Active Electronic Loads for
Radiometric Calibration
Von der Fakultät für Informatik, Elektrotechnik und Informationstechnik der
Universität Stuttgart zur Erlangung der Würde eines Doktor-Ingenieurs (Dr.-Ing.)
genehmigte Abhandlung
Vorgelegt von
Ernst Weissbrodt
geboren in Moskau, Russland
Hauptberichter: Prof. Dr. Ingmar Kallfass
Mitberichter: Prof. Dr. Gilles Dambrine
Tag der mündlichen Prüfung: 06.12.2016
Institut für Robuste Leistungshalbleitersysteme der Universität Stuttgart
2017
To my parents
Executive Summary
Although radiometer systems are widely applied in very different fields, they all have one
important requirement in common: They require a thorough radiometric calibration. Vari-
ous conventional calibration references are well established, but their bulkiness, high power
consumption, and complexity are limiting the expanding fields of application. Since novel
industrial applications such as passive millimeter-wave imaging emerge, the requirements
for calibration references have increased drastically. But also in scientific fields like radio
astronomy, cosmology or environmental monitoring, modern remote sensing radiometers do
not rely only on conventional references anymore.
In this work, millimeter-wave monolithic integrated circuits (MMICs) based on metamorphic
high electron mobility transistors (mHEMT) were designed to be used as active electronic
loads for radiometric calibration. These novel references have not only the outstanding prop-
erty, that they can be directly integrated on chip-level into the radiometer front-end, but
also, that they can exhibit cold as well as hot reference noise temperatures. Since this is
achieved without any physical cooling or heating, the power consumption is notably reduced.
By monolithic integration of field effect transistor (FET) switches, theses multiple references
can internally be routed to the receiver input without any mechanical wear. As a result, labo-
rious external references can be omitted and the repetition rate of the calibration procedure
increased, which results in a higher radiometric accuracy and allows a more compact and
cost-effective design of modern radiometer systems. This work presents the first radiometric
calibration front-end that allows to internally switch between active electronic cold and active
electronic hot loads, as well as a passive ambient load. All components are integrated on a
single MMIC, and a patent was granted for this innovation.
To predict the achievable noise temperatures of active cold loads (ACLs), different simula-
tion approaches were previously published. This work evaluates and adapts these existing
approaches to design and manufacture several W-band loads. But the required design-flow
was found to be very time-consuming because multiple iterations are necessary to succes-
sively design and optimize the input- and output matching networks, and to finally achieve
the desired low noise temperature. Therefore, a novel simulation approach is introduced
that makes efficiently use of modern optimization algorithms and the very accurate model
library of the mHEMT technology and the passive structures. With this novel simulation
method, the first active hot loads (AHLs) were designed as well as state-of-the-art ACLs up
to 140GHz.
However, the characterization of low-noise one-port devices is particularly challenging, espe-
cially at such high frequencies. Hence, a substantial part of this work is to investigate the
reliability of different noise measurement setups and the repeatability of noise temperature
results. Dedicated setups in W- and D-band are used to characterize all manufactured ac-
tive loads and some selected results are cross-checked by measuring the same circuits with
independently designed measurement systems of other research facilities. The discrepant re-
sults are discussed, concluding that high variations in measured one-port noise temperature
III
do not allow to rely on one single measurement setup. At the same time, this thorough
investigation and comparison permits to establish an accuracy range within which the results
of the manufactured active electronic loads are reliable, whereas other previously published
ACLs are typically only measured with one measurement setup.
Zusammenfassung
Obwohl Radiometer Systeme in unterschiedlichsten Anwendungsgebieten weit verbreitet sind,
haben alle eine wichtige gemeinsame Anforderung: Sie benötigen eine sorgfältige radiome-
trische Kalibration. Verschiedene konventionelle Kalibrationsreferenzen sind zwar bewährt,
doch ihre Sperrigkeit, Komplexität und die hohe Leistungsaufnahme schränken die sich ste-
tig erweiternden Anwendungsmöglichkeiten ein. Seit neuartige industrielle Anwendungen wie
passive Millimeterwellen-Kameras z.B. bei Sicherheitskontrollen aufkommen, haben sich auch
die Anforderungen an die notwenigen Kalibrationsreferenzen drastisch erhöht. Aber auch in
wissenschaftlichen Anwendungsgebieten wie der Radioastronomie, der Kosmologie oder der
Umweltbeobachtung beschränken sich moderne Fernerkundungs-Radiometer nicht mehr nur
auf konventionelle Kalibrationsreferenzen.
Im Rahmen dieser Arbeit wurden monolithisch integrierte Millimeterwellen-Schaltkreise
(MMICs), basierend auf einer metamorphen Transistor-Technologie mit hoher Elektronenbe-
weglichkeit (mHEMT) entwickelt, die als aktive elektronische Referenz-Last zur radiometri-
schen Kalibration dienen. Diese neuartigen Referenz-Lasten haben nicht nur die herausragen-
de Eigenschaft direkt auf Halbleiter-Bauteilebene mit dem Radiometereingang integriert wer-
den zu können, sondern können außerdem sowohl niedrige („kalte“) als auch hohe („heiße“)
Referenz-Rauschtemperaturen erzeugen. Da dies ohne physikalische Kühlung oder Erhitzung
erreicht wird, ist die Leistungsaufnahme besonders gering. Durch die monolithische Integra-
tion von Feldeffekttransistor-Schaltern können diese Referenz-Lasten elektronisch intern auf
den Empfängereingang geschaltet werden, ohne dass mechanische Schalt- oder Bewegungs-
vorgänge notwendig sind. Infolgedessen sind aufwändige externe Referenzen nicht mehr zwin-
gend notwendig, so dass ein kompaktes, kostengünstiges und modernes Radiometer-System
ermöglicht wird. Außerdem kann die Wiederholrate der Kalibrationsprozedur erhöht werden,
was sich in einer erhöhten radiometrischen Genauigkeit auswirkt. Diese Arbeit präsentiert
die erste radiometrische Kalibrationseinrichtung, die es ermöglicht intern zwischen aktiven
elektronischen kalten und heißen Referenz-Lasten, sowie einer passiven Referenz-Last auf
Umgebungstemperatur umzuschalten. Alle notwendigen Bauteile sind monolithisch auf ei-
nem Millimeterwellen-Schaltkreis integriert, worauf ein Patent erteilt wurde.
Um die erreichbare Rauschtemperatur aktiver kalter Referenzlasten vorherzusagen, wurden in
der Vergangenheit verschiedene Ansätze veröffentlicht. In dieser Arbeit wurden diese Ansätze
bewertet und angewandt um mehrere Referenzlasten im W-Band zu entwickeln und herzustel-
len. Der notwendige Design-Flow stellte sich jedoch als sehr zeitaufwändig heraus, da meh-
rere Iterationen notwendig sind, um sukzessiv die Eingangs- und Ausgangs-Anpassnetzwerke
zu entwerfen und zu optimieren, um letztlich die gewünschte Rauschtemperatur zu errei-
chen. Daher wurde im Rahmen dieser Arbeit eine neuartige Simulationsmethode eingeführt,
die sich die Möglichkeiten moderner Optimierungs-Algorithmen und sehr genauer Modell-
Bibliotheken der verwendeten Halbleitertechnologie und der passiven Komponenten zu Nut-
ze macht. Mit dieser Methode wurden die erste aktive heiße Referenzlast, sowie eine den
neusten Stand der Technik definierende kalte Referenzlast bei 140GHz entwickelt.
V
Allerdings ist die messtechnische Charakterisierung eines so rauscharmen Eintor-Bauteils, wie
es die kalte Referenzlast ist, besonders herausfordernd - vor allem bei solch hohen Frequenzen.
Daher besteht ein wesentlicher Teil dieser Arbeit darin, die Zuverlässigkeit von Rauschmess-
Einrichtungen und die Reproduzierbarkeit von Rauschtemperatur-Ergebnissen zu untersuchen.
Hierfür wurden spezielle Rauschmessplätze im W-Band und D-Band entworfen und zunächst
sämtliche Schaltungen damit charakterisiert. Anschließend wurden ausgewählte Schaltungen
und Module zu anderen Forschungseinrichtungen geschickt, wo sie mit deren unabhängig
entworfenen Rauschmess-Einrichtungen untersucht wurden. Die diskrepanten Ergebnisse las-
sen darauf schließen, dass die starke Variation der gemessenen Rauschtemperaturen es nicht
erlaubt, diese mit nur einem Messsystem zu bestimmen. Gleichzeitig jedoch, ermöglichte der
Vergleich der Ergebnisse verschiedener Rauschmess-Einrichtungen, eine Genauigkeitsspanne
zu bestimmen innerhalb welcher die Ergebnisse der in dieser Arbeit entwickelten aktiven
elektronischen Referenzlasten aussagekräftig und zuverlässig sind. Während in anderen Pu-
blikationen typischerweise lediglich ein Messsystem zum Einsatz kommt.
VI
Contents
Executive Summary III
Zusammenfassung V
Symbols IX
Abbreviations XI
1 Introduction 1
1.1 Radiometer Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Application Area of Radiometer Systems . . . . . . . . . . . . . . . . . . . 1
1.3 Calibration of Radiometers . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Motivation and Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 State of the Art 5
2.1 Theoretical Background of Radiometry . . . . . . . . . . . . . . . . . . . . 5
2.2 Conventional Calibration References . . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Passive Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.2 Targets with Known Brightness Temperature . . . . . . . . . . . . . 7
2.2.3 Noise Diodes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Active Electronic Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Front-End Switches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4.1 Mechanical Waveguide Switches . . . . . . . . . . . . . . . . . . . . 10
2.4.2 MEMS Switches . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4.3 FET Switches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5 Overview of Calibration Concepts . . . . . . . . . . . . . . . . . . . . . . . 11
2.6 Low-Noise Amplifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 Active Load Integrated Circuit Design 15
3.1 Operating Frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2.1 MMIC Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2.2 IAF’s mHEMT Process . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2.3 Long-Term Stability . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Stand-alone Active Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3.1 ACL Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3.2 Simulation of Noise Temperature . . . . . . . . . . . . . . . . . . . 21
3.3.3 Practical Design Flow . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3.4 Novel Approach for Improved Design Flow . . . . . . . . . . . . . . 28
3.3.5 Generic Design Schematic . . . . . . . . . . . . . . . . . . . . . . . 31
VII
3.3.6 Realized Circuits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4 Active Loads with Integrated Switches . . . . . . . . . . . . . . . . . . . . 36
3.4.1 Switching Calibration Front-End . . . . . . . . . . . . . . . . . . . . 36
3.4.2 Active Calibration Sub-System . . . . . . . . . . . . . . . . . . . . . 38
3.4.3 Active Hot-Cold Load . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5 D-Band Active Cold Loads and Low-Noise Amplifier . . . . . . . . . . . . . 41
4 One-Port Noise Measurement Setups 45
4.1 Noise Figure Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Evaluation of Hot and Cold References for Y-factor Measurements . . . . . . 48
4.2.1 Noise Diode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2.2 Cooled or Heated Waveguide Loads . . . . . . . . . . . . . . . . . . 50
4.2.3 Artificial Black Body . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3 Measurement System Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.3.1 System Noise Figure and Gain . . . . . . . . . . . . . . . . . . . . . 54
4.3.2 Interstage Mismatching . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Final Measurement Setups in W- and D-band . . . . . . . . . . . . . . . . . 57
4.4.1 W-Band Setups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.4.2 D-Band Setups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5 Characterization of W-Band Active Load MMICs and Modules 65
5.1 MMIC measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.1.1 89 GHz ACLs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.1.2 W-band ACLs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.1.3 W-band AHL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.2 Module Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.2.1 Comparison of Active Load Modules to Conventional LNAs as a Load 76
5.2.2 ACL Module as Hot-Cold Load in Noise Measurements . . . . . . . . 79
5.2.3 Temperature Dependence of ACLs . . . . . . . . . . . . . . . . . . 80
5.3 Active Loads with Integrated Switches . . . . . . . . . . . . . . . . . . . . 81
5.3.1 Switching Calibration Front-End . . . . . . . . . . . . . . . . . . . . 81
5.3.2 Active Calibration Sub-System . . . . . . . . . . . . . . . . . . . . . 84
5.3.3 Active Hot-Cold Load . . . . . . . . . . . . . . . . . . . . . . . . . 87
6 Characterization of D-Band Active Load MMICs 91
6.1 LNA Module for Measurement Setup . . . . . . . . . . . . . . . . . . . . . 91
6.2 Active Cold Load MMIC in D-Band . . . . . . . . . . . . . . . . . . . . . . 94
7 Conclusion 99
References 103
Patents and Personal Publications 111
Acknowledgments 113
VIII
Symbols
B bandwidth
Bn noise bandwidth
c speed of light
∆f bandwidth
∆G gain fluctuation
∆T radiometric resolution
∆TC calibration noise contribution to radiometric resolution
∆TG contribution of gain fluctuations to radiometric resolution
∆TS system noise contribution to radiometric resolution
F noise factor
f frequency
1
f
flicker noise
Fmin minimum noise figure
Fsys system noise figure
fT transit frequency
Γ reflection coefficient
Γ1 reflection coefficient at source terminal
Γ2 reflection coefficient at load terminal
Γin reflection coefficient at transistor input
ΓL load reflection coefficient
Γopt source reflection coefficient for minimum noise
Γout reflection coefficient at transistor output
ΓS source reflection coefficient
G gain
Ga available gain
Gij available power gains of transistor
Gij,ti terminal invariant gains
Gopt conductance for minimum noise
gm,max maximum extrinsic transconductance
Gtot total noise conductance
h Planck constant
Icor noise current correlation matrix
Id drain current
itot equivalent noise current of an otherwise noiseless device
k Boltzmann constant
IX
LA ohmic losses of the antenna and the antenna feed
Lf spectral brightness density
Llin linear loss
µ0 electron mobility
N number of calibrations
Na added noise
ne electron density
Ni input noise power
No output noise power
NFmin minimum noise figure for optimum reflection coefficient
P power
Rn noise resistance
Rtot total noise resistance
S scattering parameter matrix
Si input signal power
So output signal power
Sopt scattering parameter for minimum noise
τC calibration integration time
τS scene integration time
T0 defined temperature of 290K
T1 noise temperature at source terminal
T2 noise temperature at load terminal
TA antenna temperature
Ta alternative noise parameter
TB brightness temperature
Tb alternative noise parameter
Tc cold reference temperature
Th hot reference temperature
Tk alternative noise parameter
TL load terminal
Tn noise temperature
Tphys physical temperature
TRec receiver noise temperature
T ′Rec receiver noise temperature at the antenna input reference plane
T ′Ref reference load noise temperature at the antenna input reference plane
TS source terminal
Ts1 generated noise temperature at active cold load output
Ts1,limit theoretical lower limit of noise temperature at transistor output
TSys system noise temperature
Vg gate voltage
Vd drain voltage
Vsw switch voltage
Y admittance parameter matrix
Z impedance parameter matrix
X
Abbreviations
2DEG Two-Dimensional Electron Gas
ABCS Antimonide Based Compound Semiconductor
ACL Active Cold Load
ACSS Active Calibration Sub-System
ADS Advanced Design System by Keysight Technologies
AHCL Active Hot-Cold Load
AHL Active Hot Load
AWVR Advanced Water Vapor Radiometer
BJT Bipolar Junction Transistor
CAD Computer Aided Design
CAS Calibration Subsystem
CBR Cosmic Background Radiation
CMB Cosmic Microwave Background
COBE Cosmic Background Explorer
DC Direct Current
DLR National Aeronautics and Space Research Centre
DSP Digital Signal Processor
DUT Device under Test
ENR Excess Noise Ratio
EPW EUMETSAT Polar System
ESA European Space Agency
ESTEC European Space Research and Technology Centre
FET Field Effect Transistor
FOM Figure of Merit
GaAs Gallium Arsenide
GCPW Grounded Coplanar Waveguide
HAMSR High-Altitude MMIC Sounding Radiometer
HBT Heterojunction Bipolar Transistor
HEMT High Electron Mobility Transistor
HF High Frequency
HSB Humidity Sounder for Brazil
IAF Fraunhofer Institute for Applied Solid-State Physics
IEC International Electrotechnical Commission
IMN Input Matching Network
In Indium
XI
InAlAs Indium Aluminum Arsenide
InGaAs Indium Gallium Arsenide
InP Indium Phosphide
ISO International Organization for Standardization
ISS Impedance Standard Substrate
JMR Jason Microwave Radiometer
JUICE Jupiter Icy Moons Explorer
LN Liquid Nitrogen
LNA Low-Noise Amplifier
LO Local Oscillator
MBE Molecular Beam Epitaxy
MEMS Microelectromechanical System
MESFET Metal-Semiconductor Field Effect Transistor
mHEMT Metamorphic High Electron Mobility Transistor
MIM Metal-Insulator-Metal
MIRAS Microwave Imaging Radiometer using Aperture Synthesis
ML Matched Load
MMIC Millimeter-Wave Monolithic Integrated Circuit
MMW Millimeter-Wave
MSMR Multifrequency Scanning Microwave Radiometer
MWR Microwave Radiometer
NASA National Aeronautics and Space Administration
ND Noise Diode
NF Noise Figure
NFA Noise Figure Analyzer
NiCr Nickel-Chromium
NIR Noise Injection Radiometer
NIST National Institute of Standards and Technology
NTC Negative Temperature Coefficient
OMN Output Matching Network
PCB Printed Circuit Board
pHEMT Pseudomorphic High Electron Mobility Transistor
PIN Semiconductor with a p-type / intrinsic / n-type region
PMMW Passive Millimeter-Wave
PSR Polarimetric Scanning Radiometer
PTB Physikalisch-Technische Bundesanstalt
RADAR Radio Detection and Ranging
RF Radio Frequency
RPG Radiometer Physics GmbH
S-parameter Scattering Parameter
SCFE Switching Calibration Front-End
SiGe Silicon Germanium
SMB Subminiature Type B connector
SMMR Scanning Multichannel Microwave Radiometer
SMOS Soil Moisture and Ocean Salinity
XII
SNR Signal-to-Noise Ratio
SP3T Single-Pole Triple-Throw
SP5T Single-Pole Five-Throw
SPDT Single-Pole Double-Throw
SPST Single-Pole Single-Throw
SWI Submillimeter Wave Instrument
TMR Topex/Poseidon Microwave Radiometer
UAV Unmanned Aerial Vehicle
VNA Vector Network Analyzer
VSWR Voltage Standing Wave Ratio
VTS Variable Temperature Source
WMAP Wilkinson Microwave Anisotropy Probe
XIII

1 Introduction
1.1 Radiometer Systems
A radiometer is a device for measuring the electromagnetic radiation of a noise-like character
in a certain frequency range. In the context of this work only microwave and millimeter-
wave (MMW) radiometers will be considered. At these frequencies, radiometer systems
are similar to receiver architectures known from communication applications. But while in
communications, the information is contained in a modulated and coded signal on a narrow-
band carrier with a distinct signal-to-noise-ratio (SNR), in radiometry it is the noise itself
which is the signal of interest [1]. This, however, demands very sensitive receivers, because
the signal power level e.g. in radio astronomy receivers is in the order of 10−15 − 10−20W
and does not differ in its statistical properties from the noise originating in the receiver itself
or from the background noise [2].
There exists comprehensive literature on radiometry and radiometer systems [3], [4], [5],
where the physical fundamentals are explained and the assets and drawbacks of different
architectures are discussed. Therefore, this work will not elaborate on these fundamentals in
general, but only where it is necessary in the context of this work.
1.2 Application Area of Radiometer Systems
The physical quantities measured by radiometers and its terminology will be introduced in
the next chapter, whereas this section gives an overview of the application area of such
systems.
