the analyzer samples are not synchronized with the
incoming pulses, therefore no trigger pulse is required. The advantage of this
detection mode is that there is no lower pulse-width limit, since no matter
how broad the pulse spectrum is, most of it is filtered away, leaving only
the central spectral component. The disadvantage of narrowband detection is
that measurement dynamic range decreases as the duty cycle increases. This
phenomenon is also known as pulse desensitization and the degradation in
dynamic range can be expressed as 20 · log(DutyCycle).
At the time of commencing this project two different network analyzer
products from Agilent Technologies were available for performing pulsed-RF
measurements: the 8510C (with pulsed option 008) and the PNA. The 8510C
[30] uses the wideband detection technique. After a first superheterodyne
down-conversion to 20 MHz, the pulsed signal is down-converted to baseband
IQ pulses and then digitized (Fig. 2.4a). Each I and Q output has a bandwidth
of 1.5 MHz, for a total bandwidth of 3 MHz, yielding pulses with 300 ns rise and
fall times (tr /tf = 1/BW ). Since the pulse width must be larger than several
tr /ts in order for the detector to properly acuire the pulses, this results in a
minimum specified measurable pulse width of 1 µs [29]. The down-conversion
chain of the PNA network analyzer [29] is shown in Fig. 2.4b. After a first
down-conversion to 8.33 MHz, the pulsed signal is down-converted again to an
IF of 41.7 kHz. An anti-alias filter is placed just in front of the analog-to-digital
converter. The PNA has the possibility to operate both with a wideband and
narrowband detection technique. However, due to the low IF bandwidth of
35 kHz for the PNA and of 250 kHz for the PNA-L version, the minimum
pulse durations that can be measured with the wideband technique are 50 µs
and 10 µs respectively, therefore this mode is not suitable for performing
2.1 Pulsed measurements fundamentals 15
(a)
(b)
Figure 2.4: (a) 8510C down-conversion chain using the wideband detection
technique. (b) PNA down-conversion chain. The PNA has the possibility
to operate both with a wideband and narrowband detection technique.
Source Agilent Technologies [28].
isothermal measurements. When using the narrowband technique, instead,

digital filtering is used to select the RF carrier after the digital data acquisition.
In this case, no trigger is used for synchronizing the pulse acquisition, therefore
much lower pulse widths can be measured.
The dynamic range of the two different families of network analyzers [29] is
depicted in Fig. 2.5. The 8510C network analyzer’s dynamic range is constant
versus duty cycle, thanks to the use of the wideband detection technique, how-
ever, it is limited to about 65 dB. For this reason, a high number of averages is
required to reduce the noise floor and increase the dynamic range. This results
in extremely long measurement times. Regarding the PNA, although this type
of pulsed network analyzer has a very high dynamic range for high duty cycles
(close to 100 dB at 10 % duty cycle) and it is extremely fast, its performance
drastically decreases as the duty cycle gets lower (Fig. 2.5), becoming even
worse than that of the 8510. This is a significant disadvantage in isothermal
measurements where, depending on the device technology, the pulse width has
to be as small as possible and the duty cycle very low in order to guarantee
Figure 2.5: Dynamic range vs. duty cycle for different pulse widths for
the Agilent 8510 and the PNA pulsed network analyzers [28].
that the device will remain isothermal. For example, a pulse width of 100 ns
with a duty cycle of 0.1 % can be considered a satisfactory condition for most
devices. However, at these settings the performance of current pulsed network
analyzers is strongly reduced, as the dynamic range becomes lower than 60 dB,
as shown in Fig. 2.5.
It is important to note that further developments to the PNA have im-
proved the dynamic range of the new PNA-X for low duty cycles. In particular,
for this instrument the pulse desensitization problem, although still present, it
has now been reduced to a value of 10 · log(DutyCycle) [31], enabling the use
of this instrument for isothermal measurements. These later developments,
which occurred after this thesis work, are not discussed here.
2.2 System configuration

In the previous section, the basics of pulsed network analysis and the commer-
cially-available solutions at the beginning of the project were discussed. When
looking at the commercial instrumentation available for performing pulsed-RF
measurements, a custom, dedicated solution for performing high-speed, high-
dynamic range measurements is needed to tackle the isothermal modelling
problem.
The hardware configuration of the proposed isothermal setup is given in
Fig. 2.6. Since our application requires a dynamic range as high as possi-
ble even for low duty cycles, a wideband detection technique is employed for
performing pulsed-RF measurements. The setup utilizes a PXI based data
2.2 System configuration 17
acquisition system to implement the pulsed network analyzer, which provides

high-dynamic range (see Section 2.3.1), independent of pulse width and duty
cycle. Also the data acquisition needed for synchronized pulsed I-V measure-
ments is implemented in the PXI platform, which ensures high data transfer
rates of up to 78 MB/s and thus high measurement speed.
The pulsing of the DC biasing, the RF stimulus, as well as all the data
acquisition is synchronized by two external Agilent 8110A pulse generators,
each having two outputs, which trigger the two DC pulsers and the RF pulse
generation. The user is free to choose the delay between the pulses. The
RF pulse trigger is also used to synchronize the acquisition of all DC and RF
waveforms. All the triggering, biasing and pulse generation have been realized
using commercially-available hardware instrumentation only.
2.2.1 Pulsed VNA architecture

The improved pulsed vector network analyzer (VNA) has been realized by
replacing the acquisition part of the HP 8510C [32] with a NI PXI-5105 12-bit
A/D converter board, which features 60 MS/s sampling frequency. The new
hardware configuration is controlled by custom-developed signal processing
and filtering software.
An RF synthesizer with pulsed-RF capabilities generates the stimulus for
the device under test (DUT), and the reflected, transmitted and reference
signals are down-converted by the S-parameter test-set and subsequently ac-
quired by the two external A/D converters. The HP 85110A H05 test-set
is developed specifically for pulsed operation since it employs wideband fre-
quency mixers to work with wideband IF detection, and covers a frequency
range between 0.5 and 20 GHz. The synthesizers and the data acquisition
cards are triggered by an external reference pulse generated by the Agilent
8110A pulse pattern generator. The synthesizers, along with the test-set, are
controlled by a computer using the GPIB bus, while the A/D converters are
controlled using the PXI standard.
Since the data acquisition and the signal generation are synchronized with
an external trigger, only the “on part” of the RF pulse is acquired. The
only filtering before the A/D converters is an anti-aliasing filter with a cut-
off frequency of 24 MHz. Consequently, a wideband detection technique is
employed. By choosing a high IF, thus avoiding the IQ down-conversion as
used in the 8510C, and by making use of wideband digitizers, the traditional
limitations on the minimum pulse width are removed, and as result our setup
is only limited by the pulse generation capability of the synthesizers.
After the data acquisition a FFT is performed on the acquired waveform
to extract the magnitude and phase information of the signal components
DC 18
Subsystem DC Supply
DC Pulse
Triggers
Auriga Pulser Auriga Pulser

Agilent 8110A
Pulse Generator Vforce Imon Vforce Imon
RF Pulse
Trigger
DUT
Bias Bias
Tee Tee
Agilent 8110A
Pulse Generator V1 V2
I1 I2
b1
b2
Synthesizer CH. 0 CH. 0 CH. 0 CH. 0
LO 1 2 a1 CH. 1 CH. 1 CH. 1 CH. 1
HP 85110A
a2 TRIG TRIG TRIG TRIG
Test-Set
PXI Based
Synthesizer
DAQ
RF RF signals IF signals
10 MHz Ref MXI Link
GPIB Bus
Digital filtering
Figure 2.6: Block diagram of the custom pulsed RF - pulsed I-V setup.
TUDelft Synthetic Network Analyzer
Isothermal Measurement Systems
at the frequency of interest. Finally, a ratio between two of the measured

signals gives the S-parameter measurement. In order to ensure that the FFT
algorithm works properly in a practical implementation, coherent sampling [33]
has to be performed. This means that the FFT has to be applied on an integer
number of periods of the down-converted RF signal to avoid spectral leakage.
To establish this, the frequencies of the local oscillator (LO) and of the RF
source are synchronized to produce an IF output of 12 MHz, which results in
5 samples per period due to the 60 MHz sampling frequency. This in principle
facilitates a minimum pulse width of 83 ns. However, to allow for transients
in the RF during the start of the trigger, we use a minimal pulse width of
100 ns.
Signal processing and dynamic range considerations
The dynamic range (DR) of an ADC is the ratio of the largest to the smallest
signal that the converter can represent. The largest signal is usually taken to
be a sine wave that covers the full voltage range of the ADC, while the smallest
signal is usually taken to be the total noise level of the ADC. Consequently
the dynamic range expressed in decibels (dB) is

S
DR = 20 · log , (2.6)
R
where S is the rms amplitude of the largest signal, and R is the rms ampli-
tude of the smallest signal. In an ideal ADC, the noise is a direct result of
the quantization noise of the ADC [34], therefore the dynamic range can be
calculated directly from the number of bits N as
DR = 6.02 · N + 1.76 . (2.7)
In a real ADC, though, the presence of noise other than from quantization
is responsible for a decrease in dynamic range. In our particular situation,
the A/D converters employed are 60 MS/s NI PXI-5105 12-bit digitizers, and
their theoretical dynamic range can be calculated from the number of bits and
the input range. For an input range of 50 mV peak-to-peak, the maximum
power level that can be measured with a 50 Ω input impedance is
(V maxRM S )2

Pmax (dBm) = 10 · log 1000 · =
50 Ω
" √ #
(V maxP eak / 2)2
= 10 · log 1000 · =
50 Ω
" √ #
(25 mV / 2)2
= 10 · log 1000 · ≈ −22 dBm .
50 Ω
(2.8)
The lowest power that can be measured due to the total noise of the digitizer
as specified in the data sheet is
(V noiseRM S )2

Pmin (dBm) = 10 · log 1000 · =
50 Ω
(19 µV )2

= 10 · log 1000 · ≈ −81.4 dBm ,
50 Ω
(2.9)
yielding a dynamic range of
DR = P max(dBm) − P min(dBm) ≈ 59.4 dB . (2.10)
If we assume that the source of noise in our ADC appears as white noise, we
can use digital filtering to filter out noise components outside the bandwidth
of interest. In this case a correction factor, called process gain [34], must be
added to equation 2.7 to account for the related increase in the signal-to-noise
ratio,
fS
DR = 6.02 · N + 1.76 + 10 log . (2.11)
2 · BW
The process of sampling a signal at a rate which is greater than twice the
bandwidth is referred to as oversampling. Performing an M-point FFT over
the acquired waveform to extract information about a particular frequency
component, is equivalent to digitally filtering the signal with a bandwidth
equal to the frequency resolution of the FFT, that is fS /M . Therefore the
dynamic range due to the discrete Fourier transform is
M
DR = 6.02 · N + 1.76 + 10 log . (2.12)
2
Output Voltage
Figure 2.7: Time-domain averaging of different periods of a sine wave.
From equation 2.12 it’s clear that to improve the dynamic range it is suf-
ficient to acquire a higher number of samples. In particular, doubling the
number of FFT points drops the magnitude of the asynchronous noise compo-
nents by a factor of 2, or 3 dB on a log scale. This expected improvement drops
off somewhat for large record lengths due to round-off errors in computing the
FFT.
Time-domain averaging is another technique to improve dynamic range as
it attenuates asynchronous noise sources by averaging time-domain waveforms
from multiple triggers. Averaging in the time domain not only reduces the
white noise present on the signal acquired, but also permits to increase the
number of effective bits of the digitizer, and thus increases the dynamic range.
This technique, often referred to as “oversampling and decimation” [35], can
be achieved by oversampling the signal and then taking the average of the
subsequent samples. For each additional bit of resolution, n, the signal must
be oversampled four times,
foversample = 4n · fN iquist . (2.13)
To guarantee proper operation of this approach, the signal-component of in-

terest should not vary during a conversion, however, there should be enough
noise present over the input signal so that the results of successive acquisitions
vary the least significant bits of the A/D converter. For our application, the
signal to process is a sine wave. In this case it is also possible to apply the
same concept by acquiring different periods of the sine wave and averaging the
time samples relative to the same time reference, as depicted in Fig. 2.7. In
particular, averaging 10 different acquired sine waves yields a 10 dB increase
in the dynamic range.
140
Theoretical
Measured
Dynamic Range, dB
120
100
80
60 0 2 4 6
10 10 10 10
Number of Acquired Samples
Figure 2.8: Measured vs. theoretical dynamic range, as calculated from

equations 2.10 and 2.12, of the PXI-5105 A/D Converters.
A comparison between the theoretical dynamic range as calculated from

equations 2.10 and 2.12 and the measured dynamic range of the PXI-5105 A/D
converters is shown in Fig. 2.8. The dynamic range of the A/D converter is
measured as the difference between the power of a full-scale sinusoidal signal
and the noise floor of the converter. Note that this will be higher than the
spurious-free dynamic range as this calculation does not take into account any
spurious signals which are present within the A/D acquisition bandwidth. In
fact the influence of spurs can be avoided by carefully selecting the IF. It is
important to point out that this dynamic range analysis only takes into ac-
count the noise limitation of the receiver, and is useful only for understanding
the techniques used to improve the dynamic range. The graph of Fig. 2.8,
therefore, represents the best case scenario, as in reality the detected signal
will contain a certain amount of noise by itself that reduces the effective dy-
namic range. The dynamic range of the pulsed VNA will be discussed further
in Section 2.3.1.
2.2.2 Pulsed-DC measurements

The pulsed bias conditions to the DUT input and output are applied through
the use of two Auriga AU4102 IV Pulsers, which switch the bias provided by
an external Agilent 6629A power supply. To achieve the best pulse shape,
parasitic loading was minimized in the setup by using very short cables and
connecting the bias tees (Auriga AU0609) directly to the wafer probes. Within
our system hardware configuration this combination can provide pulse widths
as short as 200 ns. The maximum voltage and current ratings of the pulsers
Vmeas
ADC
VDS
Imeas
VGS ADC
R
R
Current Sense Resistor

(0.1, 1 or 10 Ohm)
Figure 2.9: Current sense configuration with series resistor in the DUT
ground path.
are 200 V and 10 A. The pulsed currents and voltages at the device ports
are measured using 4 channels of the same PXI NI-5105 A/D converter board
used for the RF measurements. Time-domain averaging is also used on the
DC waveforms to increase the measurement accuracy.
Two different methods are available for sensing the current. The DC
pulsers employed provide a current monitor port with an internal 0.1 Ω sense
resistor to measure high currents. This resistor is inserted in the ground path
going from the DUT ground to the negative port of the DC supply as illus-
trated in Fig. 2.9. To sense lower currents with high resolution, a 1 Ω or 10 Ω
resistor can be placed in the external ground path just before the bias-tee.
Before the actual measurement can take place, the sense resistor is calibrated
by applying a range of voltages over a set of known resistors, while measuring
the current. For the highest accuracy, the calibration resistors and voltages
have to be selected according to the intended measurement range.
One advantage of using this method for sensing the current is certainly
the high bandwidth, which will result in shorter transient times, allowing
measurement of shorter pulses. On the other end, the sense resistor will cause
a voltage drop at the device terminals. For this reason the voltage needs to be
constantly monitored and iterations are necessary to reach the desired voltage
at the DUT ports. Furthermore, due to the position of the sense resistor
in the ground path, the source providing biasing to the pulser should have
a floating ground with respect to the common ground of the measurement
setup. Failing to do so would result in a parasitic current flow which is not
directed through the current sense resistor, reducing the current measurement
accuracy. The use of a completely floating source, like an Agilent 6629A, can
solve this problem.
Another possibility to measure a pulsed current is to use a commercial

clamp-on current probe. These types of probes employ a hybrid technology
that uses a Hall sensor and an AC transformer to measure both DC as well as
AC currents up to hundreds of MHz of bandwidth. These probes can measure
currents up to tens of amperes and provide a relatively easy way to implement
a current measurement since they do not require breaking the circuit and
therefore cause no voltage drop. On the other hand, they can only measure
down to current levels of a few milliamperes with good accuracy.
2.3 System performance

Device scaling of power devices through (compact) modelling techniques is
quite often based on the characterization of relatively small device cells. In
this process, isothermal device characterization is useful in simplifying the
modelling problem, enhancing understanding and speeding up the modelling
process. Consequently, an isothermal measurement system should be capable
of measuring, with good accuracy, a wide range of pulsed voltages and currents
to allow accurate characterization of different device sizes, from the single cell
to the complete high-power device. At the same time, a high dynamic range
for the S-parameter device characterization is of fundamental importance for
two reasons:
• To guarantee accuracy, while ensuring truly small-signal device opera-

tion at low RF drive power levels.
• To accurately characterize a wide scope of device sizes / impedance

levels.
In the following paragraphs, both the accuracy of the DC current measure-

ment, as well as the RF measurement dynamic range of the proposed system
are reported.
2.3.1 Pulsed VNA dynamic range

In order to illustrate the dynamic range performance of the realized pulsed net-
work analyzer, several measurements on a bandpass filter were performed un-
der different pulse conditions in the frequency range from 2.3 GHz to 4.5 GHz.
A calibration was performed using a coaxial, 3.5 mm calibration kit. All of the
measurements were made using an averaging factor of 1,024. Fig. 2.10a shows
the measured S21 (in dB) for the pulse widths ranging from 100 ns to 100 µs
with a fixed pulse repetition period of 1 ms. As can be seen, the dynamic
2.3 System performance 25
(a) (b)
Figure 2.10: (a) S21 (dB) of a band-pass filter for different pulse widths
with a PRP of 1 ms, (b) S21 (dB) of a band-pass filter for different duty
cycles with a PW of 1 µs.
range of the setup is more than 80 dB. The use of the wideband detection
technique, which employs a trigger to synchronize the data acquisition, com-
bined with digital processing of the acquired signal, ensures a wide dynamic
range that is totally independent from the pulse width. Moreover, the high
sampling rate of the A/D converters significantly reduces the constraints on
the minimum pulse width, which can be as short as 100 ns. In Fig. 2.10b
the results obtained for different duty cycle values, with a fixed pulse width of
1 µs, are presented. As expected, the duty cycle does not affect the dynamic
range of the setup. Fig. 2.10 also shows a comparison between the dynamic
range of the synthetic pulsed VNA with that of commercially-available instru-
ments such as the HP 8510C and the Agilent PNA. It is clear that the new
setup has a 20 dB improvement in dynamic range with respect to the 8510C.
Also, for low duty cycles the new setup has a higher dynamic range than the
PNA and the PNA-X.
2.3.2 DC measurement accuracy

To investigate the accuracy of our system, the isothermal setup was calibrated
using various loading resistors ranging from 1 Ω to 2 kΩ, while using three
1
10
Range0-1 V
Range0-10 V
0
10
Minimum current [A]
-1
10
0.1 Ohm
Imax. 10 A
-2
10 1 Ohm
Imax. 1 A
-3
10
10 Ohm
Imax. 0.1 A
-4
10 0 0.2 0.4 0.6 0.8 1 1.2
Pulse width [µs]
Figure 2.11: Minimum current vs. pulse width which can be measured
with 1 % accuracy using the current sense resistors for the indicated cur-
rent ranges (Imax). Due to ringing effects the current measurement accu-
racy is dependent on the voltage range (0-1 V) and (0-10 V) used in the
calibration.
different current sense resistors (0.1, 1.0 and 10 Ω). After calibration, indepen-
dent reference resistors, which were not included in the calibration, ranging
from 5 Ω to 220 Ω are measured and the deviation of the measured value from
their nominal resistance value is calculated to quantify the relative error in
the current measurement. The loading and reference resistors were previously
measured, in a temperature-controlled environment, by using a recently cali-
brated Agilent E5270B parameter analyzer, at small applied voltages to avoid
thermal effects on the resistors. Since the accuracy of the parameter analyzer
is higher than the A/D converter, the resistance values are accurately known.
Note that the gain and offset errors of the current measurement are corrected
by the current calibration. Therefore, the error calculated in such a way in-
cludes the inaccuracies of the A/D converters, the errors in the voltage setting
and the ringing of the pulse for smaller pulse widths.
The minimum current that can be measured with 1 % accuracy is plotted
as a function of pulse width for the different sense resistors in Fig. 2.11. In
all the experiments the averaging in the data acquisition was kept constant
at 1,024. From Fig. 2.11 it’s possible to conclude that currents above 1 mA
can be measured with an accuracy of 1 % on a resistive load, provided that
the pulse width is at least 500 ns. It appears that the low-voltage range has a
2.4 Measurements examples 27
slightly better accuracy for short pulse widths. This can be explained by the
lower ringing of the bias pulse in the low voltage range. This effect disappears
when the pulse width increases.
2.4 Measurements examples

One of the most common applications that requires pulsed-RF and DC con-
ditions is the parameter extraction of active devices. To model an active de-
vice under small-signal conditions, the lumped parameters of the small-signal
model must be extracted under isothermal conditions (afterwards the thermal
node is extracted). In order to extract these values for different bias condi-
tions, all four S-parameters must be measured at a fixed frequency at different
base and collector voltages [36, 37].
In this section, we provide some measurement examples for different device
technologies to highlight the capabilities of the realized measurement setup.
2.4.1 LDMOS devices

A GEN6 NXP LDMOS device [38] with a gate width of 2 mm has been mea-
sured and the cut-off frequency extracted from the pulsed S-parameter data
(plotted in Fig. 2.12 as contours on the Id(Vds) plane).
At the input, a continuous voltage on the gate was applied, which was
stepped from 1 V to 15 V (29 points), while at the output the pulsed voltage
was changed from 0 V to 24 V (49 points). A current sense resistor of 1 Ω was
used. The cut-off frequency of the device was extracted from the measured
S-parameters at a frequency of 2 GHz. The DC pulse duration was varied from
0.2 to 150 µs, while the RF pulse was kept at 100 ns and the pulse period
was fixed to 1 ms. The applied RF power within the pulse at the device input
plane was set to -30 dBm to guarantee small-signal operation of the device
under test.
The measurement data in Fig. 2.12 clearly demonstrates the expected be-
havior of the LDMOS device with respect to self-heating, namely the drain
current reduction due to self-heating is nearly eliminated when using a very
short pulse width. There are already some differences between the measure-
ments at 1 µs and at 200 ns (both the fT and Ids go up). These results
demonstrate that the realized measurement setup is able to measure power
devices with very small pulse widths.
0.5 0.5
0.45 1
2 0.45 1
1 3 2
2 45
0.4 3 0.4 1 2 43
45 6 7 2 43 5 6
12 3 6 7 8 5 6 7
0.35 45 8 9 0.35 43 7 8
9 8 9
1
7 10 10 9 10
Ids , A
Ids , A
0.3 11 0.3 10
1
6 11 6
9 10 1 12 11
8
32 1
12
8
0.25 1 13 0.25 9 10 11 12
4312 2 1
7
13 12
9810 7 45 32 1
12
5
14 13 13
0.2 14 0.2 13
13
12 11
74 3612 312
15 14
15
0.15 0.15
53426 32
14
14
14
14
9 180
11
0.1 0.1
14 14 13
12
13 13 13 13 13 12
12 12
11 12
4 3521
0.05 11 11 12 0.05 11 11
3 65
8 10 8 10
956 4 10
9 10 10 10
4 7 3
956
24
7 31 2 7 31 8 56 6 8 75 9 6 8 75 9 6 8 75 9
0 0
0 5 10 15 20 0 5 10 15 20
Vds , V Vds , V
(a) (b)
0.5
0.45
0.4
1 1
0.35 2 2
23 4 35 4 35
4 5 6 6
Ids , A
0.3 7 8 7 7
9 8
9
7 5 321 1
0.25 8 9 10
10
6
11 10
11 11
8 93 12 4 1
0.2 12
12 12
13 13
0.15
1101
2413
0.1
6
13 13
13
13 12 12
12
34521
0.05 11 11
10
10 10
69 8 45 7 69 8 45 7 69 8 45
7
0
0 5 10 15 20
Vds , V
(c)
Figure 2.12: Measured contours of constant fT (GHz) for a GEN6 NXP

LDMOS device plotted in the Id (Vds ) plane using (a) pulse width of 0.2 µs
(pulse period = 1 ms), (b) pulse width of 1.0 µs (pulse period = 1 ms),
(c) pulse width of 150 µs (pulse period = 1 ms).
2.4 Measurements examples 29
0.4 0.4
0.3 0.3
Ic , A
Ic , A
0.2 0.2
0.1 0.1
0 0
0 1 2 3 4 5 0 1 2 3 4 5
Vce , V Vce , V
(a) (b)
0.4 0.4
0.3 0.3
Ic , A
0.2 Ic , A 0.2
0.1 0.1
0 0
0 1 2 3 4 5 0 1 2 3 4 5
Vce , V Vce , V
(c) (d)
Figure 2.13: Measured output characteristic for a GaAs HBT device for a
pulse width of (a) 300 ns, (b) 500 ns, (c) 1 µs, and (d) 5 µs, with a pulse
period = 1 ms.
2.4.2 III-V HBT devices

A GaAs HBT device with 4 emitter fingers, 3.0 µm wide and 44 µm long has
also been measured, using DC pulse durations of 0.3, 0.5, 1 and 5 µs, to high-
light the capabilities of the setup for devices with higher thermal resistances.
At the input, the current was stepped from 1 mA to 7 mA (25 points), while
at the output the voltage was pulsed from 0 V to 4.8 V (49 points). Figure
2.13a-d show the measured output characteristics of the device for different
pulse durations. This device exhibits a negative dependence of the current
gain with temperature. The results clearly show the reduction of self-heating
effects in the device with the pulse width.
2.4.3 SiGe HBT devices

The measurement of the I-V characteristics of a QUBiC4plus SiGe bipolar
transistor [39] with an area of 99 µm2 has been performed using DC pulse
durations of 0.5, 1 and 3 µs. In this example, both the base-emitter and
collector-emitter voltages are pulsed with a pulse repetition period of 1 ms,

from 0.77 to 0.91 V (0.01 V steps) and from 0.1 to 4.7 V (0.1 V steps),
respectively. The results, which are shown in Fig. 2.14, clearly show increased
self-heating effects, which eventually result in thermal runaway [40] at higher
currents and longer pulse widths.
To conclude, a measurement under pulsed-bias, pulsed-RF conditions is

reported for a pulse width of 1 µs and a pulse repetition period of 1 ms. The
base-emitter and collector-emitter voltages are pulsed from 0.7 to 0.9 V (0.02 V
steps) and from 0 to 3.5 V (0.1 V steps), respectively. The base-collector
capacitance and the cut-off frequency of the transistor are then extracted from
the pulsed S-parameter measurements, and are shown in Fig. 2.15.
0.25 0.25
0.2 0.2
0.15 0.15
Ic
Ic
0.1 0.1
0.05 0.05
0 0
0 1 2 3 4 0 1 2 3 4
Vce Vce
(a) (b)
0.5
0.4
0.3
Ic
0.2
0.1
0
0 1 2 3 4
Vce
(c)
Figure 2.14: Measured output characteristic for a QUBiC4plus SiGe HBT

device for a pulse width of (a) 500 ns, (b) 1 µs, and (c) 3 µs, with a pulse
period = 1 ms.
2.5 Conclusions 31
0.05 0.05
40
35 30
5
0.04 0.04
02
52
130
40
500
101
Ic , A
150
140
40
Ic , A
40
170
160
0.03 0.03
1000
180
30
200
35
25
0.02 35 35 0.02
1500
20
250
1015
30 30 30
0.01 0.01
300
25 25 25
2000
20 20 20
15
10 15
10 15
10
0 0
0.5 1 1.5 2 2.5 3 0.5 1 1.5 2 2.5 3
Vce , V Vce, V
(a) (b)
Figure 2.15: Measured contours of (a) constant fT (GHz) and (b) constant
Cbc (fF) for a QUBiC4plus SiGe HBT device plotted in the Ic(Vce) plane
using a pulse width of 1 µs (pulse period = 1 ms).
2.5 Conclusions
A new, low-cost, very flexible isothermal measurement setup has been pro-
posed which facilitates fast and accurate isothermal device characterization.
Measurements of different types of active devices have been performed, show-
ing that the setup is capable of measuring pulses as short as 200 ns for a wide
range of bias and power conditions, with good accuracy and measurement
speed.
The system is based on a custom pulsed network analyzer that achieves
a significantly higher dynamic range and measurement speed than currently
available commercial solutions for isothermal measurements, when a low duty
cycle is required. Moreover, the new setup does not suffer from the pulse
desensitization issue typical of a narrowband detection technique, allowing
the measurement of RF pulse widths down to 100 ns without any degradation
in the dynamic range, facilitating true isothermal measurements conditions.
Chapter 3
Source and Load-pull

Architectures
When designing an RF / microwave amplifier, it is important to properly

integrate the transistor with the rest of the circuit. It is simply not enough
to place the transistor in the circuit and expect it to perform optimally, since
the input and output impedances of the transistor deviate strongly from the
characteristic system impedance (typically 50 Ω). Improper integration of the
transistor in the system degrades gain, output power and efficiency, among
other considerations. In order to maximize these parameters, it is essential to
use matching networks at the input and output of the device. Assuming an
unconditionally stable device in a linear system, using the complex conjugate
of the small-signal input impedance as the source matching network, and the
complex conjugate of the small-signal output impedance as the load matching
network is sufficient [37]. Unfortunately, this is only applicable to small-signal
systems, and is of no assistance to the designers of power amplifiers, where
the device is driven into large-signal operation [15]. In these situations, the
technique used to determine the ideal matching impedances is referred to as
load-pull [15].
The goal of load-pull is to measure the response of a device as function
of the source and load impedances, input power, bias, temperature and other
relevant parameters. Simply stated, a load-pull system presents a set of con-
trolled impedances to the device under test (DUT), while measuring a list of
performance parameters at each impedance point. By varying the impedance
offered to the DUT, we can characterize the performance of the device and
select the best set of impedance settings for optimum large-signal operation.
Harmonic load-pull is an extension of basic fundamental load-pull, where
the loads presented to the DUT at multiple frequencies must be considered
33
34 Source and Load-pull Architectures
Bias-circuits (baseband)
IN OUT
Input match @ f0 ,2f0, etc. Output match @ f0 ,2f0, etc.
(a)
Load-pull
Provide the same test
condition to the device
without building the circuit
b2
a2
a2
L 
Reference plane
b2
(b)
Figure 3.1: (a) Typical microwave power amplifier, including matching

networks, using a packaged power transistor. (b) The physical matching
network is emulated by controlling the ratio between a2 and b2 power
waves. Load-pull techniques provide the same matching conditions to the
active device without the need for building a physical matching circuit.
simultaneously. Each parameter is then measured at various combinations of

load impedance for fundamental and harmonic frequencies. Harmonic load-
pull is often used as a method to increase efficiency [15], or to improve the
amplifier linearity when operating with modulated signals [41].
The actual load impedance (ZL ) presented to the DUT in the microwave
domain is represented by the reflection coefficient related to this load, namely
ΓL . It represents the ratio between the reflected wave from the load (a2 )
and the forward traveling wave (b2 ). The generalized formula for ΓL at all
3.1 Passive load-pull 35
Figure 3.2: A conventional passive tuner, with moving probe and slab
line.
frequency components of interest can be written as,
a2,n (fx )
ΓL,n (fx ) = , (3.1)
b2,n (fx )
in which a2,n is the wave reflected from the load, b2,n is the forward traveling
wave, n the harmonic index and fx a frequency in the band of interest. This
notation allows us to describe all the relevant circuit properties of the output
matching network and of the bias circuitry (Fig. 3.1a), while we can use a sim-
ilar formalism for the input matching. In a source or load-pull measurement
we control the reflection coefficients offered to the active device in an artificial
way and imitate the complete circuit without actually building it (Fig. 3.1b).
This yields major advantages in reducing the development time of new tran-
sistor technologies and their application in power amplifiers, while providing
a significantly better understanding of the behavior of the active device.
In the following sections the traditional load-pull architectures, along with
their strengths and weaknesses, will be described in detail.
3.1 Passive load-pull

Passive mechanical tuners [42, 43] (e.g., see Fig. 3.2), make use of sliding me-
chanical probes (or “slugs”) to create a reflection, by interrupting the electric
field of an airline. The probe is inserted into the airline at variable depth;
the further the probe penetrates into the airline and interrupts the electric
field, the greater the magnitude of reflection. The phase of this reflection is
varied by sliding the probe along the length of the slab line. Therefore, any
impedance inside the Smith chart can be presented to the DUT by selecting
the appropriate vertical and horizontal positions of the probe with respect to
RF
(a)
a1 b1 b2 a2
RF
(b)
Figure 3.3: (a) Traditional tuner based load-pull measurement system.

The tuners are placed between the DUT and the couplers. Power meters
are commonly used to measure available, incident and output power. (b)
Real-time system configuration where the couplers are placed between the
tuners and the DUT. In this configuration a VNA can be used to measure
all the scalar and vector quantities of the DUT.
3.2 Active load-pull 37
the airline. Fundamental load-pull is typically achieved by using a single tun-

ing probe. Harmonic load-pull is achieved using a combination of two, three
or more probes in either cascaded or filtered configurations.
Tuner based systems are the preferred industry large-signal test bench
choice today, due to their simplicity and high power handling capabilities (up
to 500 W of power in CW). However, passive tuners are slow because of the
time the probe takes to move from one position to another when changing
reflection coefficient. More importantly, they are constrained by losses which
limit the magnitude of the reflection coefficients that these systems can provide
to the DUT. It is clear that the wave a2 , which returns to the DUT from the
load tuner, will always be lower than b2 due to tuner losses. Also the losses
between the DUT and the tuner lower the amount of energy that can be
reflected. Typical values range between ΓL = 0.8 and ΓL = 0.92 at the device
reference plane.
For this reason, passive tuners are often placed as close as possible to the
DUT, that is, before the reflectometers used to couple the a1 , b1 , a2 and b2
waves (Fig. 3.3a). In this case, to measure all the corrected scalar (POU T ,
PIN , Gain, etc.) and vector (ΓIN , ΓS , ΓL , etc.) quantities of the DUT, the
two-port S-parameters of each tuner must be measured for each tuner position
during a time consuming pre-characterization phase. The accuracy of the load-
pull measurements will then rely on the tuner and connector repeatability.
Alternatively, very low-loss RF couplers [44, 45] can be employed to reduce
the losses and allow the use of a traditional two-port network analyzer system
architecture (Fig. 3.3b). This configuration, typically referred to in literature
as “real-time” load-pull, has all the advantages of high-speed, high-accuracy
and fast calibration which are typical of a two-port network analyzer.
3.2 Active load-pull

Active load-pull systems can, thanks to the use of amplifiers, overcome the
reflection magnitude restrictions due to losses. A close examination of equation
3.1 reveals that any arbitrary load reflection coefficient can be created by
injecting an a2 wave into the output port of the DUT. In an active load-pull
system, the magnitude restrictions due to losses are overcome by magnifying
a2 through the use of amplifiers. In this case the injected a2 wave can become
equal or bigger than b2 , so that ΓL can even exceed unity at the DUT reference
plane.
There are several advantages to an active system over traditional mechan-
ical load-pull, including gamma control and ease of integration. Since there
are no mechanical moving parts, and the system relies on electrical tuning, the
RF
Figure 3.4: Closed-loop active load-pull configuration.
tuning itself can be fast. Harmonic load-pull, or tuning impedances at multiple

frequencies simultaneously, becomes simple when using active load-pull tech-
niques. A frequency multiplexer can be used to merge multiple active tuning
paths, one per frequency, such that equation 3.1 is satisfied. Any losses inher-
ent in the use of frequency multiplexers are easily overcome by the amplifiers
added in each active tuning chain.
Over time, two basic concepts for active load-pull have been promoted,
which we briefly discuss below.
3.2.1 Closed-loop active load-pull

Closed-loop active load-pull has been referenced in IEEE publications since
the early 1980’s [46]. In a “closed-loop” or “feedback” system, an amplified
version of b2 is used as the reflected signal, a2 . To accomplish this, a coupler
or circulator is used to direct the signal from the DUT, b2 , through a variable
amplification stage with control of both magnitude and phase, and re-inject
the signal as a2 back into the device. A block diagram showing a typical
closed-loop system is shown in Fig. 3.4.
Since there are no moving parts and the system relies on electrical tuning,
speed is a major advantage of this technique. In fact the injected signal is
a direct function of the coupled wave, thus the loop gain and phase shift
determine the reflection coefficient, independent of the power and spectral
3.2 Active load-pull 39
Signal
Generator RF
DUT
Reference Active Load
planes A fA
a1 a2
Amplifier
b1 b2
Figure 3.5: Open-loop active load-pull configuration.
content of the signal. This makes the closed-loop concept suitable for fast
device characterization. On the other hand the use of a feedback topology
can cause loop oscillations [47] when the forward gain of the feedback loop
is bigger than unity, or if |ΓL | · |ΓDU T | > 1 at any frequency. This is more
likely to occur when employing wideband loop amplifiers. To avoid the risk of
oscillation, most closed-loop systems include sophisticated filtering [48]. For
this purpose, a recent publication proposes to control the reflection coefficient
at baseband frequencies [49], where it is possible to implement better frequency
selectivity.
A further disadvantage of this technique is that high-power and very linear
injection amplifiers are needed in order to maintain the loop linearity, which
can increase the overall system cost tremendously.
3.2.2 Open-loop active load-pull

The first appearance of active load-pull systems that can be found in the lit-
erature is by Takayama from the year 1976 [50]. In this topology, a part of
the input signal to the DUT is split, adjusted in amplitude and phase, and in-
jected into the output of the DUT to create a virtual load reflection coefficient
(Fig. 3.5). This technique, commonly known in literature as “open-loop”,
“two signal paths type” or “feedforward” load-pull, has been implemented,
with several modifications over time, by many authors [51–55].
The basic active tuning chain consists of a signal source, a variable phase
shifter and a variable gain stage. More advanced versions include in-phase /
quadrature (IQ) modulators to control the amplitude and phase of the injected
(a2 ) wave [54]. Since the a2 wave injected into the output of the DUT is no
longer a direct function of b2 , the reflection coefficients realized are power and
phase dependent. Therefore, iterations are needed during the measurements
to find the optimal injection signals to offer the desired reflection coefficients to
the DUT. This makes open-loop systems slower than the closed-loop ones. On
the other hand, there is no feedback path and therefore no chance of tuning-
loop oscillation compared to the closed-loop technique. Open-loop load-pull
also requires high-power amplifiers to achieve the desired reflection coefficients
when testing a high-power device. However, these amplifiers do not have to
be linear, since the user specified reflection coefficient is reached by successive
software iterations.
Furthermore, when extending the single-tone principle to wideband sig-
nals, determining the required content of the injection signals, at both the
fundamental and harmonic frequencies, becomes difficult. To tackle the prob-
lem, the use of multipliers in the signal path has been proposed [55]. However,
intermodulation distortion and memory effects will be present in practice,
making it impractical to obtain the required injection signals by a separate
analog multiplication process.
3.3 Electrical delay in load-pull systems

The introduction of wideband modulated signals as currently used in modern
communication systems, poses a serious problem for the traditional load-pull
system. Wideband signals are by definition not fixed to a discrete frequency,
rather they have some bandwidth, ranging from several MHz to tens of MHz
[56].
When working with modulated signals, the reflection coefficients offered
to the DUT should ideally be constant (not vary versus frequency) within
the modulation bandwidth at the fundamental, for well-controlled linearity
behavior of the DUT, as well as in all related frequency bands at baseband
and harmonic frequencies [41, 57]. This situation is approximated in real cir-
cuit implementations, when the matching networks are placed directly at the
reference planes of the active device. In any conventional load-pull setup, how-
ever, the actual physical impedance is always located at some distance from
the DUT (Fig. 3.6a), which is much larger than for any practical matching
network. This distance, as well as any physical length within the tuning ele-
ment itself (e.g., the length of the active feedback loop, or the position of the
3.3 Electrical delay in load-pull systems 41
(a) (b)
Figure 3.6: (a) Phase delay caused by electrical lengths of the tuning
element plus cables, adapters and probe (b) Phase rotation of the reflection
coefficient for a 2.58 MHz wideband signal for a typical load-pull system.
probe in mechanical tuners), yields very large electrical delays, causing rapid
phase changes of the reflection coefficients versus frequency. Typical values
for these phase fluctuations start from about 3◦ /MHz for a fundamental pas-
sive mechanical tuner. If high-Q resonators or impedance tuners are used to
control the harmonic terminations, the phase change with frequency is even
greater. When considering active systems, the phase rotation can reach values
of 30◦ /MHz or more for conventional closed-loop systems (Fig. 3.6b). Note
that for a W-CDMA signal with adjacent and alternate channels (total band-
width of 25 MHz), the reflection coefficient offered by a passive tuner based on
these values would vary 75◦ in phase over this bandwidth, while in a conven-
tional active closed-loop the phase change of the offered reflection coefficient
would amount to 750◦ .
It is clear that these large phase deviations represent non-realistic circuit

conditions, and will cause measurement errors such as IM3 asymmetry, spec-
tral re-growth, PAE and EVM degradation [58]. In general, maintaining the
reflections coefficients constant over frequency is more and more difficult with
the increasing modulation bandwidth of communication signals used in prac-
tical circuits and in load-pull measurement setups.
V2  Z 0  I 2
L 
V2  Z 0  I 2
Figure 3.7: Thévenin equivalent schematic of an active load-pull config-

uration. The load impedance offered to the DUT at the reference plane
is varied by adjusting the equivalent voltage source ESY S in amplitude
and phase. The related power needed to synthesize specific impedances
depends strongly on the equivalent system impedance (ZSY S ).
3.4 Injection power and load amplifier lin-

earity
As mentioned in Section 3.2 with respect to active load-pull configurations, it
is necessary to inject a certain amount of power into the DUT to synthesize
a specific ΓL at the DUT reference plane. To better analyze the problem, the
active load-pull topologies of Fig. 3.4 and of Fig. 3.5 can be described with
their Thévenin equivalent as depicted in Fig. 3.7. EDU T and ZDU T and ESY S
and ZSY S are the equivalent voltage sources and output impedances of the
DUT and of the measurement system, respectively. The equivalent voltage
sources can be expressed in terms of the transmitted and incident waves to
the DUT as,
bDU T · (ZDU T + Z0 )
q
EDU T = √ bDU T = 2 · Pb2 · (1 − |ΓDU T |2 )
Z0
(3.2)
bSY S · (ZSY S + Z0 )
q
ESY S = √ bSY S = 2 · Pa2 · (1 − |ΓSY S |2 ) .
Z0
(3.3)
With reference to the schematic of Fig. 3.7, the injected power required
to achieve a certain ΓL , or in other words a certain impedance ZL = V2 /I2
3.4 Injection power and load amplifier linearity 43
Vdut
I_Probe P_1Tone
PORT2
I_dut
Num=2
S2P_Eqn
P_1Tone Z=50 Ohm
S2P1
PORT1 P=polar(pow er,phi)
S[1,1]=0
Num=1 Freq=1 GHz
S[1,2]=1
Z=2 Ohm
S[2,1]=1
P=200
S[2,2]=0
Freq=1 GHz
Z[1]=5
Z[2]=50
Figure 3.8: ADSTM schematic testbench for the evaluation of the required
injected power by active load-pull.
can be calculated as,
(1 − |ΓDU T |2 ) |ZDU T + Z0 |2 |ZL − ZSY S |2

Pa2 = Pb2 · · · (3.4)
(1 − |ΓSY S |2 ) |ZSY S + Z0 |2 |ZDU T + ZL |2
It is clear from equation (3.4) that the injected power needed not only
depends on the output power of the DUT and the desired ΓL , but also on
the output impedance of the device. When considering high-power devices,
with output impedances in the order of a few ohms, the required injection
power to cover the desired area of the Smith chart can be extremely high in
a 50 Ω system (e.g., 2 to 10 times higher than the maximum output power of
the DUT). To overcome this problem, pre-matching is typically used, which
converts the 50 Ω impedance of the system to a value that is much closer to
the output impedance of the DUT. This widely-used technique (also applied
in passive load-pull) not only reduces the losses, but also lowers the power
requirements of the load-injection amplifier [59]. As an example, consider a
DUT with an output impedance of 2 Ω and an available output power of
200 W. To synthesize a load impedance of 1 Ω in a 50 Ω system, the required
injection power would be more than 2 kW. Reducing the system impedance to
10 Ω with a lossless pre-match fixture lowers the required injection power for
the same load condition to 360 W, while with a pre-match to 5 Ω the required
injection power is only 142.2 W.
To provide a better understanding of the problem, a simulation using Ag-
ilent’s Advanced Design Simulator (ADS), was performed using the simple
schematic testbench illustrated in Fig. 3.8. The power source at port 1 repre-
sents a DUT with 200 W of output power and an output impedance of 2 Ω,
while the power source at port 2 represents the injection amplifier in a 50 Ω
44
GammaL Source and Load-pull Architectures
GammaL
freq (0.0000Hz to 1.000GHz) freq (0.0000Hz to 1.000GHz)
(a) (b)
GammaL
freq (0.0000Hz to 1.000GHz)
(c)
Figure 3.9: ΓL at the DUT reference plane in a Smith chart normalized

to 10 Ω for a DUT delivering 200 W of output power and with an output
impedance of 2 Ω. The injected power is swept from 0 to 200 W, and
the ΓL is plotted for: (a) no impedance transformation, (b) an impedance
transformation to 10 Ω and (c) an impedance transformation to 5 Ω.
3.4 Injection power and load amplifier linearity 45
system. An impedance transformation from the 50 Ω system impedance to a

lower impedance is implemented by using an ideal two-port S-parameter equa-
tion block. The injected power is swept from 0 to 200 W, with a phase sweep
from 0 to 360 degrees. The results of the simulation are shown in Fig. 3.9,
where the ΓL that can be achieved at the DUT reference plane is shown on a
Smith chart normalized to 10 Ω for three different cases: with no pre-match
and with an impedance transformation to 10 Ω and to 5 Ω. From Fig. 3.9,
it is clear that the Smith chart coverages benefits greatly from the use of a
pre-matching circuit.
When considering multi-tone or modulated signals, the situation becomes
more complicated as the linearity of the injection amplifier needs to be taken
into account [41]. To study the linearity constraints on the injection amplifier
we can consider a two-tone test signal. The power injected by the load am-
plifier at the third-order intermodulation frequency products of the two-tone
test signal is given by,
Pa2 ,IM3 cdBm = 3 · Pa2 ,f und cdBm − 2 · IP3,a2 =

"
(1 − |ΓDU T |2 )
= 30 · log Pb2 ,f und cmW · ·
(1 − |ΓSY S |2 )
#
|ZDU T + Z0 |2 |ZL − ZSY S |2
· · +
|ZSY S + Z0 |2 |ZDU T + ZL |2
− 2 · IP3,a2 , (3.5)
where Pb2 ,f und is the available power coming from the DUT at the fundamental
tones, and Pa2 ,f und and IP3,a2 are the power injected by the load amplifier at
the fundamental tones and its output third-order intercept point, respectively.
Another harmonic balance simulation with Agilent ADS is performed using
the simple schematic illustrated in Fig. 3.10. In this schematic, an amplifier
component based on a polynomial model is used to simulate the DUT and
the injection amplifier linearity. The same DUT is used as for the single-
tone considerations, with a maximum output power of 200 W and an output
impedance of 2 Ω. The output third-order intercept (OIP3 ) is set in this
simulation to 63 dBm. For this device, the output power is set equal to 50 W
per tone in order to achieve the same peak voltage as in the single-tone case.
These conditions yield an actual third-order intermodulation (IM3 ) of the
DUT of -30.35 dBc.
The results of the simulation are shown in Fig. 3.11, where the apparent
IM3 of the DUT is plotted as a function of the decreasing OIP3 of the injection
vin vout
P_nT one P_nT one
I_Probe
PORT 1 PORT 2
I_dut2
Num=1 Amplifier2 Amplifier2 Num=2
S2P_Eqn
Z=50 Ohm DUT Inj_Amp Z=50 Ohm
S2P1
Freq[1]=2 GHz S21=38.456 S21=dbpolar(50,0) Freq[1]=2 GHz
S[1,1]=0
Freq[2]=2.1 GHz S11=polar(0,0) S11=polar(0,0) Freq[2]=2.1 GHz
S[1,2]=1
P[1]=dbmtow(7.7472) S22=polar(0.9231,180) S22=polar(0,180) P[1]=polar(dbmtow(power),180)
S[2,1]=1
P[2]=dbmtow(7.7472) S12=0 S12=0 P[2]=polar(dbmtow(power),180)
S[2,2]=0
Z[1]=50
Z[2]=5
Figure 3.10: ADSTM schematic testbench for the evaluation of the required
injection amplifier linearity by active load-pull.
amplifier for different pre-matching conditions of the system impedance. The

dotted line is the actual IM3 level as would be achieved with a passive circuit.
The dot-dash line represents the IM3 level due to the Pa2 ,IM3 as approximated
by equation 3.5. From Fig. 3.11 we can observe that the correct IM3 level is
only achieved when the injection amplifier OIP3 is sufficiently high. When the
injection amplifier is less linear, it will introduce significant IM3 products that
will cause an error in the measurements, such as IM3 increase or cancellation
effects.
Consequently, to have reliable linearity measurements in a conventional
active load-pull setup, even when pre-matching is used, the injection amplifier
linearity (and thus its peak power) needs to be at least 10 times higher than
that of the DUT. It is obvious that at high power, these amplifiers, if avail-
able, will be extremely expensive. For this reason, active load-pull systems
that can offer communication standard compliant device testing, for example,
W-CDMA at base-station power levels (100 W and above), have not been
demonstrated to date.
3.5 Conclusions
When looking at traditional load-pull topologies, it is clear that all of them
have advantages and disadvantages. Mechanical tuners are simple, relatively
inexpensive and can handle high power, but they are slow and limited by losses.
Active load-pull systems can provide ΓL ≥ 1, are compact and therefore eas-
ily integrated for on-wafer measurements, but require expensive band-limited
amplifiers for high-power devices.
By combining passive tuners and active topologies in the same “hybrid”
system, it is possible to obtain many of the advantages of both systems while
reducing the disadvantages [27]. Traditional passive mechanical tuners can be
used to reflect high power at the fundamental frequency, thus pre-matching the
3.5 Conclusions 47
−20
Z =5 Ω
sys
Z sys =7 Ω
−25 Z sys =10 Ω
Passive Impedance
IM 3 level DUT [dBc]
Calculated
−30
−35
−40
−45
75 73 71 69 67 65 63 61 59
OIP Injection Amplifier [dBm]
3
Figure 3.11: Harmonic balance simulated IM3 level of the DUT vs. de-
creasing OIP3 of the injection amplifier for different impedance pre-match
values. The dotted line is the actual IM3 level as would be achieved with
passive matching techniques. The dot-dash line represents the IM3 level
only due to the Pa2 ,IM3 as approximated by equation 3.5. A polynomial
model was used for the amplifier linearity.
system impedance to a value closer to the optimum impedance of the DUT.

In this case a much smaller power (and therefore a cheaper power amplifier) is
required by an active load to overcome losses and achieve ΓL = 1. Additionally,
active tuning can be used at higher harmonics, where ΓL = 1 is required for
optimizing PA efficiency, and the power levels are often well below the power
of the fundamental signal. In this case less-expensive wideband amplifiers may
be used to accomplish active harmonic load-pull.
No matter which combination of techniques is used, electrical delay is
something inherent to the analog nature of the measurement setup and can
only be minimized, by reducing the physical distance between the DUT refer-
ence plane and the tuning element [41]. This problem is ultimately the main
limitation that impairs the user from performing load-pull measurements us-
ing realistic wideband communication signals. For this purpose, the following
chapters of this thesis propose a new load-pull measurement system based on
mixed-signals techniques, which combines the traditional analog microwave
and RF methodologies with the advantages of digital signal generation and
acquisition. The system realized proves to have many advantages over exist-
ing techniques, such as high-speed single-tone device characterization, high Γ,
and wideband impedance control for realistic communication standard com-

pliant modulated signals, resulting in a far more realistic characterization of
the DUT.
Chapter 4
Mixed-Signal Active
Load-Pull with Realistic
Wideband Modulated Signals
The previous chapter has introduced the traditional load-pull topologies along
with their strengths and weaknesses, and has highlighted several areas where
major improvements are required in order to meet the demands of today’s
power amplifier design. The following chapters describe a novel active har-
monic load and source-pull system, which was developed at the Electronics
Research Laboratory of the Delft University of Technology during this the-
sis work, to overcome the limitations of conventional load-pull configurations
outlined previously. The proposed solution couples traditional analog and mi-
crowave techniques with low-frequency signal acquisition and generation and
the related digital signal processing, and will be referred to as “mixed-signal”
active load-pull.
This chapter describes the principles of this approach, the specific hard-
ware and software requirements and the ability of this load-pull setup to work
with realistic wideband communication signals. Measurement examples are
provided that illustrate the realized system functionality.
4.1 The wideband, open-loop load-pull ap-

proach
When testing a non-linear device with modulated signals, the measured wave-
forms will include not only the fundamental frequency band, but also the
49
Mixed-Signal Active Load-Pull with Realistic Wideband
50 Modulated Signals
Figure 4.1: Principle of the proposed wideband, open-loop active load-

pull approach. When the nonlinear DUT is excited with a user-defined
modulated signal as , it generates signals in the baseband, fundamental and
higher harmonic frequency bands. By measuring the device generated
waves (b1,n and b2,n ), as well as the incident waves, the signals to be
injected are estimated by successive iterations.
baseband, higher-order harmonics, and intermodulation distortion products

(Fig. 4.1). Since the reflection coefficient is defined as the ratio of two travel-
ing waves, the effective control of the impedances over the entire modulation
bandwidth offered to the DUT can be enforced by establishing the desired
linear ratios of the incident and device generated waves on the DUT versus
frequency. In other words, we need to inject a signal at all of the frequency
components of interest according to the linear relation
ax,n (fn ) = bx,n (fn ) · Γx,n (fn ) , (4.1)
in which ax,n and bx,n are the incident and reflected waves at port x and
harmonic index n, while Γx,n represent the user-defined reflection coefficients
versus frequency for port x and harmonic index n.
As in the classical open-loop approach, only the content of the driving
waveform (as ) is known prior to the acquisition. All other injection signals
(a1inject,n and a2inject,n ) containing all the frequency components of the signal
4.1 The wideband, open-loop load-pull approach 51
DEFINE DESIRED REFLECTION

COEFFICIENTS VS. FREQUENCY
CALCULATE FIRST GUESS

I AND Q WAVEFORM SPECTRUM OF THE
INJECTION SIGNALS
INVERSE FFT I AND Q WAVEFORMS

-
DOWNLOAD TIME DOMAIN I AND Q IN AWG
ACQUIRE & FFT IF WAVEFORMS

-
CALCULATE GAMMA VS. FREQUENCY
CALCULATE I AND Q WAVEFORM SPECTRUM
OF THE INJECTION SIGNALS BASED ON THE
GAMMA ERROR
COMPARE MEASURED GAMMA WITH

DESIRED GAMMA VS. FREQUENCY
-
CALCULATE ERROR
YES GAMMA ERROR

>
DESIRED ERROR ?
NO
EXIT OPTIMIZATION ROUTINE
Figure 4.2: Flow diagram for the optimization of the reflection coefficients.
of interest can be created from scratch. Any desired reflection coefficient

behavior vs. frequency can be created by iteratively adjusting amplitude and
phase of the injected waveforms independently at each frequency component
of interest. By monitoring the deviation of the measured reflection coefficient
with the desired one at each frequency, the spectrum of the injected wave is
optimized and found by iteration, as shown in Fig. 4.2. The error checking
and optimization are done in the frequency domain, while the actual injection
signals are loaded and acquired in the time domain. The open-loop approach
guarantees that no sustained oscillations can occur. In practice when the user-
defined, target reflection coefficients (at its input or output) force the device
to operate in an unstable region, the system will simply fail to converge to its
end solution. In all other situations normal convergence of the optimization
algorithm will occur.
Although conceptually simple, this method requires wideband, highly lin-
ear data acquisition with high dynamic range to measure the reflection coef-
ficients at every spectral component of the complex modulated signals with
its related distortion products. Secondly, the a-waves need to be generated
with a high dynamic range. Thirdly, the signal injection at the various ports
need to be phase coherent at both RF as well as at the baseband frequen-
cies. To enable operation with realistic wideband signals as currently used in
modern wireless communication systems, these requirements demand a dedi-
cated hardware and software system implementation, which we discuss in the
following sections.