The key feature of a radiometer is that it is a passive instrument, which means that - unlike
RADAR1 - it does not emit radiation of any kind. Instead, it collects the electromagnetic
radiation of objects, which can be a combination of naturally emitted, scattered and pene-
trated radiation2.
Depending on the application, objects of interest can be anything between an oscillating
molecule in our atmosphere and a remote galaxy in space. Similar to the observation of ob-
jects with optical telescopes, this allows to detect objects and to derive some of their physical
properties from a distance by measuring the amount or characteristic of the radiation. This
is called remote sensing.
In general, remote sensing radiometers can be divided into ground-based and space-based
instruments. Examples for the former are zenith-pointed radiometers, observing earth’s atmo-
sphere for meteorological and environmental sensing [6], [7]. Another example are radiome-
ters used in astronomy, often arranged into large arrays, looking into deep space to detect
1RADAR: RAdio Detection And Ranging
2In radiometry the black body model is more common, which means that technically speaking a body does
not scatter radiation from other sources, but absorbs and emits it partially
1
distant stars and galaxies [8]. Even in cosmology this technique already played an important
role: for the detection and measurement of the cosmic background radiation (CBR) with
a highly sensitive ground-based radiometer, Arno Penzias and Robert Wilson were awarded
the Nobel prize for physics in 1978 [9]. Space-based radiometers allow to take this one step
further. Instruments on board of satellites can observe earth’s surface and atmosphere on a
large scale and the gathered data is used in meteorology, climatology, oceanography, glaciol-
ogy, just to name a few applications in the field of geoscience. For astronomic applications
satellite-based systems have the advantage that the observed extraterrestrial radiation is not
attenuated or otherwise influenced by earth’s atmosphere [10]. Also, radiometers are com-
mon instruments on board of space probes, exploring other planetary bodies while orbiting
or flying by.
An emerging field which is not referred to as remote sensing but still allows contact-free
detection of objects, is the application in security-screening portals and stand-off detection at
check-points [11], [12], [13]. This field is in fierce competition with RADAR technology and
the advantages and disadvantages of both are intensively debated in the scientific community.
In addition to technical arguments, radiometry with its passive nature benefits from superior
social acceptance, since it does not expose the human body to electromagnetic radiation. The
goal to detect concealed dangerous objects from a distance is based on the principle that both
- the human body and the carried objects - emit natural radiation which penetrates fabric such
as clothing or packaging. The contrast of different amounts of measured radiation generates
an image in which suspicious objects can be identified. This is called passive millimeter-wave
imaging (PMMW) [14], [15], [16], [17]. The quality of the image substantially depends on
the radiometric resolution of the system, i.e. the smallest difference in radiation which can
still be distinguished by the radiometer. To improve the radiometric resolution and accuracy,
a crucial procedure before and during operation is the calibration of the radiometer.
1.3 Calibration of Radiometers
The terms "calibration" and "standard" have to be used with particular caution! In metrol-
ogy3, the definition of calibration involves national and international standards, whereas
the active loads presented later on in this work are merely references, not claiming to be
standards in the scientific sense. (Primary) standards are objects, systems or experiments
defined by national metrological authorities like the National Institute of Standards and Tech-
nology (NIST) in the United States or the Physikalisch-Technische Bundesanstalt (PTB)
in Germany. They are used to calibrate instruments (or secondary standards) in specified
laboratory conditions with the goal that afterwards the instrument will return a measure-
ment result instead of just some output quantity. The formal definitions of measurement,
calibration and standard can be found in the International vocabulary of metrology [18], a
document published by a joint committee of institutions like the International Organization
for Standardization (ISO) or the International Electrotechnical Commission (IEC).
In this work the term calibration will be used much less restrictively. For example, the
adjustment of the radiometer to an instrumental drift (a change over time), as well as the
use of references for this repeated procedure (instead of standards) will also be referred to
3Metrology is the science of measurement - not to be confused with meteorology, the scientific study of
physical and chemical effects in earth’s atmosphere.
2
as calibration.
To explain the basic procedure of radiometer calibration, a Dicke radiometer [1] as shown
in Figure 1.1(a) is considered. Its antenna receives a certain amount of radiation and returns
a voltage at the output of its detector which can easily be digitized and treated by a digital
signal processor (DSP). As previously mentioned, the received radiation can not be distin-
guished from the radiometer inherent noise, so the displayed result is the sum of the received
radiation and the system noise. To overcome this, the radiometer input is switched from the
antenna to a termination (a passive impedance-matched load), which allows to determine
the system-noise. Subtracting this from the previously measured total power returns a
voltage corresponding to the received radiation. Since this procedure theoretically eliminates
the varying noise contribution of the instrument itself and allows to compare independent
observations, it is referred to as relative calibration. At this point, however, one still cannot
conclude the absolute amount of received radiation. For that it is necessary to observe at
least two known sources of radiation and assign the voltage values to the respective known
amount of radiation. Assuming a linear behavior of the radiometer, two references are
sufficient to produce a generic calibration-curve, which in this case is a straight line thru the
two reference points as can be seen in Figure 1.1(b). This procedure is thus called two-point
calibration or also absolute calibration. It is only now, by the unambiguous assignment of
input radiation to output voltage with the calibration-curve, that the calibrated radiometer
can measure the received radiation in terms of quantity and unit.
(a) Dicke radiometer (b) Two-point calibration
Figure 1.1: Calibration of radiometers.
1.4 Motivation and Outline
The previous sections have shown, that radiometric calibration is essential for operating
radiometer systems in any field. While the principle procedure is straightforward, real ra-
diometers do not always operate linearly and the choice and number of suitable references
influences the measurement range and accuracy. To reduce interpolation and extrapolation
errors, multiple references are beneficial. Furthermore, references whose levels are in the
vicinity of the expected targets improve the radiometric resolution, as will be explained in
3
Section 2.3.
The highest requirements for the accuracy of the radiometric calibration are still in the field
of scientific applications such as radio astronomy or environmental monitoring. The latter is
particularly important for climate change observations, weather predictions, and early warn-
ing systems for disaster risk reduction. In the previously mentioned emerging field of PMMW
imaging in industrial and commercial applications, the main focus is on the repetition rate of
the calibration procedure. Novel techniques generate real-time images, which demand a cali-
bration and processing of the data in video-rate. Also, high resolution multi-channel systems
demand calibration references for the pixel normalization. These applications necessitate
not only compact and reliable references, but especially a front-end to switch extremely fast
between the references and several signal channels. Since these applications target a mass
market, it is needless to say, that such references and switches must be cost-effective, very
compact and lightweight, and require low maintenance and low power consumption. Except
for the cost, these are the same factors that are critical in satellite missions. The lack of such
references motivated this work, since conventional references up to now are still bulky and
complex and mostly limited to laboratory or scientific instruments. The novel active loads
with integrated switches, presented in this work, however, meet all of the above mentioned
requirements.
After presenting the state of the art of conventional calibration references and switches, the
active electronic loads will be introduced in the next chapter. Chapter 3 will guide the reader
thru the entire designing process of the developed loads, from choosing the technology and
simulating the noise temperature, to the description of the designed circuits and layouts.
Furthermore, a novel noise temperature simulation method, that significantly simplifies the
design-flow will be introduced in that chapter. The challenging task of characterizing the
resulting circuits will be the topic of chapter 4, where different dedicated noise measurement
setups will be elaborated and the critical components will be discussed. Finally, Chapter
5 and Chapter 6 will present and discuss the results of the W-band and D-band circuits,
respectively, before Chapter 7 ends this work with a conclusion.
4
2 State of the Art
This chapter will introduce the theoretical basics necessary for the understanding and further
discussion of this work. After presenting the conventional loads in general, it will mainly focus
on the advantages of active electronic loads in space-based instruments, since part of this
work was supported by the European Space Agency (ESA), specifically for the investigation
of Active Cold Loads in Post-EPS1 / MetOp-SG2 mission context [19], [20]. Furthermore, an
overview of switches will be given, which are essential parts to integrate internal calibration
references in a radiometer front-end.
2.1 Theoretical Background of Radiometry
For the sake of consistency, the Recommended Terminology for Microwave Radiometry, as
described in [21] by the NIST will be used, unless otherwise noted.
All matter above a physical temperature of 0K (absolute zero) emits natural thermal elec-
tromagnetic radiation, which is caused by atomic and molecular state transitions. It is a
broadband non-polarized non-coherent random radiation (noise). To describe this physical
phenomenon, the theoretical model of a black body was introduced in 1860 by Gustav Kirch-
hoff. It is a perfect absorber and emits the radiation with a spectral brightness density Lf
described by the Planck formula
Lf =
2hf3
c2
1
ehf/kT − 1
(2.1)
where c is the speed of light, f is the frequency, h and k are the Planck and Boltzmann
constants respectively and T is the temperature. In the microwave regime, where hf/kT ≪ 1
(which is considered for frequencies below 300GHz), this formula can be approximated with
the Rayleigh-Jeans formula
Lf ≈
2kf 2
c2
T (2.2)
which results in a linear relationship between the spectral brightness density and the physical
temperature. This, and the fact that the physical temperature of the observed scene is
typically the quantity of interest in microwave remote sensing, leads to the common use of
the term brightness temperature TB, which in theory is only valid for the black body [22].
Within the Rayleigh-Jeans limit, the microwave power received by a radiometer is P = kBTB,
where B is the bandwidth of the radiometer. Because of its analogy to the Johnson noise
(thermal noise) generated in an electrical resistor, the physical temperature of a resistor
Tphys which would deliver the same power to the radiometer as the received brightness, is
1EPS: EUMETSAT Polar System, EUMETSAT: European Organisation for the Exploitation of Meteoro-
logical Satellites
2MetOp-SG: MetOp Second Generation, the space segment of the Post-EPS
5
called noise temperature Tn. Again, in the microwave regime the approximation Tphys ≈ Tn
is valid [21]. Since this term obviously describes a quantity inside the radiometer, where
every component adds noise, it is mandatory to always refer to a reference plane within the
microwave circuit or system when using this term.
Before explaining the difference between internal and external references in the next section,
it is important to understand, that the interface between outside and inside the radiometer
- the antenna - also contributes to the measured noise temperature, just like any other
component within the system. At its aperture (which in radiometry is always the antenna
input) it receives the incident brightness temperature TB, weighted by its directivity pattern.
At its output it delivers the noise received at its aperture, attenuated by its ohmic efficiency
and the thermal noise added by its own ohmic losses, depending on its physical temperature.
This emphasizes the importance of defining reference planes and will be further discussed in
Chapter 4.
2.2 Conventional Calibration References
2.2.1 Passive Loads
After the preceding explanations, two possible references for calibration become evident,
because their physical temperature is easily determined and interrelated to brightness tem-
perature:
• An object outside the radiometer and its antenna which has properties similar to a
black body
• An impedance-matched resistive load within the radiometer.
These two generic examples at the same time classify typical references into external and
internal calibration types. A practical example of the first type would be a microwave absorb-
ing material just like used in anechoic chambers. Similar to the antenna which transforms
the characteristic impedance of the waveguide to the free space impedance, the typically
pyramidal form of the microwave absorber transforms the free space impedance of the
electromagnetic wave into an ideally fully resistive load. Hence as much as possible energy
of the incident wave is transformed into heat within the material. In reverse, by knowing the
physical temperature of the absorber and accepting all the previously discussed restrictions,
the emitted radiation of the absorber can be used for radiometer calibration. This method
will also be used and further discussed in Chapter 4. Such an absorber is sometimes referred
to as an artificial black body. In practical application, microwave absorber is either cooled
by liquid gases like nitrogen or helium or heated to have cold and hot calibration targets
respectively.
For an internal impedance-matched resistive load the relation between physical temperature
and noise temperature is straightforward. But again, the return loss of the resistor is the
crucial parameter just as for the absorber. If it is not "perfectly" matched to the character-
istic impedance of its surrounding electrical network, the emitted noise temperature will not
equal its physical temperature.
6
2.2.2 Targets with Known Brightness Temperature
Further external references can be targets whose brightness temperatures have been previ-
ously exactly measured and which are known to be constant for the duration of the mission.
Especially for space-based instruments, such targets can be for example the sun or the cold
sky. For the latter, either the whole satellite, the antenna of the radiometer or a reflective
mirror are oriented towards a region in space which is known to have few natural radiation
sources. Hence, measuring the previously mentioned, very accurately known background
radiation of approximately 2.7K [23]. This, however, necessitates orbital maneuvers of the
satellite or moving mechanical parts at the instrument and is critical in terms of time con-
sumption, mechanical wear, and reliability. For ground-based instruments, another external
calibration routine can be performed: During the tipping-curve calibration the radiometer
points towards the atmosphere at different zenith angles, measuring at frequencies where
the opacity of the atmosphere is low. This method relies on the known relationship between
the tipping angles of the radiometer and the atmospheric opacity at those angles [24], [25].
The previously described calibration methods with external targets are accurate and well
established, but are limiting the speed of calibration and reducing system sensitivity because
the calibration time goes at the expense of integration time (while observing the target).
Additional complex mechanics are necessary and are increasing mass and size of the radiome-
ter system. Orbital maneuvers consume auxiliary propellant. Furthermore, liquid gases for
cooling and additional power for heating the cold and hot calibration targets respectively
need to be held available. These factors are especially a disadvantage for air- or space-based
instruments, where payload and power consumption are critical parameters. This is why
in recent years, the feasibility of internal calibration targets for future satellite missions is
investigated [26]. In addition to the associated cost savings and the reduction of vulnerable
mechanic components, they promise performance improvements because internal calibration
can typically be performed faster and more frequently. This increases observation time and
reduces rapid gain fluctuations. Furthermore, internal targets can be better shielded against
influence from external radiation. Possible internal references can for example be the pre-
viously mentioned passive matched load or a noise diode, which is described in the next
paragraph.
2.2.3 Noise Diodes
A noise diode is an active internal calibration target, providing a hot calibration point if the
level of excess noise is known. In general avalanche diodes are used, which are small, reliable
and relatively stable. However, their performance is frequency dependent and while they
exhibit an excess noise ratio (ENR) of 20−35 dB at frequencies below 100GHz, their output
noise level degrades to below 10 dB for higher frequencies. Nevertheless, their output power
level typically exceeds the dynamic range of the highly sensitive radiometers. This leads to
the mandatory use of additional attenuators or intentionally lossy couplers when routing the
signal to the radiometer. Furthermore, since the power level of the reference noise diode is
higher than the one of the target by several orders of magnitudes, this approach necessarily
leads to high extrapolation errors. In any case, it is preferable to have reference targets with
7
noise temperatures, that encompass the expected dynamic range of the observed targets,
which is explained in the next section.
2.3 Active Electronic Loads
Active Electronic Loads are novel devices, typically consisting of a transistor with surrounding
circuitry, which can also be used as internal calibration references. Hot loads function similar
to noise diodes, generating a high noise level that results in a noise temperature far above
the measurement range. Contrary to noise diodes, however, the output power level is bias
dependent and allows to generate several different warm/hot reference levels. This facilitates
to adapt the reference level to the dynamic range of the radiometer. In this work these hot
loads will be referred to as Active Hot Loads (AHL). Previous thorough characterization of
the load and a high stability are of course mandatory in this application. Cold loads have
the notable property, that they, unlike passive or typical active devices, can exhibit a noise
temperature that is below the physical temperature of the device Tn ≤ Tphys. In this work
they will be referred to as Active Cold Loads (ACL). The theory of such an active cold
noise source was first published in [27], where the resultant circuit was called COLFET. The
principal functionality is described as followed: "The input impedance of the FET (Field
Effect Transistor) is predominantly capacitive (there is a small series resistance) with the
result that an inductance added to the source circuit is transformed to appear as a noiseless
resistor in the gate circuit. This phenomenon allows the development of FET amplifiers with
low-input VSWR (Voltage Standing Wave Ratio). Another consequence of the impedance
transformation at the input is that only part of the total input resistance contributes noise
to the circuit, that portion due to the source inductance being noiseless. This offers the
possibility that the input impedance of such an amplifier may be resistive with the apparent
temperature of the resistance being much less than ambient and in fact comparable with the
noise of the circuit when used as an amplifier." For the typical two-point calibration this offers
the following advantages: Using an ACL as the cold reference, the ambient load becomes
the hot reference. No absorber or other load needs to be physically heated, which reduces
the power consumption and avoids all problems related to thermal flux. Furthermore, the
calibration references encompass the temperature to be observed, which significantly reduces
the extrapolation error and avoids an excess of the dynamic range. Especially, the ACL offers
significant advantages compared to conventional cold loads such as liquid gas (which is an
expensive additional payload) or cold sky (which necessitates additional mechanics). But
also in the application as a Dicke-load in an unbalanced switching radiometer, the ACL has
notable advantages. Compared to an ambient reference load, the radiometric resolution of
the radiometer improves, if an ACL is used instead - especially for targets with low brightness
temperatures. This can be explained with the help of the following equations [26]. The
radiometric resolution ∆T of an unbalanced Dicke switching radiometer is composed of the
contributions due to system noise ∆TS, gain fluctuations ∆TG, and calibration noise ∆TC .
∆T =
√
∆T 2S +∆T
2
C +∆T
2
G (2.3)
Where ∆TS, in turn, is defined by
8
∆TS =
√√√√2T 2Sys + 2(T ′Ref + T ′Rec)2
Bn · τS
(2.4)
with
T ′Ref = LA ·TRef (2.5)
and
T ′Rec = LA ·TRec (2.6)
Bn is the noise bandwidth and τS the scene integration time. T
′
Ref and T
′
Rec are the noise
temperature of the reference load and the receiver noise temperature, respectively, at the
antenna input reference plane and LA the ohmic losses of the antenna and the antenna
feed. Not surprisingly, the radiometric resolution can be improved with a low system noise
temperature TSys and a low receiver noise TRec. But obviously, also the physical temperature
of the reference load TRef - or in the case of an active electronic load - its noise temperature
play an important role. The lower the reference noise temperature is, the better the system
noise dependent radiometric resolution. The same correlation is observed with
∆TC =
√√√√2T 2Sys + 2(T ′Ref + T ′Rec)2
Bn ·N · τC
(2.7)
with the difference, that the radiometric resolution can be further improved by the number
of calibrationsN , that are included in the calibration averaging. τC the calibration integration
time. For unbalanced Dicke switching radiometers, this number can be greatly increased due
to the improved stability compared to a total power radiometer.
∆TG = |TA − T
′
Ref |·
(
∆G
G
)
(2.8)
Finally, Equation 2.8 shows, that the radiometric resolution is best, if the reference tem-
perature at antenna input is close to the antenna temperature TA (which is the noise tem-
perature at antenna input while the target is observed). In this case the influence of the
gain fluctuation ∆G will be minimal. This means that for typical observation scenes and
especially for scenes with a low brightness temperatures, cold references are advantageous
for the radiometric resolution. But also for the radiometric stability the temperature differ-
ence between observed target and the reference is important. Pseudocorrelation radiometers
have been reported to have 1
f
knee-frequencies in the order of few mHz when their internal
reference is stable, and the observed brightness temperature is close to the reference [26].
An overview of published ACLs is presented in Table 2.1. Please note, that the W-band
ACLs in [28] and [29] are pre-published results of this work.
2.4 Front-End Switches
No matter which of the previously described calibration references is used, a front-end switch
is necessary to switch between the reference and the desired input signal of the radiometer.
9
Table 2.1: State-of-the-art of ACLs realized in different technologies and at different frequencies.