A simplified block diagram of the proposed measurement setup is shown in
Fig. 4.3. The test-set is based on the traditional four-coupler configuration
as used in most conventional network analyzers. This configuration allows
the simultaneous measurements of the source, input and load reflection co-
efficients at the DUT reference planes. The coupled a and b waves of the
DUT are down-converted to an intermediate frequency (IF) and then sam-
pled with wideband A/D converters (100 MS/s sampling frequency). This
architecture enables the direct measurement of the device reflection coeffi-
cients over a bandwidth of 40 MHz in a single data acquisition. With this
hardware, wider bandwidths (up to 120 MHz) and the frequency content in
the harmonic bands can be measured by stepping the frequency of the local
oscillator (LO) which drives the down-converting mixers. An open-loop archi-
tecture is used to tune the reflection coefficients at the source and load DUT
reference planes, at the fundamental and harmonics. These reflection coeffi-
Digital AWG Digital A/D Digital AWG
I/Q 2f0 signal I/Q 2f0 signal
I/Q f0 signal I/Q f0 signal

I 1
aREF
RF LO
load-pull system.
Source Source
HPR
a1,f0 a1,2f0 a1,BB b1,f0 b1,2f0 b1,BB b2,f0 b2,2f0 b2,BB a2,f0 a2,2f0 a2,BB To
LO
X2
a1,f0 a1,2f0 b1,f0 b1,2f0 b2,f0 b2,2f0 a2,f0 a2,2f0
To RF To RF
LO LO LO LO LO LO @ f0 @ 2f0
On-wafer configuration
RF @ f0 PA @ f0 RF @ f0
To DC To DC PA @ f0
DUT
Bias Bias
Tee Tee
Input section Output section
Reference
BaseBand I V Planes I V BaseBand Load
Reference
PA @ 2f0 Plane PA @ 2f0
RF @ 2f0 RF @ 2f0
To DC To DC
DUT
Reference Planes
BaseBand I V I V BaseBand
High power fixture with bias decoupling
Figure 4.3: Simplified block diagram of the mixed-signal active harmonic

53
cients are synthesized by injecting signals into the DUT which are generated
by baseband arbitrary waveform generators (AWG) and up-converted using
in-phase/quadrature (IQ) modulators. All the AWGs and the A/D converters
are integrated in a PXI express platform. They share the same time-base and
are fully synchronized. Since data generation and data acquisition of both
RF signals and DC parameters are handled through the PXI-based D/A and
A/D instrumentation, no mechanical tuners, VNA or DC-parameter analyzer
is needed, yielding a cost-effective, high-end characterization solution. Custom
bias-Tees with low inductance are placed directly at the DUT reference planes
in order to minimize the electrical delay of the baseband (BB) impedance,
which is implemented for now as a passive impedance switch bank [41]. Low-
frequency couplers for the baseband impedance measurement are also imple-
mented on the baseband board. Note that in a more extended version, the
baseband impedance can also be made active and controlled by an additional
AWG (see Section 7.1.2).
4.2.1 Signal generation

The source and all injection signals needed to create the user-defined reflec-
tion coefficients at the DUT reference planes originate from fully-synchronized
(400 MS/s) arbitrary waveform generators. When using the open-loop load-
pull concept to control the reflection coefficient provided to the DUT, it is
important that the source and all the injected signals at fundamental and har-
monics are phase coherent, or in other words, are perfectly locked in phase
and exhibit no phase drift among each other. To guarantee phase coherency,
IQ up-conversion is preferred over digital IF techniques [60]. By employing
IQ up-conversion, a single RF source can be used to drive the local oscillator
port of the IQ mixers that up-convert the baseband signals. In this setup,
frequency multipliers are used to obtain the LO signals which drive the IQ
mixers used for generating the higher harmonics (e.g., a x2 multiplier is used
for generating the 2nd harmonic). This approach guarantees that the active
loads and the driving signal are fully phase coherent.
Finally, computer-controlled attenuators and high-power amplifiers are
placed after the IQ up-converters, in the signal path, in order to level the
power of the injection signals. This approach allows us to make full use of the
maximum dynamic range of the arbitrary waveform generators at all times,
something that will prove essential for meeting the spectral requirements of
modern wideband communication signals.
Another advantage of the IQ approach, compared to other known sig-
nal generation techniques, is the relatively limited length of the data records
needed to fulfill the standard model requirements of complex modulated sig-
nals, yielding a significant speed advantage in practical measurement situa-

tions.
4.2.2 Data acquisition

To obtain an accurate representation of the spectral content of the RF waves
at fundamental and harmonic frequencies, the measured power waves need to
be down-converted to a lower frequency prior to data acquisition. This is a
common technique used in super-heterodyne receivers, and yields the highest
possible dynamic range. However, to do this correctly for modulated signals,
it is very important that the detection path is free from nonlinear errors that
cannot be corrected by linear calibration techniques. To maximize the detec-
tion dynamic range of the harmonic frequency components, power splitters
and high-pass filters are used at the detection ports of the couplers in the
input and output sections, as shown in Fig. 4.3 [61]. By high-pass filtering
the higher harmonic components, the mixer used for the down-conversion in
this signal path is protected from the (high) power of the fundamental signal.
This drastically relaxes the mixer linearity requirements for the second and
higher harmonics, improving the quality of the acquired signals. By stepping
the LO frequency, the frequency band centered around the harmonic of inter-
est is down-converted to its low IF representation for data acquisition. The
detection dynamic range of the system in the fundamental frequency band
is maximized by using variable attenuators in the RF path and high-power,
high-linearity mixers. Note that DUT nonlinearities also give rise to baseband
signals, which do not need frequency conversion prior to the data acquisition.
The resulting low IF signals and the baseband signals from the DUT are fed to
high-speed multiplexing switches in order to reduce the number of wideband
A/D converters needed.
4.2.3 System operating frequency and bandwidth

The system operating frequency has in principle no strict limitation, since
it is only bounded by the frequency handling capabilities of the IQ modula-
tors, test amplifiers and multipliers, which are widely available up to K band
commercially. The standard test-set implementation allows measurement and
control of the DUT reflection coefficients from 400 MHz to 18 GHz. The only
band-limiting components of the system are the frequency multiplexers and
the high-pass filters for the harmonic mixers, which cover one octave band-
width (e.g., 1.5 GHz to 3 GHz). These components are external to the test-set
and can be easily replaced to change the frequency of operation.
When considering the modulation bandwidth, two different bandwidths

should be considered, namely the signal generation bandwidth and the signal
detection bandwidth. The maximum analog frequency of the arbitrary wave-
form generators sets the upper limit for the input signal as well as the band-
width over which the reflection coefficients can be controlled. In the proposed
setup, the arbitrary waveform generators have a sampling rate of 400 MS/s
and can produce signals up to 160 MHz, which in IQ up-conversion results in
a maximum controllable bandwidth of 320 MHz. The A/D converters were
selected for maximum dynamic range and therefore have a slightly lower ana-
log bandwidth of about 40 MHz. However since the only information needed
for the optimization process is the reflection coefficients versus frequency at
the DUT reference planes, no phase coherence is required between the data
acquisition and the signal generation. Therefore the LO frequency used for
down-conversion can be stepped to cover larger bandwidths, just like in a tradi-
tional network analyzer. In this way, different parts of the frequency spectrum
are brought into the range of the A/D converter. However, to avoid mixer
down-conversion of the undesired sidebands, the maximum signal bandwidth
is limited to 120 MHz, due to the 40 MHz bandwidth of the A/D converters
used.
4.2.4 System calibration

The system calibration is relatively straightforward and is a combination of
the techniques described in [62–64]. A traditional two-port calibration (SOLT,
LRM, TRL, etc.) [62] is performed at the DUT reference planes to measure
the device reflection coefficients. Afterwards, an absolute power calibration is
done by connecting a power meter at either one of the DUT reference planes.
An additional calibration step can be performed at the load reference plane
to allow on-wafer measurements. During this step “short”, “open” and “load”
standards are connected in turn to the load reference plane, when a “thru” is
used as a DUT. In this case, since the power meter cannot be directly con-
nected to the wafer probes, the power calibration is also performed at the
load reference plane [63]. An additional calibration step is necessary for mea-
suring the baseband impedance and is performed by use of “short”, “open”
and “load” standards at the DUT input and output reference planes [64].
Finally, an optional calibration step can be performed to correct for IQ modu-
lator imperfections. During this step the IQ modulator’s leakage is minimized
by adjusting DC offsets, while balance and quadrature errors are corrected
through digital pre-compensation [65].
4.3 Signal processing for modulated signal measurements 57
4.3 Signal processing for modulated signal

measurements
The following sections describe the signal processing required to generate and
measure the waves which are used during modulated signal measurements.
4.3.1 Signal generation

When working with complex modulated signals, a good place to start is the
modulation test standard [9, 66]. According to the standard, a test signal is
created which consists of a finite sequence of IQ data samples specified in
the time domain. In conventional lab instrumentation such as vector-signal
generators, this sequence for a given standard (e.g., W-CDMA) is typically
embedded in the instrument. During testing these signals are uploaded in
the internal arbitrary waveform generators (AWG) and up-converted with IQ
mixers yielding the modulated RF signal. In practice, these test records are
repeated sequentially yielding a large but finite number of discrete spectral
components in the frequency domain. More precisely, the number of samples,
in combination with the sampling speed at which the signal is generated,
result in an effective frequency bin size (∆fAW G ), or frequency resolution of
the generated signal
f sAW G 1
∆fAW G = = , (4.2)
NAW G TM OD
where ∆fAW G represents the frequency bin size of the generated signals, and
f sAW G and NAW G are (respectively) the sampling frequency and the number
of samples used by the arbitrary waveform generators to construct the wave-
form. TM OD is the minimum period of the source signal in the time domain,
that is needed to comply with the specifications of the standard test model.
To provide the reader with an example, a W-CDMA signal has a chan-
nel bandwidth of 5 MHz, a chip rate of 3.84 Mcps, 2560 chips/slot and
15 slots/frame. When considering one frame, the complex waveform is 10 ms
long, or in other words, it will have a frequency resolution of 100 Hz. If we
then consider a single slot, the frequency resolution becomes 1.5 kHz. This
frequency representation allows us to analyze modulated communication sig-
nals like “classical” multi-tone signals, but now with a very large number of
frequency tones (e.g., more than 23,000 frequency tones when considering a
bandwidth of 35 MHz).
Amplitude [dBm]
Amplitude [dBm]
0 0
I Q
∆f
-40 -40
BW/2
-80 -80
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
Frequency [MHz] Frequency [MHz]
(a)
Amplitude [dBm]
-30
-60 IM3 BW IM3

IM5 IM5
-90 I
0 20 40 60 80 100
Frequency [MHz]
(b)
Amplitude [dBm]
Amplitude [dBm]
-30 -30
I Q
-60 -60 IM3
IM3 IM5
IM5
-90 -90
0 10 20 30 40 0 10 20 30 40
Frequency [MHz] Frequency [MHz]
(c)
Figure 4.4: Illustration of the generated and acquired signals in the pro-
posed load-pull system. Each circle corresponds to a frequency bin. (a)
Frequency-binned spectral content of the I and Q waveforms for generat-
ing the drive signal of the DUT. (b) Down-converted low IF representation
of the spectrum in the fundamental band at the output of the DUT. (c)
Spectral content of the I and Q waveforms for generating the active load
injection signal to achieve the user-defined reflection coefficient over the
fundamental band.
4.3.2 Data acquisition

A fraction of the waves incident and reflected by the DUT is taken by the
directional couplers and fed to high-linearity mixers that down-convert the
signals for digitization, after which an FFT is applied to obtain their spectral
content. The sampling speed of the A/D converter and the time span used
for the data acquisition also set an effective frequency bin size here, which
must be compatible with the original applied test signal, as described by the
following expression
f sA/D ∆fAW G 1
∆fA/D = = = , (4.3)
NA/D k k · TM OD
where ∆fA/D is the resulting frequency bin size of the acquired signals, f sA/D
and NA/D are (respectively) the sampling frequency and the number of samples
used by the A/D converters, and k is an integer.
The frequency bins of the acquisition and the generation should match for
a correct measurement, thus the frequency resolution of the A/D converter
should be set equal (k=1), or an integer factor larger (smaller frequency bin
size) than that of the generated signals.
4.3.3 Working example

To provide the reader with some intuitive insight into how these spectra appear
when operating with modulated signals, Fig. 4.4 depicts the frequency binned
spectral content of the I and Q signals to be delivered to the IQ modulator
(Fig. 4.4a). This block generates the RF source signal which drives the DUT
with a given modulation (e.g., W-CDMA). Due to the always-present non-
linearities of the active device under test, the DUT generated waves (b1,f und
and b2,f und ) will contain intermodulation sidebands besides the desired funda-
mental signal. Moreover, spectral content generated by nonlinearities will also
be present in the baseband and harmonic frequency bands. When consider-
ing fundamental operation (same reasoning applies also for the harmonic fre-
quency bands), the down-converted RF signal with intermodulation sidebands
is given in Fig. 4.4b. In order to realize the desired reflection coefficients over
the total bandwidth where spectral content is present, the I and Q injection
signals must now include the third- and fifth-order intermodulation distortion
(IM3 and IM5 ) sidebands (Fig. 4.4c). Failing to provide the proper signal at
the IM3 and IM5 frequency bands would create an unrealistic 50 Ω termina-
tion for those DUT generated signals, invalidating any linearity performance
measurement.
4.4 System performance

To demonstrate the functionality of the setup, a test signal composed of 161
sinusoidal tones in the bandwidth between 2060 MHz and 2220 MHz is fed to a
“thru”, while the output active load is set to provide an open condition over the
whole 160 MHz bandwidth. Fig. 4.5 shows the measured reflection coefficient
at the output reference plane of the “thru” as a function of frequency. It is
clear from the plot that the desired reflection coefficient (ΓL = 1) is achieved
for all frequencies without any phase delay or amplitude unbalance.
Figure 4.5: Measured reflection coefficient at the output reference plane

of the DUT for a signal composed of 161 sinusoidal tones in a 160 MHz
bandwidth.
−60
Receiver Attenuator = 0 dB
−65 Receiver Attenuator = 10 dB
−70
−75
IM 3 [dBc]
−80
−85
−90
−95
−100
−15 −10 −5 0 5 10 15 20 25 30
Output Power [dBm]
Figure 4.6: IM3 measurements on a “thru” for different receiver attenua-

tion settings. The results show the system measurement dynamic range.
4.4.1 Measurement linearity and dynamic range

Since mixer based down-conversion is used to measure the reflection coeffi-
cients at the reference planes of the DUT in a wideband fashion, the dynamic
range of the signal detection will be limited by the system noise floor at lower
powers and by the mixers nonlinearities at very high power levels.
Consequently, we have optimized our system for low noise and made use
of high-linearity mixers in the signal detection path. Furthermore, stepped
attenuators, controllable from 0 to 70 dB of attenuation in steps of 10 dB, are
placed in the receiver path, just before the down-converting mixers.
Fig. 4.6 shows the IM3 measured on a “thru” for a two-tone stimulus, and
for different attenuation settings, versus output power. For each attenuation
setting, the resulting dynamic range is about 75-80 dB in the signal detection,
over a 15 dB span in output power.
4.4.2 Active load dynamic range

In order to control and optimize the reflection coefficient offered to the DUT
for a particular frequency bin, it is necessary to measure the power waves for
that frequency bin with high-accuracy and repeatability. For this reason, one
should avoid the use of noisy data since this would make the optimization of
the reflection coefficient unstable. Therefore, the reflection coefficient is con-
trolled only for those frequency bins where the power exceeds a user-specified
threshold level. The selection of these power levels sets the dynamic range over
which the load reflection coefficient can be controlled. We will refer to this as
the “active load dynamic range”. As mentioned in Section 4.2.1, computer-
controlled stepped attenuators are used to guarantee full use of the voltage
swing produced by the arbitrary waveform generators at all times. Therefore,
the active load dynamic range will be limited only by the measurement noise,
the IQ mixer imperfections and the robustness of the optimization algorithm
used.
To give the reader an estimate of the “active load dynamic range” that can
currently be achieved, a W-CDMA input signal centered at 2.14 GHz is fed to
a heterojunction bipolar transistor (HBT) with an emitter area of 528 µm2 .
The power of the signal is chosen purposely high so that the active device
clips and generates intermodulation sidebands. The active load at the DUT
output for the fundamental is set to provide a constant reflection coefficient
(ΓL = |0.283| ∠ − 135o ) across the entire 20.4 MHz bandwidth. Fig. 4.7 shows
the reflection coefficient and power spectrum measured at the DUT output.
Note that the reflection coefficient can be controlled accurately over the whole
(broadened) spectrum over a 60 dB power range, and the loss of fidelity at
0.5
Magnitude
0.4
0.3
0.2
0.1
2.13 2.135 2.14 2.145 2.15
(a) Frequency [GHz]
-90
Phase [degrees]
-120
-150
-180
2.13 2.135 2.14 2.145 2.15
(b) Frequency [GHz]
Magnitude [dBm/Hz]
0 60 dB
-20
-40
-60 Control threshold
-80
2.13 2.135 2.14 2.145 2.15
(c) Frequency [GHz]
Figure 4.7: Measured load Γ in (a) magnitude and (b) phase, and (c) power
spectrum (resolution bandwidth of 6 kHz) at the output reference plane of
a HBT overdriven with a W-CDMA signal yielding spectrum broadening.
The user-specified reflection coefficient is ΓL = |0.283|∠−135o . The power
threshold level for the load reflection coefficient is set to -65 dBm.
the edge of the band corresponds to those frequencies where the signal power
falls below the user-specified threshold.
4.5 Measurement results

To demonstrate the unique capabilities of the active open-loop load-pull setup,
a NXP GEN6 LDMOS device with a gate width of 1.8 mm is measured us-
ing wideband modulated signals in load-pull test conditions. In all of the
experiments a drain current and voltage of 13 mA and 28 V respectively are
used. For this device, the optimum fundamental load and source matching
conditions are found using conventional single-tone load-pull measurements,
4.5 Measurement results 63
1 1
-1 -1
-1 1 -1 1
Γsource,f , delay Γload,f , delay

1
Γload,f , no delay
1
Γsource,f , no delay 1
Γload,f , delay
1
Γsource,f , delay 2
Γload,f , no delay
2
Γsource,f , no delay 2
2
Figure 4.8: Source and load reflection coefficients at the device refer-
ence plane in the fundamental (2.1225 GHz - 2.1575 GHz) and harmonic
(4.245 GHz - 4.315 GHz) frequency ranges, with electrical delay (open
symbols) and without electrical delay (filled symbols).
namely: ΓL,f 1 = |0.6| ∠45o and ΓS,f 1 = |0.5| ∠90o . The input and output
baseband impedances enforce a short condition, and the input and output 2nd
harmonics are set to an open circuit condition (ΓL,f 2 = ΓS,f 2 = |0.95|) in or-
der to optimize the efficiency [67]. To highlight the excellent wideband control
and the delay-free electrical operation of the new measurement setup, a com-
parison is made with a previously developed state-of-the-art active harmonic
load-pull system [41], which was optimized especially for minimum electrical
delay. For this purpose, we use a two channel W-CDMA signal (centered at
2.135 GHz and 2.145 GHz) and set the input and output reflection coefficients
in the newly developed setup to the following two cases:
1. Without electrical delay.
2. With an electrical delay of 4.85◦ /MHz for the fundamental source and
load and 4.6◦ /MHz for the 2nd harmonic source and load.
Fig. 4.8 illustrates the source and load matching conditions provided to
the active device under test for the two different cases. Note that the filled
markers represent the source and loading conditions for the two-carrier W-
CDMA signal without any electrical delay, yielding completely overlapping
points in the Smith chart. As shown in Fig. 4.8, the fundamental load trajec-
tory has been shifted for the case with electrical delay such that the optimum
matching condition is now centered at 2.135 GHz. This was required to avoid
the unstable region of the active device.
It is important to note that this is a comparison to the “best known case” of
a classical closed-loop active load-pull system, since closed-loops are subject to
amplitude variations within the control frequency bands in practice. Moreover,
oscillation conditions in closed-loop systems for these very large bandwidths
are difficult to avoid due to the use of wideband loop filters. Passive load-pull
systems with harmonic tuning will have a comparable or even worse phase
variation of the reflection coefficients versus frequency than the closed-loop
system used in the comparison.
The measurement results are summarized in Table 4.1. There is significant
performance degradation for the active device when measured with electrical
delay present in the reflection coefficients. This is also evident from Fig. 4.9a
and 4.9b which show the power spectral density at the device output reference
plane for the fundamental and 2nd harmonic frequency bands. Note that a
5 dB output power drop and close to 8 % degradation of the power-added
efficiency (PAE) can be observed when compared to the situation with no
electrical delay. Furthermore, due to the electrical delay, the output power
spectrum shows significant asimmetry both at fundamental and 2nd harmonic
frequency bands.
MEASUREMENT RESULTS
Without With
electrical delay electrical delay
PAE 24.2 % 16.3 %
POUT Ch. 1 20.3 dBm 20.5 dBm
POUT Ch. 2 20.6 dBm 15.4 dBm
ACLR1 Ch. 1 -43.9 dBc -43.0 dBc
ACLR2 Ch. 1 -42.2 dBc -41.6 dBc
ACLR1 Ch. 2 -42.1 dBc -41.8 dBc
ACLR2 Ch. 2 -39.6 dBc -39.2 dBc
Table 4.1: Comparison of measurement results with and without electrical

delay.
4.6 Conclusions 65
-10 -20
Output Power Spectral Density [dBm/Hz]
Output Power Spectral Density [dBm/Hz]

-20 -30
-30 -40
-40 -50
-50 -60
-60 -70
Without delay Without delay
With delay With delay
-70 -80
2.125 2.13 2.135 2.14 2.145 2.15 2.155 4.25 4.26 4.27 4.28 4.29 4.3 4.31
Frequency [GHz] Frequency [GHz]
(a) (b)
Figure 4.9: Measured output power spectral density [dBm/Hz] vs. fre-
quency [GHz] of a NXP GEN6 LDMOS device (gate width 1.8 mm) in
the proposed load-pull setup (a) at the fundamental frequency band using
a 3 kHz resolution bandwidth. (b) at the 2nd harmonic frequency band
using a 6 kHz resolution bandwidth. The measurement is shown for the
two cases with (dashed line) and without electrical delay (drawn line).
The reflection coefficients offered to the device under test are given in
Fig. 4.8.
4.6 Conclusions
In this chapter, a novel active harmonic load-pull setup has been presented
which is suitable for large-signal device characterization under “real life” mod-
ulated signal stimulus. The new system is capable of synthesizing arbitrary
source and loading conditions at the fundamental and harmonic frequencies
over a frequency bandwidth of 120 MHz, and therefore permits testing of ac-
tive devices under realistic (circuit-like) conditions. Furthermore, even user-
defined reflection coefficients versus frequency can be downloaded to the mea-
surement system in order to perform active device testing.
The IQ signal up-conversion and low IF down-conversion provide a very
high dynamic range of about 80 dB in the signal detection and about 60 dB in
the modulated active load impedance control. This proves to be sufficient for
the communication standards used today in industry. Furthermore, this ap-
proach can be easily adjusted to any frequency band of interest (e.g., X-band)
since all of the parts required are commercially available. The IQ-based open-
loop approach in combination with the frequency binning technique described
resolves all of the conventional drawbacks of current load-pull techniques, while
still being competitive in terms of measurement speed.

The unique system capabilities have been demonstrated by “load-pulling”
a NXP GEN6 LDMOS device under a two carrier W-CDMA input stimulus.
Note that up-to-date two-carrier W-CDMA testing, which is one of the most
severe tests in industry for base-station amplifiers, is normally only performed
at the circuit board level, since conventional load-pull systems fail to handle
these extreme bandwidths. The advantage of controlling the electrical delay
has been made evident by comparing the difference in output power and power-
added efficiency of an LDMOS device with a state-of-the-art system.
Finally, the selected hardware configuration still leaves room for numer-
ous improvements and innovations which will be discussed in the following
chapters.
Chapter 5
High-Speed, High-Power,
Fully-Controlled,
Multi-dimensional Load-Pull
Parameter Sweeps
Traditional load-pull measurement techniques are considered to be an essential

tool in the large-signal characterization of high-frequency devices, since they
facilitate direct determination of the optimal loading conditions of the device
under test (DUT) for any give large-signal quantity (e.g., gain, output power,
POU T , or power-added efficiency, PAE). Nevertheless, both passive and active
techniques are highly demanding in terms of measurement time, especially
when it is desired to monitor and control multiple parameters simultaneously
(e.g., fundamental and harmonic terminations) at multiple input power levels.
This is usually reflected in extremely long measurement times that slow down
the PA design/optimization process. Recently, there have been several reports
in literature proposing “real-time” (high-speed) load-pull measurements by
synthesizing load reflection conditions through analog modulation techniques
[68–70]. Although very good results have been achieved with these techniques,
their analog nature still imposes limitations on the number, and how well these
parameters (including the non swept ones) can be controlled for their value
and/or sweeping range.
In this chapter, a simple mixed-signal approach to perform fully-controlled,
multi-dimensional parameter sweeps, is presented. This approach enables
sweeping multiple parameters at the same time (e.g., input power, fundamental
and 2nd harmonic load impedances), while being able to control all other non-
67
High-Speed, High-Power, Fully-Controlled, Multi-dimensional
68 Load-Pull Parameter Sweeps
swept parameters (e.g., fundamental and 2nd harmonic source impedances)

to their user-defined values. For this purpose, the system capability to co-
herently synthesize any arbitrary signal at the fundamental, as well as at the
harmonics, is used to generate multiple input power levels and source and
load conditions in one single-shot measurement, as explained in detail in the
following sections. Such functionalities result in an extremely fast large-signal
device characterization tool that can dramatically speed up the PA design
process.
5.1 Generation and measurement of load

and source terminations
In an open-loop active load-pull system (Fig. 3.5), a signal with controlled
phase and amplitude is injected to present to the DUT with a synthetic, user-
specified, reflection coefficient. Since in our fully coherent system it is possible
to generate any arbitrary signal, we can define the waveforms to be injected
into the DUT at the fundamental and harmonics frequencies, such that they
contain multiple sinusoidal time-segments with different amplitude and phase
information. This approach allows the generation of any arbitrary succession
of synthesized load and source terminations, both at the fundamental and at
harmonic frequencies. Also, the test signal driving the DUT is generated in
the same way (i.e., with multiple wave segments) in order to provide the DUT
with different power levels. Fig. 5.1 shows an example of a time-segmented
RF wave, where four different loads are presented to the DUT at two different
power levels.
The entire ensemble of injected and device-reflected time-domain wave-
forms are then acquired, while the different time-segments are processed sepa-
rately. In this way, it is possible to measure the entire set of loading conditions
that are generated, and all of the large-signal device parameters of interest rel-
ative to each time segment in a single shot.
The major advantage of this method is that multi-dimensional parameter
sweeps can be executed within a single measurement, while the coherency
of the system and the measurement software allow tracking the information
embedded in each wave segment. This approach eliminates the overhead of
instruments settling times completely, which (in general) dominate the total
measurement time in conventional load-pull setups, while providing a greater
control than what is possible with analog load modulation methods.
Although a single measurement execution of such parameter sweeps takes
only a fraction of a second, practical situations quite often require accurately-
5.1 Generation and measurement of load and source
terminations 69
RF Input Signal
Power 2
1 Power 1
Voltage [V]
−1
0 5 10 15 20 25 30
(a) Time [µs]
RF Load Injection Signal
Load 2 Load 3 Load 4
1 Load 1
Voltage [V]
−1 Load 2 Load 3 Load 4

Load 1
0 5 10 15 20 25 30
(b) Time [µs]
Figure 5.1: Time-segmented RF waves for (a) multiple input power levels
and (b) load termination control. In this example four different loads over
a range of two power levels are presented to the DUT.
controlled power and loading conditions, which in an open-loop system can

only be achieved through successive iterations. In view of this, the coherent
mixed-signal IQ generation in our system is exploited to optimize the indi-
vidual time-segments of the IQ waveforms independently, offering full control
of the applied loading and power conditions. This feature can, for example,
be used to target any given area in the Smith plane, while at the same time
support sweeping of the input power. These “real-time” multi-dimensional
parameter sweeps are at least two orders of magnitude faster than the conven-
tional approach of using measurement of only one signal condition at a time
to construct a parameter sweep.
The total number of different measurement points is only limited by the

size of the AWG’s memory, and it is higher than 40,000 points with the cur-
rent system configuration. The DC voltage and current waveforms needed
for power efficiency calculation are also measured for each time-segment, us-
ing the same A/D converters as for the reflection coefficient measurements.
In particular, the DUT current is sensed by a current probe, or across an
external resistor inserted in the ground path just before the bias-tee. The
resistor method requires a simple additional calibration step to accurately
pre-characterize the sense resistor, as described in [71].
1
0.13
0.14 ΓLf0 1
0.12 0.13
0.14 ΓSf0
1.0
0.15 0.15
0.1
6
0.1
ΓLf0 Target 0.1
6
0.1
ΓSf0 Target
7 7
0.8 0.1
ΓL2f0 0.8 0.1
ΓS2f0
2.0
2.0
8 8
0.1
0.1
ΓL2f0 Target ΓS2f0 Target
9
9
0.6
0.2
0.6
0.2
0
0
0.2
0.2
1
1
0.4 5.0 0.4 5.0
1.0
1.0
0.22
0.22
0.23
0.23
0.2 0.4
10
0.2 0.4
10
0.24
0.24
2.0
5.0
1.0
2.0
5.0
10
20
50
10
20
50
0.25
0.25
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(a) (b)
Figure 5.2: Load-pull example with controlled fundamental and harmonic

terminations. The user-specified targets for the reflection coefficients are
represented with squares, while the actual measured ones are depicted by
a cross. (a) load plane (b) source plane.
5.1.1 Measurement examples and results

To demonstrate the functionality of the new measurement methodology, a
NXP GEN6 LDMOS device with a gate width of 1.8 mm has been measured.
In these experiments the device has been biased at a drain current (ID ) of
13 mA and a drain voltage (VDS ) of 28 V, while the fundamental frequency of
the RF stimulus is 2.14 GHz.
A first measurement example is given in Fig. 5.2, which shows the capa-
bility of the system to obtain, with excellent precision, any arbitrary source
and load impedance state at the fundamental and harmonic frequencies. In
this example the fundamental load termination provided to the DUT is swept,
covering a circle centered at ΓL = |0.5| ∠40o with a radius of 0.3. The source
fundamental is kept constant at the value ΓS = |0.5| ∠90o , while the 2nd
harmonic source and load terminations are kept constant in an open circuit
condition.
The ability to select any arbitrary range of reflection coefficients is partic-
ularly useful when it is desired to characterize only selected areas of the device
source or load terminations in the Smith chart, for example, when one wants
to avoid regions where the device is potentially instable, or loading conditions
5.2 High-power, real-time pulsed-RF measurements 71
that can yield device failure (voltage or thermal breakdown); something that
is far from trivial when using analog load-modulation methods. An example
is given in Fig. 5.3, showing the constant PAE contours for a given PAV S as
measured with the proposed technique.
In this example, the output stability circle was first obtained from the (pre-
viously measured) device small-signal S parameters [37], then only the stable
region was addressed in the actual measurement by a fundamental load sweep
at various (much higher) power levels. In this measurement the output power
of the device has been obtained for 90 different load terminations, keeping ΓS
fixed to the previously specified value, while at the same time sweeping the
source and load harmonic terminations between open and short conditions,
and the power available from the source from 1 to 16 dBm. The total time for
this measurement, which consists of more than 5,000 controlled measurement
points, is below 5 minutes.
Given the importance of second harmonic source and load termination con-
trol in PA design, which has been addressed several times in literature [72,73],
the capability of the presented system to simultaneously sweep the 2nd har-
monic impedances presented to the DUT can prove to be very useful when in-
vestigating the optimal device terminations for high-efficiency / high-linearity
operation. The constant PAE contours plotted in Fig. 5.3 show that the
highest PAE values are obtained with the 2nd harmonic load impedance set
to an open, and the 2nd harmonic source impedance set to a short. For this
latter case, the measured PAE as a function of POU T and the measured trans-
ducer power gain (GT ) as a function of PAV S are plotted for all the different
fundamental loading conditions in Fig. 5.4a and Fig. 5.4b, respectively.
Finally, the constant PAE contours shown in Fig. 5.5a and 5.5b allow the
reader to compare the results obtained using the proposed “real-time” tech-
nique and a traditional open-loop load-pull technique under the same PAV S
and loading conditions. The excellent agreement between the two methods
indicates that the much higher measurement speed of the new approach does
not affect accuracy.
5.2 High-power, real-time pulsed-RF mea-

surements
For single-tone CW signal conditions, the system configuration described is
able to generate and measure thousands of source and load conditions at dif-
ferent power levels in a single acquisition. This feature provides ultra-fast
load-pull device characterization. However, for high-power devices, the use
0.12 0.13 0.12 0.13

0.14 0.14
1 L−Plane Stability Circle 1 L−Plane Stability Circle
1.0
1.0
0.15 0.15
0.1 0.1
6 6
0.1 0.1
7 7
0.8 0.1
0.8 0.1
2.0
2.0
8 8
0.1
0.1
9
9
0.6
0.2
0.6
0.2
0
0
0.2
0.2
0.70
1
1
0.4 5.0 0.4 0.65 5.0
1.0
1.0
0.22
0.22
0.65
0.6
0.6
0.55
0.23
0.23
0.55
0.2 0.4 0.4 0.5 10
0.2 0.4
0.5 10
0.45 0.45
0.24
0.24
0.4
1.0
2.0
5.0
1.0
2.0
5.0
10
20
50
10
20
50
0.25
0.25
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(a) (b)
0.12 0.13 0.12 0.13

1 0.14
1 0.14 L−Plane Stability Circle
L−Plane Stability Circle
1.0
1.0
0.15 0.15
0.1 0.1
6 6
0.1 0.1
7 7
0.8 0.1
0.8 0.1
2.0
8 2.0 8
0.1
0.1
9
9
0.6
0.2
0.6
0.2
0
0
0.2
0.2
1
0.4 5.0 0.4 5.0

1.0
1.0
0.65
0.22
0.22
0.6
0.6
0.55
0.23
0.23
0.55
0.4
0.5
0.2 0.4 10
0.2 10
0.4
5 0.4
0.
4 0.5
0.24
0.24
0.45
1.0
2.0
5.0
1.0
2.0
5.0
10
20
50
10
20
50
0.25
0.25
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(c) (d)
Figure 5.3: Constant PAE load-pull contours for a NXP GEN6 LDMOS
device (PAV S = 14 dBm). In this measurement the power available from
the source is swept from 1 to 16 dBm for 90 different load terminations.
The source fundamental termination is set to ΓS = |0.5| ∠90o , while load
and source 2nd harmonic terminations are swept to: (a) open - short (b)
open - open (c) short - open (d) short - short using the proposed tech-
nique. Note that the unstable area for the DUT is avoided directly during
the measurement (i.e., no measurement points taken in the potentially
unstable region).
5.2 High-power, real-time pulsed-RF measurements 73
0.8 21
0.7
20
0.6
0.5 19
GT [dBm]
PAE
0.4 18
0.3
17
0.2
0.1 16
0 15
18 20 22 24 26 28 30 32 34 0 2 4 6 8 10 12 14 16 18
POUT [dBm] PAVS[dBm]
(a) (b)
Figure 5.4: (a) PAE as a function of POU T , (b) GT as function of PAV S ,

plotted for 90 different load terminations. The source fundamental termi-
nation is set to ΓS = |0.5| ∠90o . The 2nd harmonic load and source are
set to an open and a short, respectively.
0.12 0.13 0.12 0.13

0.14 0.14
1 L−Plane Stability Circle 1 L−Plane Stability Circle
1.0
1.0
0.15 0.15
0.1 0.1
6 6
0.1 0.1
7 7
0.8 0.1
0.8 0.1
2.0
2.0
8 8
0.1
0.1
9
9
0.2
0.2
0.6 0.6
0
0
0.2
0.2
1
0.6 0.6
0.4 5.0 0.4 5.0
0.22
0.22
1.0
1.0
0.5
0.5
0.23
0.23
0.4 0.4
0.2 0.4
10
0.2 0.4
10
0.24
0.24
0.3
0.3
1.0
2.0
5.0
1.0
2.0
5.0
10
20
50
10
20
50
0.25
0.25
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
(a) (b)
Figure 5.5: Comparison of load-pull PAE contours with swept fundamen-

tal load terminations (PAV S = 11 dBm). The source fundamental termi-
nation is set to ΓS = |0.5| ∠90o , 2nd harmonic terminations are set to open
circuit with (a) the proposed technique, (b) the traditional technique.
of CW conditions needs to be omitted to avoid self-heating, which in ex-

treme conditions can even yield device failure. This is especially true for
base-station devices which are optimized for operation with complex modu-
lated signals. These signals reach their peak values only occasionally. As a
result the active device operates most of the time in power back-off. Conse-
quently, pulsed-RF operation is required in order to create realistic load-pull
testing conditions. For this reason we extend the original concepts described
in Section 5.1 to pulsed operation at much higher peak power levels (above
100 W). These additional features allow the user to perform accurate high-
power, ultra-fast device characterization, while providing full control over the
maximum operating conditions of the active device, thus avoiding voltage and
thermal breakdown conditions.
All of the waveforms to be injected into the input and output ports of
the DUT are defined such that they contain multiple sinusoidal time-segments
with different amplitude and phase information. As a result, the device will
experience a sequence of time segments with different input powers and loading
conditions. When working under pulsed-RF conditions, the desired duty cycle
is achieved by adding a sufficient “idle time” after the stimulus representing
the different power and load states. An example is shown in Fig. 5.6, where
the input waveform to the DUT and the load injected waveforms are depicted.
In this simplified case the DUT will “experience” two input power levels, and
three different load impedances.
The desired injection signals are then optimized independently for each
segment to synthesize the desired impedances seen by the DUT. Note that
the use of iterations to optimize the injection signals opens up the possibil-
ity to introduce several features that ensure safe device operation during the
measurement, something that is extremely useful, especially at high power
levels. For example, it is possible to exclude from the measurement a range
of impedances that might cause device instability [74]. Furthermore, it is pos-
sible to check the gain compression of the device for every impedance during
each iteration, thus limiting the input power at impedances where the DUT
gain reduces by more than a user-specified amount.
5.2.1 High-power measurement examples

As application example of the proposed active load-pull concept, a complete
load-pull and power sweep for a NXP BLF7G22LS-130 device was carried out
at a frequency of 2.14 GHz under pulsed-RF conditions. In this example,
a pulse width of 10 µs and 10 % duty cycle were used. The power gain
compression was limited during this measurement to 4 dB to avoid device
degradation due to extreme gain compression, e.g., in higher load line regions.
5.3 High-power measurements with modulated signals 75
RF Input Signal
1
PW POWER 1 POWER 2
Voltage [V]
0.5
−0.5
PRP
−1
0 40 80 120 160 200 240
(a) Time [ µs]
RF Load Injection Signal
1
L1 L2 L3 L1 L2 L3
Voltage [V]
0.5
−0.5
−1
0 40 80 120 160 200 240
(b) Time [ µs]
Figure 5.6: Time-segmented pulsed-RF waves for (a) multiple input power
levels and (b) load termination control with pulsed-RF. In this simplified
example, three different loads for two input power levels are presented to
the DUT.
The results are shown in Fig. 5.7, where the maximum output power contour
and the PAE at 3 dB of gain compression are plotted. Also in Fig. 5.8, the
PAE versus output power calculated at a gain compression level of 3 dB is
shown. Note that the complete device characterization with 25 power levels
at each of the 50 load impedances takes less than 3 minutes.
5.3 High-power measurements with mod-

ulated signals
The mixed-signal active harmonic load-pull system is unique in that it can per-
form active load-pull device testing with communication-standard-compliant
wideband modulated signals like W-CDMA. This feature has been described in
detail in Chapter 4 and demonstrated for low-power levels. In this section the
capability of the system to work with complex modulated signals is extended
to the power levels that are typically in use for base-station applications (e.g.,
peak envelope power of 200 W).
As previously explained in Section 3.4, to synthesize a specific ΓL at the
DUT reference plane in an active load-pull system, it is necessary to inject a
certain amount of power into the DUT. An extra complication when working
0
Max PAE [%]
Pout @ compression [W]
−0.1
77.
−0.2 8
63.3
88.
.6
61.9
9
64
−0.3
99.
9
59.
11
1.0
3
−0.4
55 51.2
12
2
.2
13 .1
3.2
1
15 44.3
−0.5
45 16 .35
.9 6
−0.6 177 .4
.5
40.
−0.7 5
−0.8
−0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0 0.1
Figure 5.7: Load-pull contours of maximum power added efficiency and

output power at 3 dB of gain compression for a NXP GEN7 LDMOS
device at different load states and input powers, using pulsed single-tone
conditions (10 µs pulse width and 10 % duty cycle).
70
65
Power Added Efficiency [%]
60
55
50
45
40
35
60 80 100 120 140 160 180 200
Output Power @ compression [W]
Figure 5.8: Measured PAE vs. output power at 3 dB power gain compres-
sion level for a NXP GEN7 LDMOS device at different load states and
input powers, using pulsed single-tone conditions (10 µs pulse width and
10 % duty cycle).
5.3 High-power measurements with modulated signals 77
0 0
Average PAE [%] @ 3 dB PAR reduction
Average PAE [%] for 30 W output power Average Pout [W] @ 3 dB PAR reduction
ACPR [dB] for 30 W output power
−0.1 −0.1
−0.2 −0.2
−0.3 −0.3
33.
−0.4 7 −0.4
32.
30. 3
22.5
32
23.7
9 .3
29
24.9
30
28 .5
26.1
−0.5 29 .9
27.3
−0.5
26 .1 2 .5
28.5
26 8.0
29.8
25 .8
30.9
32.2
2 .6
33.4
.
2 4 −34.9 2 5
22 3.8 .2
−0.6 22. 4 −0.6 2 .
19 1.0 3
21. 6 −34.3 18 .5
19. 2 −33.8 .1
8 −33.3
−0.7 −32.7 −0.7
−31−.232.2
−0.8 −0.8
−0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 −0.6 −0.5 −0.4 −0.3 −0.2 −0.1 0 0.1
(a) (b)
Figure 5.9: (a) Load-pull contours of average power added efficiency and
ACPR for an average output power of 30 W. (b) Load-pull contours of
average power added efficiency and average output power at 3 dB of peak-
to-average ratio reduction. The related peak-to-average power (PEP) is
as high as 150 W.
at high powers with wideband signals arises because the linearity of the injec-
tion amplifier needs to be taken into account. When the injection amplifier is
not sufficiently linear, it will introduce intermodulation products in a “conven-
tional” active load-pull system, that will cause significant measurement errors,
such as IM3 increase or cancellation effects. To have reliable linearity mea-
surements in a “conventional” active load-pull setup (even when pre-matching
is used), the injection amplifier linearity (and thus its peak power) needs to
be at least 10 times higher than that the of the DUT, as concluded from the
simulation shown in Fig. 3.11. When working with devices designed for base-
station applications where peak envelope powers are as high as 200 W, using
such amplifiers becomes extremely expensive and impractical.
Instead, an iteration process is performed in the mixed-signal active load-
pull system, to optimize the reflection coefficient of each individual frequency
component of the wideband signal (e.g., 23,362 tones with 1.5 kHz spacing
for a W-CDMA considering a total bandwidth of 35 MHz). Due to these
iterations, the injection amplifier is pre-distorted for its own non-linearities.
This allows the use of injection amplifiers with much lower linearity than what
is typically required in conventional active load-pull systems.
As an example consider Fig. 5.9, which shows ACPR, average PAE and
output power for a single-channel W-CDMA signal at 2.14 GHz with a peak-
to-average ratio of 9.5 dB. It should be stressed that in these experiments the
maximum saturated power rating of the injection amplifier is only 200 W, with
an associated 60 dBm output IP3 . In these measurements, the non-linearity
of the injection amplifier does not affect the measurement results because ΓL
is controlled to the user-defined value in and out-of-band.
5.4 Conclusions
In this chapter, the measurement principles that enable the mixed-signal ac-
tive harmonic load-pull system to be compatible with the requirements of
high-speed, high-power, high-linearity base-station applications have been pre-
sented. The system provides ultra-fast large-signal device characterization in
both CW and pulsed conditions. For the latter, both the duty-cycle and the
gain compression of the DUT during the measurement can be controlled. All
of these features are crucial in guaranteeing the safe operating conditions for
high-power DUTs (> 100 W).
In addition, high-power device characterization with realistic W-CDMA
signals has been demonstrated. It was shown that the realized system can
compensate for nonlinearities of the injection amplifiers, which normally would
obscure the linearity and ACPR measurements. This property allows the use
of cheaper injection amplifiers, providing a lower PSAT than required in con-
ventional active load-pull systems. The ability to eliminate losses and electrical
delay while being completely free to define the source and load reflection coef-
ficients vs. frequency, allows perfect mimicking of in-circuit situations, making
the system an extremely valuable tool for the RF power amplifier developer.
Chapter 6
Measurement of
Time-Domain Waveforms
The measurement of time-domain voltage and current waveforms has proven

to be of fundamental importance for the design and optimization of active
devices operating under large-signal conditions. Engineers use time-domain
waveform measurements for several purposes: amplifier design [53, 75], device
model extraction [76, 77] and validation [78, 79], as well as for ruggedness
evaluation [80] of active devices.
Time-domain waveform measurements are easily made using a conven-
tional real-time oscilloscope at low frequency, while at RF and microwave
frequencies the situation is more complicated. This is due to the fact that the
parasitic effects of the measurement system can no longer be neglected or re-
duced to simple low-pass filtering effects, thus requiring extensive calibration
procedures such as those commonly used in VNAs.
In the following sections, the fundamentals of time-domain waveform mea-
surements are introduced. The performance of a commercially-available phase
reference and a newly-realized phase reference prototype are examined. Their
impact on the quality of the phase calibration is estimated. Finally, an ap-
proach for time-domain waveform analysis of multi-tone signals which are
closely spaced in frequency is introduced.
6.1 Time-domain waveform measurement

fundamentals
Due to their non-linear operation, power amplifiers generate harmonics at
frequencies N × f0 at their output when driven by a CW signal at frequency,
79
80 Measurement of Time-Domain Waveforms
f0 . When considering a periodic signal, the time-domain voltage and current

waveforms can be reconstructed from the incident and reflected waves (i.e.,
a and b) measured in the frequency domain at the input and output of the
non-linear DUT [81] as
p
vi (t) = Z0 · [ai (t) + bi (t)] (6.1)
1
ii (t) = √ · [ai (t) − bi (t)] , (6.2)
Z0
where i is the port number, and
N
X
ai (t) = ai,n · cos(nω0 t + φai ,n ) (6.3)
n=0
N
X
bi (t) = bi,n · cos(nω0 t + φbi ,n ) . (6.4)
n=0
Variables ai,n , bi,n and φai ,n , φbi ,n are the amplitude and phase coefficients
of the power waves measured at the fundamental and harmonic frequencies,
respectively.
The errors in a measurement system have to be removed by means of a
system calibration when working at microwave frequencies. In VNAs, all of
the error contributions of the measurement system are described by an error
model [62, 82]. As an example, consider the flow graph in Fig. 6.1 which
depicts the error model at the input port of the DUT. The four error terms
ed1 , es1 , i01 and i10 represent the error network between the DUT reference
plane and the measurement plane. By measuring several electrical standards,
the error terms of the model can be calculated and the measurement corrected
up to the DUT reference plane.
Conventional network analysis, however, allows only the measurement of
the ratio of power waves (e.g., b1 /a1 ). In fact, only the product term i01 · i10
is derived when referring to the classical 12 term error model used for VNA
calibration. This is better explained when considering the flow graph in Fig.
6.1. Equations 6.5 and 6.6 can be derived from the flow graph, which allow
the calculation of the calibrated incident and reflected waves at the input port
of the DUT from the measured quantities aM M
1 and b1 . It is clear that the
two terms i01 and i10 have to be known independently both in amplitude and
phase in order to calculate a1 and b1 according to
es1 · bM M
1 − ∆ · a1
a1 = (6.5)
i01
6.1 Time-domain waveform measurement fundamentals 81
a 1M i10 a1
ed1 es1 ΓIN
b 1M i01 b1
ΓINM DUT reference plane

Figure 6.1: Error flow graph at the input port.
bM M
1 − ed1 · a1
b1 = , (6.6)
i01
where ∆ = ed1 · es1 − i01 · i10 .
The addition of an extra calibration step with a power meter [63] allows
the calculation of the amplitude of the separate terms i01 and i10 from the
product term i01 · i10 , enabling the measurement of the absolute power of the
a and b waves. Finally, when the phase information of the input and output
signals are also required, an extra phase calibration step is needed to acquire
the phase information of the i01 and i10 terms.
Traditionally two techniques can be employed to generate a signal to be
used for a phase calibration:
• A comb generator (known as a phase reference) is stimulated with a

low frequency tone (e.g., 10 MHz) to create a homogeneous multi-tone
signal with known phase relationship between the tones [83, 84].
• An RF multi-sine is generated with known phase relations between

the various sinusoids (e.g., by a wideband AWG or by means of up-
conversion) and used as a reference signal [85].
In terms of hardware, the first systems capable of measuring time-domain

waveforms started to appear in the late ’80s. In the literature, mainly two
techniques have been described:
• Measurement of the phases of different harmonic products of the periodic

signal in the frequency domain with an extended VNA architecture.
• Sampling of the periodic waveform in the time-domain using sub-sam-

pling techniques (e.g., sampling oscilloscope).
6.1.1 Waveform reconstruction with VNAs

As previously mentioned, a conventional VNA can only measure the ratio of
power waves (i.e., reflection coefficients). This is not only related to calibra-
tion (only the product term i01 · i10 is known) as explained in the Section 6.1,
but depends on the hardware capabilities of a network analyzer. When trying
to extract the absolute phase information of the a and b waves, a conventional
VNA is not capable of providing consistent values because the measured ab-
solute phase at each frequency varies over time and over different acquisitions.
This is due to the fact that the signal generated by the local oscillator syn-
thesizer (LO), which is used to down-convert the measured signal, will have a
different phase, in respect to a common reference, every time one of the differ-
ent frequency components is measured. In order to understand this, a closer
look must be given to the architecture of modern factional-N phased locked
loops (PLL) which are employed in all commercial synthesizers. The output
frequency (fV CO ) of a fractional-N PLL is a fractional portion of the reference
frequency (fR ) and is given by fV CO = fR · (N + K/F ), where N, K
and F are integers [86]. Since the divider is changed in the loop dynamically,
and multiple fractional solutions exist which yield the same output frequency,
the phase of the output signal can change every time the same frequency is
selected (i.e., returning to the same frequency after a sweep). Therefore, every
time a signal at the same frequency is acquired, the local oscillator will have
a different delay with respect to the RF signal to measure. As a consequence,
the measured phase will vary at each acquisition, even if the phase of the signal
has not changed.
In order to properly extract the phase information, it is necessary to mod-
ify the hardware to ratio all of the a and b waves (i.e., at fundamental and
harmonics) against a reference signal (aref ) containing all of the frequency
components of interest (Fig. 6.2). This reference signal is created by feeding
a low frequency component, coherent with the RF signal source used for the
measurement, to a comb generator that creates all the harmonic products.
Finally, an additional phase calibration step has to be performed to calcu-
late the phase of the i10 and i01 error terms in the error model.
A first implementation of this approach was proposed in 1989 [83]. Since
then, several characterization setups have been developed based on this con-
6.1 Time-domain waveform measurement fundamentals 83
aREF a1 b1
LO
Source
Comb
Generator
DUT
RF
Source
Figure 6.2: Simplified schematic of a VNA down-converting receiver. To

measure the phase of a1 and b1 , all spectral components must be ratioed
against the reference signal aref created by the comb generator.
cept [81, 87, 88].

While this approach is relatively complex and requires stepping of the LO
to measure all the harmonics of interest, it allows the use of high dynamic
range receivers from conventional network analyzers (up to ≈ 120 - 130 dB).
6.1.2 Waveform reconstruction with sub-sampling

techniques
In these systems a sampling oscilloscope or similar data acquisition instru-
ment is used to acquire the time-domain traveling waves by sub-sampling the
periodic waveform. The sub-sampling procedure creates a replica of the pe-
riodic waveforms at low frequency that can be acquired by A/D converters.
The waves are first acquired in the time-domain, then transformed into the
frequency domain where the additional phase calibration step is performed.
The first system to measure time-domain waveforms used a two channel
sampling oscilloscope and was introduced in 1988 [89]. Successively more
advanced systems [53, 90–93] based on commercial receivers such as the mi-
crowave transition analyzer (MTA) [94] and the large-signal network analyzer
(LSNA) [95] appeared, coupling time-domain waveform analysis with load-pull
capabilities.

I/Q 2f0 signal I/Q 2f0 signal
I/Q f0 signal I/Q f0 signal

I 1 aREF
RF LO
Source Source
HPR
a1,f0 a1,2f0 a1,BB b1,f0 b1,2f0 b1,BB b2,f0 b2,2f0 b2,BB a2,f0 a2,2f0 a2,BB
Figure 6.3: Simplified block diagram of the phase reference channel needed
to enable waveform reconstruction in the mixed-signal load-pull system.
Detail from Fig. 4.3.
In general, systems based on these kind of receivers are relatively sim-

ple since no reference channel is necessary. Furthermore, depending on the
receiver used, they can be relatively fast since all the harmonics needed for
the waveform reconstruction are acquired at once. However, their dynamic
range is lower compared to the much more narrowband heterodyne architec-
ture (≈ 60 - 70 dB) due to the use of wideband samplers that also acquire a
prominent amount of unwanted noise.
6.2 Waveform reconstruction on the mixed-

signal load-pull system
The ability to measure time-domain waveforms combined with source and
load-pull tuning capability is of the utmost importance when performing am-
plifier design or studying amplifier classes. In addition to the previously de-
scribed features, the mixed-signal active load-pull system introduced in Chap-
ters 4 and 5 also allows the extraction of time-domain waveform data. Since
the system receiver is based on a super-heterodyne down-conversion scheme
(similar to conventional VNAs) a reference channel is also required to keep
track of the LO phase changes as described in Section 6.1.1. This is accom-
plished by driving a comb generator integrated inside the system with the CW
RF source signal used in the IQ up-conversion block (Fig. 6.3).
In the following sections, the system phase measurement repeatability is
presented. Then the phase reference specifications are discussed and the results
for different comb generators are shown. Finally, an example of a class-B
amplifier design on the mixed-signal load-pull system is provided as a working
example of the setup’s capabilities.
6.2 Waveform reconstruction on the mixed-signal load-pull
system 85
1
Phase Error [deg]
−1
−2
0 5 10 15
Frequency [GHz]
Figure 6.4: Standard deviation of the phase measurement of b2 for the

different measured harmonics.
6.2.1 System phase measurement repeatability

The phase measurement stability for the characterization system developed
in this thesis is verified by performing 30 repeated measurements on a comb
generator stimulated by a 2 GHz input signal, and by measuring the phase
of the output traveling wave up to the 7th harmonic. The measurements are
performed with an acquisition time of 2 ms. The results are shown in Fig.
6.4, where the standard deviation of the phase measurement of b2 is depicted
for the different measured harmonics.
6.2.2 Phase reference specifications

As described in Section 6.1, a phase reference device with known phase re-
lationship between the fundamental and harmonic components is required in
order to calibrate the system for measuring time-domain waveforms. Such a
device should work as a comb generator, or in other words, when stimulated
with a sinusoidal signal at frequency f0 it generates sufficient power at all of
the harmonics of interest. The most common implementations of comb gen-
erators make use of high-speed diodes in shunt configurations [83] acting as
rectifiers, non-linear transmission lines [96, 97], or digital circuits generating

very narrow pulses [98].
To use a comb generator as a phase calibration standard, the phase of
the output wave of the comb generator is measured on a sampling scope cal-
ibrated with a nose-to-nose calibration [99], or by an electro-optical calibra-
tion [100]. This calibration is ideally done by a specialized institute (e.g., a
national metrology institute). The resulting calibrated phase reference should
be portable from one system to another. For this reason there is the require-
ment that it needs to be insensitive to a wide set of external conditions. In
particular, it is essential that its behavior is independent of input power (in a
sufficiently wide range), and source and loading conditions at the fundamental
and harmonics, since these might change between the primary calibration of
the unit and the one performed on the large-signal setup.
In this section, the performance of different commercially-available comb
generators are evaluated as potential phase reference standards. The results
of a new comb generator prototype developed during this thesis work are also
presented for comparison. In particular, the phase stability of these devices
versus power, and versus fundamental and second harmonic load is analyzed
by measuring the phase of the b2 -wave while sweeping the input power and
performing fundamental and second harmonic load-pull. In all of the following
experiments a fundamental frequency of 2 GHz is used and 7 harmonics are
measured.
Marki A-0030 square-wave amplifier

First, a Marki Microwave A-0030 saturated amplifier is evaluated s a phase
reference standard. This particular model is an overdriven double-stage trav-
eling wave amplifier working from 5 MHz to 30 GHz, which generates a square
wave. The output spectrum of the amplifier is shown in Fig. 6.5, where it can
be seen that the output power of each spectral tone is sufficiently high up to
the 6th harmonic.
Fig. 6.6 shows the phase variation of the b2 -wave versus input power. Due
to the large phase variation of b2 versus power, the use of this device as a phase
calibration standard would require the knowledge of the absolute input power
provided to the amplifier in both the characterization and calibration phases.
Furthermore, Fig. 6.7 and Fig. 6.8 show the phase variation versus funda-
mental and 2nd harmonic output impedance up to a |Γ| = 0.3. The very large
variation versus load makes it impossible to transfer this amplifier together
with its phase calibration information from one setup to the other without
losing measurement accuracy. For these reasons such a device is clearly not
suitable to be used as a phase standard.
system 87
30
20
Output Power [dBm]
10
−10
−20
−30
0 2 4 6 8 10 12 14
Frequency [Hz]
Figure 6.5: Output power spectrum of a Marki A-0030 amplifier.
50
0 f0
Phase Variation [deg]
2f0
−50 3f0
4f
0
−100 5f0
6f0
−150 7f0
−200
0 1 2 3 4 5
Input Power [dBm]
Figure 6.6: b2 phase variation for the different harmonics versus input
power for a Marki A-0030 amplifier.
2nd harmonic 3rd harmonic

0.4 0.4
0.3 0.3
0.2 0.2 22
2 2
4 10
0.1 0.1 18
146
25 0
1
2 102
2 5
0 0 8
6
10 1 12 4
20 2
−0.1 −0.1 14
5 6
1 5 108
−0.2 15 0 −0.2
25 16
20 18
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(a) (b)
th th
4 harmonic 5 harmonic
0.4 0.4
0.3 0.3
0.2 25 0.2 25
0.1 20 0.1 20
15 15
10 10
0 0
10 10 5
5
−0.1 −0.1 15
5 5
−0.2 15 −0.2
20 20
25 25
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(c) (d)
6th harmonic 7th harmonic

0.4 0.4
0.3 0.3
0.2 5 35 0.2
30
0.1 25 0.1
5 20 5
0 15 0 4405
20 10 233505
10
5
25
−0.1 −0.1 125
5 15 230 50
20
1510 35
−0.2 30 −0.2
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(e) (f)
Figure 6.7: b2 phase variation of a Marki A-0030 amplifier vs. fundamental

output load Γ at: (a) 2nd harmonic, (b) 3rd harmonic, (c) 4th harmonic,
(d) 5th harmonic, (e) 6th harmonic, (f) 7th harmonic.
system 89

0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
4.5 4
0 0
2
1.5
3.5
0.5
3
2.5 2
3
1.5
−0.1 −0.1
3.5
1
2.5
4
2.53
0.5
1.5
1.52
0.5
1
2
1
0.5
−0.2 −0.2
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(a) (b)
th th
0.4 0.4
0.3 0.3
65 3
4
0.2 0.2 1 32
1
2
2
1.5
0.5
2.5
0.5
1
1.5 1
0.1 0.1
1
2
2
3
0 0 3
6
3
2
2.5
−0.1 −0.1
1
5 2
4
3
−0.2 −0.2 1 1
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(c) (d)

0.4 0.4
0.3 0.3 25
20
0.2 0.2
2.5 3
0.5 15
0.5 51
1. 2
1 10
3.5
0.1 1.5 0.1

2
3.5
2
4
3.5
0 0 5
5
4
−0.1 −0.1 10
15
−0.2 0.5 −0.2
20
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(e) (f)
Figure 6.8: b2 phase variation of a Marki A-0030 amplifier vs. 2nd har-
monic output load Γ at: (a) 2nd harmonic, (b) 3rd harmonic, (c) 4th
harmonic, (d) 5th harmonic, (e) 6th harmonic, (f) 7th harmonic.
HHFT MD-CG1 comb generator

A commercial comb generator implemented with a digital circuit was also eval-
uated. The output spectrum for this component is shown in Fig. 6.9, which
shows that the higher harmonic power levels of this module are sufficiently
high. The b2 phase variation versus input power (shown in Fig. 6.10) for this
comb generator is much lower in comparison to the previously described non-
linear amplifier, although the higher harmonics phase variation versus power
is still significant. Moreover, as shown in Fig. 6.11, the phase variation versus
fundamental load impedance is still considerably large, and would result in
a high phase measurement uncertainty when using such a device as calibra-
tion standard. The variation versus 2nd harmonic load impedance (shown in
Fig. 6.12) is much more contained in this case (below 5 to 6 degrees) and is
acceptable.
−15
−20
Output Power [dBm]
−25
−30
−35
−40
−45
0 2 4 6 8 10 12 14
Frequency [GHz]
Figure 6.9: Output power spectrum of a HHFT MD-CG1 comb generator.
−5 f0
2f0
−10 3f0
4f0
−15 5f0
6f0
−20 7f0
−25
0 2 4 6 8 10
Input Power [dBm]
Figure 6.10: b2 phase variation for the different harmonics versus power
for a HHFT MD-CG1 comb generator.
system 91

0.4 0.4
0.3 0.3
7 6 5 10 8
0.2 0.2
3
2 1
0.1 0.1 6
4 2
2 4
0 0
1
3 4
4 6
5 2 8
−0.1 7 6 −0.1
1
−0.2 −0.2 2 10
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(a) (b)
th th
0.4 0.4
0.3 0.3
14 12 10 16 14 12 0
0.2 0.2 1
6 6
4 2
2 4
0.1 0.1
8
2 8 4
0 4 0 6
6 8
8 1102
−0.1 10 −0.1 2 14
2 12 2 16
−0.2 14 −0.2 18
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(c) (d)

0.4 0.4
0.3 0.3
15 20
0.2 10 0.2 15
5
5
0.1 5 0.1
10 5
0 10 0
10
15 15
−0.1 20 −0.1
20
−0.2 25 −0.2 25
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(e) (f)
Figure 6.11: b2 phase variation of a HHFT MD-CG1 comb generator vs.

fundamental output load Γ at: (a) 2nd harmonic, (b) 3rd harmonic, (c)
4th harmonic, (d) 5th harmonic, (e) 6th harmonic, (f) 7th harmonic.