Cit., year Technology Frequency [GHz] Noise Temp. [K]
[27], 1981 GaAs MESFET 1.4 50
[30], 1997 InP HEMT 18 105
[31], 2000 GaAs mHEMT
2− 10 90
10-26 125
[32], 2008 GaAs pHEMT 4− 8 100
[33], 2009 GaAs pHEMT 10.69
56
77
[34], 2010 SiGe HBT 1.4 65
[35], 2010 GaAs mHEMT 23.8 72
[28], 2012 GaAs mHEMT 87− 110 155
[29], 2015 GaAs mHEMT 31.4 75
[29], 2015 GaAs mHEMT 52.0 141
[29], 2015 GaAs mHEMT 89.0 170
Even though in laboratory conditions it is possible to disconnect and and connect several
front-ends such as an antenna or different loads to the same radiometer to perform calibration
measurements, this is not practical in a typical operation mode. Depending on the type of
radiometer, a maximum calibration interval needs to be maintained, for example <10 s for
total power radiometers [26]. For switches used in PMMW imaging, especially at video-rate
or in multi-channel systems, even much faster switching is required.
2.4.1 Mechanical Waveguide Switches
The most mature technology for radio frequency (RF) and microwave switches are mechan-
ical or electromechanical waveguide switches. The latter can be controlled by an electrical
signal and both have the best performance in terms of insertion loss and isolation. How-
ever, their major drawback is a limited reliability due to mechanical wear. If thousands of
switching procedures are necessary, degradation and failure are imminent. While acceptable
in laboratory conditions or with accessibility for maintenance or replacement, they are not
used for radiometer calibration on board of satellites or space probes [36]. On such vehicles
they are merely used for one time operation, for example to switch to redundant systems in
case of an instrument defect. Likewise, they are not suited for the application in PMMW
imaging systems, due to the limited switching speed.
2.4.2 MEMS Switches
An emerging technology are Microelectromechanical System (MEMS) switches, which are
superior to the established electromechanical switches in terms of size, weight, cost, and
power consumption. Due to the physical separation or contacting of the conducting struc-
ture, the performance of MEMS switches is remarkable in terms of isolation and insertion
loss, respectively. However, their performance is typically demonstrated in a stand-alone
configuration i.e. as a Single-Pole Single-Throw (SPST) switch with simple on/off capabil-
ity. Applications in a SPDT, SP3T, etc. which are based on a topology with impedance
10
transforming lossy transmission lines to a common star point are rare. In this respect, a
performance comparison to other switch types is often biased. Furthermore, the technology
is also not yet considered to be mature enough for reliable implementation in critical systems
such as satellite instruments.
2.4.3 FET Switches
"Solid state switches are more reliable and exhibit a longer lifetime than their electromechan-
ical counterparts due to their superior resistance to shock, vibration and mechanical wear.
They also offer a faster switching time. However, the higher innate ON resistance of the
of the solid state switches gives them higher insertion loss than electromechanical switches.
Therefore solid state switches are preferred in systems where fast switching and long lifetime
are essential." [37] There are two main types of solid state switches: PIN diode (a semicon-
ductor with a p-type, an intrinsic, and a n-type region) switches and FET switches, as well as
combination of both (hybrid). While PIN diodes are current-controlled, the conductivity of
the channel in the FET is depending on the electric field which is voltage-controlled, hence,
FET switches having the lower current consumption. On the other hand, PIN diode switches
achieve even faster switching speeds and exhibit a lower insertion loss than FET switches.
However, the focus of this work will be on FET switches because it is possible to implement
them in the same technology as the presented active electronic loads and to combine them
to an integrated switching calibration front-end (SCFE).
2.5 Overview of Calibration Concepts
An exemplary overview of past, on-going and future air- or space-based vehicles, their on-
board instruments and its calibration method are summarized in Table 2.2. For most past
and on-going satellite missions, conventional external references such as the cold sky and an
external black body are still widely spread and well established. Some spaceborn microwave
remote sensing radiometers replace the external hot reference by an internal matched load
at ambient temperature, still relying on the cold sky as a cold reference. In recent satellite
radiometers, noise diodes have been applied as a hot reference, allowing to use the internal
matched load as a cold reference. By that, the radiometer does not rely on external targets
for the calibration of the instrument itself (i.e. instrumental drift or gain and noise variations).
However, it should be noted that these internal calibration procedures do not include the
influence of the antenna and feeding networks. To include these factors into the calibration
loop, a complementary calibration with external references is necessary. Since these other
factors, however, do not fluctuate as fast as the instrument, much longer calibration intervals
are possible. For airborne or ground-based radiometers, the requirements for the calibration
procedures might be less strict, depending on the application. Typical data acquisition
flights of unmanned aerial vehicles (UAVs) or aircrafts are very limited in time, compared to
space flights. And ground based instruments do normally not have a limitation of payload
or power consumption, thus even more extensive procedures as cooling or heating artificial
black bodies or other loads are possible. In conclusion, the appropriate calibration concept
needs to be chosen with regard to the planned mission profile and often a combination of
different references is chosen. The pixel normalization as it is necessary for PMMW imaging
11
was not taken into account in this overview, because typically this application requires no
absolute calibration.
2.6 Low-Noise Amplifiers
Low-noise amplifiers (LNAs) are elementary components of every radiometer system. To
amplify the very weak signals, relevant in radiometry and radio astronomy, a high gain is
as important as the low-noise performance. Equally, LNAs are most important for noise
measurement systems, as will be seen in Chapter 4. Since this work does not focus on the
design of LNAs in general, a state of the art overview is omitted at this point. However,
a custom-designed broadband D-band LNA was part of this work, as it was necessary for
the characterization of the D-band ACLs and no commercially available LNAs did satisfy
the necessary requirements. Therefore, a comparison of the developped D-band LNA to the
state of the art will be presented in Chapter 6.
Summary and Conclusion
After a short introduction into radiometry, conventional calibration references and active
electronic loads were introduced. Different front-end switch types were evaluated and an
overview of common calibration concepts, used in scientific instruments, was given. Up to
now, loads and switches were separate components in internal calibration systems.
The following chapter will now focus on the design of Millimeter-Wave Monolithic Integrated
Circuits (MMICc), that will allow a novel calibration concept in the future. The MMICs
will incorporate the front-end switch together with active loads into a complete internal
calibration front-end.
12
Table 2.2: Vehicle, radiometer instruments and applied calibration references.
Cit. Vehicle Instrument Cold Ref. Hot Ref.
Earth Observation
[38]
SMOS
CAS / MIRAS Cold Sky Noise Diode
[39] NIR
[40] Aqua HSB Cold Sky External
Black Body
[41] OceanSat-1 MSMR Cold Sky External
Black Body
[42] SeaSat SMMR Cold Sky Matched
Load
[43] Topex / TMR Cold Sky Matched
Poseidon Load
[44] Jason-1 JMR Mached Noise Diode
Load
Space Probes
[45] Juno MWR Matched Load Noise Diode
Airborne
[46] Global Hawk HAMSR External External
Black Body Black Body
[47] NASA PSR External External
Aircraft Black Body Black Body
& Noise Diode
Ground Based
[48] - AWVR Matched Load Noise Diode
13

3 Active Load Integrated Circuit
Design
For the design of active electronic loads, several circuit properties need to be taken into
account. Beside the main goal to achieve the lowest possible noise temperature, further
emphasis must be placed on circuit stability, sensitivity, reliability and the possibility to
integrate the loads with other components such as switches or LNAs. These factors rely on
a trade-off in the choice of technology and circuit topology, available simulation models, and
of course careful choice of circuit design parameters, which will be discussed in this chapter.
3.1 Operating Frequency
The choice of frequencies for the circuits presented in this work is based on two considera-
tions. The atmospheric microwave and millimeter-wave attenuation is frequency dependent
and increases with higher frequency. Furthermore, it has local maximums, that are caused
by the resonance frequencies of the O2 and H2O molecules contained in our atmosphere.
These maximums limit the usable range in the electromagnetic spectrum. Between these
maximums however, so called atmospheric windows allow the penetration of electromagnetic
waves through the atmosphere. Such windows are located around the frequencies of 90GHz
and 140GHz and are highlighted in blue in Figure 3.1. Therefore, the W-band (75−110GHz)
and the D-band (110 − 170GHz) are the targeted frequency bands for applications in re-
mote sensing and imaging, where a trade-off between high geometrical resolution (i.e. high
frequency) and low attenuation is desired. Another motivation is the previously mentioned
intended application of ACLs in the calibration of radiometers for PostEPS / MetOp SG
missions. Within the scope of [19], circuits at 31.4GHz, 50 − 54GHz and 89GHz where
developed and manufactured, of which some of the W-band activity will be presented in this
work and compared to alternative approaches.
3.2 Technology
In a rough initial approximation, the lowest achievable noise temperature of a well designed
ACL is limited by the minimum noise temperature of its internal amplifier [26]. Hence,
the choice of a low-noise transistor technology is most important. In a dedicated ACL
technology trade-off study from 2005, several transistor technologies have been evaluated,
based on existing literature and ACL publications [49]. In this study, Indium Phosphide (InP)
High Electron Mobility Transistors (HEMTs) or InP pseudomorphic HEMTs (pHEMTs) were
still considered too immature and too exotic for the robustness needs of an ACL used as a
calibration load. However, availability and technological maturity of some of the investigated
manufacturing processes have significantly changed since then. Also, the targeted frequency
15
Figure 3.1: Atmospheric attenuation.
range (f ≤ 10GHz) of the mentioned study differs from the circuits presented in this work.
Table 3.1 summarizes the considerations of the report with regard to recent semiconductor
technology developments and the frequency requirements of this work. FET technologies
unlike Bipolar Junction Transistors (BJTs) and Heterojunction Bipolar Transistors (HBTs),
suffer from 1
f
-noise due to surface and interface (trapping) effects of the semiconductor.
However, [49] states, that 1
f
-noise is not the relevant parameter for ACLs. Furthermore,
especially at higher frequencies, the benefit from high electron mobility and velocity favors
the use of HEMT technology. This is supported by the fact that all ACLs above 2GHz listed
in Table 2.1 are manufactured either in HEMT, pHEMT or metamorphic HEMT (mHEMT)
technology. A more recent study from 2010 also compares the simulation results of ACLs
with different types of transistors [50]. In the investigated frequency bands around 23.8GHz
and 36.5GHz it clearly comes to the conclusion, that - compared to Silicon Germanium
(SiGe) HBTs and pHEMTs - the lowest noise temperature can be achieved with mHEMT
transistors. It should be noted, that InP HEMTs were not investigated in this study. While
the limitation of SiGe HBT technology already becomes evident around 23GHz, where no
noise temperature below 210K can be achieved, the report states, that to achieve noise
temperatures below 180K, mHEMT technology is believed to be suitable even at frequencies
much higher than the investigated 36.5GHz.
3.2.1 MMIC Technology
While most of the previously published loads had to fall back on discrete packaged transistors
in a hybrid design with external feed- and matching networks, this work benefits from the
advancement in MMIC technology. This allows not only a very compact design of the active
loads, especially at higher frequencies, but also the integration with a switch for an Active
Calibration Sub-System (ACSS) or even with the radiometer front-end. Furthermore, MMIC
technology facilitates mass production with repeatable performance and can be qualified for
space application [50].
16
Table 3.1: Transistor technology trade-off in regard of suitability for ACL design.
Technology Characteristics
BJT + current flow perpendicular to semiconductor
surface and therefore considerable lower 1f -noise
than FETs
− high noise figure
HBT − in general III/V HBTs have inferior noise
(InP, GaAs, performance at high frequencies than III/V FETs
SiGe) + InP HBTs are suitable for ultra-low power
applications
+ GaAs HBTs have better noise performance than
InP HBTs
− SiGe HBTs have comparable noise performance to
GaAs MESFETs and GaAs HEMTS but only at low
frequencies
GaAs MESFET − compared to other recent FET technologies the
highest minimum noise figure
GaAs HEMT + 2-DEG creates high concentration and mobility of
electrons thus, higher maximum operating frequency
and lower noise figure than GaAs FETs
GaAs pHEMT + better noise figure than GaAs HEMT
+ better frequency performance than GaAs HEMT
GaAs mHEMT + even higher doping levels possible than in
pure InP HEMTs
+ equivalent performance as InP HEMTs
+ lower cost than InP and larger wafer size
InP HEMT + best III/V FETs for microwave applications
in terms of operating frequency and noise figure
− fragile substrate, structures like vias are difficult
to manufacture
− lower process yield, limited wafer size
17
Prediction of Achievable Noise Temperature
To estimate the achievable ACL noise temperature in W-band with the above mentioned
rough approximation, some published MMIC LNAs with their respective manufacturing tech-
nology shall be shortly reviewed. [51] and [52], report on InP-based HEMT amplifiers with
noise figures of 2.7 dB (100 nm gate) and 2.5 dB (50 nm gate) at 94GHz, respectively. In
[53], an antimonide based compound semiconductor (ABCS) amplifier with InAs channel is
presented, exhibiting a noise figure of 2.5 dB at 90GHz. The conversion between noise figure
in dB (NF (dB)) to noise temperature (Tn) is performed by applying Equation 3.1, where
T0 is per definition always 290K.
Tn = T0·
(
10
NF (dB)
10 − 1
)
(3.1)
The achieved NF (dB) in the cited work of 2.5 dB thus corresponds to a Tn of 226K.
In [54], an InP HEMT LNA with 35 nm gate length is presented, which even achieves a
measured Tn in a waveguide package of 200K. With IAF’s 50 nm mHEMT technology
which is presented in the next subsection, an LNA with Tn of 160K was published in [55].
This confirms the suitability of HEMT based MMIC technology for the design of low noise
circuits and gives an estimation of the expected achievable Tn for ACLs. Further evidence is
presented in [26], where noise temperatures of 160K and 180K on MMIC level are predicted
at 89GHz and 100GHz , respectively. This report also states, that MMIC-based ACLs up
to 101GHz are realistic for the use in Post-EPS context. However, it should be pointed out
that these estimations, based on published LNA performances, are only rough approximations.
The MMIC matching- and feed networks of LNAs contribute to the noise of the amplifier
and are different in topology than for active loads. The main difference is, that an LNA
is an amplifying two port device, typically consisting of multiple transistor stages, where
the matching at the output port has very little influence on the noise generated in the first
transistor stage. Whereas active loads are typically one-stage, one-port devices where the
noise temperature on gate side of the transistor is of interest and very sensitive to reflections
on the drain side. A detailed discussion of the simulated noise temperature of active loads
and the influence of matching networks is presented in Section 3.3.
3.2.2 IAF’s mHEMT Process
For the design of the presented MMICs in this work, IAF’s in-house mHEMT technology
was used. Indium Gallium Arsenide/Indium Aluminum Arsenide heterostructures (InGaAs/I-
nAlAs) with high In-content in the electron transport channel are grown on 4 inch semi-
isolating GaAs wafers by molecular beam epitaxy (MBE). This approach allows to process
highly integrated ultra low noise circuits on cost-efficient GaAs wafers. To adapt the lattice
constants of the heterostructure and the substrate, a metamorphic buffer layer is grown in be-
tween. Parasitic substrate modes are supressed, by applying a backside process to the wafer,
where it is thinned down to a final thickness of 50µm. Subsequently, through-substrate vias
with 20µm contact diameter to the front side are processed by dry etching and finally, the
wafer backside is gold plated.
18
Passive Components
In addition to the active devices, metal-insulator-metal (MIM) capacitors, NiCr thin film
resistors and airbridge technology are provided for circuit layout, as well as grounded coplanar
waveguides (GCPW) as transmission lines. With these passive components, input matching
networks (IMN) and output matching networks (OMN) can be realized, as well as DC bias
and RF blocking structures. IAF’s mHEMT process features gate lengths from 20 nm to
100 nm, which differ in their technological properties. Details can be found in [55] (100 nm
and 50 nm), [56] (50 nm and 35 nm), [57] (35 nm) and [58] (20 nm). Different gate-lengths
were chosen for the active loads, switches and LNA and will be stated in the respective
section.
Small-Signal Model
To chose the ideal transistor size and bias point for optimum noise performance (and gain
in the case of LNA), a versatile small-signal model of the used transistors was available
[59]. It consists of an intrinsic part and a parasitic shell, is bias dependent, and features the
prediction of noise parameters. It is parametrized, so that different gate widths and number
of gate fingers can be simulated. The characterization of active and passive test structures,
as well as the modeling itself is done in-house. Hence, an extensive IAF proprietary model
library was available for designing and optimizing the MMICs, using the computer aided
design (CAD) software Advanced Design System (ADS) by Keysight Technologies.
3.2.3 Long-Term Stability
If ACLs are intended to be used as calibration references in radiometer systems, especially
on board of satellites, not only the short-term, but also the long-term stability of the noise
temperature is most crucial. Similar to the gain stability of (low-noise) amplifiers in radiome-
ters, the noise stability of active electronic loads is dependent on the utilized manufacturing
technology of the device. For existing ACLs, few investigations have been published, proving
the feasibility of active loads in different technologies.
In [60], [61], and [62], the short-term and intermediate-term stability of a SiGe HBT based
ACL at 1.4GHz was investigated. Over a duration of one week, the noise temperature varied
less than 40mK. During a four month period of continuous operation the maximum variation
of the ACL is reported to be 0.35K and assuming a linear drift versus time, a value of less
than 1K/year is extrapolated. In [63] the results of the SiGe HBT ACL were compared to a
GaAs pHEMT ACL at the same frequency, showing a similar performance with a variation
of 0.26K during two and a half months. [64] investigated the stability of a 18GHz noise
source, which is referred to as a variable temperature source (VTS). Presumably, this is the
InP HEMT based ACL presented in [30]. On a short-term period, they report a slope of
0.05K/day around a mean noise temperature of 127.2K and a measured long-term increase
of the noise temperature of 2K/year around a mean noise temperature of 126.1K. In [65] the
noise temperature of a GaAs pHEMT ACL at 10.69GHz was monitored in a temperature
stabilized test-bed over a time period of one year. Around a mean noise temperature of
77K, a drift rate in the order of 1.5K/year during the first eight months of operation
were observed and around 2K/year over the entire measurement period. The only studies
19
found, analyzing the noise stability of a GaAs mHEMT ACL were [35] and [50]. However,
the investigated 23.8GHz ACL was also at a much higher operating frequency than any of
the listed above, that were investigated over a longer time period in a stabilized test-bed.
[50] reports a variation of the noise temperature of less than 0.5K around 72K during one
month. In conclusion, the utilized GaAs mHEMT technology for the circuits developed in
this work is believed to satisfy the necessary requirements for the stability of the active loads.
Especially, since all of the previously published ACLs above 20GHz were manufactured in
this technology.
3.3 Stand-alone Active Loads
Even though a limited set of active cold loads was previously published (see Table 2.1),
there exist no standard guidelines for the design of active loads. The primary design goals
are evidently to obtain the lowest possible noise temperature for cold loads and a high noise
temperature for hot loads at the output of the load, while achieving a very high return loss
in the frequency range of interest. In this work several active hot and cold loads, as well as
integrated hot-cold loads were developed, using different simulation and design approaches,
which will be presented in the following.
3.3.1 ACL Topology
As explained in Section 2.3, the principle functionality of an ACL is based on an impedance
transformation of the capacitive input impedance of the FET with an inductive source, so that
the resistive portion of the source inductance appears noiseless and the total input resistance
appears to generate less noise. This is a similar approach as for the design of LNAs and as
generally known, the challenge in LNA design is to achieve a simultaneous noise and power
matching at the input. The common methods to modify the transistor’s input impedance
without modifying its optimum source reflection coefficient for low-noise are either feedback
circuits or a deliberate output mismatch. The latter, however, is not an option for ACL
design: Any mismatch at the output would cause to reflect the generated noise on drain side
back thru the transistor to the input and increase the ACL’s noise temperature. Therefore,
as will be explained in the next section, some of the simplified noise temperature simulation
approaches even assume a perfect match on drain side and/or a minimum backward gain of
the transistor.