0.4 0.4
0.3 0.3
1
0.2 0.2
1.8
1
1
211.61.4.2
6
0.
0.4
0.8
3
5
1
0.1 0.1 0.2
6
0.8
1.
0
0..4
2
1.4
6
6
0.2
0 0
0.6
6
7
0.
4
0.6
0.
−0.1 −0.1
00.
1
1
4
11. .86
3
0.
2
2
4
11..8.642
1
−0.2 −0.2 0.4
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(a) (b)
th th
0.4 0.4
0.3 0.3
5
1
1.5
0.
0.2 0.2 2
2
1
1 1.5
2
1
0.1 1.5 0.1

0.
5
1.
5
5
1
0.5
0 0
0.5
0.
1
1
1
0.5
2
1
0.5
−0.1 0.5 −0.1 1.5 1

1.5 1 2.5 2
−0.2 2.52 −0.2 3.5 1
3
1 2
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(c) (d)

0.4 0.4
0.3 0.3
2 1
0.2 0.2
3
5.5 3.5
434.5 1
3
2.5
0.1 2 1 0.1 2
2
23.5 1 1
1.5
5
0 0
1
32
1.
1
−0.1 0.5 −0.1 1 1
2
1
32
1.52
0.5
1 2
−0.2 1.5 1 −0.2 4
1
5
2 6
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(e) (f)
Figure 6.12: b2 phase variation of a HHFT MD-CG1 comb generator vs.

2nd harmonic output load Γ at: (a) 2nd harmonic, (b) 3rd harmonic, (c)
system 93
New comb generator concept

A new comb generator concept has been realized during this thesis work by
combining several commercially-available components in order to minimize
the phase variation of the b2 output wave versus input power and loading
impedance terminations. The output spectrum of the implemented prototype
is shown in Fig. 6.13. Fig. 6.14 shows the phase variation versus power of
the implemented comb generator, while Fig. 6.15 and 6.16 show the b2 phase
variation versus fundamental and 2nd harmonic load impedances, respectively.
These results clearly show that the realized prototype is very insensitive to
input power and changing loading conditions, making this new comb generator
an excellent candidate for being used as a phase reference standard.
−5
Output Power [dBm]
−10
−15
−20
−25
−30
−35
0 2 4 6 8 10 12 14
Frequency [GHz]
Figure 6.13: Output Power Spectrum of the new comb generator proto-
type.
4 f0
2f0
2 3f0
4f0
0 5f0
6f0
−2 7f0
−4
10 11 12 13 14 15 16 17
Input Power [dBm]
Figure 6.14: b2 phase variation for the different harmonics versus power
for the new comb generator prototype.

0.4 0.4
0.3 0.3
0.2
0.2 0.2 0.4
1.4
11.2 1.21
0.
1
1
8
0.1 0.5 0.1 0.8
0.6 0.4
0.5 0.6
0.81
1.2
0 1 0 0.2
1 1.21
11.8.6
0.4 .6
0 .8
−0.1 −0.1 . 4 00.6
1 0.4
1.5
1.6
1
0.2
−0.2 −0.2 1.4 1.
1.5 2
1 2 1
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(a) (b)
th th
0.4 0.4
0.3 0.3
1 1
1 2
0.2 0.2 1.5 2.5
0.5
1.
2.
5 1.51 .5 1.25 1
0.1 0.1 2 2 0.5
1
0.5 1.5
1.5
2
35 2
0 2 0
1
0.5
2.5
3
2
0.
2.
1 5
1.
1
−0.1 −0.1
1.5
−0.2 −0.2 2
2
1.5 2 1 1.5 2.5 1

3 3
1
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(c) (d)

0.4 0.4
0.3 0.3 1
2
1 1.5 2
1
0.2 1 0.2 3
0.5 2
0.1 1 2 0.1
1.5
3
1
2
0.5
0 2 1.5 1
2.5 0 1
4
2
5 53
3.4
1. 3
2
2
−0.1 −0.1
4
−0.2 0.5 −0.2 3 1

54
3
2
1
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(e) (f)
Figure 6.15: b2 phase variation of the new comb generator prototype vs.
fundamental output load Γ at: (a) 2nd harmonic, (b) 3rd harmonic, (c)
system 95

0.4 0.4
0.3 0.3
0.6
0.2 0.4 2 0.2 0.24
1. 1 0.
0.6
0..86
0
0.81 1
0.4
0.1 8 0.1
0. 0.6 0.04.61
10.8
1.2
1
0.6
2
4
0 0 2 0.2
1.
0.2 11. 0.8
0.4 0.2
−0.1 0.6 −0.1 0.4 .60.81
0.4 0 111.4
0.2
1.2 .2
1.4 0.8 .6
−0.2 −0.2
1
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(a) (b)
th th
0.4 0.4
0.3 0.3
0.5 1
0.2 0.2 .5
12
0.5
1
0.1 0.1
0.5
0.5 0.5
1 0 1.5
1 .5 0 1 1
1
.5 3
2 1.5
1.5
0 0 2.5
5
1.
1 2
1
−0.1 −0.1
11.5
1
1.5
2
1.5
−0.2 1.5 −0.2

1
1
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(c) (d)

0.4 0.4
0.3 0.3
1 1.25
2
1
00.5.5
0.2 1 1.5 0.2

3 1.5
2.5
2 1.5
1
5
0.1 0.1
1.
1
1.5
5 5 1 1. 2
2 0. 1.2 5 5
4 1.
23.5
0 2 0 3 1
2
122.5
5
1
1.
−0.1 −0.1 2 0.5

1 2.521.5
3.53
1
11.5
−0.2 −0.2 2
4
1.5
23.52
2
2 1
−0.3 −0.3
−0.4 −0.4
−0.4 −0.2 0 0.2 0.4 −0.4 −0.2 0 0.2 0.4
(e) (f)
Figure 6.16: b2 phase variation of the new comb generator vs. 2nd har-
monic output load Γ at: (a) 2nd harmonic, (b) 3rd harmonic, (c) 4th
harmonic, (d) 5th harmonic, (e) 6th harmonic, (f) 7th harmonic.
Figure 6.17: Package, bond wires and output capacitance de-embedding

for the GaN HEMT with Wg = 12 mm.
6.2.3 Measurement example

The system capability to measure time-domain waveforms is demonstrated by
the example of a 50 W-rated, packaged NXP GaN HEMT with a gate width
of 12 mm measured under class-B operation at a frequency f0 of 2 GHz.
In order to get insightful time-domain waveform data similar to what is
commonly shown in textbook discussions of power amplifiers [15], it is neces-
sary to monitor the voltage and current waveforms at the level of the device
intrinsic current generator, that is the device internal drain node. For this
reason, de-embedding of the package, bond wires, and the device output ca-
pacitance are needed. The simple model used for this de-embedding is shown
in Fig. 6.17. Although the output capacitance is non-linear, and as such has
a strong dependence on the output voltage, the de-embedding approach in
this work is kept simple and assumes a 3 pF linear capacitor. More sophis-
ticated approaches using non-linear de-embedding techniques [101] could also
be implemented.
The device under test is biased at VDD of 50 V with a quiescent current
IDQ of 150 mA. To load-pull the DUT into class-B operation, the fundamental
load impedance termination is controlled over several values (ranging from 15
to 40 Ω, real), while the second harmonic input and output terminations are
fixed at a short condition. The time-domain waveforms are reconstructed from
the measurement of 5 harmonic waves in this particular case. The resulting
waveforms are shown in Fig. 6.18, where the dynamic load-line for differ-
ent resistive fundamental loading conditions are given at 6 dB input power
back-off with respect to the 2 dB power-gain-compression point. The classical
sinusoidal voltage waveform and the half-rectified sine-wave current waveform
can be observed in Fig. 6.18a and 6.18c.
Finally, the voltage and current waveforms are plotted as a function of
system 97
80
3
70 2.5
60 2
v2 [V]
1.5
50
i2 [A]
1
40
0.5
30
0
20 −0.5
0 0.2 0.4 0.6 0.8 1 20 30 40 50 60 70 80
t [s] −9
x 10 v2 [V]
(a) (b)
3.5
2.5
20 Ω 30 Ω 40 Ω
2
i2 [A]
15 Ω 25 Ω 35 Ω
1.5
0.5
0
0 0.2 0.4 0.6 0.8 1
t [s] −9
x 10
(c) (d)
Figure 6.18: (a) Voltage (VDS ) time-domain waveform measured at the

device internal drain node (current generator plane) without output ca-
pacitance and bond-wire connections of a 50 W GaN HEMT for different
resistive fundamental loading terminations. Each color corresponds to a
fundamental loading termination of Fig. 6.18d. (b) Dynamic load line
plotted for different resistive fundamental loading terminations. (c) Cur-
rent (ID ) time-domain waveform measured at the output current generator
of the DUT for different resistive fundamental loading terminations. (d)
Fundamental loading terminations provided to the DUT.
120
100
80
v2 [V]
60
40
20
0
0 0.2 0.4 0.6 0.8 1
t [s] −9
x 10
(a)
3
i2 [A]
−1
0 0.2 0.4 0.6 0.8 1
t [s] −9
x 10
(b)
Figure 6.19: (a) Voltage (VDS ) and (b) current (ID ) time-domain wave-
forms measured as a function of power at the output current generator
of the DUT for a fundamental loading termination of 25 Ω. The source
available power (PAV S ) is swept from 15 dBm to 37 dBm in 1 dB steps.
power for a fundamental loading impedance of 25 Ω in Fig. 6.19. In this

example it is easy to see how the peak voltage approaches 2 · VDD (the the-
oretical value for class B) before the DUT becomes overdriven. From these
measurements we can also note that there are ripples in the current waveform
which are reaching even negative values. Initial experiments using a simulator
with a detailed compact model of the GaN device indicate that the use of
non-linear de-embedding techniques can remove these effects.
6.3 Waveform reconstruction for closely–spaced multi-tone
signals 99
6.3 Waveform reconstruction for closely–

spaced multi-tone signals
In the characterization and optimization of non-linear devices, it is often re-
quired to excite the device with complex modulated signals (e.g., W-CDMA,
IEEE 802.11, etc.). In practical test conditions these signals are represented
using a large number of closely spaced sinusoids [102] with amplitude and
phase set to mimic the cumulative distribution function (CDF) of the commu-
nication standard under consideration correctly.
As mentioned in Section 6.1, two methods are commonly used when per-
forming a phase calibration, and both of them exhibit severe limitations when
a large frequency range (e.g., from fundamental up to the 7th harmonic) is re-
quired in combination with a narrow tone spacing (e.g., below 10 MHz). When
using an AWG, the AWG bandwidth typically limits the number of harmon-
ics covered, while its memory depth restricts the minimum tone spacing that
it can provide. When considering the homogeneous grid method, the phase
reference needs to be stimulated at a very low frequency in order to create a
sufficiently dense frequency grid. Due to the limited output power capabilities
of the phase reference and the large number of frequency tones generated, the
power per tone will be significantly reduced at higher frequencies. Moreover,
the phase error arising when measuring these tones (due to phase noise / jitter
multiplication) increases.
In the following sections, a novel phase calibration procedure is proposed
which facilitates wideband phase calibration of non-linear VNAs (NVNAs)
when employing closely spaced tones. At the same time, the proposed method
has the advantage that it reduces the maximum order of harmonic components
generated by the phase reference, resulting in an increased power per tone and
a reduced phase error.
6.3.1 System description

The hardware configuration used in this work is based on an Agilent N5242A
PNA-X platform. However, the proposed solution is general and can be em-
ployed on any NVNA platform, including the one described in Chapters 4
and 5. A block diagram of the extended hardware configuration is given in
Fig. 6.20. In the proposed approach, a Marki A-0030 square-wave amplifier is
used to generate the static cross-frequency phase relations. These are needed
to keep track of the phase changes when stepping the local oscillator (LO) of
the PNA. The calibration harmonic phase reference (cHPR) is a calibrated
Agilent U9391C unit [103], which is capable of providing the calibrated cross-
Custom BiasTee
To LF coupler
PNA-X
Source 1 Source 2
OUT 1
OUT 1
Rear panel
R1 R3 R4 R2
A C D B
Port 1 Port 3 Port 4 Port 2

Cal. plane for
Couplers ports are connected to
S.O.L. power
HPR external IF inputs
and phase
To external Ext C Ext R3
Two-tones receiver A To LF source
LF coupler
Routing for meas. HPR DUT

Ext D Ext R4
Custom BiasTee
Routing for DUT/cHPR Out. IF Cal.
LF coupler plane for
IF routing S.O.L.
Out. RF Cal. plane
External hardware for S.O.L.
Plane for To LF coupler
eventual Tuner
Figure 6.20: Simplified block diagram of the system, where HPR is the
harmonic phase reference (e.g., square-wave amplifier), and cHPR is the
calibration harmonic phase reference (e.g., Agilent U9391C).
signals 101
Figure 6.21: Measured S21 and S31 of the custom bias-Tee representing
the RF path and baseband path, respectively.
frequency phase relations at the desired reference plane of the device under
test (DUT).
The hardware of the PNA-X is designed to work in the frequency band
ranging from 10 MHz to 26.5 GHz. When exciting a non-linear device with a
multi-tone signal, intermodulation products will appear in the high frequency
band (e.g., 2f1 , 3f1 , etc.) as well as in the baseband (e.g., f2 − f1 , etc.).
When the stimulus is composed of closely-spaced multi-tones, the baseband
components will appear at frequencies below the PNA-X detection band. For
this reason, external hardware is required to properly sample these waves in
order to support correct waveform reconstruction for the overall signal. In the
proposed setup the a and b baseband waves (e.g., below 10 MHz) are sensed
using external low-frequency bridges (e.g., Minicircuits ZFDC-10-6).
The resulting coupled waves are then routed directly to the wideband A/D
converter (e.g., 16 MHz) of the PNA-X. Custom designed bias-Tees are em-
ployed to separate the low-frequency components (e.g., below 100 MHz) from
the high frequency components (e.g., above 1.5 GHz), because the low fre-
quency bridges would present too high loss for the high frequency components
above 10 MHz. Fig. 6.21 shows the measured transmission parameters for
both the RF and baseband path of the bias-Tees.
Baseband Fundamental Second Third
f1 f2
DC f1+f2
2f1-f2 2f2-f1 2f1+f2 2f2+f1

f2-f1 2f1 2f2
3f1 3f2
Frequency
Figure 6.22: Frequency list selected by the user, based on harmonic order
(H) and intermodulation order (M). In this figure H=3 and M=3.
6.3.2 Phase calibration

In the proposed approach, the measurement frequency grid (referred as inter-
modulation grid) is generated based on the desired tone spacing, the harmonic
order and the intermodulation order, as shown in Fig. 6.22 (e.g., as in a har-
monic balance simulator). The two-tones at frequencies f1 and f2 are applied
simultaneously to the non-linear device used as the measurement HPR, which
will generate all the intermodulation products needed for reference, as shown
in Fig 6.20. Note that these tones will be constantly applied to the HPR
during both calibration and measurements, and are used to “keep track” of
the phase change when the local oscillator is stepped to measure the various
frequency components. The same frequencies (e.g., f1 and f2 ) will also be ap-
plied to the DUT during measurements using external sources (e.g., E8257D
PSG, see Fig. 6.20). The phase calibration procedure is then divided in two
main steps, as outlined in the following two subsections.
1. First step
A single-tone at baseband frequency fin = f2 − f1 is applied to the calibra-
tion harmonic phase reference (cHPR), generating a homogeneous grid at the
output of the cHPR (Fig. 6.23). The first step of the phase calibration (e.g.,
the extraction of the phase component of the i10 error term [63]) is performed
up to the highest tone in the fundamental band of the selected frequency grid
(e.g., 2f2 − f1 in Fig. 6.22). Note that the phase calibration will give cor-
rect results only for the frequency components which are also generated by
the measurement harmonic phase reference generator (e.g., spectral compo-
nents of Fig. 6.22). For all of the other frequency components, the data are
discarded.
signals 103
fin (n-1)fin nfin (n+1)fin (n+2)fin

Where:
fin=f2-f1 (e.g. 10 MHz)
(n-1)fin=2f1-f2 (e.g. 1.99 MHz)
nfin=f1 (e.g. 2 GHz)
(n+1)fin=f2 (e.g. 2.01 GHz)
(n+2)fin=2f2-f1 (e.g. 2.02 GHz)
Frequency
angle i10 first step
Figure 6.23: Frequency grid generated by the calibration HPR during the
first step of the calibration procedure.
angle i10 first step
2f1-f2 f1 f2 2f2-f1
fin Frequency
Df*2 Df*n
angle i10 second step
f1 2f1 nf1
Frequency
Figure 6.24: Alignment of the error terms measured in the second step to
the correspondent frequency components measured during the first step
of the phase calibration procedure.
2. Second step
During the second step, a single-tone is applied to the input of the cHPR at
frequency f1 in the fundamental frequency band. At this stage, the phase of
the error terms (e.g., i10 ) is acquired for all of the harmonic components of f1 ,
up to the highest frequency selected in the user-defined frequency grid (e.g.,
3rd harmonic in Fig. 6.22), with the upper limit being the bandwidth of the
instrument.
The phase of the error terms acquired during this second step will not be
consistent with the result of the first step (Fig. 6.24), since a new tone is
now applied to the phase reference that is not coherent with the stimulus of
the first step (e.g., f2 − f1 ). By comparing the phase of the i10 error term
obtained in the second step with the one measured in the first step, the phase
offset to be applied to all the error terms measured in the second step can be
calculated, as shown in Fig. 6.24.
This procedure is repeated when applying a new tone at f2 . In this way,
the phases of the error terms measured in the successive steps can be “aligned”
with the ones measured during the first calibration step. The result is a con-
sistent set of error terms for all the frequency components of interest. Note
that the phases of the error terms for the frequency components that cannot
be generated as an integer multiple of the tones present in the fundamental
frequency band (e.g., f1 +f2 ), are obtained via interpolation to the neighboring
components. This completes the calibration procedure, allowing the measure-
ment of corrected power waves at the input and output of the DUT for the
entire (user-defined) frequency grid, including the baseband components.
6.3.3 Phase stability results

The proposed calibration procedure achieves better measurement phase accu-
racy for the following reasons:
1. Higher tone power for the higher-order harmonics.
2. Lower phase noise and jitter of the spectral components generated, due
to the lower multiplication order.
To illustrate the improved phase measurement accuracy, a direct phase

comparison of raw waves is made between the proposed method and the tra-
ditional homogeneous grid method.
The proposed method uses the two internal PNA-X signal sources to drive
both the measurement HPR as well as the DUT (e.g., a Minicircuits 2x mul-
tiplier, ZX90-2-50-S+) via a power divider. This avoids the use of external
signals 105
(a)
(b)
Figure 6.25: Comparison of normalized amplitude and mean Gaussian

distribution based on the measured standard deviation of the phase of the
b2 wave using the two methods: a) measurement at 10 GHz (5th harmonic),
b) measurement at 14 GHz (7th harmonic).
Figure 6.26: Comparison of normalized amplitude and mean Gaussian

distribution based on the measured standard deviation of the b2 phase
using the new method for different ∆f tones: 10 MHz, 1.25 MHz and 100
KHz.
sources which phase drift over time. In the homogeneous grid measurement
setup, the 10 MHz reference of the PNA-X is used to drive the Agilent U9391C
phase reference, while the two internal sources of the PNA-X drive the DUT.
Note that the 10 MHz reference is phase locked with the internal signal sources
and has very low phase noise, providing maximum output power for all of the
harmonics generated by the U9391C phase reference.
By measuring the phase of the input/output waves (e.g., b1 /b2 ) 100 times,
the standard deviation is computed and a Gaussian distribution (with nor-
malized mean and amplitude) is used to compare the data. Fig. 6.25 shows
the results of the two methods, assuming a ∆f of 10 MHz and the f1 tone at
2 GHz in both cases. In Fig. 6.25a, the measured phase variation is shown
at the 5th harmonic (10 GHz), while Fig 6.25b shows the measured results for
the 7th harmonic (14 GHz). As shown in these figures, the phase variation
of the homogeneous grid method rises with increasing harmonic order, while
the proposed method provides a constant phase deviation. Finally, the phase
deviation is shown in Fig. 6.26 for the new method when measuring the 5th
harmonic of a two-tone signal with f1 equal to 2 GHz when the applied tone
signals 107
0.3
0.2
Voltage, V
0.1
−0.1
−0.2
0 0.5 1 1.5 2
Time, s −7
x 10
(a)
0.3
0.2
Voltage, V
0.1
−0.1
−0.2
0 0.5 1 1.5 2
Time, s x 10
−7
(b)
Figure 6.27: time-domain waveform of the b2 wave measured for the square
wave LO amplifier, a) multi-step calibration procedure b) PNA-X hard-
ware and conventional calibration.
spacing is 10 MHz, 1.25 MHz and 100 kHz. As expected, the accuracy of
the phase measurement does not reduce at narrower tone spacings, as would
happen for the homogeneous grid method.
6.3.4 Waveform reconstruction

To validate the multi-step phase calibration procedure, the waveform of a
square-wave amplifier (e.g., Marki Microwave A-0010) was measured and com-
pared with the data obtained using the PNA-X NVNA software employing the
homogeneous grid approach, Fig. 6.27.
As can be seen clearly, there is good agreement between the two results. In
this experiment the frequency of the two carriers is 1.34 GHz and 1.35 GHz (the
10 MHz frequency grid is required for comparison with the PNA-X software
and hardware). This proves that the proposed calibration procedure provides
results consistent with the conventional, well-established, homogeneous grid
phase calibration procedure.
6.4 Conclusions
In this chapter extensions to the mixed-signal active load-pull system that al-
low time-domain waveform measurements have been discussed. The require-
ments and the performance of commercially-available phase references and of
a newly-developed prototype have been presented. Finally, an approach for
time-domain waveform analysis dedicated to multi-tone signals closely spaced
in frequency has been introduced.
Chapter 7
Application Examples
In the previous chapters of this thesis, an innovative load-pull system has been
proposed. In this chapter, several examples of applications that highlight the
most unique capabilities of the system are presented.
7.1 Out-of-band linearity optimization

The demand for higher data rates in communication systems has led to the
introduction of 3G and 4G communication standards over the past decade,
which make use of higher bandwidths and greater peak-to-average ratios. At
the same time, there is an increased need, based on cost and logistic consid-
erations, to handle multiple communication channels over a single transmit
/ receive chain. These trends require that the underlying hardware must be
very linear over a large frequency span and power range to avoid channel-to-
channel interference and EVM degradation. Combined with the ever-present
need for higher efficiencies, these demands have driven the optimization of
power devices.
From theory based on Volterra series analysis [104], it can be seen that two
kinds of third-order mixing products are present in an active circuit. They are
the direct 3rd order product, and the indirect IM3 mixing product resulting
from secondary mixing between the fundamental signal and the IM2 products
at baseband and 2nd harmonic. These two components can add constructively
or cancel, determining the device linearity.
This effect is illustrated in Fig. 7.1 for the case of a complex modulated
signal. Since this secondary mixing process is influenced by the baseband and
2nd harmonic impedances, control at the device terminals is not only required
in the fundamental band, but also in the baseband and in the 2nd harmonic
109
110 Application Examples
direct mixing Ch: Desired channel

CB: Channel Bandwidth
IML/U: Intermod. distortion
CB lower/upper band
Pout [dBm]
BB: Baseband
B2nd: 2nd harmonic band
IML IMU
BB Ch B2nd
f
2CB
indirect mixing Linearized modulated
signal
Figure 7.1: Basic principle of the out-of-band linearization technique. By

controlling the out-of-band impedances at the baseband and 2nd harmonic
frequencies, the indirect third-order intermodulation products are used to
cancel out the direct ones.
band. Providing a short circuit condition in the baseband or 2nd harmonic

band blocks secondary mixing, while the use of high impedances (e.g., a poorly
decoupled bias circuit) tends to enforce the impact of secondary mixing.
7.1.1 HBT device linearity optimization
Experimental proof of this theory is given in [105], where the active harmonic
load-pull system described in this thesis work is used to measure the linearity
of SiGe and GaAs heterojunction bipolar transistors (HBT) under two-tone
stimulus. The fundamental and 2nd harmonic source and load impedances are
actively controlled, and the baseband source and load impedance are controlled
passively using real impedances ranging from 0.5 to 2,048 Ω with a resistive
switch bank [64], as illustrated in Fig. 7.2.
Four different HBT devices are measured, as shown in table 7.1.
7.1 Out-of-band linearity optimization 111

I/Q 2 0 I/Q 2 0
I/Q 0 I/Q 0
I 1
RF @ 2f0 RF @ 2f0
From From
LO LO
RF @ f0 RF @ f0
On-wafer configuration
0 0
DUT
Bias Bias
Tee Tee
0 0
Passive BB Passive BB
impedance DC impedance DC
board board
Figure 7.2: The mixed-signal based active load-pull setup with passive
baseband control extension.
Measured HBT devices

Technology Size Measured Peak
(µm2 ) ft (GHz)
A SiGe (C) 99 45
B GaAs (AlGaAs) 48.4 43
C GaAs (AlGaAs) 48.4 35
D GaAs (InGaP) 44 100
Table 7.1: Overview of measured HBT devices.
Firstly, the fundamental source and load impedances are optimized for
PAE under single-tone excitation. From two-tone measurements, the baseband
impedance, second harmonic impedance and base-emitter voltage were swept,
and the optimal combination of Zs,bb , Zs,2f and collector current were found.
Finally, Zs,bb , and Zs,2f are fixed to their optimal values, and the input power
and Vbe are swept. The output third-order intercept point (OIP3 ) contours
are plotted in Fig. 7.3 on the output power and quiescent current plane. The
practical application of this method becomes apparent if the bias point for
achieving optimum linearity is found from these plots. The resulting IM3
level versus PAE is shown in Fig. 7.4. In particular, Fig. 7.4a shows the IM3
vs. the two-tone PAE for the collector quiescent currents that result in the
highest OIP3 value. Fig. 7.4b depicts the IM3 vs. the two-tone PAE for the
quiescent current that provides the highest two-tone linearity near the 1-dB
compression point, which was close to 3 mA for all devices shown.
Note that to guarantee low uncertainties in the linearity measurement,
the harmonic distortion products generated by the system should be at least
18 dB below the harmonic distortion level of the DUT. For this reason, the
measurements reported are only possible thanks to the high dynamic range of
the system, and the capability to control the impedance for a wideband signal
in and out-of-band.
8 P−1dB P−1dB
10
18
Collector Quiescent Current (mA)
18
21
20
20
22
9
7
optimal linearity 24
locus 24 8 optimal linearity 22
locus
6 26
21
7 24
28 26
20
18
18
30
26
22
20
22
5 28
24
26 26 24
24 5 22
21
24 20
22 18 21
4 20 4
18 16
20
3 18
20
14
18
20 16
22
20
3 12
22
16 18
−2 0 2 4 6 8 −4 −2 0 2 4 6 8 10
Output Power (dBm) Output Power (dBm)
(a) (b)
~P−1dB P−1dB
16
22
9
26 2
18
10 optimal linearity
18
20
30
20
24
22
locus 32 optimal linearity

8
24
8
26
9 locus
34
32
28 32
8 7 30
30
22
26
16
7 28
32
18
26 6
26 28
24
18
36
20
20
34
22
26 28
24
6 24 30 30
32
22 5 2
5 22 22 4
20 20 26
18 20 4
4
16
16 18
18
20 22 24
20
22
18
18
24
3 14 16 3 16
−6 −4 −2 0 2 4 6 0 2 4 6 8
Output Power (dBm) Output Power (dBm)
(c) (d)
Figure 7.3: Measured OIP3 hi (dBm) contours on the output power and
quiescent current plane for: (a) device A, (b) device B, (c) device C and
(d) device D from Table 7.1. The 1-dB compression points and optimal
linearity loci are indicated. Courtesy of K. Buisman [105].
0 0
−10 −10
−20 −20
IM3hi (dBc)
IM3hi (dBc)
−30 −30
−40 −40
−50 SiGe A −50 SiGe A

III−V B III−V B
−60 III−V C −60 III−V C
III−V D III−V D
−70 −70
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Two−tone PAE Two−tone PAE
(a) (b)
Figure 7.4: IM3 (dBc) versus two-tone PAE (%) for: (a) the bias current
for which the highest OIP3 was measured, and (b) the bias current for
which the highest two-tone linearity was measured near the 1 dB com-
pression point, which was 3 mA for all devices measured. Courtesy of K.
Buisman [105].
7.1.2 Device characterization for LTE applications

with baseband, fundamental and harmonic
wideband impedance control
In the previous section, baseband impedance control is offered by using a re-
sistor switch bank. When a passive baseband impedance control is placed
very close to the RF probe or test fixture, the minimum electrical delays are
too large, and it is therefore only useful in the characterization of low-power
devices for relatively narrow modulation bandwidths (e.g., BW < 10 MHz).
When moving to higher power levels (which demand lower impedance lev-
els) or to higher bandwidths, the use of passive baseband terminations will
impair the linearity characterization. In fact, under these conditions passive
baseband terminations cannot provide sufficiently low impedance levels across
the whole frequency range to avoid electrically-induced memory effects. In
addition, it is not possible to mimic realistic baseband impedances vs. fre-
quency which are present in real power circuits. An example here could be
the notorious resonances in the impedance behavior of the bias network, which
can severely violate the desired baseband decoupling, and cause severe prob-
lems for the digital predistorters presently used in base-station applications.
Consequently, the ability to control and define the baseband impedance vs.
frequency arbitrarily is highly desirable in the linearity characterization of
Mixed-signal active harmonic

load-pull system
AWG AWG AWG ADC ADC ADC ADC AWG AWG AWG
Trigger
and Clock
I Q I Q I Q I Q
2f0 f0 f0 2f0
I1 V1 V2 I2
a1 b1 b2 a2
LO LO LO LO
0 0
0 0
Transistor test fixture

DUT
DC coupled DC coupled
driver driver
amplifier amplifier
I1 V1 V2 I2
Baseband current and voltage sensing
Figure 7.5: The mixed-signal-based active load-pull setup with active

baseband control extension.
devices for wireless applications.

The mixed-signal active load-pull system described in Chapter 4 has been
extended to provide arbitrary control of the baseband impedance termination
vs. frequency for a wideband modulated signal by means of active injection. In
this setup, depicted in Fig. 7.5, a two-channel arbitrary waveform generator
is used to control both the source and the load impedance at baseband to
the DUT. The signal is then amplified by two DC-DC converters / baseband
drivers that are used to provide the active bias to the DUT. These active loops
are directly connected to the bias paths of a dedicated wideband test fixture,
which is optimized for the characterization of packaged high-power devices,
and supports both direct and indirect water cooling to control the operating
temperature of the active DUT. Note that in contrast to many other works
that focus on device linearity, the controlled baseband impedance is truly
DC-coupled. These baseband loops act like DC supplies with an internal
impedance that can be user-defined in software. Failing to do so yields non-
realistic discontinuous behavior in the baseband impedance vs. frequency,
while phenomena that occur in almost any power amplifier, such as self-biasing
and memory effects induced by the baseband impedance, cannot be mimicked
1 Optimum f load Γ
0
0.8 Optimum 2f 0 load Γ
0.6
0.4
0.2
−0.2
−0.4
−0.6
−0.8
−1
−1 −0.5 0 0.5 1
Figure 7.6: Optimum load fundamental and 2nd harmonic reflection coef-
ficients providing maximum PAE under pulsed-RF single-tone excitation.
correctly.
The measurement of the baseband impedance seen at the DUT reference
plane is obtained by acquiring the voltage and the current from DC to 48 MHz
using a current and voltage probe and analog-to-digital converters with a
sampling speed of 100 MS/s. To obtain a calibrated measurement of the
source and load baseband impedances seen by the DUT at the reference plane,
a conventional SOL calibration is performed. For this purpose, short, open
and load calibration standards compatible with the device test fixture have
been designed, implemented and modelled.
To illustrate the importance of controlling the baseband impedance prop-
erly when characterizing active devices with realistic communication signals,
a NXP GEN7 LDMOS device with Wg = 10 mm is measured with a 20 MHz
wide LTE signal using a peak-to-average ratio of 9.8 dB. Initially, the funda-
mental and 2nd harmonic load impedances yielding maximum PAE are found
by performing harmonic load-pull measurements under pulsed-RF, single-tone
excitation. The optimum values for the load reflection coefficients are depicted
in Fig. 7.6. Afterwards, the DUT is measured with an LTE input signal for the
(previously found) fundamental and harmonic conditions, and its ACPR and
EVM performance are evaluated as a function of the baseband load impedance.
Three different cases are evaluated in the following experiments, which
0.8 30 MHz
0.6
RF choke of
0.4 150 nH
0.2
0
10 kHz
−0.2
Passive baseband
−0.4 impedance
−0.6 Baseband short
−0.8
−1
−1 −0.5 0 0.5 1
Figure 7.7: Measured impedance at the DUT reference plane for three
different cases.
yield different baseband load impedances:
• Passive baseband impedance as provided by the test fixture (no BB

impedance control).
• Baseband injection signal set to mimic a baseband short at the load

side.
• Baseband injection signal at load side set such that it mimics a feed
inductance of 150 nH in the DUT bias line.
The baseband impedance provided to the load plane at the DUT measured
up to the 30 MHz controlled bandwidth is illustrated in Fig. 7.7. The results
reported in Fig. 7.8 clearly highlight that proper termination of the baseband
load impedance to a short condition improves both the out-of-band (ACPR)
and the in-band (EVM) linearity of the active device. In particular an im-
provement of 10 dBc in the ACPR and 5 % in EVM is observed for an average
output power of 35 dBm. Note that the system is capable of providing any
arbitrary, user-defined impedance vs. frequency in the fundamental, 2nd har-
monic and baseband frequency bands. In these experiments, a bandwidth of
60 MHz is controlled at f0 to include the 20 MHz signal span and its adjacent
−15 30
Passive BB Passive BB
BB short BB short
−20 BB 150 nH inductance
BB 150 nH inductance 25
−25
ACPR low [dBc]
20
−30 EVM [%]
−35 15
−40
10
−45
−50 5
25 30 35 40 25 30 35 40
P [dBm] P OUT [dBm]
OUT
(a) (b)
0
Passive BB
BB short
BB 150 nH inductance
Power Spectral Density [dBm/Hz]
−10
−20
−30
−40
−50
−60
1.97 1.98 1.99 2 2.01 2.02 2.03
Frequency [GHz]
(c)
Figure 7.8: Measured: (a) ACPR (dBc) low, (b) EVM (%), and (c) power
spectral density (dBm/Hz) at 30 dBm POU T for three different baseband
load impedance terminations.
1.1
Magnitude
0.9
0 5 10 15 20 25 30
200
Angle, deg
180
160
0 5 10 15 20 25 30
Frequency, MHz
Figure 7.9: Magnitude and phase (deg) of the controlled baseband load
reflection coefficient vs. frequency (MHz) at the DUT reference plane.
0.8
0.6
8.2
8.4
0.4
8.1
8.
3
0.2
8.
2
0
8
8.1
−0.2
−0.4 7.9
8
7.
8
−0.6 7.
7
−0.8
−1
−1 −0.5 0 0.5 1
Figure 7.10: EVM (%) contour at a fixed output power of 31 dBm as

function of baseband impedance termination. The contours are plotted
on a Smith chart (20 Ω reference impedance).
7.2 High-power device measurements for base-station
applications 119
channels. This will result in a 120 MHz bandwidth controlled at the 2nd har-
monic, while the baseband impedance is controlled over a 30 MHz frequency
span. An example is shown in Fig. 7.9, where the magnitude and angle of
the baseband load reflection coefficient vs. frequency is plotted. It can be
seen that an excellent control can be achieved throughout the measurement
bandwidth.
Finally, Fig. 7.10 shows contours of EVM for a 20 MHz LTE signal at
a constant average output power of 31 dBm as a function of the baseband
load impedance, plotted in a Smith chart with a reference impedance of 20 Ω.
It is clear that both the imaginary as well as the real parts of the baseband
impedance at the DUT reference plane affect the linearity of the active device,
and therefore require careful attention in both circuit implementation and
characterization.
7.2 High-power device measurements for

base-station applications
The transistor device size used in modern macro base-stations has been in-
creasing drastically, and recently powers of several hundred Watts have been
reached. When looking at load-pull measurements of such devices, the limi-
tation in active load-pull lies in the amount of power required to provide the
optimum reflection coefficient to the DUT at these power levels. For this rea-
son, high-power device testing in a non-50 Ω environment has been limited to
passive load-pull topologies.
The power requirements of active load-pull have already been discussed in
Section 3.4, while in Section 5.3 it has been highlighted that high linearity in
the injection amplifier is not required for the active load-pull system described
in this thesis work.
In this section, load-pull measurement results are presented for two ex-
tremely high-power devices, proving the validity of the load-pull characteriza-
tion system for base-station applications.
In a first test case, an NXP 7th generation LDMOS 200 W-rated transistor
was measured at 2.14 GHz in a pulsed-CW condition with pulse width of 50 µS,
duty cycle of 10 %, and using a prematch to 7 Ω via a transformer test fixture.
The power amplifier used to drive this DUT up to a POU T of 360 W and down
to 0.5 Ω was a NXP Doherty amplifier demo board with a saturated output
power of 500 W. Fig. 7.11a shows the measured PAE vs. output power for this
test case. Fig. 7.11b shows the measured PAE vs. output power, evaluated
at 3 dB of gain (Gp ) compression for all of the load impedances provided.
65
60
60
50
55
40
PAE [%]
PAE [%]
50
30
45
20
10 40
0 35
0 50 100 150 200 250 300 350 100 150 200 250 300 350
PL [W] PL [W]
(a) (b)
PAE@GainCompression
0.1 PL_f0@GainCompression
−0.1
0 160
14
−0.2 180
55
50
200
−0.3 220
240
45
−0.4
60
−0.5
−0.6 300
280
40
35260
−0.7
−0.8
−0.8 −0.6 −0.4 −0.2 0
(c)
Figure 7.11: (a) Power-added efficiency (%) versus output power (W),
(b) power-added efficiency (%) versus output power (W) at 3 dB of gain
compression, and (c) POU T (W) and power added efficiency (%) contours
at 3 dB of gain compression for an NXP 7th generation LDMOS 200 W-
rated transistor.
7.2 High-power device measurements for base-station
applications 121
65
60
60
50
55
40
PAE [%]
PAE [%]
50
30
45
20
10 40
0 35
0 100 200 300 400 500 200 250 300 350 400 450 500
PL [W] PL [W]
(a) (b)
0.4
PAE@GainCompression
PL_f0@GainCompression
0.3
0.2
450
0.1
0
40
−0.1
−0.2
−0.3
60 400
45
25
55
50
−0.4
0
350
30
0
−0.5
−0.8 −0.6 −0.4 −0.2 0
(c)
Figure 7.12: (a) Power-added efficiency (%) versus output power (W),
(b) power-added efficiency (%) versus output power (W) at 3 dB of gain
compression, and (c) POU T (W) and power-added efficiency (%) contours
at 3 dB of gain compression for an NXP 8th generation LDMOS 360 W-
rated transistor.
This particular device reaches a maximum output power of over 320 W at

3 dB compression (highlighted in red) and a maximum efficiency of over 64 %
at 3 dB compression (highlighted in blue). Fig. 7.11c shows POU T and PAE
contours at 3 dB gain compression using the 7 Ω pre-match fixture (center of
Smith chart is 7 Ω). Note that impedances as low as 0.5 Ω (or Γ = 0.98 or
V SW R = 100 : 1 with imaginary component) were presented to the DUT
(in the DUT reference plane) and measured.
In a second measurement example, an NXP 8th generation LDMOS 320
W-rated transistor was measured with the same measurement settings. The
same 500 W Doherty amplifier demo board as in the previous example was
used as the injection amplifier to drive the DUT up to a POU T of 500 W and
down to 1.5 Ω fundamental output loading. Fig. 7.12a shows the measured
PAE vs. output power for this test case, with maximum power and maximum
efficiency curves highlighted in red and blue, respectively. Fig. 7.12b shows
the measured PAE vs. output power evaluated at 3 dB of gain compression,
with a maximum power of over 470 W (red) and a maximum efficiency of over
62 % (blue) at 3 dB compression (not concurrently). Fig. 7.12c shows POU T
and PAE contours at 3 dB gain compression using a 7 Ω pre-match fixture
(center of Smith chart is 7 Ω).
It is important to note that all of the above measurements were performed
in an active load-pull environment with no passive mechanical tuners used in
any hybrid configuration.
7.3 Device characterization for high effi-

ciency power amplifier design
The continuous struggle towards higher efficiencies has led to the extensive
use in industry of power amplifiers working in higher classes of operation [15].
Even though significant progress is made in the modelling of such devices,
conventional models may not predict the device behavior under large-signal
conditions properly. For this reason, power amplifier designers quite often
rely on data from load-pull systems. With this perspective, the capability to
control the higher harmonics terminations provided to the device under test,
and the ability to measure the time-domain voltage and current waveforms
developed in this thesis work prove essential to the power amplifier designer
in order to achieve higher and higher efficiencies.
In this section, an example of high-efficiency amplifier design is presented.
For this purpose, the mixed-signal active load-pull is used to find the optimum
fundamental and second harmonic load impedances on the 50 W-rated, pack-
7.3 Device characterization for high efficiency power amplifier
design 123
(a) (b)
Figure 7.13: (a) Fundamental load Γ sweep, (b) second harmonic load Γ
sweep on a Smith chart (10 Ω reference impedance).
aged NXP GaN HEMT with a gate width of 12 mm, already used in Section
6.2.3. Also, the device in this example is measured at a frequency of 2 GHz
and is biased at a VDD of 50 V with a quiescent current IDQ of 150 mA. Once
again, the simple model shown in Fig. 6.17 has been used to de-embed the
voltage and current waveforms at the level of the internal drain node.
In order to find the optimum tuning impedances yielding the highest effi-
ciency, the fundamental and second harmonic load Γ are swept simultaneously
in the range depicted in Fig. 7.13.
The results are shown in Fig. 7.14, where the efficiency versus output
power is reported at a gain compression level of 3 dB. The gain is plotted
versus output power and the dynamic load line, measured at 35 dBm and at
45 dBm of output power, is depicted for every combination of fundamental
and 2nd harmonic load Γ.
It may be noted from Fig. 7.14c that the device is operating into class J
for the combination of Γ highlighted in yellow. This is more clearly visible
by looking at the voltage and current waveforms versus time as a function of
power, shown in Fig. 7.15a and 7.15c. Here, the peak voltage approaches the
theoretical value of 2.92 · VDD for class J very closely.
In Fig. 7.14a, however, it is easy to identify that the highest efficiency
reaches 80 % for the fundamental and 2nd harmonic Γ highlighted in red. In
80 24
22
70
20
60
Gain [dB]
18
Eff [%]
50
16
40
14
30 12
20 10
0 20 40 60 80 20 25 30 35 40 45 50
P [W] P [dBm]
OUT OUT
(a) (b)
6
2
5
1.5 4
3
1
i2 [A]
i2 [A]
0.5 1
0
0
−1
−0.5 −2
0 20 40 60 80 100 −50 0 50 100 150
v2 [V] v2 [V]
(c) (d)
Figure 7.14: (a) Efficiency (%) versus output power (W) at 3 dB of gain
compression, (b) power gain (dB) versus output power (dBm), (c) dy-
namic load line measured at POU T = 35 dBm and (d) dynamic load line
measured at POU T = 45 dBm plotted for every combination of funda-
mental and 2nd harmonic load Γ. The highighted points correspond to
the combination of fundamental and 2nd harmonic load Γ which are also
highlighted in Fig. 7.13.
7.3 Device characterization for high efficiency power amplifier
design 125
150 140
120
100 100
80
v2 [V]
v2 [V]
50 60
40
0 20
−50 −20
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t [s] x 10
−9 t [s] x 10
−9
(a) (b)
5 5
4 4
3 3
2 2
i2 [A]
i2 [A]
1 1
0 0
−1 −1
−2 −2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
t [s] −9 t [s] −9
x 10 x 10
(c) (d)
Figure 7.15: (a) Voltage (VDS ) and (c) current (ID ) time-domain wave-
form measured as a function of power at the output current generator of
the DUT for the combination of fundamental and 2nd harmonic load Γ
highlighted in yellow in Fig. 7.13. (b) Voltage (VDS ) and (d) current (ID )
time-domain waveform measured as a function of power at the output cur-
rent generator of the DUT for the combination of fundamental and 2nd
harmonic load Γ highlighted in red.
this case, it is not easy to identify what class the device is operating in from the
load-line, while from Fig. 7.15b and 7.15d it can be seen that the peak voltage
for this particular point is slightly lower than for pure class-J operation.
This example clearly highlights that the ability to measure time-domain
waveforms gives powerful insight into the device behavior, while providing
valuable information such as peak voltage and current which can be helpful for
considerations such as ruggedness. However, the high speed of the system that
allows to quickly search through a wide range of fundamental and harmonic
impedances proves to be the real added value when using such a large-signal
measurement system for power amplifier design.
7.4 Conclusions
Several examples have been presented which highlight the most important
features of the mixed-signal load-pull system developed during this thesis work.
Experimental results obtained using this setup show that active device lin-
earity, when operating with wideband-modulated signals (e.g., LTE) is severely
influenced by the baseband impedance offered to the device under test. The
ability to control these baseband impedances vs. frequency allows the user
to optimize the linearity of active devices for complex modulated signals, and
to troubleshoot memory effects due to the baseband impedance terminations
provided by a realistic matching network implementation as well.
Furthermore, the system is flexible enough to serve many different kinds of
applications, ranging from low-power, high-linearity measurements to the char-
acterization of extremely high-power devices for base-stations applications.
Finally, the high speed of the system coupled with the ability to measure
time-domain waveforms at the device intrinsic current generator plane, makes
the mixed-signal load-pull system a valuable tool for high-efficiency power
amplifier design.
Chapter 8
Conclusions and Future Work
8.1 Conclusions
The focus of this thesis is the creation of new characterization techniques
that support technology development, (compact) model device extraction /
validation and design activities, that address the specific needs of advanced
wideband communication systems. Since the transmit path in these next-
generation wireless systems presents the largest challenges in terms of effi-
ciency, linearity and bandwidth, special attention is given to extending the
current state-of-the-art in pulsed and large-signal characterization techniques
in terms of duty-cycle, accuracy, power, impedance control, bandwidth, mea-
surement speed and functionality.
One of the key inventions in this work is a novel time gating and data
alignment approach that, when applied in an isothermal measurement system,
enables pulsed-DC and RF measurements down to 200 ns with an excellent
dynamic range which is independent from duty-cycle. As such, the realized
setup facilitates the characterization of RF / microwave devices under truly
isothermal conditions. This functionality gives significant advantages in defin-
ing, extracting and verifying (compact) models for various RF power devices
that typically suffer severely from bias and operation-dependent self-heating
and/or trapping effects.
The second and dominant part of this thesis relates to the development
and realization of a revolutionary active harmonic load-pull system. The sys-
tem capabilities are summarized and compared to state-of-the-art load-pull
systems in Fig. 8.1. This system provides, for the first time, the capability
to synthesize truly arbitrary source and loading conditions vs. frequency at
the fundamental and harmonic frequencies over a large bandwidth (currently
127
128 Conclusions and Future Work
120 MHz). This setup enables testing of active devices under realistic (circuit-
like) conditions with wideband modulated signals. The ability to control up
to three harmonics at high power levels with an extremely high speed (up to
1,000 measurement points per minute), dramatically enhances the process of
developing new transistor technologies and their application in very efficient
and linear power-amplifiers. Moreover, the option to measure time-domain
voltage and current waveforms can provide significant insight into the actual
device behavior, which benefits power amplifier design, ruggedness evaluation,
as well as (database) model extraction and validation. The usefulness of the
system has been demonstrated by applying this newly developed load-pull
characterization system to several relevant application examples.
All measurement setups and techniques introduced in this thesis are based
on the innovative mixed-signal measurement concept, which replaces tradi-
tional analog solutions with digital data acquisition and digital signal gener-
ation. This approach, sometimes also referred to as “synthetic” or “software-
defined” instruments [106], provides much higher flexibility, functionality, per-
formance and speed in many different applications when compared to tradi-
tional techniques. It is expected that mixed-signal systems will change the
current landscape of RF/microwave characterization.
8.2 Future work

The flexibility of the mixed-signal approach, and of the resulting measurement
setups developed during this thesis work, provides many opportunities for
future work. In this concluding section some of the future trends and topics
are identified that can motivate future research activities and developments in
this field.
8.2.1 Supporting new generation signals and sys-

tems
The introduction of new modulation standards, such as LTE Advanced and
802.11ac, which make use of wider bandwidths to provide higher data rates, is
pushing the bandwidth requirements of the underlying hardware beyond cur-
rent capabilities. At the same time, the growing interest in handling multiple
standards in one single transmit / receive chain demands the design of power
amplifiers that maintain high efficiency over multiple bands or across a wide
frequency range while accommodating multiple standards.
Another trend in the R&D world is defined by the presence of large seg-
ments of unlicensed spectrum at mm-wave frequencies (e.g., 57 to 64 GHz),
8.2 Future work 129
Figure 8.1: Comparison of state-of-the-art load-pull (LP) methods.

which has initiated a tremendous effort to develop broadband communication

systems capable of achieving multi gigabit data transfer rates at mm-waves
frequencies.
To support these design developments, there is need for broadband load-
pull test-benches and general purpose measurement systems, that are capable
of handling ultra-wide modulation bandwidths, not only in the telecommuni-
cation bands, but also at higher frequencies. A first effort in this direction
has already been made in [107], where the mixed-signal load-pull concept
discussed in this thesis is used to enable active load-pull at mm-wave frequen-
cies, with a modulation bandwidth up to 1 GHz. Nevertheless, the desire for
greater bandwidth, coupled with the ever-present need for high accuracy and
dynamic range, brings about the need for innovative system architectures and
measurement techniques.
8.2.2 Supporting high-efficiency and high-linearity

advanced PA design
The high peak-to-average ratio of 3G and 4G modulation standards has re-
quired the introduction of advanced PA topologies which are able to perform
efficiently not only at peak output power, but also in power back-off where (sta-
tistically) most of the energy is consumed. A topology like the Doherty PA [12]
has found wide use in industry, while concepts like envelope-tracking [14] and
outphasing [13] amplifiers are being studied intensively at the research level.
Designing these advanced classes of amplifiers is a challenging task. From
this perspective, measurement tools and techniques to enable the load-pull
characterization of multi-input amplifiers would be extremely valuable. More-
over, the capability to modulate the bias to the DUT directly in a load-pull
environment would be essential to support envelope tracking amplifiers design.
Another opportunity for research is given by the fact that many of these
amplifier topologies are extremely non-linear, and require predistortion of the
input signal to operate linearly and comply with the communication standard
specifications. The possibility to predistort the input signal directly on the ac-
tive load-pull test bench would be extremely useful to verify directly whether
the DUT can comply with the linearity specifications after predistortion, be-
fore actually implementing the power amplifier board.
8.2.3 Supporting next generation device modelling

The introduction of new transistor technologies with higher performance (i.e.,
power, efficiency) brings with it the challenge of creating new and more ac-
8.2 Future work 131
curate models for circuit simulation. In a fast growing industry like telecom-
munication, where time-to-market is extremely important, enabling the rapid
and simple extraction of new device models directly from measurements has
received significant interest. For this reason, significant effort has been put
into the development of behavioral models in recent years, such as the poly-
harmonic distortion (PHD) model [108], which enable the extraction of device
models directly from large-signal measurements.
These models employ the measurement of time-domain voltage and cur-
rent waveforms under the different boundary conditions which affect the DUT
performance in order to create a database describing the behavior of the DUT.
For this reason, they require the device to be measured throughout the multidi-
mensional space of all of the different tunable parameters, such as fundamental
and harmonic source and load impedance, frequency, bias, etc.
Thanks to its very high measurement speed of up to 1,000 measurement
point per minute, the load-pull system described in this thesis can prove to
be an extremely valuable asset for the extraction of these type of models.
Furthermore, the development of additional measurement techniques to extend
their range of validity (e.g., to include memory effects) is also a topic of interest.
In terms of compact modeling, an issue is represented by the continuous
scaling in device size which is necessary to achieve higher cut-off frequencies
[109]. For this reason, the thermal resistance of some high-frequency devices
can reach values as high as 10,000 K/W [110,111]. To model these new device
generations, faster pulsed-DC measurements than what is currently available
are necessary to achieve truly isothermal conditions.
Conventional pulsed bias approaches employ analog pulsers to provide cur-
rent drive and to deliver the required voltage shape to the DUT. Nevertheless,
these methods cannot compensate for the distorting effect on the voltage wave-
form, arising from the interconnects and parasitic loading, due to the analog
nature of the driver. Therefore the minimum DC pulse widths achievable with
conventional pulsers are severely limited (to approximately 200 - 300 ns).
A more elaborate approach, based on a mixed-signal technique, can achieve
significantly shorter DC pulse widths (e.g., 10 ns or lower) by using a high-
speed AWG and a DC-coupled driver amplifier to generate the DC pulse, a
high-speed A/D converter to measure the calibrated DC pulse at the DUT,
and an iterative approach to optimize the DC pulse provided to the DUT [112].
However, in order for such an approach to be successful, the design of a high-
power DC-coupled amplifier, capable of providing high voltage and current
(e.g., 200 V and 30 A) with a wide bandwidth (e.g., 100 MHz), is required.
8.3 Future trends

The aforementioned demand for higher bandwidth, efficiency and functional-
ity of next generation communication systems requires more complex design
approaches to the power amplifier, which is now shifting further away from
the traditional “linear” class-AB operation.
For this reason, RF characterization can no longer be limited to simple
S-parameter and spectral distortion measurements. The transmitter output
stage with its surrounding circuitry (e.g., voltage supply modulators) should
now be considered as a dynamic, non-linear sub-system, which can no longer be
described or characterized easily by conventional modelling and measurement
tools.
To fully address these upcoming design challenges, innovative and more
advanced, test and modelling solutions will be required. The mixed-signal
techniques introduced in this thesis provide an enabling platform for a new
generation of measurement tools, which can encompass many of the functions
of an advanced circuit or system simulator at an hardware level. The device
or sub-system under test can now be placed in a virtual circuit environment,
where the DUT itself constitutes the actual (most complete) model, and the
mixed-signal core of the measurement system is capable to mimic a realistic
circuit or system implementation accurately. It is the author’s belief that this
novel approach will open up new possibilities and measurement applications,
which will push microwave and RF characterization to unprecedented levels.
Bibliography
[1] http://newsroom.cisco.com/release/1114955.
[2] http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/
ns537/ns705/ns827/white paper c11-520862.html.
[3] G. Fettweis and E. Zimmermann, “ICT energy consumption-trends and

challenges,” in Proceedings of the 11th International Symposium on
Wireless Personal Multimedia Communications, vol. 2, no. 4, 2008, p. 6.
[4] E. Oh, B. Krishnamachari, X. Liu, and Z. Niu, “Toward dynamic energy-

efficient operation of cellular network infrastructure,” Communications
Magazine, IEEE, vol. 49, no. 6, pp. 56–61, 2011.
[5] W. Vereecken, W. Van Heddeghem, M. Deruyck, B. Puype, B. Lannoo,