As stated in [50] and [66], two different topologies for the design of active loads are possible,
to achieve the above mentioned design goals: an inductive or a parallel feedback, as illustrated
in Figure 3.2. The latter, however, would not be suitable in a hybrid circuit, because to
achieve a specific transconductance value for optimum input matching, the gate dimensions
of the transistor have to be varied. In all the presented circuits in this work, exactly that
was investigated, to achieve optimum performance. Nevertheless, the inductive feedback
topology was chosen, just as in most previously published work, because the possible variation
of the feedback inductance gives an additional degree of freedom in the design of the circuits.
This simplifies to simultaneously achieve the different design goals.
20
(a) Inductive feedback (b) Parallel feedback
Figure 3.2: Different topologies for the active load design as described in [50].
3.3.2 Simulation of Noise Temperature
For the description of the circuits and the simulation process, standard terminology of scat-
tering parameters (S-parameters), gain G, and reflection coefficients Γ, as described in [67],
[68], and [69] will be used. The noise characteristics of noisy two-port are described by its
noise parameters, the noise resistance Rn, the minimum noise figure Fmin and the real and
imaginary value of the source impedance for minimum noise, which typically is described
by the source reflection coefficient Γopt [70]. With the available small-signal model of the
transistor, ADS calculates the network parameter matrices S, Z, and Y , as well as the noise
parameters Rn, Sopt, NF , NFmin, Te, and the noise current correlation matrix Icor. The lat-
ter is particularly important for the novel approach in this work and will be further explained
in Section 3.3.4. With these parameters available, different approaches are possible to design
active electronic loads with the desired noise level and the appropriate matching networks,
yielding optimum return loss at the output port. Three different simulation methods were
used to design the circuits presented in the following sections:
• The general approach of Weatherspoon and Dunleavy which is described in [71], is
valid for an arbitrary non-matched output and was utilized to design the active loads in
the framework of the ESA project Calload [19]. By subsequently designing output- and
input matching networks , the noise temperature calculation can be further simplified.
The practical design flow and the discrepancy in the simulation results is presented in
this work on the basis of the 89GHz ACLs for the Active Calibration Sub-System.
• The simplified approach by Wait and Engen [72] assumes perfect match on drain side
of the transistor and was utilized to design the ACL in the integrated hot-cold noise
source published in [73]. By similarly designing an input network for optimum noise
matching on gate side, this approach results in a noise temperature simulation very
similar to the above.
• A novel, further simplified numerical method, derived from standard noise power cal-
culation, using noise-model parameters and ADS optimization algorithms to design
AHLs and ACLs in W-band [28], [74], and ACLs in D-band.
Since the nomenclature of the previously introduced noise parameters used by ADS and
common literature are somewhat different, Table 3.2 compares the equivalent denotations.
21
Table 3.2: Denotation of noise related parameters in literature and in ADS.
Literature ADS parameters and functions
Γopt Sopt
Gopt real
(
1
Z0
·
Sopt−1
−1−Sopt
)
Te,min T0·
(
10
NFmin
10 − 1
)
In the following description of the noise calculations throughout this work, the standard
nomenclature which is common in literature will be used. Further ADS specific parameters
like Icor, which will be necessary for the explanation of the novel numerical simulation method,
will be introduced in the accordant sections.
Figure 3.3: The transistor described by its S-parameters in a standard two-port noise representation with
the accordant reflection coefficients and the external terminals.
In general, an active load is a one-port device which provides the wanted noise temperature,
depending on its internal circuitry. The internal circuitry consists of an active device, typically
a transistor, some kind of In Figure 3.3 the core element of the active load, the transistor, is
described by its S-parameters and connected to external terminals. Since this representation
is based on the design of low-noise amplifiers, the orientation of the source on the left side
(TS) and the load on the right side (TL) will be maintained throughout this work. Likewise,
matching networks on the source side will be labeled as input matching networks (IMN) and
on the load side output matching networks (OMN), as shown in Figure 3.4. However, in the
case of active loads, the relevant terminal for the calculation of the noise temperature is TS.
Without any matching networks, the generated noise temperature Ts,1 depends only on the
transistor parameters and the reflection coefficients and can be calculated by
Ts1 = Tb +
((
T1
(
1− |Γ1|
2
)
+ Ta
)
G21|Γ2|
2 + T2
(
1− |Γ2|
2
))
G12 (3.2)
where Tk, Ta and Tb are alternative noise parameters as introduced in [71]:
Tk = 4·T0·Rn ·Gopt (3.3)
Ta = Te,min +
Tk · |Γ
′
opt|
2
1− |Γ′opt|2
(3.4)
Tb =
Tk
1− |Γ′opt|2
− Te,min (3.5)
and Γ′opt is given by
22
Γ′opt =
S∗11 − Γopt
ΓoptS11 − 1
(3.6)
T1 and T2 are the noise temperatures at the terminals TS and TL, respectively. The
available power gains G12 and G21 of the transistor are per definition [75] also dependent
on the matching between itself and the external terminals. They can be expressed by the
S-parameters of the transistor and the reflection coefficients:
G12 =
|S12|
2
| (1− S22Γ2) |2· (1− |Γin|2)
(3.7)
G21 =
|S21|
2
| (1− S11Γ1) |2· (1− |Γout|2)
(3.8)
Figure 3.4: Design schematic of the transistor with appropriate matching networks to the external terminals.
So far, the noise calculation is valid for a general not necessarily matched circuit. Obviously,
appropriate matching networks will reduce the noise temperature. An estimation of the
lowest achievable temperature with a theoretically perfect matching is presented later on.
Moreover, assuming proper matching networks the above introduced available power gains
can be described independent of the external terminals. An accordant schematic is depicted
in Figure 3.4. Parameters referring to this terminal invariant gain will have the subscript ti
in the following. With Γ1 = 0 and Γ2 = 0, equations 3.7 and 3.8 can be written as:
G12,ti|Γ2=0 =
|S12|
2
1− |Γin|2
(3.9)
G21,ti|Γ1=0 =
|S21|
2
1− |Γout|2
(3.10)
Simultaneously, the general definitions of Γin and Γout
Γin = S11 +
S21S12Γ2
1− S22Γ2
(3.11)
Γout = S22 +
S12S21Γ1
1− S11Γ1
(3.12)
being simplified to
23
Γin|Γ2=0 = S11 (3.13)
Γout|Γ1=0 = S22 (3.14)
Now, Equations 3.9 and 3.10 can be written as
G12,ti|Γ2=0 =
|S12|
2
1− |S11|2
(3.15)
G21,ti|Γ1=0 =
|S21|
2
1− |S22|2
(3.16)
modifying 3.2 to
Ts1,ti = Tb +
((
T1
(
1− |Γ1|
2
)
+ Ta
)
G21,ti|Γ2|
2 + T2
(
1− |Γ2|
2
))
G12,ti (3.17)
Based on Equation 3.17, the theoretical lower limit of the noise temperature can be
estimated, which can be achieved with a certain transistor technology. Since for a properly
designed amplifier, the backward gain G12 can be assumed to be minimal, the estimation
Ts1,ti ≥ Tb is valid. Furthermore, a good matching network means that |Γ
′
opt|
2 ≤ 1, so that
in Equation 3.5 the estimation Tb ≥ Tk − Te,min is allowed. This leads to the concluding
estimation
Ts1,ti ≥ Tb ≥ Tk − Te,min (3.18)
The theoretical lower limit is now only dependent on the noise parameters of the transistor
and can therefore be calculated as
Ts1,limit = 4·T0·Rn ·Gopt − Te,min (3.19)
To recapitulate, this theoretical lower limit of the achievable noise level is what the
transistor gate exhibits if it would be simultaneously perfectly noise matched, input power
matched, and load matched to its external terminals, which is of course not possible. Even
physical networks that would perfectly provide the necessary impedance would add noise
depending on their physical temperature, due to their own resistive losses.
If the assumption Γ1 = 0 and Γ2 = 0 is not only applied to the available gains, but also to
the noise temperature calculation 3.17, a further simplified expression of 3.2 can be written
as
Ts1|Γ1,2=0 = Tb + T2·G12,ti (3.20)
This equation was used to simulate the noise temperatures for the 89GHz ACLs within
the Calload project [19]. It was found to be a reasonable compromise, because it accounts
for the not negligible backward gain of the transistor and the not perfect noise match at the
transistor gate, but assumes negligible reflections at the external ports due to the matching
networks. While the practical design flow for these matching circuits is presented in the
24
following section on the basis of the 89GHz ACL, the discrepancy of noise temperatures
calculated with the different simulation methods shall be presented beforehand. Figure 3.5
shows the different simulated noise temperatures of a 89GHz ACL, consisting of a single
transistor with inductive source feedback and input and output matching networks. Obviously,
the ACL was optimized to exhibit its lowest noise temperature in the vicinity of 89GHz. The
noise parameter Te,min and Ts1,limit, calculated using Equation 3.19, represent the upper
and lower limit of the achievable noise temperature and enclose the other three calculation
methods over a wide frequency range (Figure 3.5(a)). In Figure 3.5(b) the relevant frequency
range is enlarged. While there is a difference of around 100K between the upper and lower
limit, Ts1, Ts1,ti, and Ts1|Γ1,2=0 only differ by around 10K at 89GHz. Three conclusions are
worth emphasizing:
• The curves confirm, that it is a valid approximation and a good compromise to use
Ts1|Γ1,2=0 for the simulation of noise temperature, if an error of around 5% is acceptable.
• Despite the matching networks, the backward gain of the transistor is not entirely
independent of the external terminals.
• Even for properly designed matching networks the reflection coefficients at the external
terminals have an influence on the generated noise temperature at the transistor gate.
3.3.3 Practical Design Flow
Following the design flowchart in Figure 3.6, the S-parameter matrix and Γopt of the bare
transistor device with inductive source feedback are determined. With this information the
optimum load reflection coefficient ΓL can be calculated using Equation 3.21. The value
of ΓL strongly depends on the transistor dimension (number of gate fingers, gate length)
and the length of the inductive source feedback line. While in general smaller transistors
yield a lower noise temperature, they complicate the realization of the matching networks,
because the optimum reflection coefficient moves closer to the edge of the Smith chart. This
introduces higher losses for the matching network and increases the final noise temperature.
Depending on these parameters a trade-off has to be found between minimum achievable
noise temperature, realization of the matching network and stability of the device. The
stability of the circuit is determined using the µ-factor as defined in [76] and is taken into
account for the investigation of optimal transistor dimension and source line length.
ΓL =
Γ∗opt − S11
S12S21 + S22
(
Γ∗opt − S11
) (3.21)
In the following, ΓL, the noise temperature and the stability of the device is analyzed
at 89GHz for different transistor dimensions and different source line lengths. With regard
to the optimum load reflection coefficient at this frequency, all chosen devices facilitate
designing an appropriate output matching network. Table 3.3 summarizes the investigated
transistor dimensions and the simulation results.
While the 4-finger transistors demonstrate lowest noise temperature with no source line
at all, they demonstrate strongly increasing noise temperature with increasing source line
length. This complicates the stabilization of the device which requires some inductive source
25
0 20 40 60 80 100 120 140 160 180 200
100
150
200
250
300
350
400
450
500
 Te,min
 Ts1,limit
 Ts1,|
 Ts1
 Ts1,ti
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
(a) Simulation results for 89GHz ACL.
80 85 90 95 100
120
140
160
180
200
220
240
260
Ts1
Ts1,tiTs1,|
Ts1,limit
Te,min
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
(b) Noise temperatures in the relevant frequency range.
Figure 3.5: Comparison of differently simulated noise temperatures.
Figure 3.6: Design flow as described in [71].
26
Table 3.3: Best achievable results at 89GHz for different transistor dimensions at different source line
lengths.
Transistor Minimum |ΓL| Minimum Ts1|Γ1,2=0 Maximum µ-factor
size
2× 10µm 0.268 @ 75µm 89.413K @ 45µm 0.997 @ 100µm
2× 15µm 0.087 @ 35µm 97.308K @ 30µm 1.013 @ 75µm
2× 20µm 0.096 @ 30µm 107.014K @ 23µm 1.036 @ 45µm
2× 30µm 0.066 @ 30µm 128.797K @ 18µm 1.108 @ 20µm
4× 10µm 0.186 @ 7µm 101.895K @ 0µm 0.988 @ 45µm
4× 15µm 0.147 @ 5µm 118.626K @ 0µm 0.991 @ 40µm
4× 20µm 0.107 @ 10µm 136.519K @ 0µm 0.999 @ 25µm
4× 30µm 0.064 @ 27µm 171.031K @ 0µm 1.023 @ 17µm
0 20 40 60 80 100 120 140 160 180 200
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Source Line Length [ m]
L
(a) Magnitude of ΓL.
0 20 40 60 80 100 120 140 160 180 200
60
80
100
120
140
160
180
200
Source Line Length [ m]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Te,min
Ts1|
(b) Noise temperature.
Figure 3.7: 2× 15µm transistor parameters vs. source line length.
0 20 40 60 80 100 120 140 160 180 200
0.80
0.85
0.90
0.95
1.00
1.05
1.10
Source Line Length [ m]
-F
ac
to
rs
'
(a) µ-factors.
0 20 40 60 80 100 120 140 160 180 200
-30
-25
-20
-15
-10
-5
0
5
10
15
20
Frequency [GHz]
S
 - 
P
ar
am
et
er
 [d
B
]
S12
S21
S11
S22
(b) S-parameters of single transistor.
Figure 3.8: 2× 15µm transistor parameters vs. source line length and frequency.
27
feedback. For some of the 2-finger transistors a compromise in source line length can be
found to achieve an acceptable trade-off in noise temperature and stability. The transistor in
the second line of Table 3.3 is chosen for further analysis. While it has the highest µ-factor
at a source line length of 75µm (see Figure 3.8(a)), it still demonstrates a noise temperature
of below 110K at this source line length (see Figure 3.7(b)). Figure 3.8(b) demonstrates
the S-parameters of this device versus frequency. The S11 and S22 parameters indicate no
instabilities (see Figure 3.8(b). Figure 3.9(a) shows the µ-factors which are above unity at
the desired frequency range.
0 20 40 60 80 100 120 140 160 180 200
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Frequency [GHz]
-F
ac
to
rs
'
(a) µ-factors vs. frequency.
0 20 40 60 80 100 120 140 160 180 200
0
50
100
150
200
250
300
350
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Ts1|
Te,min
(b) Increased noise temperature.
Figure 3.9: Transistor parameters with a 75µm 50Ω source line.
The increase of the inductive source feedback leads to a slightly increased noise tempera-
ture, which can be seen in Figure 3.9(b). This noise temperature will further increase when
the input matching network is implemented. Figure 3.10(a) demonstrates the optimum load
reflection coefficient for the output matching network to obtain minimum noise temperature
at the input. With this requirement, an output matching network was designed, following
the design-flow in Figure 3.6. The output of the network is connected to a 50Ω termina-
tion and the OMN designed to provide the required ΓL on transistor side. Figure 3.10(b)
compares the required ΓL to the actual reflection coefficient of the OMN. Subsequently the
input matching network was designed under the conditions for a simultaneous optimum noise
match and input power match. The match of IMN to the 50Ω terminal at the input can
be seen in Figure 3.11(a), while the match at transistor side to the required Γopt can be
seen in Figure 3.11(b). Finally input and output matching networks are connected to the
transistor with feedback source line and the 50Ω termination at the output is replaced by an
ADS simulation port to calculate all S-parameters. The result can be seen in Figure 3.12(a).
The crucial parameter is the input return loss of the ACL, which is shown in detail in Figure
3.12(b) and achieves below −40 dB.
3.3.4 Novel Approach for Improved Design Flow
As the description of the practical design flow in the previous section has shown, the noise
temperature of the complete ACL circuit can not be directly simulated in ADS, using the
available transistor and transmission line models, but needs to be simulated in an iterative
way. First, [S] and Γopt need to be optimized by varying the feedback inductance. This
yields the necessary ΓL to design an OMN in the second step. Then, in the third step
28
0 20 40 60 80 100 120 140 160 180 200
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Frequency [GHz]
L
(a) Required ΓL.
 achieved L 
 required L
 
85 GHz ... 94 GHz
(b) Achieved ΓL of OMN.
Figure 3.10: Required and achieved ΓL .
70 75 80 85 90 95 100 105 110
-55
-50
-45
-40
-35
-30
-25
-20
-15
-10
-5
0
Frequency [GHz]
In
pu
t r
et
ur
n 
lo
ss
 [d
B
]
(a) Towards 50Ω at input
70 75 80 85 90 95 100 105 110
-55
-50
-45
-40
-35
-30
-25
-20
-15
-10
-5
0
Frequency [GHz]
In
pu
t r
et
ur
n 
lo
ss
 [d
B
]
(b) Towards ΓOpt at transistor gate
Figure 3.11: Input return loss of the input matching network.
0 20 40 60 80 100 120 140 160 180 200
-50
-40
-30
-20
-10
0
10
Frequency [GHz]
S
 - 
P
ar
am
et
er
 [d
B
]
S12
S21
S11
S22
(a) S-parameters.
75 80 85 90 95 100 105 110
-50
-45
-40
-35
-30
-25
-20
-15
-10
Frequency [GHz]
In
pu
t r
et
ur
n 
lo
ss
 [d
B
]
(b) Input return loss.
Figure 3.12: S-parameters of transistor with IMN and OMN.
29
the IMN can be designed. What is not shown in the design flow chart in Figure 3.6, is
that after connecting the designed OMN to the drain of the transistor, [S] and Γopt change,
because the required ΓL can not be perfectly achieved with a physical OMN over a large
bandwidth (see Figure 3.10(b)). This means, that before designing the IMN, Γopt needs
to be recalculated. Moreover, two parallel projects need to be handled in ADS, because to
determine the reflection coefficients, so called Gamma-Probes need to be placed between
the stages of interest. But if the signal flow between two simulation terminals is interrupted
by a Gamma-Probe, ADS can not calculate the noise parameters of the transistor anymore.
Therefore, any change of the matching networks in one project needs to be copied to the
other project, which can be a major source of error and makes it impossible to use automated
optimization algorithms.
To overcome this iterative and very time-consuming procedure, and to benefit from the
possibilities of modern numerical simulation methods and optimization algorithms, a novel
approach was developed in this work. Based on the theory of noisy four-poles, as presented
in [70], the Nyquist formula in the form
|itot|2 = 4· kB ·T ·∆f ·Gtot (3.22)
with itot being the equivalent noise current of an otherwise noiseless device, and Gtot the
total noise conductance, is transformed to calculate the noise temperature T :
T =
|itot|2·Rtot
∆f · 4· kB
(3.23)
The resistance Rtot is the reciprocal value of the conductance Gtot. Now, two implemented
functions in ADS are used to compute the generated noise inside the active loads: Icor
and stoz(S). Icor is the noise current correlation matrix, in units of Amperes squared. It
describes the short circuit noise currents squared at each port, and the correlation between
noise currents at different ports. The function stoz(S) transforms a scattering matrix [S] to
an impedance matrix [Z]. Next, the active load as a whole, i.e. with its matching networks
and the inductive feedback, is considered as a noisy two-port device and placed between two
simulation terminals. The circuit is now characterized by its scattering matrix [S] and the
noise currents Icor. Expression 3.24 now allows to directly simulate the noise temperatures
at the external terminals. For the noise temperature only the real part of the resulting matrix
is of relevance and the missing bandwidth ∆f can be omitted, because for spectral noise
simulation the bandwidth is set to 1Hz.
[T ] = Re
(
[stoz(S)]· [Icor]
4· kB
)
(3.24)
Having this expression available, it can be used as a goal for automated optimization
algorithms, which are implemented in ADS. Another simultaneous goal must be the input
return loss. After defining the frequency range and the targeted values of the goals, the
parameters which the algorithm is allowed to vary and the limits within which they can be
varied must be specified.
30
3.3.5 Generic Design Schematic
The described design flow in Section 3.3.3 was the basis for several active load circuits
at different frequencies. Hence, in successive iterations, details of the matching networks
were varied to optimize the performance for a specific frequency range or to balance different
parameters like maximum return loss, minimum noise temperature, and stability against each
other. To avoid multiple slightly different schematics for the circuits presented in following
sections, a general design schematic shall be discussed beforehand.
Figure 3.13: Generic design schematic of an ACL.
Figure 3.13 depicts the transistor as the core element of the active load. After choosing
the number and length of gate fingers, the S-parameters and ΓOpt of the transistor can
be varied by adopting the feedback inductance. This is achieved with at least one shorted
transmission line which is connected to the source of the transistor and labeled as inductive
source feedback in the schematic. To attain symmetry in the layout, the feedback can be
applied at both source connections of the transistor. However, for some of the designs it was
beneficial to use one single transmission line because two parallel transmission lines require
more length to achieve the necessary inductance. All other optional elements that were
not implemented in all designs are marked with blue dotted circles in the schematic. The
variation of line length and line impedance of the series transmission line Lso and the parallel
transmission line Lpo1 yields the necessary ΓL for the output matching. For some designs
it was helpful to include another open stub Lstub to reach this goal. Lpo1, Rstab2, and Lpo2
are in addition used to supply the drain voltage to the transistor and with additional parallel
capacitors to ground, to decouple the voltage supply from the microwave signal. Both series
capacitors serve as a DC-block towards the external terminals. In case of the OMN, the
circuit is terminated with an optional 50Ω load to ground in which case the DC-block is not
necessary. But in case of the ACL for the integrated hot-cold load, this terminal is used as
the hot output terminal, which makes the DC-block necessary. In analogy to the OMN, the
variation of line length and line impedance of the series transmission line Lsi and the parallel
transmission line Lpi1 yields the simultaneous noise and power matching at the input. Again,
the transmission lines Lpi1, Lpi2, and Lsi, and the resistor Rstab1 are at the same time used
to provide the gate voltage to the transistor and the parallel capacitors to ground to decouple
the microwave signals. In the case of the IMN however, the implementation of Rstab1 which
31
is improving the circuit stability, is very critical, because the noise generated in this resistor
at ambient temperature directly adds to the noise of the ACL. Therefore, it was replaced in
most designs by a transmission line. The influence of Rstab2 in the OMN was investigated
as well, but due to the limited backward gain of the transistor, it has no significant impact
on the noise temperature. All unlabeled transmission lines in the schematic have merely
connection purposes.
3.3.6 Realized Circuits
89 GHz ACLs for Active Calibration Sub-System
In Figures 3.14(a) and 3.15, the chip photographs of the two 89GHz ACLs that were realized
in the framework of the ESA project Calload [19] are shown. Both measure a size of 1×1mm2.
It can clearly be seen, that their matching networks differ significantly. While the main focus
of the first version was to achieve lowest possible noise temperature, the second design was
optimized to improve the input return loss, which for the first version turned out to be worse
than simulated (see Chapter 5). Another goal was a more compact layout, facilitating better
integration into the ACSS. Both designs use GCPW transmission lines with different line
impedances, which are recognized by the width of the transmission line. The more narrow
ones are 70Ω lines, while the more wider ones are 50Ω lines. The transistor, the NiCr thin
film resistors in the OMN and the source line inductance are labeled and marked by arrows.
Furthermore, the MIM capacitors are recognized as the square structures that are connected
to the transmission lines or the voltage supply. The smaller square structures that are spread
on the chip are the through-substrate vias. For voltage supply, the IMN and OMN are
connected to DC pads at the top of the chips. They can either be contacted by DC-probes
for on-wafer measurements or gold wires can be bonded to the pads for module assembly.
(a) 89GHz ACL V 1,variant for ACSS V 1. (b) 89GHz ACL V 2, variant for stand-
alone module assembly.
Figure 3.14: Chip photographs of two variants of the first 89GHz ACL design, measuring a size of 1×1mm2,
each.
For the OMN of the 89GHz ACL V 4 in Figure 3.15, the open stub and the middle one
of the parallel capacitors in the DC supply were omitted. The first parallel capacitor was
32
replaced by a MIM capacitor, yielding a higher capacitance while having a more compact
layout. While the lengths of Lpi1 and Lpo1 are predetermined due to their impedance trans-
forming function and can not be significantly reduced, the size of the elements above was
drastically reduced yielding a compact layout. In the IMN of V 4, all three parallel capacitors
were included but reduced capacitance values also allowed a more compact layout.
Figure 3.15: Chip photograph of the re-designed 89GHz ACL for the second wafer run, measuring a size
of 1 × 1mm2. This MMIC will be referred to as 89GHz ACL V 4 in the following and was
integrated into the ACSS V 2.
In Figure 3.14(b), a variant of the ACL V 1 is depicted. It has an identical ACL core,
including identical values of the components relevant for the matching networks, therefore
a very similar performance to ACL V 1 is expected. The reason for this modification was,
to achieve compatibility to a module package that was originally designed for the W-band
ACL and AHL. Hence, an additional stand-alone 89GHz ACL module could be realized.
Therefore, the core of the ACL was shifted towards the center of the MMIC to have the
signal port in the middle of the side edge. On the IMN side the parallel stub just snugly
fits into the available die space. Merely the MIM capacitor needed to be placed next to
the parallel stub line. On OMN side however, a mirroring of the parallel stub towards
to bottom of the die was necessary. There it was bent with a 90 degree bend towards
the vacated space below the inductive source feedback line. A minor adjustment of the
stub line length was made to compensate for the bend structure and to have an identical
electrical length. In the simulation no difference between ACL V 1 and ACL V 2 was observed.
W-band ACL for Active Hot-Cold Load
A further simplification of Equation 3.20 is applied to simulate the noise temperature of the
W-band ACL for the Active Hot-Cold Load (AHCL). It is based on the approach of [72] which
is valid for a perfect match on drain side of the transistor. Accordingly assuming perfect
noise matching on gate side, S∗11 equals Γopt and Γ
′
opt becomes 0 (see Equation 3.6). As was
demonstrated in the previous section, a sophisticated input matching network can achieve
a return loss of below −25 dB (see Figure 3.11(b)) towards the transistor gate, supporting
33
this assumption. In this case, the alternative noise parameter Tb equals the theoretical lower
noise level Ts1,limit of Equation 3.19 and Equation 3.20 can now be written as
Ts1,simp. = Ts1,limit + T2·G12,ti (3.25)
Figure 3.16: Chip photograph of the ACL measuring 0.75× 0.75mm2. The left-side port provides the cold
noise temperature when the right-side port is terminated with 50Ω.
The resulting circuit is depicted in Figure 3.16 and is in its topology very similar to the
89GHz ACLs presented above. The series transmission lines on gate side have an impedance
of 30Ω and the parallel stub which is also used for the DC supply has an impedance of 70Ω.
On the drain side towards the hot output 50Ω impedance transmission lines are used and
the parallel stub is also realized with 70Ω impedance. Further differences of this W-band
ACL to the above are, that it uses two symmetrical source lines to realize the inductive
feedback and that another parallel capacitor in the gate voltage supply line was omitted.
More important, however, is that this ACL is realized as a two-port device. This allows on
one hand to verify the achieved matching of the output port to a 50Ω termination, and
to measure the full S-parameter. On the other hand, to implement this circuit into the
integrated AHCL presented in Section 3.4.3, where the drain side output is used to generate
a hot temperature when an appropriate input impedance is present on the gate side terminal.
It should be noted, that as a stand-alone device this ACL is not designed to exhibit a hot
output temperature on drain side. The full S-parameters allow to investigate the influence
of the backward gain of the ACL, while for the measurement of the cold noise temperature
on gate side, the output is terminated with an RF-probe that is connected to a matched
calibration waveguide load. Since the hot output on the right side of the active load is not
terminated with a constant and matched termination as the circuits in Figure 3.14, but when
integrated in the AHCL will be confronted with different impedances, both, Rstab1 and Rstab2
must be implemented to ensure stability. As previously mentioned, this will slightly increase
the cold output temperature.
W-band Active Loads for Switching Calibration Front-End
Based on the generic ACL schematic depicted in Figure 3.13, the novel approach presented
in Section 3.3.4 was applied to design the active loads for the SCFE. Expression 3.24 was
used to find the optimum design parameters for minimal noise temperature of the circuit.
34
One of these parameters is of course the length of the shorted source line which generates
the feedback inductance. Even the transistor dimension in terms of number and width of
the gate fingers can be included and varied in discrete steps. These parameters influence
Γopt. To simultaneously optimize the OMN and IMN, further parameters to vary are the
transmission line lengths of Lsi, Lso, Lpi1, Lpi2, Lpo1, Lpo2, Lstub as defined in the design
schematic in Figure 3.13, and the values of Rstab1, Rstab2, and of the blocking capacitors.
Theoretically, even the line impedance of every single transmission line could be permuted
between the discrete possible values 30Ω, 50Ω, and 70Ω. The resulting MMIC is shown in
3.17(a).
(a) Active Cold Load. (b) Active Hot Load.
Figure 3.17: Chip photographs of W-band active loads, measuring a size of 1× 1mm2, each.
For the AHL, the approach is very similar. Since the schematic in Figure 3.13 is quite
symmetrical in its general form, merely the orientation of the transistor needs to be reversed.
This results in the AHL topology shown in Figure 3.18. To maintain consistency in the
nomenclature, the IMN is still defined on the left side (drain) and the OMN on the right side
(gate). Now, the same optimization routine can be run, except that the new primary design
goal is to achieve a high noise temperature. A good input return loss of the AHL is of course
equally important as for the ACL. Due to the expected high output noise temperature, the
absolute power of the noise reflections between the AHL output and the subsequent stages
will be higher than for ACLs. However, its influence on the noise measurement accuracy is
also less critical. In Figure 3.17(b) a chip photograph of the AHL MMIC is shown.
Figure 3.18: Topology of the 94GHz AHL.
35
A drawback of this automated optimization approach is, that the parametric simulation
models of capacitors or transmission lines can yield large geometries or line lengths to achieve
the optimum values. The parallel capacitor in the IMN of the AHL for example could have
been easily replaced by a parallel MIM capacitor, yielding the same capacitance with smaller
geometry. Or the conspicuous length of the parallel stub line between the blocking capacitors
in the IMN of the ACL, which should have very little influence on the actual matching of the
circuit. If the optimizer finds a local gradient towards an improvement of the design goal,
it will vary this parameter, even if the absolute influence on the desired noise temperature
or matching is small. To facilitate an integration of ACL and AHL in close vicinity to each
other in the SCFE presented in the next section, the parallel stubs of the matching networks
were bent to run geometrically parallel to the signal line.
3.4 Active Loads with Integrated Switches
With the active loads presented in the previous sections, different circuits are realized that
allow to switch between one or two input ports that can for example be connected to a
receiving antenna, and two or more internal loads. In addition to active loads, two of the
designed circuits also include a passive matched load, just as in a Dicke-radiometer front-end.
3.4.1 Switching Calibration Front-End
H
active
cold
load
active
hot
load
passive
load
V
radiometer
receiving
antennas
SCFE
Figure 3.19: Principle operation mode of SCFE and example of application in a polarimetric radiometer
system, where the SCFE is placed between the radiometer and the receiving antennas.
The principle operation mode of the SCFE is demonstrated in Figure 3.19. A SP5T
switch allows routing the signal from five different sources to a common output port. While
two branches are connected to two input ports (e.g. two orthogonal polarized receiving
antennas), the three remaining branches are connected to active and passive loads within
the same MMIC. The output of the SCFE is connected to the first stage LNA of a radiometer
system.
Figure 3.20 shows a chip photograph of the manufactured SCFE MMIC with the dimen-
sions 2 × 2.5mm2. The SP5T is marked by the white dotted area at the right half of the
MMIC. The transmission lines from the five signal sources and the output port all intersect at
a common star point S , which is marked by an arrow in the photograph. At λ/4 distance of
this point FETs (labeled T1 to T5) are placed in shunt to the signal lines (labeled L1 to L5).
The channel conductance of the transistors can be controlled by the applied gate voltage.
For above-threshold bias conditions, the transistor shortens the signal line to ground and the
36
Figure 3.20: Chip photograph of W-band SCFE MMIC with integrated active loads (2× 2.5mm2).
short circuit is transformed by the λ/4-line to an open circuit at the star point S, isolating
this specific branch. For sub-threshold bias conditions the FET channel is pinched-off, its
parasitic capacitance is compensated by a shorted stub in close vicinity of the FET and the
signal is routed to the output [77]. The shorted stub not only compensates for the parasitic
capacitance, but also provides power matching and zero drain bias.
Since the function of this type of switches is not dependent on the performance of the FET
itself, but on the proper design of the passive circuits, the repeatability and stability of the
switching operation is warranted. However, measurements presented in Chapter 5 show,
that the correct bias of the FET switches is crucial, to achieve proper isolation between the
channels. The switching process involves no mechanics and is not subject to degradation
like today’s MEMS-based switches still are. Moreover, the switching rate is merely limited
by the transistor speed. These facts are crucial for fast, continuous and reliable switching to
calibration loads.
At the left half of the MMIC, the ACL and the AHL presented in the previous section (see
Figures 3.17) are connected to channel three and four of the SP5T, respectively. Further-
more, a matched ambient load, realized with a 50Ω NiCr resistor to ground, is connected to
channel one. The two remaining channels are routed to the opposite side of the SCFE output
(port 2). This is necessary to allow on-wafer measurements with the VNA setup, where the
signal probes are typically placed on opposite sides, while the DC probes are at one or both
orthogonal edges. This necessitates very long transmission lines, which of course increase
the insertion loss. For the evaluation of the principle SCFE performance this loss can be
de-embedded, but for an optional re-design intended for module assembly, these channels
would be routed to the top and bottom edge, while the DC-pads would be placed at the left
edge.
37
3.4.2 Active Calibration Sub-System
For the ACSS, an SP3T switch was designed by Mikko Kantanen fromMillilab. The principal
design procedure is similar to SP5T of the SCFE. Again, short circuits to ground at the FETs
marked by white circles in Figure 3.21 are transformed to an open circuit at the common star
point S, isolating this path from the intended signal path, while the FET of the signal path
is open. In contrast to the SP5T, two transistors are placed at each branch, which improves
the isolation between the channels, but also increases the insertion loss. In the first wafer
run, the SP3T was placed as a stand-alone circuit to verify its performance and to estimate
how its insertion loss will increase the noise temperature of the ACL. For the second run,
the core structure of this SP3T was utilized and its three input ports modified as part of
this work to connect to the respective ACL. The goal was to have the least insertion loss on
the thru path to have as little influence on the radiometer performance. The second goal
was to have symmetrical paths between the ACL and the internal matched load to ensure
comparable conditions. The additional loss of the transmission lines and bends to connect
the upper path to the ACL also needs to be taken into account. Both ACSS contain the
identical SP3T, as can be seen in Figure 3.21, but differ by the integrated ACL version. For
ACSS V 1, the 89GHz ACL V 1 depicted in Figure 3.14(a) was utilized, while the 89GHz
ACL V 4, shown in 3.15 was integrated into ACSS V 2. Chip photographs of both ACSS are
presented in Figures 3.21(a) and 3.21(b). As previously mentioned, one of the design goals
of ACL V 4 was to yield a more compact design. In the chip photograph of ACSS V 2 it
can be seen, that some space above the ACL is not used (except for the generously placed
MIM capacitors). It would have been possible to reduce the size of the die by 0.25mm of
the longer edge length. However, to allow compatibility between the two ACSS versions in
terms of module packaging, the dimensions of ACSS V 2 were kept identical to ACSS V 1.
(a) 89GHz ACSS V 1. (b) 89GHz ACSS V 2.
Figure 3.21: Chip photographs of the two 89GHz ACSS versions, measuring a size of 1.75 × 1.25mm2,
each.
38
3.4.3 Active Hot-Cold Load
As shown in the AHL in Section 3.3.6, the drain-connection of an ACL can be used as an
active hot load. Contrary to the approach of the SCFE with two separate active loads, the
main idea for the AHCL is to use only one noise-source and to switch between the two ports
of the source to realize a cold and hot load. Although an additional transistor is needed to
route the signal and the total number of transistors remains constant, only one transistor
operates as active device. This leads to a reduction of the DC power dissipation by half.
Moreover, only one transistor has to be power-matched, which allows for the reduction of
chip-size, that is dominated by the transmission lines needed for the routing of the signal
and impedance-matching.
Figure 3.22: Schematic representation of the integrated AHCL. The dotted green box and dashed blue boxes
frame the SP3T and the additional switching network, respectively. The Triangle stands for
the transistor circuit operating as hot and cold load.
The schematic of the integrated AHCL is shown in Figure 3.22. The AHCL has two RF-
ports - the left port (P1) is the output port going to the radiometer. The upper right port
is connected to the receive antenna. The core of the AHCL is the ACL, which is illustrated
by the amplifier symbol. Internally, the left and the right side of it are connected to the
gate and drain of the ACL, respectively. The routing of the signal is done by a SP3T switch,
which is marked by the green dotted box. The SP3T makes use of three transistor-switches
(S1, S2 and S3) and three λ/4 length transmission lines (l1, l2 and l3). The behavior of the
AHCL can be changed by a variation of the control voltage of these three switches and one
additional switch (S4) on the drain side of the transistor. Three switching configurations
exist, that are described in the following:
• Receive: When S2 and S3 create a short, the transmission lines l2 and l3 transform
the short into an open and isolate the branches from P1. When in addition to that,
S1 is biased to provide a through, the signal can travel from the input port to the
output port P1, without being influenced by the other branches.
• Cold: The additional switch S4 is required to provide a 50Ω termination to the ACL
in the cold-load state, and to isolate this termination in the hot-load state. When S1
and S3 are biased to provide a short and S2 and S4 are biased for a through, the
output port P1 is terminated with a low noise load. The short generated by S3 is
transformed by the transmission line l5 with a length of 3/4λ into an open at node
39
Nx. The AHCL in cold operation has the same performance as the stand-alone ACL,
with the difference that its noise temperature is increased by the SP3T losses.
• Hot: The hot load termination is realized, when S1, S2 and S4 are biased for short
and S3 is biased to provide a through. In that case, the λ/4 length transmission line
l4 transforms the S4-short into an open at node Nx and isolates the 50Ω termination.
This is necessary since it would otherwise decrease the noise temperature of the active
hot load. In this switching configuration, the AHCL noise temperature cannot be
described by Equation 3.25, which mandates matching of all ports. Thus, according
to [72], and [71], the hot noise temperature will be dominantly dictated by the power
generated in the transistor output, the amplified noise adjacent at the transistor input
and by the noise power reflected at the transistor output.
Table 3.4 summarizes the switching states and the resulting AHCL behaviour at the
output port.
Table 3.4: Biasing configurations of the AHCL switches for the desired output signal.
Switch state
Output signal S1 S2 S3 S4
Receive Through Short Short N/A
Cold Short Through Short Through
Hot Short Short Through Short
Figure 3.23: Chip photograph of the AHCL having a size of only 1.75 × 1.25mm2. For radiometer appli-
cation, the left port is the output for radiometer connection and the top port is for receive
antenna connection.
Again, the implementation of the SP3T on the MMIC is equivalent to the procedure de-
scribed for the ACSS and the SCFE. Figure 3.23 shows the chip photograph of the developed
AHCL. The nomenclature of the lines and switches in the photograph correspond to the one
40
used in the above schematic. As for the ACSS, to improve isolation two switching transistors
were placed at each branch, marked as Sx1 and Sx2 in the photograph. The shorted 70Ω
stubs, that compensate for the parasitic capacitance are placed between the two transistors
in each branch and marked as lxp. The three λ/4 lines with 70Ω impedance all intersect at a
common star point and the respective path is routed to output port of the AHCL at the left
edge. The core of the AHCL, marked by the red dotted box, is the ACL previously depicted
in Figure 3.16 with its drain side output connected to the node Nx. For the switching
network to the right of the ACL, only one shunt FET was used, since a low insertion loss at
this point has priority over isolation. The 50Ω load is marked R in the figure.