W. Joseph, D. Colle, L. Martens, and P. Demeester, “Power consump-
tion in telecommunication networks: overview and reduction strategies,”
Communications Magazine, IEEE, vol. 49, no. 6, pp. 62–69, 2011.
[6] O. Blume, D. Zeller, and U. Barth, “Approaches to energy efficient wire-

less access networks,” in Communications, Control and Signal Process-
ing (ISCCSP), 2010 4th International Symposium on, 2010, pp. 1–5.
[7] I. Stevanovic, A. Skrivervik, and J. R. Mosig, “Smart antenna systems

for mobile communications,” Ecole Polytechnique Federale De Lausanne,
Technical Report, 2003.
[8] J. Louhi, “Energy efficiency of modern cellular base stations,” in

Telecommunications Energy Conference, 2007. INTELEC 2007. 29th In-
ternational, 2007, pp. 475–476.
[9] “Base station conformance testing (FDD),” Technical Specification

Group Radio Access Networks - 3rd Generation Partnership Project,
3GPP TS 25.141 V4.1.0 (2001-06), 2001.
133
134 BIBLIOGRAPHY
[10] LAN/MAN standards Committee and others, “Part 11: Wireless LAN
medium access control (MAC) and physical layer (PHY) specifications,”
IEEE-SA Standards Board, 2003.
[11] J. Sevic, “Statistical characterization of RF power amplifier efficiency for

CDMA wireless communication systems,” in Wireless Communications
Conference, 1997., Proceedings, 1997, pp. 110–113.
[12] W. H. Doherty, “A new high efficiency power amplifier for modulated

waves,” Radio Engineers, Proceedings of the Institute of, vol. 24, no. 9,
pp. 1163–1182, 1936.
[13] H. Chireix, “High power outphasing modulation,” Radio Engineers, Pro-

ceedings of the Institute of, vol. 23, no. 11, pp. 1370–1392, 1935.
[14] F. Raab, P. Asbeck, S. Cripps, P. Kenington, Z. Popovic, N. Pothecary,

J. Sevic, and N. Sokal, “Power amplifiers and transmitters for RF and
microwave,” Microwave Theory and Techniques, IEEE Transactions on,
vol. 50, no. 3, pp. 814–826, 2002.
[15] S. Cripps, RF power amplifiers for wireless communications. Artech

House Norwood, MA, 2006.
[16] H. Ku, M. D. McKinley, and J. S. Kenney, “Quantifying memory effects

in RF power amplifiers,” IEEE Trans. Microwave Theory Tech., vol. 50,
no. 12, pp. 2843–2849, 2002.
[17] J. Paasschens, W. Kloosterman, and R. vd Toorn, “Model derivation of

Mextram 504,” dNA, vol. 510, pp. 2–30, 2005.
[18] C. McAndrew, J. Seitchik, D. Bowers, M. Dunn, M. Foisy, I. Getreu,

M. McSwain, S. Moinian, J. Parker, D. Roulston, M. Schroter, P. van
Wijnen, and L. Wagner, “VBIC95, the vertical bipolar inter-company
model,” Solid-State Circuits, IEEE Journal of, vol. 31, no. 10, pp. 1476
–1483, Oct. 1996.
[19] M. Schroter, HICUM, a scalable physics-based compact bipolar transistor

model, level 2 ed., http://www.iee.et.tu-dresden.de/iee/eb/eb homee.
html, 2005.
[20] V. Cuoco, M. van den Heijden, and L. de Vreede, “The “Smoothie”

data base model for the correct modeling of non-linear distortion in
FET devices,” in Microwave Symposium Digest, 2002 IEEE MTT-S In-
ternational, vol. 3, 2002, pp. 2149 –2152.
BIBLIOGRAPHY 135
[21] P. Roblin, D. E. Root, J. Verspecht, Y. Ko, and J. P. Teyssier, “New

trends for the nonlinear measurement and modeling of high-power RF
transistors and amplifiers with memory effects,” IEEE Trans. Microwave
Theory Tech., vol. 60, no. 6, pp. 1964–1978, 2012.
[22] J. P. Teyssier, M. Campovecchio, C. Sommet, J. Portilla, and R. Quere,

“A pulsed S-parameters measurement setup for the non-linear charac-
terization of FETs and bipolar power transistors,” in Microwave Con-
ference, 1993. 23rd European, Sept. 1993, pp. 489 –493.
[23] B. Taylor, M. Sayed, and K. Kerwin, “A pulse bias/RF environment

for device characterization,” in ARFTG Conference Digest-Fall, 42nd,
vol. 24, Dec. 1993, pp. 57 –60.
[24] J. P. Teyssier, P. Bouysse, Z. Ouarch, D. Barataud, T. Peyretaillade,

and R. Quere, “40-GHz/150-ns versatile pulsed measurement system for
microwave transistor isothermal characterization,” Microwave Theory
and Techniques, IEEE Transactions on, vol. 46, no. 12, pp. 2043 –2052,
Dec. 1998.
[25] S. J. Doo, P. Roblin, S. Lee, D. Chaillot, and M. Vanden Bossche,

“Pulsed-IV pulsed-RF measurements using a large signal network an-
alyzer,” in ARFTG Conference Digest, 2005. Spring 2005. 65th, June
2005, p. 7 pp.
[26] C. Baylis, L. Dunleavy, and J. Martens, “Constructing and benchmark-

ing a pulsed-RF, pulsed-bias S-parameter system,” in ARFTG Confer-
ence Digest-Fall, 66th, Dec. 2005.
[27] V. Teppati, A. Ferrero, and U. Pisani, “Recent advances in real-time

load-pull systems,” Instrumentation and Measurement, IEEE Transac-
tions on, vol. 57, no. 11, pp. 2640 –2646, Nov. 2008.
[28] Application note, 1408-11: accurate pulsed measurements, Agilent Tech-

nologies, Feb. 2004.
[29] Application note, 1408-12: pulsed-RF S-parameter measurements using

wideband and narrowband detection, Agilent Technologies, May 2006.
[30] Agilent Technologies, “8510 pulsed-RF network analyzer user’s guide,”

Mar. 1995.
[31] L. Betts, “Tracking advances in pulsed S-parameter measurements,” Mi-

crowaves & RF, Sept. 2007.
136 BIBLIOGRAPHY
[32] 8510 pulsed-RF network analyzer user’s guide , Agilent Technologies,

Mar. 1995.
[33] I. Kollár, “Evaluation of sine wave tests of ADCs from windowed data,”
Computer Standards & Interfaces, vol. 22, no. 4, pp. 261–268, 2000.
[34] W. Kester, A. Devices, et al., Data conversion handbook. Newnes, 2005.
[35] M. Hauser, “Principles of oversampling A/D conversion,” J. Audio Eng.
Soc, vol. 39, no. 1/2, pp. 3–26, 1991.
[36] M. Reisch, High-frequency bipolar transistors: physics, modelling, appli-
cations. Springer Verlag, 2003.
[37] G. Gonzalez, Microwave transistor amplifiers: analysis and design.
Prentice hall New Jersey, 1997, vol. 2.
[38] F. van Rijs and S. Theeuwen, “Efficiency improvement of LDMOS tran-
sistors for base stations: towards the theoretical limit,” in Electron De-
vices Meeting, 2006. IEDM ’06. International, Dec. 2006, pp. 1 –4.
[39] P. Deixler, R. Colclaser, D. Bower, N. Bell, W. De Boer, D. Szmyd,
S. Bardy, W. Wilbanks, P. Barre, M. v Houdt, J. Paasschens, H. Veen-
stra, E. v d Heijden, J. Donkers, and J. Slotboom, “QUBiC4G: a
fT/fmax = 70/100 GHz 0.25 µm low power SiGe-BiCMOS production
technology with high quality passives for 12.5 Gb/s optical networking
and emerging wireless applications up to 20 GHz,” in Bipolar/BiCMOS
Circuits and Technology Meeting, 2002. Proceedings of the 2002, 2002.
[40] N. Nenadovic, “Electrothermal behavior of high frequency silicon-on-
glass transistors,” Ph.D. dissertation, Delft University of Technology,
2004.
[41] M. Spirito, M. Pelk, F. van Rijs, S. Theeuwen, D. Hartskeerl, and
L. de Vreede, “Active harmonic load-pull for on-wafer out-of-band de-
vice linearity optimization,” Microwave Theory and Techniques, IEEE
Transactions on, vol. 54, no. 12, pp. 4225 –4236, Dec. 2006.
[42] http://www.maurymw.com.
[43] http://www.focus-microwaves.com/.
[44] V. Teppati and A. Ferrero, “A new class of nonuniform, broadband, non-
symmetrical rectangular coaxial-to-microstrip directional couplers for
high power applications,” Microwave and Wireless Components Letters,
IEEE, vol. 13, no. 4, pp. 152 – 154, Apr. 2003.
BIBLIOGRAPHY 137
[45] F. De Groote, J. Verspecht, D. Barataud, and J.-P. Teyssier, “An im-

proved coupling method for time domain load-pull measurements,” in
Microwave Conference, 2005 European, vol. 1, Oct. 2005, p. 4 pp.
[46] G. Bava, U. Pisani, and V. Pozzolo, “Active load technique for load-pull
characterisation at microwave frequencies,” Electronics Letters, vol. 18,
no. 4, pp. 178 –180, Feb. 1982.
[47] J. E. Muller and B. Gyselinckx, “Comparison of active versus passive

on-wafer load-pull characterisation of microwave and mm-wave power
devices,” in Microwave Symposium Digest, 1994., IEEE MTT-S Inter-
national, May 1994, pp. 1077 –1080 vol.2.
[48] A. Ferrero, F. Sanpietro, U. Pisani, and C. Beccari, “Novel hardware and

software solutions for a complete linear and nonlinear microwave device
characterization,” Instrumentation and Measurement, IEEE Transac-
tions on, vol. 43, no. 2, pp. 299 –305, Apr. 1994.
[49] T. Williams, J. Benedikt, and P. Tasker, “Experimental evaluation of

an active envelope load pull architecture for high speed device charac-
terization,” in Microwave Symposium Digest, 2005 IEEE MTT-S Inter-
national, June 2005, p. 4 pp.
[50] Y. Takayama, “A new load-pull characterization method for microwave

power transistors,” in Microwave Symposium, 1976 IEEE-MTT-S In-
ternational, June 1976, pp. 218 –220.
[51] K. Nagatomo, S. Koike, M. Shigaki, N. Okubo, H. Takahashi, and

T. Nakatani, “38-GHz band high-power MMIC amplifier design using
improved load-pull method,” in Microwave Conference, 1990. 20th Eu-
ropean, vol. 1, Sept. 1990, pp. 151 –156.
[52] F. van Raay and G. Kompa, “A 40 GHz large-signal double-reflectometer

waveform measurement system designed for load-pull applications,” in
Microwave Conference, 1996. 26th European, vol. 2, Sept. 1996, pp. 657
–661.
[53] J. Benedikt, R. Gaddi, P. Tasker, and M. Goss, “High-power time-

domain measurement system with active harmonic load-pull for high-
efficiency base-station amplifier design,” Microwave Theory and Tech-
niques, IEEE Transactions on, vol. 48, no. 12, pp. 2617 –2624, Dec.
2000.
138 BIBLIOGRAPHY
[54] B. Bunz and G. Kompa, “Active load pull with fourth harmonic tuning
based on an IQ modulator concept,” in Microwave Conference, 2003.
33rd European, Oct. 2003, pp. 359 –361.
[55] H. Arthaber, M. Mayer, and G. Magerl, “A broadband active harmonic

load-pull setup with a modulated generator as active load,” in Microwave
Conference, 2004. 34th European, vol. 2, Oct. 2004, pp. 685 – 688.
[56] E. McCune, “Modern cellular wireless signals,” in Microwave Measure-

ments Conference (ARFTG), 2010 75th ARFTG, May 2010, pp. 1 –7.
[57] M. van der Heijden, H. de Graaff, and L. de Vreede, “A novel frequency-

independent third-order intermodulation distortion cancellation tech-
nique for BJT amplifiers,” Solid-State Circuits, IEEE Journal of, vol. 37,
no. 9, pp. 1176 – 1183, Sept. 2002.
[58] M. Marchetti, M. Pelk, K. Buisman, W. Neo, M. Spirito, and

L. de Vreede, “Active harmonic load-pull with realistic wideband com-
munications signals,” Microwave Theory and Techniques, IEEE Trans-
actions on, vol. 56, no. 12, pp. 2979 –2988, Dec. 2008.
[59] Z. Aboush, C. Jones, G. Knight, A. Sheikh, H. Lee, J. Lees, J. Benedikt,

and P. Tasker, “High power active harmonic load-pull system for charac-
terization of high power 100-watt transistors,” in Microwave Conference,
2005 European, vol. 1, Oct. 2005, p. 4 pp.
[60] W. C. E. Neo, J. Qureshi, M. J. Pelk, J. R. Gajadharsing, and L. C. N.

de Vreede, “A mixed-signal approach towards linear and efficient N-way
Doherty amplifiers,” Microwave Theory and Techniques, IEEE Trans-
actions on, vol. 55, no. 5, pp. 866 –879, May 2007.
[61] D. Poulin, J. Mahon, and J.-P. Lanteri, “A high power on-wafer pulsed
active load pull system,” Microwave Theory and Techniques, IEEE
Transactions on, vol. 40, no. 12, pp. 2412 –2417, Dec. 1992.
[62] D. Rytting, “Network analyzer error models and calibration methods,”

White Paper, September, 1998.
[63] A. Ferrero and U. Pisani, “An improved calibration technique for

on-wafer large-signal transistor characterization,” Instrumentation and
Measurement, IEEE Transactions on, vol. 42, no. 2, pp. 360 –364, Apr.
1993.
BIBLIOGRAPHY 139
[64] M. Pelk, L. de Vreede, M. Spirito, and J. Jos, “Base-band impedance

control and calibration for on-wafer linearity measurements,” in ARFTG
Conference Digest Spring, 2004. 63rd, June 2004, pp. 35 – 40.
[65] R. Marchesani, “Digital precompensation of imperfections in quadrature

modulators,” Communications, IEEE Transactions on, vol. 48, no. 4,
pp. 552 –556, Apr. 2000.
[66] “UTRA (BS) FDD; Radio Transmission and Reception,” Technical

Specification Group Radio Access Networks - 3rd Generation Partner-
ship Project, 3GPP TS 25.104 V4.1.0 (2001-06), 2001.
[67] D. Hartskeerl, I. Volokhine, and M. Spirito, “On the optimum 2nd har-
monic source and load impedances for the efficiency-linearity trade-off of
RF LDMOS power amplifiers,” in Radio Frequency integrated Circuits
(RFIC) Symposium, 2005. Digest of Papers. 2005 IEEE, June 2005, pp.
447 – 450.
[68] F. Verbeyst and M. Bossche, “Real-time and optimal PA characteriza-

tion speeds up PA design,” in Microwave Conference, 2004. 34th Euro-
pean, vol. 1, oct. 2004, pp. 431 –434.
[69] T. Williams, J. Benedikt, and P. Tasker, “Fully functional “real time”

non-linear device characterization system incorporating active load con-
trol,” in Microwave Conference, 2006. 36th European, sept. 2006, pp.
1610 –1613.
[70] P. Roblin, S. J. Doo, X. Cui, G. Jessen, D. Chaillot, and J. Strahler,

“New ultra-fast real-time active load-pull measurements for high
speed RF power amplifier design,” in Microwave Symposium, 2007.
IEEE/MTT-S International, jun 2007, pp. 1493 –1496.
[71] M. Marchetti, K. Buisman, M. Pelk, L. Smith, and L. de Vreede, “A

low-cost pulsed RF & I-V measurement setup for isothermal device char-
acterization,” in ARFTG Conference Digest Fall, 2007. 70th, Nov. 2007.
[72] P. Colantonio, F. Giannini, E. Limiti, and V. Teppati, “An approach

to harmonic load- and source-pull measurements for high-efficiency PA
design,” Microwave Theory and Techniques, IEEE Transactions on,
vol. 52, no. 1, pp. 191 – 198, jan. 2004.
[73] F. van Rijs, R. Dekker, H. Visser, H. Huizing, D. Hartskeerl, P. Mag-

nee, and R. Dondero, “Influence of output impedance on power added
140 BIBLIOGRAPHY
efficiency of Si-bipolar power transistors,” in Microwave Symposium Di-

gest., 2000 IEEE MTT-S International, vol. 3, 2000, pp. 1945 –1948
vol.3.
[74] M. Squillante, M. Marchetti, M. Spirito, and L. de Vreede, “A mixed-

signal approach for high-speed fully controlled multidimensional load-
pull parameters sweep,” in Microwave Measurement Conference, 2009
73rd ARFTG, jun 2009, pp. 1 –5.
[75] D. Barataud, F. Blache, A. Mallet, P. Bouysse, J.-M. Nebus, J. Villotte,

J. Obregon, J. Verspecht, and P. Auxemery, “Measurement and con-
trol of current/voltage waveforms of microwave transistors using a har-
monic load-pull system for the optimum design of high efficiency power
amplifiers,” Instrumentation and Measurement, IEEE Transactions on,
vol. 48, no. 4, pp. 835 –842, aug 1999.
[76] D. Morgan, G. Edwards, A. Phillips, and P. Tasker, “Full extraction

of PHEMT state functions using time domain measurements,” in Mi-
crowave Symposium Digest, 2001 IEEE MTT-S International, vol. 2,
2001, pp. 823 –826 vol.2.
[77] G. Simpson, J. Horn, D. Gunyan, and D. Root, “Load-pull + NVNA

= enhanced X-parameters for PA designs with high mismatch and
technology-independent large-signal device models,” in ARFTG Mi-
crowave Measurement Symposium, 2008 72nd, dec. 2008, pp. 88 –91.
[78] O. Jardel, F. De Groote, T. Reveyrand, J.-C. Jacquet, C. Charbonni-

aud, J.-P. Teyssier, D. Floriot, and R. Quere, “An electrothermal model
for AlGaN/GaN power HEMTs including trapping effects to improve
large-signal simulation results on high VSWR,” Microwave Theory and
Techniques, IEEE Transactions on, vol. 55, no. 12, pp. 2660 –2669, dec.
2007.
[79] R. Gaddi, J. Pla, J. Benedikt, and P. Tasker, “LDMOS electro-thermal

model validation from large-signal time-domain measurements,” in Mi-
crowave Symposium Digest, 2001 IEEE MTT-S International, vol. 1,
2001, pp. 399 –402 vol.1.
[80] M. Spirito, L. de Vreede, L. Nanver, S. Weber, and J. Burghartz, “Power

amplifier PAE and ruggedness optimization by second-harmonic con-
trol,” Solid-State Circuits, IEEE Journal of, vol. 38, no. 9, pp. 1575 –
1583, sept. 2003.
BIBLIOGRAPHY 141
[81] D. Barataud, C. Arnaud, B. Thibaud, M. Campovecchio, J.-M. Nebus,

and J. Villotte, “Measurements of time-domain voltage/current wave-
forms at RF and microwave frequencies based on the use of a vec-
tor network analyzer for the characterization of nonlinear devices-
application to high-efficiency power amplifiers and frequency-multipliers
optimization,” Instrumentation and Measurement, IEEE Transactions
on, vol. 47, no. 5, pp. 1259 –1264, oct 1998.
[82] S. Rehnmark, “On the calibration process of automatic network ana-
lyzer systems (short papers),” Microwave Theory and Techniques, IEEE
Transactions on, vol. 22, no. 4, pp. 457 – 458, apr 1974.
[83] U. Lott, “Measurement of magnitude and phase of harmonics generated
in nonlinear microwave two-ports,” Microwave Theory and Techniques,
IEEE Transactions on, vol. 37, no. 10, pp. 1506 –1511, oct 1989.
[84] D. Gunyan and Y.-P. Teoh, “Characterization of active harmonic phase
standard with improved characteristics for nonlinear vector network ana-
lyzer calibration,” in Microwave Symposium Digest, 2008 IEEE MTT-S
International, june 2008, pp. 73 –79.
[85] A. Barel and Y. Rolain, “A microwave multisine with known phase for
the calibration of narrowbanded nonlinear vectorial network analyzer
measurements,” in Microwave Symposium Digest, 1998 IEEE MTT-S
International, vol. 3, jun 1998, pp. 1499 –1502 vol.3.
[86] C. Barrett, “Fractional/integer-N PLL basics,” 1999.
[87] P. Blockley, D. Gunyan, and J. Scott, “Mixer-based, vector-corrected,
vector signal/network analyzer offering 300 kHz-20 GHz bandwidth and
traceable phase response,” in Microwave Symposium Digest, 2005 IEEE
MTT-S International, june 2005, p. 4 pp.
[88] A. Ferrero, V. Teppati, and B. Noori, “A generalized time-domain wave-
form test-set,” in Microwave Conference, 2008. EuMC 2008. 38th Eu-
ropean, oct. 2008, pp. 749 –752.
[89] M. Sipila, K. Lehtinen, and V. Porra, “High-frequency periodic time-
domain waveform measurement system,” Microwave Theory and Tech-
niques, IEEE Transactions on, vol. 36, no. 10, pp. 1397 –1405, oct 1988.
[90] G. Kompa and F. van Raay, “Error-corrected large-signal waveform mea-
surement system combining network analyzer and sampling oscilloscope
capabilities,” Microwave Theory and Techniques, IEEE Transactions on,
vol. 38, no. 4, pp. 358 –365, apr 1990.
142 BIBLIOGRAPHY
[91] J. Scott, J. Verspecht, B. Behnia, M. Vanden Bossche, A. Cognata,

F. Verbeyst, M. Thorn, and D. Scherrer, “Enhanced on-wafer time-
domain waveform measurement through removal of interconnect disper-
sion and measurement instrument jitter,” Microwave Theory and Tech-
niques, IEEE Transactions on, vol. 50, no. 12, pp. 3022 – 3028, dec
2002.
[92] A. Ramadan, A. Martin, T. Reveyrand, J.-M. Nebus, P. Bouysse,

L. Lapierre, J. Villemazet, and S. Forestier, “Efficiency enhancement of
GaN power HEMTs by controlling gate-source voltage waveform shape,”
in Microwave Integrated Circuits Conference, 2009. EuMIC 2009. Euro-
pean, sept. 2009, pp. 487 –490.
[93] M. Thorsell and K. Andersson, “Fast multiharmonic active load-pull

system with waveform measurement capabilities,” Microwave Theory
and Techniques, IEEE Transactions on, vol. 60, no. 1, pp. 149 –157,
jan. 2012.
[94] Agilent Technologies, “The microwave transition analyzer: a versatile

measurement set for bench and test. Product note 70820-1,” 1991.
[95] J. Verspecht, “Large-signal network analysis,” Microwave Magazine,

IEEE, vol. 6, no. 4, pp. 82 – 92, dec 2005.
[96] http://www.aeroflex.com/ams/metelics/micro-metelics-prods-comb.
cfm.
[97] http://www.picosecond.com/product/category.asp?pd id=22.
[98] http://www.hhft.de/index.php?page=rf&subpage=generators p2.
[99] J. Verspecht and K. Rush, “Individual characterization of broadband

sampling oscilloscopes with a nose-to-nose calibration procedure,” In-
strumentation and Measurement, IEEE Transactions on, vol. 43, no. 2,
pp. 347 –354, apr 1994.
[100] T. Clement, P. Hale, D. Williams, C. Wang, A. Dienstfrey, and

D. Keenan, “Calibration of sampling oscilloscopes with high-speed pho-
todiodes,” Microwave Theory and Techniques, IEEE Transactions on,
vol. 54, no. 8, pp. 3173 –3181, aug. 2006.
[101] F. Vanaverbeke, W. De Raedt, D. Schreurs, and M. Bossche, “Real-

time non-linear de-embedding,” in Microwave Measurement Conference
(ARFTG), 2011 77th ARFTG, june 2011, pp. 1 –6.
BIBLIOGRAPHY 143
[102] J. Pedro and N. De Carvalho, “On the use of multitone techniques for
assessing RF components’ intermodulation distortion,” Microwave The-
ory and Techniques, IEEE Transactions on, vol. 47, no. 12, pp. 2393
–2402, dec 1999.
[103] Agilent Technologies, U9391C/F Comb generators 10 MHz to 50 GHz

technical overview” (2009) Agilent Technologies.
[104] M. P. van der Heijden, “RF amplifier design techniques for linearity
and dynamic range,” Ph.D. dissertation, Delft University of Technology,
2005.
[105] K. Buisman, “Device realization, characterization and modeling for lin-

ear RF applications,” Ph.D. dissertation, Delft University of Technology,
2011.
[106] J. Bains, “Modular systems for RF and microwave measurements,”

in Modern RF and Microwave Measurement Techniques, V. Teppati,
A. Ferrero, and M. Sayed, Eds. Cambridge University Press, 2013,
ch. 7.
[107] L. Galatro, M. Marchetti, and M. Spirito, “60 GHz mixed signal active
load-pull system for millimeter wave devices characterization,” in Mi-
crowave Measurement Symposium (ARFTG), 2012 80th ARFTG, 2012,
pp. 1–6.
[108] J. Verspecht and D. Root, “Polyharmonic distortion modeling,” Mi-

crowave Magazine, IEEE, vol. 7, no. 3, pp. 44–57, 2006.
[109] B. Heinemann, R. Barth, D. Bolze, J. Drews, G. Fischer, A. Fox,

O. Fursenko, T. Grabolla, U. Haak, D. Knoll, R. Kurps, M. Lisker,
S. Marschmeyer, H. Rucker, D. Schmidt, J. Schmidt, M. Schubert,
B. Tillack, C. Wipf, D. Wolansky, and Y. Yamamoto, “SiGe HBT tech-
nology with fT/fmax of 300GHz/500GHz and 2.0 ps CML gate delay,”
in Electron Devices Meeting (IEDM), 2010 IEEE International, 2010,
pp. 30.5.1–30.5.4.
[110] V. D’Alessandro, I. Marano, S. Russo, D. Céli, A. Chantre, P. Chevalier,

F. Pourchon, and N. Rinaldi, “Impact of layout and technology param-
eters on the thermal resistance of SiGe:C HBTs,” in Bipolar/BiCMOS
Circuits and Technology Meeting (BCTM), 2010 IEEE, 2010, pp. 137–
140.
144 BIBLIOGRAPHY
[111] M. Schroter, J. Krause, N. Rinaldi, G. Wedel, B. Heinemann, P. Cheva-

lier, and A. Chantre, “Physical and electrical performance limits of high-
speed Si GeC HBTs - Part II: lateral scaling,” Electron Devices, IEEE
Transactions on, vol. 58, no. 11, pp. 3697–3706, 2011.
[112] K. Manjanna, K. Buisman, M. Spirito, M. Marchetti, M. Pelk, and

L. de Vreede, “Synthesized pulsed bias for device characterization,”
in Microwave Measurement Symposium (ARFTG), 2013 81st ARFTG,
2013, pp. 1–4.
Summary
Title: Mixed-Signal Instrumentation for Large-Signal Device Char-

acterization and Modelling
By: Mauro Marchetti
This thesis concentrates on the development of advanced large-signal measure-

ment and characterization tools to support technology development, model
extraction and validation, and power amplifier (PA) designs that address the
newly introduced third and fourth generation (3G and 4G) wideband com-
munication standards. By exploiting an innovative mixed-signal approach,
the measurement systems developed within this thesis work extend the limits
of the current state-of-the-art large-signal characterization in terms of band-
width, power range, speed and functionality. As described in Chapter 1
these activities are needed to address the demands that follow from the ever-
growing data transfer rates in modern telecommunication systems. Here the
introduction of the new 3G and 4G communication standards, which make
use of higher bandwidths and peak-to-average ratios, in combination with the
necessity to reduce the power consumption of mobile networks, put very strin-
gent demands on the power amplifier in the wireless transmitter. Since the
PA is one of the dominant sources of energy consumption in the wireless net-
work, it needs to be very efficient and linear over a wide frequency range and
over a wide power span. Therefore, to accomplish this difficult mission, the
PA designer has to rely on either an accurate nonlinear model of the active
device to perform the design in a circuit simulator, or on device data resulting
from load-pull measurements, which can accurately characterize the transistor
performance parameters as a function of the load and source impedances at
all frequencies of interest.
In order to support the development of compact device models which in-
clude self-heating and trapping effects, isothermal measurement systems are
required. Chapter 2 introduces the theory and the requirements for pulsed-
RF and pulsed-DC measurements. Moreover, a new isothermal measurement
system is presented, which provides the ability to measure with DC and RF
145
146 Summary
pulses as short as 200 ns, while featuring a very high dynamic range (≈ 85 dB)
under pulsed-RF conditions, which is independent on the duty-cycle. The sys-
tem performance is discussed in detail through a set of benchmarks, and some
examples on isothermal active device characterization are provided.
The second and dominant part of this thesis introduces a revolutionary
active harmonic load-pull approach. Load-pull device characterization is fun-
damental to all activities related to PA design and PA development, from tech-
nology development, to model extraction and validation, to the actual power
amplifier design. For this reason Chapter 3 reviews conventional passive and
active source and load-pull architectures, and discusses their basic limitations,
with particular attention to the problems arising when characterizing devices
with wideband complex modulated signals. Moreover, the requirements of
active load-pull systems to perform high power measurements with complex
modulated signals are also explained.
To solve the problems of conventional load-pull systems when dealing with
wideband modulated signals, a novel active harmonic load-pull system based
on a mixed-signal approach is described in detail in Chapter 4. The system
developed during this thesis work enables the measurement of active devices
up to 120 MHz of modulation bandwidth, and allows arbitrary control of the
reflection coefficient in this band. Measurement data highlighting the system
performance, and measurement results on active devices are presented.
To enhance the process of developing new transistor technologies and
their application in very efficient and linear PAs, in Chapter 5, a new ap-
proach for enabling high-speed multidimensional source and load-pull param-
eter sweeps is introduced. The method described allows any combination of
multiple parameters (e.g., input power and/or fundamental and harmonic load
impedance) to be swept, at a very high speed, while maintaining all other pa-
rameters (e.g., second harmonic source impedance) accurately controlled to a
user-defined value. Moreover, several measurements are reported, with par-
ticular emphasis on the high-power capabilities of the system, both in CW as
well as under modulated signal excitations.
The option to measure time-domain voltage and current waveforms can
provide significant insight into the actual device behavior, which benefits to
power amplifier design, ruggedness evaluation, as well as to (database) model
extraction and validation. In Chapter 6 the basic theory behind the mea-
surement of high frequency time-domain voltage and current waveforms at the
device reference planes are discussed, and an extension to the mixed-signal
load-pull system described in the previous chapters is presented, with particu-
lar attention on the requirements of the calibration device used for the system
calibration. Furthermore an approach for time-domain waveform analysis of
Summary 147
multi-tone signals which are closely spaced in frequency is introduced.