For the developed circuits of active loads with integrated switches presented in this section,
a patent was granted [78].
3.5 D-Band Active Cold Loads and Low-Noise Amplifier
ACLs were also designed in D-band to push the state of the art to the limit and to demon-
strate that the concept of active loads is still feasible in this frequency range. However, as
will be presented in the next chapter, the requirements for active loads characterization are
particularly challenging. Especially the low noise performance and good input return loss of
the first amplifying stage in the measurement setup is essential. Due to limited commercial
availability of appropriate LNAs in D-band, a dedicated broadband LNA MMIC was designed
as part of this work [79].
D-band ACLs
(a) D-band ACL. (b) D-band ACL with integrated SP2T
and matched load.
Figure 3.24: Chip photographs of D-band MMICs, measuring a size of 1× 1mm2, each.
For the design of the D-band ACLs the same principal topology was used as for the W-
band circuits. The layout of the 140GHz ACL V 1 can be seen in the chip photograph in
Figure 3.24(a). Parts of the signal line and the parallel stub line of the IMN were realized
41
with 30Ω line impedance, while the parallel stub line of the OMN was realized with 70Ω line
impedance. For stability at these high frequencies, NiCr resistors were placed in both DC
supply paths, accepting the increase of noise figure associated with that. For the inductive
source feedback, two symmetrical 70Ω line are used. For the termination of the OMN, the
signal path is divided by a T-structure into two parallel transmission lines, of which one is
necessary for the power supply and the other to connect to the 50Ω termination. This layout
change was necessary to allow the integration with a SP2T switch into a very compact die,
which is shown in Figure 3.24(b). The left white dotted rectangle marks the ACL core as
seen in Figure 3.24(a), merely with smaller MIM capacitors in the DC supply lines, due to the
restricted space. The right white dotted rectangle marks the SP2T with its two shunt FETs
T1 and T2 and the two λ/4-lines L1 and L2, that connect at the star point S. The principal
functionality is identical to the one explained for the SCFE. The performance of the SP2T at
D-band frequencies was demonstrated in [80] by the implementation as a Dicke-switch into
an LNA MMIC. As for the switching references presented in Section 3.4, the SP2T allows
to switch between a passive matched load at ambient temperature and the ACL.
D-band LNA
D-band LNAs, using IAF’s mHEMT process with 100 nm and 50 nm gate length were pub-
lished in [81] and achieved noise figures on MMIC level as low as 3 dB, but in a very narrow
bandwidth around 130GHz. For the purpose of covering the frequency range of noise mea-
surement setup presented in the following section, the goal of this LNA design was to
achieve a broadband good noise performance [79]. The LNA was designed using transistors
with 35 nm gate length. This mHEMT process exhibits a channel electron mobility µ0 and a
channel electron density ne as high as 9800 cm
2/Vs and 6.1· 1012 cm−2, respectively. The
maximum extrinsic transconductance gm,max is 2500mS/mm and a transit frequency fT of
over 500GHz is achieved [57]. The chosen topology is a four-stage design in common source
configuration with gate widths of 2× 15µm, each. Inductive source degeneration with sym-
metric 30Ω transmission lines was used to increase stability and improve noise behavior [68].
This is especially important at the first stage. 50Ω GCPW lines are used to connect the four
transistors and series capacitors placed between the stages to decouple the DC bias. The
input and output matching as well as the inter-stage matching is achieved with parallel stub
lines with 30Ω and 50Ω impedance. These stub lines are also used to provide the DC bias.
A serial NiCr resistor and two parallel capacitors in each stub line are placed to decouple the
RF signal from the DC source. The first two and the second two stages of the LNA can be
biased separately at the DC pads at the upper edge of the MMIC. A schematic of the LNA
is shown in 3.25 and a chip photograph is depicted in Figure 3.26.
Figure 3.27 shows the manufactured LNA MMIC mounted into an open split-block module.
The module is equipped with an integrated programmable DC-control board. Note, that the
chip layout allows the first two transistor stages and the subsequent two stages to have
separate DC-supplies. While all four transistors are operated with the same gate voltage
in the module, the DC-control board allows to allocate different drain voltages for the first
two and the subsequent two stages. This permits to fine-tune the LNA for optimum gain-
to-noise-ratio. Gold bond wires are used to connect the signal port pads to microstrip lines
and the DC pads to external pads on a PCB. The microstrip lines are processed on a quartz
substrate and terminated with feeding patches on the waveguide side. The waveguide routes
42
Figure 3.25: Schematic of D-band LNA.
Figure 3.26: Chip photograph of the processed D-band LNA MMIC (2 × 1mm2) with grounded coplanar
waveguide transmission lines and large on-chip capacitors.
43
the signal in a 90 degree bend to the external standard flanges on opposite sides of the
LNA module. The complete module can be seen in Figure 4.15 as part of the D-band noise
measurement setup presented in the following chapter.
Figure 3.27: View of an open D-band split-block package with mounted LNA MMIC and transitions to
waveguide.
Summary and Conclusion
After frequency considerations and the introduction of the utilized technology, the main
emphasis of this chapter was the design of stand-alone active loads and active loads with in-
tegrated switches. For that, different simulation approaches and design-flows were compared
to each other and several MMICs in W-band and D-band were designed for manufacturing.
It was shown, that the general design-flow is very elaborate, iterative and time-consuming,
so that other authors have already investigated simplified approaches. This work has demon-
strated a way to effectively make use of today’s numerical computation methods. Based on
the transistor and passive components models and ADS features, a very direct calculation
method of the noise temperature was presented, that allows to apply automated optimization
algorithms. By that, the normal optimization effort was drastically reduced.
The designed circuits were manufactured but before they can be characterized, an appropri-
ate measurement setup needs to be evaluated and assembled. This will be discussed in the
following chapter.
44
4 One-Port Noise Measurement
Setups
For the characterization of the designed MMICs presented in the previous chapter, several
noise temperature measurement approaches were investigated as part of this work. Possible
error sources were analyzed analytically and different measurement methods experimentally
tested. The findings concluded in two dedicated one-port noise measurement setups for W-
and two for D-band, which are presented in this chapter. The measurement results achieved
with these setups, are demonstrated and compared in Chapters 5 and 6. In addition, some
of the W-band circuit- and module measurement results were verified by measurements of
third parties. The setups used by Radiometer Physics GmbH (RPG) and Millilab for this
purpose are also shortly introduced in this chapter.
4.1 Noise Figure Measurements
For a better understanding of the following discussion about the components and factors
that influence the measurement setup, a schematic diagram of the finally established setup
is shown beforehand in Figure 4.1. All of the components illustrated in this schematic will be
discussed in this chapter and a detailed description of the final setup will be given in Section
4.4.
Figure 4.1: Setup of custom on-wafer measurement system for one-port noise measurements.
To distinguish the task of measuring the absolute noise temperature of a one-port noise
source from the prevalent procedure of noise figure measurements, it is important to intro-
duce some common fundamentals. The noise figure measurement of a two-port DUT is
straightforward and well explained in numerous textbooks and application notes. However,
requirements increase with frequency, while the availability of accurate measurement equip-
ment drastically decreases. The common definition of noise figure was introduced in [82]
45
as the ratio of the signal-to-noise ratio (SNR) at the input of a network to the SNR at the
output:
NF = 10· log10(F ) = 10· log10
(
SNRi
SNRo
)
(4.1)
It is common to differentiate between noise factor F , which is the linear ratio, and the
noise figure NF , which is the decadic logarithm of the noise factor with the unit dB. The
relation between noise figure and noise temperature was already introduced in Equation 3.1.
The SNR is deteriorated by the added noise Na of the device, which becomes evident when
transforming Equation 4.1 into
F =
Si/Ni
So/No
=
Si/Ni
GSi/(Na +GNi)
=
Na +GNi
GNi
=
Na + kT0BG
kT0BG
, (4.2)
where G is the gain of the DUT, k the Boltzmann constant, and B the bandwidth. Hereby,
one fundamental difference to the presented one-port noise measurements setups becomes
obvious. The investigated DUTs have no gain G and no input signal Si, but only exhibit an
inherent noise which is the quantity of interest to be measured. The lack of gain in the DUT
is a crucial challenge in the accuracy of the one-port noise measurements. While the gain
of typical DUTs like LNAs drastically reduce the influence of the measurement setup’s own
noise figure and mitigate the requirements, the noise contribution of the setup in one-port
measurements is particularly critical. To make things worse, the expected noise temperature
of the DUT is considerably lower than the one of the measurement setup itself. Because all
errors in the determination of the measurement system’s noise figure are introduced into the
results, a thorough characterization of the setup is necessary. This is done by the Y-factor
method, which is basically the same procedure as the calibration of radiometers described in
Section 2.
Y-factor Method
The Y-factor method is the basis of most noise figure measurements, whether they are
manual or automatically performed internally in a noise figure analyzer [83]. The Y-factor is
the ratio between two measured power levels at the output of the system with two different
noise levels at its input.
Y =
N2
N1
=
NON
NOF F
(4.3)
Due to the ratio, the absolute power level accuracy of the measuring device is not im-
portant. The relative level accuracy within the noise figure analyzer (NFA) however, is
fundamental and requires very linear internal power detectors.
In addition to the measured noise levels and the calculated Y-factor, the excess noise ratio
(ENR) of the two input noise levels needs to be known in order to calculate the system
noise figure. For commercially available noise measurement setups, the NFAs come with a
calibrated noise diode as a noise source and include the ENR table either stored internally or
provided as hard-copy. In this case the measured noise levels in the above described Y-factor
correspond to the noise source being turned ON and OFF. In general, the ENR can be
calculated for any two noise temperatures at the input by
46
ENRdB = 10· log10(ENR) = 10· log10
(
Th − Tc
T0
)
, (4.4)
where Th is the hot, Tc the cold noise temperature, and T0 equals as always 290K.
Note, that therefore an ENR of 0 dB still describes a noise temperature difference of 290K
between hot and cold. Furthermore, it should be noted that suppliers, when providing the
ENR table for their noise diode, assume that the physical temperature of the noise source
during measurement is T0. Which leads to Tc = T
OF F
ND = T0. In many cases laboratory
conditions differ from this assumption and the error needs to be corrected.
With the measured Y-factor and the known ENR of the noise source, the system noise factor
can be easily calculated [84], using
Fsys =
ENR
Y − 1
(4.5)
Errors in the ENR of the noise source are a major source of measurement uncertainty in
noise figure measurements. In some cases, the uncertainties are similar to the measured
quantities [85]. As can be derived from Equation 4.5, any ENR error transfers directly to
the noise figure [86]: NF = ENRdB − 10 log10(Y − 1). This applies of course also for
noise sources other than noise diodes. Especially errors during calibration with the hot
and cold references will inflict subsequent errors during measurements. However, errors
can be reduced by carefully choosing the proper noise source for this measurement method
and especially for this kind of low-noise one-port DUT. Depending on the characteristic of
the DUT, either a high or low ENR is advantageous [87]. This will be discussed in Section 4.2.
To better understand the implications of the Y-factor and the ENR on the performance
of the measurement setup, Figure 4.2 depicts the parameters, that influence the slope and
the range of the two-point curve. This curve characterizes the setup and is defined by the
choice of the measurement setup components and hot and cold references. It is important
to understand this graph, since during the evaluation of components and references in the
following sections, it will repeatedly be referred to.
-100 0 100 200 300
0
2
4
6
8
10
12
14
m
1
=k
B
BG
1
m
2
=k
B
BG
2
N1
A
N2
A
-T
DUT
T
cold
Y=2
Temperature [K]
Li
ne
ar
 P
ow
er
 [a
rb
itr
ar
y 
un
it]
Y=2
T
hot
Figure 4.2: Diagram showing the Y-factor variables.
47
The blue and the green lines in Figure 4.2 are exemplary for two different measurement
setups. Both curves are determined by using the same references Tc = 77K and Th = 298K
which are plotted on the x-axis. The y-axis represents the measured noise power at the NFA
on a linear scale. Obviously, the slopem of the curve determines in which power measurement
range of the NFA the signal is received. The slope m, in turn, is dependent on the total
(conversion) gain of the components between the DUT or the references and the NFA. The
two exemplary curves and the numbers in the diagram are chosen on purpose, to point out
one crucial fact: If the Y-factor for both curves is calculated by applying Equation 4.3, it
becomes evident that both measurement setups have the same Y-factor (in this example
Y = 2). Furthermore, they both were characterized with the same references Tc and Th,
yielding an identical ENR. According to Equation 4.5 this means, that they both have the
same system noise figure Fsys. This can also be concluded from the fact that both curves
intersect with the negative x-axis at the same point, which is the negative system noise
temperature (or in general −TDUT ). What needs to be pointed out here is, that two setups,
though having the same system noise figure, can project the same temperature range in
different power measurement ranges of the NFA, which are marked blue and green at the
y-axis in Figure 4.2. This, however, can have a major influence on the accuracy of the
measurement, as will be discussed in the following sections.
Second-Stage Effect
During the typical noise figure measurement procedure of a two-port DUT, first the mea-
surement system is characterized by connecting the noise source directly to the system. This
results in the noise factor F2 of the measurement system alone. After that, the DUT is
placed between the noise source and the measurement system and the combined noise fac-
tor Fsys of both, DUT and system, is measured. In analogy to the Friis formula [82], the
cascaded noise factor can be rewritten to find F1, the noise factor of the DUT:
F1 = Fsys −
F2 − 1
G
, (4.6)
where G is the gain of the DUT. It is obvious, that if the gain of the DUT is sufficiently
high, the so-called second stage contribution of the measurement system will be small and
can be neglected. If not, a second stage correction should be performed, which is typically
done automatically by the NFA. As previously pointed out, the DUTs in this work exhibit no
gain. Nevertheless, this automated correction is not possible, because, due to the nature of
a one-port DUT, the noise source can not be connected to the input of the DUT. Therefore,
the combined noise factor Fsys in Equation 4.6 can not be measured with the Y-factor
method and it is not possible to perform an automatic second stage correction.
4.2 Evaluation of Hot and Cold References for Y-factor
Measurements
4.2.1 Noise Diode
Noise diodes (ND) typically provide a high level of ENR, which is desirable in many ap-
plications. If the noise figure of the DUT is much higher than the ENR, the device noise
48
tends to mask the noise source output. In this case the Y-factor will be very close to 1 and
accurate measurements of small ratios become difficult. As a rule of thumb, the noise figure
should not be more than 10 dB above the ENR of the noise source [83]. So in general, a
high ENR is beneficial, when not measuring very low-noise devices. On the other hand, it is
recommended to use low ENR whenever possible, especially when the DUT has a very low
noise figure.
The following paragraph is cited from Keysight’s application note [87]: "A low ENR noise
source will minimize error due to noise detector non-linearity. This error will be smaller if the
measurement is made over a smaller, and therefore more linear, range of the instruments
detector. [...] A low ENR noise source will require the instrument to use the least internal
attenuation to cover the dynamic range of the measurement, unless the gain of the DUT is
very high. Using less attenuators will lower the noise figure of the measurement instrument,
which will lower the uncertainty of the measurement."
Referring to the diagram in Figure 4.2, this not only means, that Tc and Th need to be
carefully chosen, but also, that it is not necessarily beneficial to have the highest possible
gain in the measurement setup. This will be further discussed in Section 4.4. Available
noise diodes for the setups in W-band and D-band were the NC5110 from NoiseCom with
an ENR varying between 9 and 20 dB over the band and the ISSN-06 from Elva with an
ENR of 11 − 15 dB, respectively. Due to the previously cited guideline, most manufacturer
of noise diodes boast about the high ENR their diodes can provide. But an ENR of 15 dB
corresponds to a noise temperature of almost 9000K, which is far above the expected noise
temperature of the DUTs in this work. Since the measurement setup’s gain is optimized to
measure the very weak noise power of the ACL - as will be explained in Section 4.4 - this
would drive the setup into saturation or cause the NFA to automatically switch to higher
internal attenuation to prevent this.
As previously mentioned, it is necessary to regularly have the ND re-calibrated by the supplier
to reduce ENR uncertainties. Typically, in data sheets of ND the nominal ENR and the ENR
flatness over the frequency band are provided. Sometimes also the voltage standing wave
ratio (VSWR), and the stability over temperature or time are provided [88], [89]. What
typically is not provided is the accuracy of the delivered ENR table after the (re-)calibration.
On specific request after a commissioned re-calibration, the distributor of Elva NDs forwarded
the information from the manufacturers engineers, that the estimated accuracy of the ISSN
calibration is ±0.5 dB for W-band and ±1 dB for D-band [90]. As we remember from
Equation 4.5, this uncertainty directly transfers to the calculation of the noise temperature,
resulting in an additional error of ±35K and ±75K in W-band and D-band respectively. This
relatively large error is due to the fact, that the method used to determine the ND’s ENR
is to calibrate it versus hot and cold standards, just like the ones presented in the following
sections. Since the slope of the hot cold calibration curve is defined over a small temperature
range (typically between 77 and 373K) this introduces a significant extrapolation error at
the ND’s temperatures around several thousand Kelvin. This is less critical when measuring
DUTs with NF in the vicinity of the ENR, because the same error is inevitably introduced
when characterizing such DUTs directly with hot cold standards. But when using NDs to
measure a very low noise DUT, a second extrapolation error is introduced. In this case other
references than a noise diode are preferable, which encompass the expected noise temperature
49
of the DUT. Such references will be presented in the following. Another disadvantage is the
large impedance change of the ND when switching between the ON and OFF state, which
is partially counteracted by incorporating an isolator. But again, the passive references
presented in the following (waveguide loads and antennas) are advantageous, because of
their negligible impedance change between hot and cold.
4.2.2 Cooled or Heated Waveguide Loads
Matched passive loads emit a noise temperature equivalent to their physical temperature.
So a very straightforward way is to cool or heat standard waveguide calibration loads to use
them as cold and hot references. By this, either cold and ambient, ambient and hot, or cold
and hot loads can be generated to create a two-point calibration curve.
(a) Container for liquid nitrogen with integrated
waveguide load.
(b) Dissipator for thermal decoupling.
Figure 4.3: Components for setup with cooled waveguide matched load.
Cooled Waveguide Load
A waveguide matched load was soldered into the outer casing of a container so that the load
is inside the container and the standard waveguide flange outside can be connected to the
measurement system (see Figure 4.3(a)). Now the container can be filled with liquid gas.
The use of liquid gases as a cold reference is based on the fact that its boiling temperatures
are well defined and known. For the commonly used liquid nitrogen (LN) this temperature
is at 77K. The Container was thermally isolated with foam to minimize boiling of the
LN. A thermal equilibration needs to be awaited, even though the boiling itself is not as
critical as if looking with an antenna from above into the LN as presented in the next
section. However, thermal conductivity of the waveguide is a massive error source. In both
directions: the first stages of the measurement setup consequently get cooled down, while
the thermal capacity of the measurement system leaks into the container to the matched
load and avoids that it can be cooled down to the actual temperature of the LN. The main
problem with this is, that no stable and defined temperature at any reference plane of the
setup can be achieved. A possible solution is to monitor accurately the temperature along
the waveguide with several sensors, modeling the heat flow and to extract afterwards the
error from the measurement results. A more practical approach is the use of a piece of
50
stainless steel waveguide instead which has notable lower thermal conductivity as copper or
brass. Also, one solution presented in [91], was to physically separate the waveguides and
leave an isolating air gap between two flanges. However this has a massive filtering effect,
so that the distance of the gap and additionally needed structures to compensate for the
effect need to be modeled and simulated and the customized interconnection needs to be
manufactured. For the evaluated setup in this work, the heat flow was strongly reduced by
the use of a customized heat dissipator as depicted disassembled in Figure 4.3(b). When
assembled, it sourrounds a piece of waveguide and the thermal conductivity is optimized by
conductance paste. In the measurement setup it is placed between the waveguide flange
of the matched load in the cooling container and the first stage isolator of the receiver. A
fan, pointed at the dissipator, amplifies the convection and pulls the temperature of the cold
waveguide towards the ambient temperature. By this, a thermal barrier is generated between
the physically cold waveguide load and the measurement receiver. In Figure 4.4 the effect of
this barrier can be observed: while the condensed air humidity generates a layer of frost on
the waveguide bend left of the dissipator, the isolator on the right was only slightly colder
than the ambient temperature.
Figure 4.4: Setup for measuring with cooled waveguide load.
However, a residual heat transfer and therefore an influence on the receiver can not be
prevented by this method. Whatever action is taken to minimize the thermal conductivity
along the waveguide, it can not be avoided, that the air exchange inside the waveguide due
to the temperature gradient leads to a heat flow. Furthermore, this leads to condensation
inside the waveguide, which of course has an unpredictable influence on the wave propagation
inside the waveguide. Therefore, this option was not chosen for the calibration of the final
measurement setup.
Heated Waveguide Load
When in contrast heating the waveguide load, the issue of condensing air humidity is omitted.
In addition, from a practical and safety point of view, it is not as challenging as the handling
with liquid gases. However, the heat transfer issue remains and has an influence on the
receiver. Furthermore, the achieved ENR is limited by the fact that a waveguide load should
not be heated to more than 100 ◦C to avoid damage of the load.
51
4.2.3 Artificial Black Body
The concept of an artificial black body was introduced in Section 2.2. A microwave absorber
is used to radiate a defined brightness temperature, depending on its own physical temper-
ature. Connecting an antenna to the front-end of the measurement system and facing it
towards the absorber, the setup can be calibrated with two known absorber temperatures,
just as a radiometer. The necessary components for this procedure are presented in the
following.
Microwave Absorber
The absorber used as an artificial black body was unpainted pyramidal C-RAM from Cuming
Microwave [92]. It is a high performance broadband absorber from specially treated low
density polyurethane foam, which is typically used for anechoic chambers. Even though it is
only characterized up to 50GHz by the manufacturer, yielding a typical reflectivity of −50 dB
at normal incidence, it is known to provide premium performance even at higher frequencies
and off-normal incident angles. An extensive investigation of similar pyramidal absorber was
performed up to 600GHz in [93] and confirmed its usability. Furthermore, the absorber foam
is resistant to cold temperatures like being soaked in LN. Therefore it can not only be used
as an ambient load, but also as a cold load.
Antenna
To calibrate the measurement setup with hot and cold microwave absorber, an antenna needs
to be connected to the measurement system front-end. This has the inherent advantage,
that by the physical separation of the measurement system and the loads, the previously de-
scribed issue of heat transfer and thermal conductivity is practically solved. Another notable
advantage of this approach is, that in contrast to noise diodes, the antenna impedance does
not change between the hot and cold state. The return loss of the antenna is in fact a very
crucial parameter that will be discussed in Section 4.3. The influence of the ohmic resistance
of the antenna, however, is only negligible as long as it is at the same temperature as the
measurement system. This is typically the case, but when working with LN as described in
the following, some precautions must be taken. The beam efficiency of the antenna might
have some influence but is not critical because it is assumed, due to the setup’s geometry,
that from any relevant receiving angle the same temperature is seen. In [50], the optimal
distance between a horn antenna and cold target is stated to be in the range of 8− 8.5 cm
for the test-beds at 23.8GHz and 36.5GHz. For the investigated setups in W-band and
D-band, a similar distance was found to be optimal with a corrugated horn antenna. The
main influence is thus not directly dependent on the frequency, but on the measurement
setup geometry. This was experimentally confirmed and applied for the setups used in this
work.
Liquid Nitrogen
Having the two previously described components at hand, the measurement of the ambient
reference temperature is straightforward. To generate a cold reference target, the absorber
needs to be cooled with LN. To achieve reliable results, many details need to be taken into
52
account. In the applied setup, the LN was filled into a Dewar vessel which was covered with
absorber pyramids at the bottom. When pouring the liquid nitrogen into the Dewar vessel
this results in a strong evaporation, due to the high temperature difference. Temperature
equilibration needs to be awaited, so that the surface of the LN becomes flat. The air
temperature inside the Dewar vessel in the space between the upper edge of the vessel and
the LN surface becomes significantly cold. Therefore it needs to be taken care of, that the
antenna does not immerge into that space, but is above the edge of the Dewar. Only then
the thermal isolation is ensured and the ohmic resistance of the antenna does not influence
the calibration result. Figure 4.5 illustrates such a setup geometry.
Figure 4.5: Utilized setup for the measurement system calibration with liquid nitrogen.
Due to the small opening angle of the main lobe of the antenna and the reduced side-
lobes, this arrangement is not critical. Furthermore, the outer casing of the Dewar is made
of metal. This way all unwanted radiation that would leak from any angle from above into
the vessel would be reflected into the absorbing bottom. On the other side, any side-lobe of
the antenna is pointed at a flat angle towards the outer casing and will therefore also receive
the emitted cold radiation of the absorber reflected by the casing. The physical temperature
of the antenna was monitored to be at room temperature.
Another critical point is the filling level of the LN. If it is far above the microwave absorber,
the radiation emitted from the absorber will be attenuated. The radiation of the LN itself
is not too crucial, because its characteristic is far from a black body and its emissivity very
low. Its reflectivity, however, could cause some spurious radiation due to reflections on its
surface. In a worst-case scenario, the emitted low temperature of the absorber would be
strongly attenuated and instead the antenna would measure spurious noise from the room
which is reflected at the LN surface. But due to the chosen geometry, the conducting
casing of the Dewar, the above mentioned antenna pattern, and an appropriate LN filling
level, this influence can be neglected. With much more sensitive radiometer systems, which
use a lot of integration time and measure with an accuracy of a tenth of Kelvin, these
parameters can have some influence. Also, the boiling temperature of the LN is dependent
on the actual air pressure. While this was accounted for during the measurements with
the temperature stabilized test-bed at DA-Design to avoid misinterpretation of drifts during
continuous measurement, this was not as critical for the short-term measurements in this
work. Several filling levels were experimentally tested, as well as different angles of the
antenna in relation to the LN surface, but no significant influence was observed with the
applied setup.
An alternative setup would be to have an antenna pointed to a styrofoam container above
the measurement setup. The container is covered with absorber at the bottom so that the
53
antenna looks directly through the styrofoam at the absorber, which is permanently cooled by
the LN filled in the container. The styrofoam itself has negligible influence for the penetration
of the radiation. The advantage is, that there is no influence due to reflection of the LN
surface, and that the filling level of the container has no relevance at all. This way it can be
generously filled and closed with a lid for better isolation and limiting evaporation, which is
especially beneficial for long-term measurements. Figure 4.6 illustrates such an alternative
setup.
Figure 4.6: Alternative setup for the measurement system calibration with liquid nitrogen.
Disadvantage is, that despite the isolating capability of the styrofoam, the bottom of
the container becomes colder than the surrounding air, which leads to condensation of the
air moisture. Water drops at the bottom of the styrofoam container have on one hand
significant influence on the radiation path and on the other hand risk to drop inside the
antenna. Another issue with this setup is, that the air below the container is cooled down
due to convection. If the setup is placed below the container, this will have an effect
on the temperature stability of the measurement system. And long waveguides to move
the receiver outside the reach of influence yield additional losses, also the waveguides are
influenced by the cooling. While this setup, despite its drawbacks, would be favorable for
long-term measurements and stability analysis, it was assessed to be too elaborate for the
purpose of this work. Besides, it is not practicable for on-wafer measurements.
4.3 Measurement System Analysis
As it was emphasized in Section 4.1, a minimal overall noise figure of the measurement
system is crucial to mitigate large influence on the measurement results. With this goal,
several components were cascaded in the setups. A trade-off between minimal noise figure,
optimum gain, and reduction of interstage mismatch had to be found. These factors will be
discussed in the following.
4.3.1 System Noise Figure and Gain
The system noise figures of the setups can be characterized thoroughly and the results for
the accordant setup can be taken into account for the interpretation of the measurements.
However, the noise figure is not the only parameter which influences successful measurements.
As was pointed out in Section 4.1, the choice of the proper gain in the setups front-end is
equally important, because it dictates in which dynamic power range of the NFA the noise
temperature is projected. The NF or losses of the first stages in the receiver chain (e.g.
54
the first LNA or isolator) are the relevant components that influence the overall system
noise figure [82], while the noise introduced by the mixer or further amplifiers for example
has little influence (this only applies for the influence on the system noise figure - not on
possibly occurring reflections, as will be derived in Chapter 5). However, the gain or loss
of any component within the receiver chain has a direct influence on the overall gain of
the measurement setup. As was shown in Figure 4.2, for two setups with identical system
noise figure, this yields different slopes of the hot-cold curve and, as a consequence, another
dynamic range of the NFA is addressed.
Over a larger dynamic range, the NFA’s power detector can exhibit a different linearity,
yielding a different measurement result for the same noise temperature. In the application
note [84] it is stated, that the power range in which the noise is measured in the NFA should
be rather small because this favors the linearity of the detector. Because it is not known
in which range the linearity of the detector is optimal, two different setups were used to
characterize the circuits. A major focus of this work is therefore, to point out the influence
of the measurement setup on the results and to explain diverging results of identical DUTs.
4.3.2 Interstage Mismatching
To have the best possible system noise figure of the measurement setup, it would be beneficial
to have an amplifying component as a first stage. This is easily explained by applying the
Friis formula [82]. Therefore, in a first approach, the reference plane of the measurement
system was directly at the first stage amplifier input. This, however, resulted in strong
oscillations in the measurement results, due to reflections between the LNA module and the
probe. These reflections superpose with others that originate at the discontinuity between
probe tip and the MMIC. At these stages, before the first LNA, where the power levels to
be measured are very weak, this has a major influence. Since none of any components used
in a measurement setup has an ideal matching, especially because noise is an extremely
broadband effect, always some amount of the noise is reflected at its input or output. These
re-reflections happen between any stage within the receiver chain and, even worse, can be
carried forward through the subsequent stages where they are superposed with the other
reflections. Therefore, the effect of mismatch on noise figure measurements is extremely
complicated to analyze. Some effort was made to subtract this effect from the measurement
result, by characterizing the critical components in magnitude and phase and subtracting
their influence from the result. The outcome however, was not satisfying.
Isolators
The best way to minimize these reflections, is to place isolators between critical stages. While
its influence on the overall system noise figure and gain is negligible if they are placed after
the first LNA, its insertion loss directly adds to the system noise figure if it is placed at
the first stage. However, especially at this position its use is mandatory to achieve a good
improvement in the ripple of the results.
Figure 4.7(a) depicts the measured S-parameters of the isolator used as a first stage com-
ponent. It has an input return loss of around 20 dB and an isolation of more than 30 dB.
This significantly reduces the amount of noise waves traveling back to the DUT, however,
as results in Chapter 5 will show, this is not enough to completely isolate the DUT from the
55
influence of the measurement setup. If the antenna, which is used for the calibration causes
re-reflections due to non-ideal matching, this error will persist in all subsequent measure-
ments after the calibration. Therefore a good matching of the antenna is equally important.
Figure 4.7(b) depicts the input return loss of the W-band horn antenna used for the system
calibration. With an input return loss of more than 22 dB on antenna side and 20 dB on
the isolator side, this effect can be significantly minimized, at least during calibration. For
the following measurements however, the amount of interstage reflections depends on the
matching of the DUT and, for on-wafer measurements, additionally on the matching of the
wafer probe.
75 80 85 90 95 100 105 110
-50
-40
-30
-20
-2.5
-2.0
-1.5
-1.0
-0.5
0.0
 
Frequency [GHz]
S
 - 
P
ar
am
et
er
 [d
B
]
Return Loss
Insertion Loss
Isolation
(a) Measured S-parameters of the W-band first stage
isolator from HP (Ser.No.178).
75 80 85 90 95 100 105 110
-40
-35
-30
-25
-20
-15
-10
 
 
Frequency [GHz]
In
pu
t R
et
ur
n 
Lo
ss
 [d
B
] W-band horn antenna
(b) Measured return loss of the W-band horn an-
tenna at the waveguide flange.
Figure 4.7: S-parameters of the most critical front-end components of the W-band measurement setup.
RF-probe
The previously described isolator can significantly reduce the reflections between the mea-
surement setup and the module DUT and also between the setup and the RF-probe if the
probe is sufficiently matched at the waveguide side. Figure 4.8 shows the S-parameters of
the used Picoprobe for W-band (measured) and the Cascade Microtech probe for D-band
(provided by the supplier). While the return loss on waveguide side is better than 20 dB at
the respective center frequencies of both RF-probes and better than 10 dB within the respec-
tive waveguide band, the return loss on the probe-tip side for the W-band probe is less than
10 dB at some frequencies. This has two consequences. For one this yields re-reflections
between the probe-tip side and the on-wafer DUT, which can not be avoided. Furthermore,
the loss of the probes was only accounted for by subtracting the temperature increase due to
its scalar loss (red curve in Figure 4.8) from the noise temperature results. This is normally a
sufficiently valid approach. For very low absolute noise temperatures and for high reflection
coefficients, however, the accurate way would be to account for the available gain of the
wafer-probe, defined by Equation 4.7.
Ga =
(1− |ΓS|
2)|S21|
2
|1− S11ΓS|2(1− |S22 +
S12S21ΓS
1−S11ΓS
|2)
(4.7)
This means, that the amount of loss that the RF-probe inflicts on the measurement signal,
is dependent on the matching and the reflection coefficients of its surrounding stages.
56
75 80 85 90 95 100 105 110
-30
-20
-10
-3,0
-2,5
-2,0
-1,5
-1,0
-0,5
0,0
 
 
Frequency [GHz]
S
 - 
P
ar
am
et
er
 [d
B
]
Return Loss Waveguide Side
Insertion Loss
Return Loss Probe Side
(a) W-band Probe
110115120125130135140145150155160165170
-30
-20
-10
-3,0
-2,5
-2,0
-1,5
-1,0
-0,5
0,0
 
Frequency [GHz]
S
 - 
P
ar
am
et
er
 [d
B
]
Return Loss Waveguide Side
Insertion Loss
Return Loss Probe Side
(b) D-band Probe
Figure 4.8: S-parameters of the microwave probes by GPP industries (W-band) and Cascade Microtech
(D-band).
4.4 Final Measurement Setups in W- and D-band
4.4.1 W-Band Setups
For the noise temperature characterization of the reference loads, a dedicated hot-cold mea-
surement setup with the possibility of on-wafer one-port measurements was established. A
schematic diagram of the setup was already depicted at the beginning of this chapter in Fig-
ure 4.1. The reference plane for the hot-cold calibration with a corrugated horn antenna and
microwave absorber at 398 and 77K temperature is behind the RF-probe, which is mounted
on a wafer probe station. To cool down the absorber, LN was used. Since the probe is not
within the calibration loop (please see Figure 4.1), the resistive part of its loss adds noise to
the measured noise temperature of the DUT. This additional noise is taken into account and
subtracted from the result after measurement. The receiver consists of a LNA module with
20 dB small-signal gain and a receiver module containing an integrated LNA-mixer-MMIC
with a conversion gain of approximately 10 dB. The LO signal for the mixer is provided by a
HP 83650B swept signal generator and an HP 83558A millimeter-wave source module for
W-band. The IF output of the mixer is connected to an Agilent 8975A NFA. To minimize
the influence of noise reflections, isolators were placed at critical stages. A photograph of
the wafer probe station and the above mentioned components of the on-wafer measurement
setup can be seen in Figure 4.9. Please be aware that the components necessary for the
calibration of the setup (compare with Figure 4.1) are not present in the photographed
scenario.
The return losses of the first stage isolator and the on-wafer measurement probe are both
around 20 dB and the return loss of the horn antenna lies even well above 20 dB. Since this
is crucial for accurate measurement of the DUT with the Y-method, the degradation of the
system noise figure due to a lossy first stage is accepted. The system noise figure of the
measurement setup at the reference plane is shown in Figure 4.12. It is around 4 dB in the
frequency range of 75− 95GHz. It should be noted that this slightly worsens if referred to
the reference plane on chip, due to the mentioned losses of the probe.
57
Figure 4.9: Photograph of the wafer probe station and the on-wafer measurement setup.
W-Band Temperature Chamber Measurement Setup
While the noise temperature of passive loads has a linear relation to its physical temperature
(1K/◦C), investigations in [34] (for SiGe HBT ACLs) and [65] (for GaAs FET ACLs) have
shown, that ACLs are less sensitive to temperature variations than passive loads (0.3K/◦C
and less than 0.4K/◦C, respectively). It should be noted, that the investigated ACLs in these
publications were at much lower frequencies (1.4GHz and 10.69GHz, respectively), than the
ACLs in this work and that in both publications only physical temperatures above 0 ◦C
were analyzed (0-50 ◦C). To analyze the temperature dependence of the W-band ACL design
integrated in the SCFE, an ACL module was characterized in a temperature chamber. The
ACL module was placed into a WEISS WKL series climate and temperature test chamber
and connected with a WR−10 waveguide to the measurement receiver outside the chamber
(see Figure 4.10). The measurement setup was calibrated at the reference plane at the ACL
module before and after the test series. Because a long piece of waveguide had to be placed
between the first LNA outside the chamber and the DUT inside the chamber, the reference
plane moves inside the chamber, yielding a higher system noise figure than the regular setup
(see blue curve in Figure 4.12). To ensure a stable operation of the receiver and to minimize
the influence of thermal conduction through the waveguide, a fan was placed at the front-end
and a thermistor (NTC) was attached to the first stage LNA, monitoring its temperature.
To avoid condensation inside the ACL module, the temperature ramp was run from cold to
hot. The temperature was increased from -40 ◦C to +80 ◦C in steps of 10K. Since the ACL
is not an amplifying element and has negligible power dissipation, it can be assumed that
after an appropriate time, the temperature of the module is equal to the temperature set
inside the test chamber.
Millilab Setup
For the measurements performed at Millilab, a standard waveguide load was used with a
heating resistor attached to its body. A separate control box measured the temperature at
58
Figure 4.10: Temperature chamber measurement setup.
Figure 4.11: Photograph of the temperature chamber measurement setup.
75 80 85 90 95 100 105 110
3
4
5
6
7
8
 
 
Frequency [GHz]
S
ys
te
m
 N
oi
se
 F
ig
ur
e 
[d
B
]
Setup Temperature Chamber
Setup 1
Setup 2
Figure 4.12: System noise figure of the measurement systems at their reference plane.
59
the waveguide load with a sensor and controlled the current in the resistor. The temperature
oscillated with a ±2K amplitude around the set temperature. Four temperatures between
ambient and 100 ◦C were used to fit a hot-cold curve between the calibration points. All
points were very close to the resulting line, and essentially a calibration with only ambient and
100 ◦C would have yielded the same results. Measurements at all temperatures took about
15-20 minutes when starting from the lowest temperature towards the higher temperatures.
A 50mm long stainless steel waveguide was used to minimize the heat flux between the
heated load and the measurement setup. While the isolator did not get warmer during the
measurements, a temperature gradient could be observed on the solid steel waveguide.