To highlight the most unique capabilities of the mixed-signal load-pull
system developed in this thesis, Chapter 7 reports several relevant examples
of significant applications. In particular an out-of-band linearity optimization
of an HBT device, the characterization of a GaN device for high efficiency
PA design, and some very high-power device measurements for base-station
applications are described.
Chapter 8 finishes the thesis and gives the most important conclusions
and the recommendations for future work.
148 Summary
Samenvatting
Titel: “Mixed-Signal” Meetsystemen voor de Grootsignaal Karak-

terisatie en Modellering van Componenten
Door: Mauro Marchetti
Dit proefschrift richt zich op de ontwikkeling van geavanceerde grootsignaal

meet- en karakterisatietechnieken t.b.v. innovaties in: transistortechnologie,
model-extractie / verificatie en RF vermogensversterkers voor de derde en
vierde generatie (3G, 4G) breedbandcommunicatiestandaarden. Door gebruik
te maken van een innovatieve “mixed-signal” aanpak (een combinatie van
analoog en digitaal) zijn, in het kader van dit proefschrift, nieuwe oplossingen
ontwikkeld die de limieten van de reeds bekende grootsignaalmeettechnieken
overtreffen in termen van bandbreedte, vermogen, snelheid en functionaliteit.
Hoofdstuk 1 geeft de achtergrond waarom de activiteiten, beschreven in dit
proefschrift, noodzakelijk zijn om aan de vraag van alsmaar hogere datasnel-
heden in draadloze netwerken te blijven voldoen. De introductie van de 3G en
4G communicatiestandaarden, welke gebruik maken van signalen met hogere
bandbreedtes en amplitude variaties, leveren samen de wens om het energie-
verbruik van mobiele netwerken te verminderen. Dit leidt tot zeer strenge eisen
voor de RF vermogensversterkers in de zenders van deze netwerken. Aangezien
deze versterkers de grootste energieverbruikers zijn in een draadloos netwerk,
moeten deze RF vermogensversterkers zeer efficiënt en lineair zijn over een
voldoende breed frequentiebereik en uitgangsvermogen. Om aan deze hoge
eisen te voldoen, vertrouwt de vermogensversterker ontwerper op niet-lineaire
modellen van de actieve componenten in een netwerk simulator of hij maakt
gebruik van ’load-pull’ meetdata. Deze laatste aanpak kan nauwkeurig de op-
timale “werkcondities” van de transistor bepalen als functie van belasting- en
bron-impedanties op alle vereiste frequenties.
Om de modelontwikkeling voor een actieve component te ondersteunen,
welke “self-heating” and “trapping” omvat, is een isotherm meetsysteem nood-
zakelijk. Hoofdstuk 2 introduceert de theorie en eisen aan zo’n gepulst
RF/DC meetsysteem. Met deze kennis wordt een nieuw isothermmeetsys-
149
150 Samenvatting
teem concept geı̈ntroduceerd, welke de mogelijkheid heeft om zowel DC en RF

met zeer korte pulsbreedtes (200 ns) te pulsen. Dit, met een zeer hoog dy-
namisch bereik (≈ 85 dB) welke onafhankelijk is van de “duty cycle” (aan-uit
verhouding). De prestaties van dit systeem zijn gedemonstreerd d.m.v. ver-
schillende benchmarkmetingen en de isotherme karakterisatie van een aantal
actieve componenten.
Het tweede en meest belangrijke onderdeel van dit proefschrift betreft de
introductie van een revolutionaire actieve harmonische “load-pull” meetmeth-
ode. “Load-pull” component karakterisatie is het fundament waarop alle ac-
tiviteiten m.b.t. vermogensversterker ontwerp en ontwikkeling zijn gebaseerd.
Dit geldt zowel voor de ontwikkeling van transistortechnologie zelf, als voor de
bijbehorende model extractie / verificatie en het uiteindelijke vermogensver-
sterkerontwerp. Hoofdstuk 3 bediscussieert de conventionele passieve en
actieve “load-pull” architecturen met hun beperkingen, specifieke aandacht
wordt gegeven aan de problemen die optreden wanneer de componenten wor-
den gekarakteriseerd m.b.v. breedbandige complex-gemoduleerde signalen.
Ter aanvulling worden de eisen verklaard die gesteld worden aan hoogvermo-
gensmetingen met complex-gemoduleerde signalen. Om de problemen van con-
ventionele “load-pull” systemen m.b.t. het gebruik van breedbandige complex-
gemoduleerde signalen op te lossen, wordt een er nieuw actief harmonisch
“load-pull” meetsysteem geı̈ntroduceerd in Hoofdstuk 4. Dit “mixed-signal”
meetsysteem maakt metingen van actieve componenten met 120 MHz modu-
latie bandbreedte mogelijk, terwijl de reflectiecoëfficiënt in deze bandbreedte
vrij kan worden gekozen en gecontroleerd. De mogelijkheden van dit nieuwe
meetsysteem worden gedemonstreerd d.m.v. meetdata m.b.t. het systeem
zelf, alsmede door metingen aan actieve componenten.
Om de ontwikkelingen van nieuwe transistortechnologieën en hun gebruik
in hoog-efficiënte en lineaire vermogensversterkers te bespoedigen, is in Hoofd-
stuk 5 een nieuwe methode geı̈ntroduceerd die zeer snelle ’load-pull’ metin-
gen met meerdimensionale parametervariaties mogelijk maakt. De geı̈ntro-
duceerde methode staat het toe om iedere combinatie van parameters (bi-
jvoorbeeld: ingangsvermogen en/of fundamentele en harmonische belasting
impedantie) te variëren met zeer hoge snelheid, terwijl andere parameters
(bijvoorbeeld: de tweede-harmonische bronimpedantie) nauwkeurig worden
vastgehouden op een ingesteld waarde. Ook worden diverse hoogvermogens
metingen beschreven, zowel voor continue signalen (CW) als voor complex-
gemoduleerde signalen.
De mogelijkheid om tijddomein spanning- en stroom golfvormen te meten,
geeft aanvullend inzicht in het gedrag van de actieve component. Dit biedt
voordelen voor het vermogensversterkerontwerp, de robuustheid evaluatie, als-
Samenvatting 151
mede voor (database) modelextractie en verificatie. In Hoofdstuk 6 wordt

de theorie voor het meten van tijddomein spanning- en stroomgolfvormen bij
hoge frequenties aan een specifiek referentievlak geı̈ntroduceerd. Om deze
metingen te ondersteunen is het meetsysteem verder uitgebreid en wordt er
gebruik gemaakt van een extra kalibratiecomponent. Extra aandacht wordt
gegeven aan de specifieke eisen van dit kalibratiecomponent om nauwkeurige
meetresultaten te kunnen garanderen. Verder wordt er een tijd-domein anal-
yse / kalibratie methode beschreven welke geschikt is voor signalen met veel
spectrale componenten binnen een beperkte bandbreedte.
Om de unieke mogelijkheden van de in dit proefschrift ontwikkelde meet-
technieken te belichten, geeft Hoofdstuk 7 diverse voorbeelden van relevante,
praktische toepassingen. Beschreven zijn: de ’out-of-band’ lineariteitsoptimal-
isatie van een HBT transistor, de karakterisatie van een GaN transistor voor
een hoog-efficiënte vermogensversterker en enige metingen bij zeer hoge ver-
mogensniveaus aan een transistor voor base-station applicaties.
Hoofdstuk 8 sluit dit proefschrift af met de belangrijkste conclusies en
aanbevelingen voor toekomstig werk in dit vakgebied.
152 Samenvatting
List of Publications
Journal Papers
M. Marchetti, “Mixed–signal active load pull: the fast track to 3G
and 4G amplifiers,” Microwave Journal, pp. 108–118, Sep. 2010.
C. Huang, P. J. Zampardi, K. Buisman, C. Cismaru, M. Sun, K. Stevens,
J. Fu, M. Marchetti and L. C. N. de Vreede, “A GaAs junction varac-
tor with a continuously tunable range of 9 : 1 and an OIP3 of 57 dBm,”
IEEE Electron Device Letters, vol. 31, no. 2, pp. 108–110, Feb. 2010.
J. H. Qureshi, M. J. Pelk, M. Marchetti, W. C. E. Neo, J. R. Ga-

jadharsing, M. P. van der Heijden and L. C. N. de Vreede, “A 90–W
peak power GaN outphasing amplifier with optimum input signal con-
ditioning,” IEEE Trans. Microwave Theory Tech., vol. 57, no. 8, pp.
1925–1935, Aug. 2009.
C. Huang, K. Buisman, M. Marchetti, L. K. Nanver, F. Sarubbi, M.

Popadić, T. L. M. Scholtes, H. Schellevis, L. E. Larson and L. C. N. de
Vreede, “Ultra linear low–loss varactor diode configurations for adaptive
RF systems,” IEEE Trans. Microwave Theory Tech., vol. 57, no. 1, pp.
205–215, Jan. 2009.
M. Marchetti, M. J. Pelk, K. Buisman, W. C. E. Neo, M. Spirito

and L. C. N. de Vreede, “Active harmonic load–pull with realistic wide-
band communications signals,” IEEE Trans. Microwave Theory Tech.,
vol. 56, no. 12, pp. 2979–2988, Dec. 2008.
C. Huang, L. C. N. de Vreede, F. Sarubbi, M. Popadić, K. Buisman,

J. H. Qureshi, M. Marchetti, A. Akhnoukh, T. L. M. Scholtes, L.
E. Larson and L. K. Nanver, “Enabling low–distortion varactors for
adaptive transmitters,” IEEE Trans. Microwave Theory Tech., vol. 56,
no. 5, pp. 1149–1163, May 2008.
153
154 List of Publications
Conference Papers
A. Kumar Manjanna, M. Marchetti, k. Buisman, M. Spirito, M. J.
Pelk and L. C. N. de Vreede, “Device characterization for LTE applica-
tions with wideband baseband, fundamental and harmonic impedance
control,” in Proc. 43rd European Microwave Conference, 2013, Nurem-
berg, Germany, Oct. 2013.
K. Buisman, M. Marchetti, M. P. van der Heijden, P. J. Zampardi
and L. C. N. de Vreede, “Evaluation of HBT device linearity using
advanced measurement techniques,” in Proc. 43rd European Microwave
Conference, 2013, Nuremberg, Germany, Oct. 2013.
A. Kumar Manjanna, K. Buisman, M. Spirito, M. Marchetti and M.

Spirito, M. Pelk and L. C. N. de Vreede, “Synthesized pulsed bias for
device characterization,” in Proc. 81st Automatic RF Techniques Group
Conf. (ARFTG2013), Seattle, WA, Jun. 2013.
L. Galatro, M. Marchetti and M. Spirito, “60 GHz mixed signal active

load–pull system for millimeter wave devices characterization,” in Proc.
80th Automatic RF Techniques Group Conf. (ARFTG2012), San Diego,
CA, Nov. 2012.
M. S. Alavi, F. van Rijs, M. Marchetti, M. Squillante, Tao Zhang, S.

J. C. H. Theeuwen, Y. Volokhine, H. F. F. Jos, M. P. v.d Heijden, M.
Acar and L. C. N. de Vreede, “Efficient LDMOS device operation for
envelope tracking amplifiers through second harmonic manipulation,”
in IEEE MTT-S Digest, Baltimore, MD, June 2011.
M. Marchetti, R. Heeres, M. Squillante, M. Pelk, M. Spirito and L.

C. N. de Vreede, “A mixed–signal load–pull system for base-station ap-
plications,” in Radio Frequency Integrated Circuits Symposium (RFIC),
Anaheim, CA, May 2010, pp. 491–494.
M. Mirra, M. Marchetti, F. Tessitore, M. Spirito, L. C. N. de Vreede

and L. Betts, “A multi–step phase calibration procedure for closely
spaced multi–tone signals,” in Proc. 75th Automatic RF Techniques
Group Conf. (ARFTG2010), Anaheim, CA, May 2010.
M. Squillante, M. Marchetti, M. Spirito and L. C. N. de Vreede, “A

mixed–signal approach for high–speed fully controlled multidimensional
load-pull parameters sweep,” in Proc. 73rd Automatic RF Techniques
Group Conf. (ARFTG2009), Boston, MA, Jun. 2009.
List of Publications 155
L. C. N. de Vreede, M. Pelk, E. Neo, J. Qureshi, M. Spirito, M. Squil-

lante, M. Marchetti, “Enhanced RF power amplifiers and device char-
acterization setups using coherent mixed–signal techniques,” in 11th
Annual Wireless and Microwave Technology Conference (WAMICON),
Melbourne, FL, April 2010.
M. Marchetti, K. Buisman, M. Pelk, L. Smith and L. C. N. de Vreede,

“A low–cost pulsed RF & I-V measurement setup for isothermal device
characterization,” in Proc. 70th Automatic RF Techniques Group Conf.
(ARFTG2007), Tempe, AZ, Nov. 2007.
K. Buisman, C. Huang, A. Akhnoukh, M. Marchetti, L. C. N. de

Vreede, L. E. Larson and L. K. Nanver, “Varactor topologies for RF
adaptivity with improved power handling and linearity,” in IEEE MTT-
S Digest, Honolulu, HI, June 2007, pp. 319–322.
M. Marchetti, M. Pelk, K. Buisman, M. Spirito and L. C. N. de

Vreede, “A pulsed network analyzer for high dynamic range isothermal
measurements,” in Proc. 68th Automatic RF Techniques Group Conf.
(ARFTG2006), Broomfield, CO, Dec. 2006.
Patents
M. Marchetti, M. J. Pelk, L. C. N. de Vreede, “Open loop load pull ar-
rangement with determination of injections signals,” World Intellectual
Property Organization Patent No. 2009131444, 30 Oct. 2009.
Book Chapters
M. Spirito and M. Marchetti, “Broadband large signal measurements
for linearity optimization,” in Modern RF and Microwave Measurement
Techniques, V. Teppati, A. Ferrero, and M. Sayed, Eds. Cambridge
University Press, 2013, ch. 14.
Workshop and Short Courses

M. Spirito and M. Marchetti, “Mixed–signal active load pull, the fast
track to 3G and 4G amplifiers,” in Short Course on Microwave Mea-
156 List of Publications
surements, 81st Automatic RF Techniques Group Conf. (ARFTG2013,

San Diego, CA, Nov. 2012.
M. Marchetti, M. Squillante, M. Pelk, M. Spirito and L. C. N. de
Vreede, “Active harmonic load–pull for LDMOS devices in base station
applications,” in Workshop on Silicon Characterization from MHz to
THz, 40th European Microwave Conference, 2010, Paris, France, Sep.
2010.
M. Marchetti, M. Squillante, M. Pelk, M. Spirito and L. C. N. de

Vreede, “A mixed–signal approach for large–signal device characteri-
zation with wideband modulated signals,” in Workshop on Advanced
Non-Linear Characterization of RF and Microwave Components, 39th
European Microwave Conference, 2009, Rome, Italy, Sep. 2009.
Acknowledgments
My personal experience as a Ph.D. could very well be summarized with this

old Neapolitan say: “Dicette ’o pappice vicino ’a noce: damme ’o tiemp ca te
spertoso” (“The worm told the walnut: give me time and I will pierce you”).
That is to say that with perseverance, a good degree of stubbornness and a
sufficient amount of time, everything can be accomplished. Since I started as
a Ph.D. student in July of 2006, it did take quite some time to complete this
journey. Through all this time, I had the chance to meet a lot of exceptional
people, who contributed to this thesis in many different ways, and without
whom I could have never reached this goal. To all of you goes my unlimited
gratitude.
First of all I would like to thank my promotor Prof. John Long for his
many advices and for his precious feedback on my thesis. Furthermore, I
would like to thank my co-promotor Dr. Leo de Vreede. Without his constant
guidance, support and scientific creativity I would have never made it this
far. Thank you for believing in me and for giving me the opportunity first to
pursue a Ph.D., then to start a company together. A special acknowledgment
also goes to my other advisor and also co-founder of Anteverta, Dr. Marco
Spirito. Since I arrived in Delft to carry out my master thesis, you have proved
to be a constant source of knowledge, a good friend, and the most talented
person I know in making the most amazing plots and figures.
For accepting to participate in my Ph.D. defense and for taking the time
to review my thesis, I would like to thank all the members of my doctoral
committee: Prof. Valeria Teppati, Prof. Rik Jos, Prof. Dominique Schreurs,
Dr. Marc Vanden Bossche, Prof. Lis Nanver and Prof. Andrea Neto. I am
also grateful to Prof. Lina Sarro for frequently reminding me, and all my
co-promotors, that some things in life do need to end (especially the book you
are now reading).
Some time after I commenced my Ph.D., my doctoral research went hand
in hand with my adventure as an entrepreneur. In this perspective I have to
express my gratitude to all the people not mentioned so far that have been
involved with successfully starting up and growing Anteverta.
157
158 Acknowledgments
First and foremost I have to thank Michele Squillante for too many things
to mention. To name a few, thank you for developing big part of the current
load-pull software, for sharing most of the burden and many late evenings of
work without ever complaining, for the sausage with beans and the occasional
mozzarella, and most of all for being a good friend. I would also like to ac-
knowledge the other two members of the current Anteverta team for their hard
work, commitment and good spirit, Ajay Kumar Manjanna and Giampiero “’o
’mericano” Esposito (who has joined us and found the new continent). This
is the proper moment to also thank Marco Pelk for being able to solve most
of the practical technical issues we encountered and for knowing the answer
to some of the most difficult questions.
I also want to express my appreciation for Han Oey, Paul Althuis and
Ronald Gelderblom and all the other people of the Delft Valorisation Center
for their guidance during the start-up of Anteverta and for their continuous
support.
A special acknowledgment goes to all the people at NXP that have be-
lieved in us and in our system when we were still writing lines of code to
run a measurement. Particularly, I am grateful to Rob Heeres for sharing his
knowledge, for the continuous feedback, for his patience with the early ver-
sions of the software and for being one of the most sincere supporters of our
system. I would also like to thank Steven Theeuwen for generously providing
many test devices for us to write papers, test new software and sometimes to
destroy. A special acknowledgment also goes to Dr. Mark van der Heijden for
many discussions and for even bringing customers to us. Furthermore, I have
to thank Dr. Martino Lorenzini for the many measurement sessions and for
providing the GaN devices, Rob Bubeck, Angelo Andres, Petra Hammes and
Lex Harm for their feedback and for many useful discussions.
I would also like to thank all the people at Maury Microwave for helping
us in turning the research shown in this thesis into a commercial product. In
particular, I have to thank Steve Dudkiewicz for the many advices and almost
daily Skype calls.
Carrying out my Ph.D. first in the HiTec group, then in the ELCA group,
has proven to be a very enjoyable experience. I have had the pleasure to work
with some of the brightest and most interesting people I know. First of all I am
indebted to Dr. Koen Buisman for being always the first user and debugger
of my measurement systems, for answering my many questions about device
physics, for translating the summary and propositions of this thesis in Dutch,
and for a whole lot of other things. I have also had the pleasure to work
closely with Dr. Edmund Neo, Dr. Jawad Qureshi and Dr. Cong Huang to
whom goes my gratitude. Furthermore, for the enjoyable atmosphere and the
Acknowledgments 159
many fun times I would like to acknowledge my colleagues Morteza Alavi, Rui
Hou, David Calvillo Cortes, Luca Galatro, Gennaro Gentile, Yi Zhao, Akshay
Visweswaran. I would also like to thank Atef Akhnouk for his help in the
measurement room, for his phone calls to the Dutch tax office on my behalf,
and for continuously reminding me to be careful. I would also like to thank
Marysia Lagendijk, Marian Roozenburg, Bianca Knot and Marion de Vlieger
for their support.
Being far from home is never easy, therefore I would like to thank all
the friends that have made me feel at home both in Delft and in Eindhoven:
Luigi “van der Appels” Mele, Agata Ŝakić who shares my same chocolate
addiction, Riccardo “Materassi” Donatantonio, Luis Alberto Cusati nonos-
tante non sia il mio presidente, Roberto “Johnny” Amabile with whom I’ll
always be happy to go surfing, Francesco Vitale and all his wooden spoons,
Salvatore “rum e cotechino” Russo my MATLAB GUI teacher, Fabio San-
tagata, Elina Iervolino, Alessandro “the wall” Baiano, Maria De Biase, An-
drea Ingenito and his wonderful light blue Vespa, Daniel Tajari Mofrad, Ben-
jamin Mimoun, Luigi La Spina, Giovanna Razzano, Francesco Sarubbi, Gian-
paolo Lorito, Theodoros Zoumpoulidis, Olindo Isabella, Joelle Olivet, Ghaz-
aleh Nazarian, Vladimir “the professor” Jovanovic, Yann “purpette” Civale,
Giuseppe Fiorentino, Bruno Morana, the Andricciola’s, the De Maio’s, the
Tripodi’s, Fabio Sebastiano the master of beer, Elisa “Dr. House” Buonanno,
Salvo Drago, Lisa de Vries, Alberto Fazzi, Laura Pirani, Gerard Villar Pique,
Muhammed Bolatkale, the Van der Weide’s, the “6 o’ clock heroes” football
team.
Thanks to my friends back at home (who are now spread throughout
the world) because when I am back at home it feels like I never left: Gia-
como “o’ So’” Sozio, Alessandro “l’avvocato” Caprio, Ciro Pileggio, Marco
”o’ lion’” Leone, Michele “sapone” Esposito, Francesco Miglino, Francesco
Liquido, Claudio “culo” Esposito, Andrea “o’ So’ grande” Sozio, Luca Russo,
Alfonso “pivell” Losco, Roberta “Enzo” Esposito.
Being myself an Italian man from the South, you should know that for us
(and probably not only for us), family is the most important thing (and no,
not in “the Godfather” kind of way). Truly, life without family is like a babà
(typical Neapolitan dessert) without rum.
I would like to thank all my relatives, my aunts and uncles, my parents
in law Carmela e Antonio Spella, Claudio “maestro spigola” Spella, Simona
“a milanese” De Toni, Francesca Spella, Giovanni Colacurcio e la piccola Gio-
vanna. Thanks to them all for their sustained encouragement and support.
I am grateful to my sister for bringing some mess (and a lot more fun) into
my super organized world while growing up together, for being my fashion
160 Acknowledgments
advisor, and for coming to visit me in the Netherlands more than I have gone
to visit her in Gorizia. I would also like to thank the new addition to the
family Salvo Pennisi, who some day will hopefully catch some fresh fish in
Palinuro when I am there.
I will always be grateful to my parents for raising me with love, for their
continuous encouragements, unconditional support and for making me what I
am today. Although we are geographically so distant now, we will always be
close to each other.
To conclude, I want to express my love and profound appreciation to my
wife Maristella. Thanks for lighting up my existence and for facing life with
me while supporting me every step of the way: without you no achievement
would be possible. Finally, thanks to my wonderful little son Matteo, because
you are my purpose of life and whatever you will grow up to be, you will
always be my biggest achievement.
About the Author
Mauro Marchetti was born in Naples, Italy, in 1981. He received his B.S. and
M.Sc. degrees (cum laude) in electrical engineering from the “Università degli
studi di Napoli Federico II”, Naples, Italy, in 2004 and 2006 respectively.
In 2006 he joined the Electronic Research Laboratory (ELCA) of Delft
University of Technology, The Netherlands, where he carried out his Ph.D.
research on the development and implementation of mixed-signal instrumen-
tation for large signal device characterization and modelling.
In 2010 he co-founded and was appointed CEO of Anteverta-mw B.V., a
spin-off company of Delft University providing mixed-signal harmonic load-
pull measurement systems.
161

5a 059 PDF

Uploaded by

Copyright:

Available Formats

5a 059 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

5a 059 PDF

Uploaded by

Copyright:

Available Formats

Mixed-Signal Instrumentation

for Large-Signal Device

ter verkrijging van de graad van doctor

op maandag 25 november 2013 om 15.00uur

Rector Magnificus, voorzitter

Keywords: load-pull, isothermal, large-signal, complex modulated signals, dy-

Copyright c 2013 by Mauro Marchetti

Cover design by twoeight studio, www.twoeight.it.

2 Isothermal Measurement Systems 11

3 Source and Load-pull Architectures 33

4 Mixed-Signal Active Load-Pull with Realistic Wideband Mod-

5 High-Speed, High-Power, Fully-Controlled, Multi-dimensional

6 Measurement of Time-Domain Waveforms 79

7 Application Examples 109

8 Conclusions and Future Work 127

List of Publications 153

About the Author 161

1.1 Trends in wireless communication

points than people on this planet. Moreover, driven by applications such as

CO2 emissions and its contribution is growing by 16 − 20 % yearly, doubling

• Operational strategies such as dynamic shut down of some base-stations

• Network architectures based on low power microcells, instead of high

• The increase of the transmission range of base-stations by using smart

Nevertheless, when considering the architecture of any base-station, the

1.2 Requirements on the power amplifier

Figure 1.3: Probability Density Function (PDF) for a downlink W-CDMA

Figure 1.4: Example of a spectral mask of a 802.11n WiFi signal. The

1.3 The need for advanced measurement

to advanced load-pull setups with large-signal vector measurements, there is

1.4 Thesis objectives

tions with wideband modulated signals. By further exploiting this technique,

1.5 Thesis outline

Isothermal measurement systems are typically used to evaluate the perfor-

Pulsed width (PW)

Pulse repetition interval (PRI)

Figure 2.1: Pulsed-RF signal.

dynamic range under pulsed-RF conditions. The realized system performance

2.1 Pulsed measurements fundamentals

Y (s) = [P W · sinc(P W · s) ∗ X(s)] · [P RF · shah(P RF · s)] (2.3)

Y (s) = [P W · sinc(P W · s)] · [P RF · shah(P RF · s)] (2.4)

Y (s) = DutyCycle · sinc(P W · s) · shah(P RF · s) . (2.5)

-1/2PW 0 1/2PW -2/PRF -1/PRF 0 1/PRF 2/PRF -1/PRF 0 1/PRF

Figure 2.2: Mathematical representation of a pulsed-RF signal. Source

Pulse repetition frequency

Figure 2.3: Frequency spectrum of the pulsed-RF signal of Fig. 2.2.

isothermal measurements. When using the narrowband technique, instead,

2.2 System configuration

acquisition system to implement the pulsed network analyzer, which provides

2.2.1 Pulsed VNA architecture

Auriga Pulser Auriga Pulser

LO 1 2 a1 CH. 1 CH. 1 CH. 1 CH. 1

10 MHz Ref MXI Link

at the frequency of interest. Finally, a ratio between two of the measured

Signal processing and dynamic range considerations

DR = 6.02 · N + 1.76 . (2.7)

power level that can be measured with a 50 Ω input impedance is

yielding a dynamic range of

DR = P max(dBm) − P min(dBm) ≈ 59.4 dB . (2.10)

Figure 2.7: Time-domain averaging of different periods of a sine wave.