RPG Setup
Additional measurements were performed at RPG. For the measurement of the ACL and AHL
modules different first stage LNAs were used to adapt for the large difference in expected
noise temperature. For the ACL measurement setup an LNA with a small-signal gain of
40 dB was used, while for the AHL measurement setup an LNA with only 25 dB was used,
to avoid the need of an additional attenuator. The respective system noise figures differ
from each other and are shown in Figure 4.14. For the ACL setup the measurement system
has a noise figure of below 5 dB (red curve) and for the the AHL setup a noise figure of
around 6 dB (blue curve). Instead of a NFA, a Rohde&Schwarz FSU Spectrum Analyzer was
used to measure the noise power from the IF output of the measurement setup’s mixer. A
Rohde&Schwarz SFM 100A Signal Generator was used to generate the LO signal for the
mixer. The receiver front-end is placed on a temperature stabilized Peltier-testbench, which
reduces the gain variations of the measurement setups and yields very accurate results. As
in the IAF setups, an isolator and LNA are used as first stages. A photograph of the setup
is shown in Figure 4.13.
Figure 4.13: Module measurement setup with temperature stabilized radiometer at RPG.
60
75 80 85 90 95 100 105 110
3
4
5
6
7
8
9
 
Frequency [GHz]
S
ys
te
m
 N
oi
se
 F
ig
ur
e 
[d
B
]
Setup for AHL module
Setup for ACL module
Figure 4.14: System noise figures of the measurement system at RPG for the characterization of ACL and
AHL modules.
4.4.2 D-Band Setups
As previously mentioned, the D-band LNA, presented in Section 3.5, was specially developed
for the purpose of characterizing the D-band ACL circuits. Therefore, several LNA MMICs
were assembled into split-block modules and characterized. While the overall performance of
the modules was very equal to each other, one of the modules has shown particularly good
results in terms of noise figure. Consequently, this module LNA 667 was always chosen to
be the first amplifying stage in the following setups. Its performance is presented in Chapter
6. Other available modules were the LNA 666 and LNA 670. Due to the limitation of the
mixer in the receiver, only the lower part of the D-band (110 − 150GHz) can be analyzed.
The general architecture of the setup (see Figure 4.15(a)) is similar to the ones presented in
W-band. Again, the reference plane is always at the input of the first isolator. After hot-cold
calibration with a D-band horn antenna, the antenna is replaced by the Cascade Microtech D-
band probe and accounted for the probe loss (see Figure 4.8(b)) during the noise temperature
calculation. Furthermore, the setup consists of two consecutive LNA modules, connected by
a piece of waveguide. Together, they achieve a total gain of 44 dB. Different permutations
of the sequence of LNAs and isolators were investigated by comparison of the flatness of the
system noise figure over frequency. This gives an indication about the unwanted reflections
of the power waves between the stages. Typically best performance is achieved if a lossy
component is placed between two amplifying stages. Hence, the two consecutive LNAs
were believed to be the most critical combination within the setup. The measurement
of the system noise figure presented in Figure 4.16 (blue curve), however, confirms that
this sequence achieved the wanted flatness in the frequency range above 130GHz. While
this is the relevant frequency range for the intended characterization of the D-band ACL
MMICs, which were designed at a center frequency of 140GHz, this setup was not suitable
to investigate the performance of the W-band ACL module above 110GHz. All setups that
were hot-cold calibrated with the available D-band horn antenna, exhibited a more or less
strong ripple below 130GHz, independent of the sequence or number of LNAs and isolators.
Therefore a modified assembly was used, especially for the characterization of the W-band
61
ACL module. It is presented in Figure 4.15(b) and allows a hot-cold calibration with the
previously described W-band horn (see Figure 4.7(b)). For this, the waveguide transition to
W-band and the HP isolator (see Figure 4.7(a)) are placed directly in front of the LNA 667.
After hot-cold calibration with the W-band horn, the DUT is connected at the reference
plane, as shown in Figure 4.15(b). The result of this measurement is presented in Section
5.2.
(a) D-band measurement setup for
on-wafer characterization of ACLs
and module measurements.
(b) Modified D-band measurement
setup for characterization of W-band
ACL-module.
Figure 4.15: Different measurement setups for the lower D-band (110− 150GHz).
110 115 120 125 130 135 140 145 150
4
5
6
7
8
9
10
11
 
Frequency [GHz]
S
ys
te
m
 N
oi
se
 F
ig
ur
e 
[d
B
]
standard setup
modified setup
Figure 4.16: System noise figures of the two D-band measurement setups.
The system noise figure of the modified D-band setup is also shown in Figure 4.16 (green
curve). Compared to the chosen general setup (blue curve) and all other investigated setups,
it exhibits an adequate flatness and a noise figure below 7 dB in the frequency range of
110 − 120GHz, which is suitable for the characterization of the W-band ACL module. In
62
return its performance deteriorates above 140GHz compared to standard setup. In general,
the covered frequency range by both setups can be divided into four sections:
• From 110 − 120GHz, where the calibration with the W-band horn is advantageous
(green curve).
• From 120 − 130GHz, where both setups have a strong fluctuation between the mea-
surement points, indicating a common inherent error source other than the antenna.
This could be the receiver or more likely the first stage LNA 667, which has an output
return loss of less than −10 dB in the range of 115− 128GHz (see Figure 6.2).
• From 130− 140GHz where both setups have a very similar system noise figure.
• From 140−150GHz where the standard D-band setup with two LNAs is advantageous
(blue curve).
110 115 120 125 130 135 140 145 150
150
200
250
300
350
calibration wg load
W-band ACL-module 
w/ wg transistion
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
D-band LNA-module @ input
Figure 4.17: Measurement of W-band ACL-module with a waveguide transition to D-band, and the D-band
LNA module 666 at its input port.
To verify the suitability of the standard setup for the characterization of the D-band
ACLs, the D-band LNA module 666 and the W-band ACL module were measured. After
hot-cold calibration with the D-band horn, the LNA was connected with its input port to
the reference plane at the D-band isolator, while the output port was terminated with a
waveguide matched load. Likewise, the W-band ACL-module was connected at the reference
plane, but together with the waveguide transition from W- to D-band. The result can be seen
in Figure 4.17. As a reference, a waveguide calibration load was measured (black curve). The
red curve and the green curve are the measured noise temperatures of the ACL-module and
the LNA without bias, respectively. The shaded area indicates the frequency range, where
due to the previously determined system noise figure no correct results can be expected. The
measured noise temperature of the ACL module (blue curve) is consistent with the module
measurement that was performed with the modified setup (see Figure 5.11 except for the
strong ripple above 135GHz which is due to the waveguide transition (which was not part
of the calibration loop) and no reason for concern. The LNA (orange curve) exhibits a
63
noise temperature around 250K from 120−135GHz and around 275K from 135−150GHz.
However, the ripples on the curves of the calibration load and the DUTs without bias from
120− 135GHz indicate, that the strong ripple of the system noise figure in this range (see
Figure 4.16) has a not negligible influence on the measurement. This has of course more
consequences at temperatures that are in the vicinity of the temperatures with which the
setup was calibrated (in this case the ambient temperature) and less at temperatures that
are encompassed by the two calibration references (like the ACL module). Nevertheless, due
to this verification, only measurement values above 135GHz are assumed to be reliable with
this setup. Therefore, the area below this frequency is also shaded in the result figures in
Chapter 6.
Summary and Conclusion
In this chapter it was elaborated, that it is not sufficient to assemble a supposedly standard
measurement setup, but that one-port noise measurements - especially at these frequencies
and such expected low noise temperatures - are a very challenging task. Several different
approaches were evaluated, which concluded into few dedicated measurement setups in W-
and D-band, that were hot-cold calibrated with absorber at ambient temperature and at
77K. One common most critical influence however, was found to be the reflected noise
power between the multiple stages due to non ideal matching. Even though this problem is
also known from standard two-port noise measurements, there exists no analytical way to
satisfactorily account for this reflections or to eliminate its influence on the noise measure-
ment result. Therefore, as crucial part of this work, critical components were identified and
measures taken to reduce their influence as much as possible. However, the conclusion of
comparing very similar setups against each other is, that it is not sufficient to just apply
these best practices and then to rely on the result. But that in order to have reliable results
for such impedance sensitive DUTs, measurements should be performed with different setups,
yielding different results, but narrowing down the correct values. Measurement results of
this kind should only be interpreted under consideration of the applied measurement setup,
which will be further discussed in the following result chapters.
64
5 Characterization of W-Band Active
Load MMICs and Modules
The setups presented in the last chapter allow a thorough noise characterization of the cir-
cuits designed for this work, in addition to the standard S-parameter measurements. While
the latter were performed in a common procedure with standard measurement equipment
and unambiguous results, all the measured noise temperatures presented in this chapter
need to be discussed with regard to the used measurement setup. Especially the comparison
of identical DUTs, measured with different setups, or the difference between on-wafer and
module measurement results with the same setup will give further insight into the interdepen-
dency of the noise measurement setup and the DUT. This shall be discussed in this chapter,
along with the comparison of the used noise simulation method with the measurement result.
Furthermore, the performance of the manufactured ACL modules will be compared against
other loads and its suitability as a hot-cold reference for noise measurements demonstrated.
5.1 MMIC measurements
5.1.1 89 GHz ACLs
Because the evaluation and assembly of the appropriate measurement setup described in the
previous chapter was not yet completed at the early stage of the Calload project, the stand-
alone ACL circuits for this project, were initially shipped to the project partner Millilab for
characterization. The measurement results of the 89GHz ACL V 1 were awaited to confirm
the simulation approach before initiating a re-design. Later, comparative noise measurements
were performed for some of the circuits with the setups at IAF. The S-parameters of the
circuits were measured with a VNA, an appropriate W-band extension and GSG on-wafer
probes. In Figure 5.1(a), the measured input return loss (red curve) is compared to the
simulated value (blue curve) and it can be noticed that the measurement result at 89GHz
is 10 dB worse than than the predicted value. At the same time, in Figure 5.1(b), the
comparison of the measured result (red curve) and the noise temperature simulated with
Equation 3.20 (blue curve), yields a better result than predicted at 89GHz. Therefore, the
design goal of the re-design was to focus on an improved return loss, while accepting a higher
noise temperature. For the measurements, the ACLs were biased with a drain voltage of
Vd = +1.0V and the gate voltage Vg was set to yield a drain current of Id = 400mA/mm,
which corresponds to the bias settings in the simulation.
The results of the re-design, 89GHz ACL V 4, are depicted in Figures 5.2. Again, the
same bias was applied for the measurements as for the simulation (same bias as for ACL
V 1). The simulated input return loss (blue curve) in Figure 5.2(a) was much more ambitious
than for ACL V 1, yet, an even larger discrepancy of over 20 dB between simulation and
65
50 60 70 80 90 100 110
-30
-25
-20
-15
-10
-5
0
 
Frequency [GHz]
In
pu
t R
et
ur
n 
Lo
ss
 [d
B
] Measurement
Simulation
(a) Input return loss.
75 80 85 90 95 100 105 110
0
50
100
150
200
250
300
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
] Simulation
Measurement
(b) Noise temperature.
Figure 5.1: Measured and simulated S-parameter and noise temperature of 89GHz ACL V 1 MMIC, mea-
sured at Millilab.
measurement (red curve) can be observed. However, an improvement of over 5 dB was
achieved compared to ACL V 1. For ACL V 4, the measured (red curve) noise temperature
in Figure 5.2(b) was higher than predicted in the simulation (blue curve), even though the
simulation yielded similar results as for ACL V 1. By the time ACL V 4 was manufactured,
the noise measurement Setup 2 was operational, allowing a comparative measurement which
is represented by the green curve in Figure 5.2(b). Different cells were probed and biased
with different voltages, yielding a variation of maximum 10K at 89GHz. For the green
measurement curve shown in Figure 5.2(b), Vg = 0.0V and Vd = +0.6V was set, to achieve
the same drain current of Id = 400mA/mm.
50 60 70 80 90 100 110
-50
-40
-30
-20
-10
0
 
Frequency [GHz]
In
pu
t R
et
ur
n 
Lo
ss
 [d
B
]
Simulation
Measurement
(a) Input return loss.
75 80 85 90 95 100
0
50
100
150
200
250
300
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Polynomial fit
Measurement
Millilab setup
Measurement
setup 2
Simulation
(b) Noise temperature.
Figure 5.2: Simulated S-parameter and noise temperature of 89GHz ACL V 4, compared to measurements
at Millilab and at IAF.
The difference of the measurement results achieved with the two different setups, confirms
the concerns that initially led to the investigation of one-port noise measurement setups in
Chapter 4. At some frequencies, both setups yield identical results, while at others, a worst
case difference of around 40K can be observed. Furthermore, the red curve exhibits a
strong ripple, which is an indicator for the noise power reflections described in Chapter 4. It
also shows, that by the taken measures to reduce this impact, this ripple can be significantly
reduced. However, if the effect of the ripple is masked by applying a second order polynomial
66
fit through the measurement points (black curve), the principal trend of both measurements
is very similar. This is of course a very nonscientific approach and only done at this point
to emphasize the importance of a thorough investigation of the measurement setup. Still,
the fact that in average, one setup yields consistently (referring to the fitted curve) lower
temperatures, than another remains. This can be explained with the described influence
of the setup’s front-end gain on the linearity of the NFA in the dynamic range. Further
observations of this effect will be discussed in the course of this chapter. At this point,
it can be stated, that the final measurement setups, derived from the investigations in
Chapter 4, did successfully reduce the influence of noise power reflections. And that the
results measured with Setup 2 yield around 30K higher noise temperature for on-wafer
measurements at 89GHz.
75 80 85 90 95 100
0
50
100
150
200
250
300
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
ACL89 V1
ACL89 V4
Figure 5.3: Comparison of noise temperatures of 89GHz ACLs V 1 and V 4, measured at Millilab.
In Figure 5.3 the noise temperatures of both 89GHz ACL versions are compared against
each other. As expected, the noise temperature of V 4 is around 40K higher as consequence
of the trade-off for an improved input return loss. What’s more, it can be seen, that in the
vicinity of 85GHz, where the input return loss of ACL V 1 is worse than −15 dB, the ripple
on the noise temperature is particularly strong. Simultaneously, for ACL V 4, where the input
return loss is better than −15 dB over the entire W-band, the ripple is slightly smaller.
5.1.2 W-band ACLs
The noise temperature of the stand-alone W-band ACL circuit that was intended for integra-
tion into the SCFE, as well as for module assembly, was measured on-wafer with Setup 1.
The input return loss was measured with a VNA, two different port extensions for W- and D-
band and the respective on-wafer probes. The ACL bias was Vg = +0.1V and Vd = +1.0V,
yielding an Id = 300mA/mm.
Figures 5.4 show the comparison between the simulation (red curves) and the measurement
(black curves). For the simulation of the noise figure, Equation 3.24 was used. In Figure
5.4(a), a perfect match between simulation and measurement at the desired center frequency
of 94GHz can be observed, yielding −30 dB. Above the center frequency, the input return
67
60 70 80 90 100 110
-50
-40
-30
-20
-10
0
 
Frequency [GHz]
In
pu
t R
et
ur
n 
Lo
ss
 [d
B
]
Simulation
Measurement
(a) Input return loss.
75 80 85 90 95 100 105 110 115
0
50
100
150
200
250
300
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
] Measurement
Simulation
(b) Noise temperature.
Figure 5.4: Comparison of simulated and measured input return loss of the ACL MMIC and noise temper-
ature.
loss is better than expected, yielding around −20 dB up to 110GHz and more than −10 dB
up to 140GHz. The measured noise temperature above 90GHz in 5.4(b) matches very well
with the simulated values, but drastically increases for frequencies below. This coincides
exactly with the degradation of the input return loss. This emphasizes the fact, that an
extraordinary good input return loss is equally important as the supposed main goal of a low
noise temperature. Further evidence of this correlation will be presented in Section 5.2.
(a) S-parameter. (b) Noise figure.
Figure 5.5: Comparison of simulation and measurement results of the ACL S-parameters and noise figure.
Compared to the previous ACLs, the W-band ACL, that was intended for the AHCL, has
the notable difference that the drain side is also routed to an RF-signal pad. Thus, a full
S-parameter analysis can be performed, as well as a standard two-port NF measurement.
This allows to verify, if the desired design goals for the simplified approach of [72] were
achieved. Remember, that Equation 3.25 is only valid for a perfect match on the drain
side of the transistor. Furthermore, the added noise from the output port is minimized,
if the backward gain is reduced. Therefore, the focus of this design was to achieve the
lowest possible S22 and S12, while still achieving noise matching at the gate side. The
measurements were performed with a bias of Vg = 0.0V and Vd = +0.8V, yielding a drain
current of 100mA/mm. As a hot-cold source for the NF measurement, a Noise Com noise
diode was used. As can be seen in Figure 5.5(a), measured and simulated S-parameters are
68
in good agreement. With a measured backward gain of less than −10 dB and a S22 of below
−25 dB at 90GHz, the validity of 3.25 is assured. Figure 5.5(b) shows the comparison of
the NF measurement with simulation. An achieved NF of below 3 dB confirms a successful
noise matching.
75 80 85 90 95 100 105 110
100
150
200
250
300
350
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Frequency [GHz]
Ts1,simp.
Ts1,limit
measurement
ambient
Figure 5.6: Comparison of measured noise temperature with Setup 1 to two simulation methods. The
dotted line indicates ambient temperature.
Figure 5.6 shows, that the measured noise temperature between 85 and 98GHz is only
slightly higher than simulated. Above 100GHz and even more towards 110GHz, it strongly
increases, which coincides with the deteriorating S22 in Figure 5.5(a). This confirms the
relevance of a good S22 for the validity of 3.25. Although a low S12 was achieved to
minimize the backward gain, its influence is yet not negligible: Ts1,limit in Figure 5.6 depicts
the noise temperature that could be achieved with zero backward gain.
The absolute noise temperature achieved with this approach is higher than for the previously
presented ACLs. This is due to the different design goal prioritization, which yields a non
ideal S11. The influence of S11 on the achievable noise temperature was pointed out above.
5.1.3 W-band AHL
The noise temperature of the stand-alone W-band AHL circuit that was intended for inte-
gration into the SCFE, as well as for module assembly, was measured on-wafer with Setup 1.
The input return loss was measured with a VNA, two different port extensions for W- and
D-band and the respective on-wafer probes. The AHL bias was Vg = 0.0V and Vd = 0.8V,
yielding an Id = 384mA/mm.
Figures 5.7 show the comparison between the simulation (red curves) and the measurement
(black curves). Again, Equation 3.24 was used, to simulate the noise temperature. While
a fairly good agreement between simulation and measurement is demonstrated in Figure
5.7(a), the achieved noise temperature in Figure 5.7(b) decreases faster than predicted by
the simulation towards higher frequencies. However, the difference of almost 2500K over
the entire band due to this gradient is not much if compared to noise diodes. We remember,
that as explained in Section 4.2, the ENRdB of the Noise Com noise diode NC5110 varies
between 9 and 20 dB, yielding a noise output flatness of ±5.5 dB. If the maximum (3950K)
69
60 70 80 90 100 110 120 130 140
-35
-30
-25
-20
-15
-10
-5
0
 
Frequency [GHz]
In
pu
t R
et
ur
n 
Lo
ss
 [d
B
]
Simulation
Measurement
(a) Return loss.
75 80 85 90 95 100 105 110 115
0
1000
2000
3000
4000
5000
6000
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Measurement
Simulation
(b) Noise temperature.
Figure 5.7: Comparison of simulated and measured input return loss of the AHL MMIC and noise temper-
ature.
and minimum (1570K) noise temperatures from the measurement shown in Figure 5.7(b)
are used to calculate the achieved ENRdB with Equation 4.4, the resulting ENRdB are
11 dB and 6.4 dB, respectively. This corresponds to a flatness twice as good as for the
mentioned ND. However, the absolute noise temperature of the AHL is considerably lower
than of the ND. The packaging of the AHL MMIC has not much influence on the absolute
noise temperature, as will be shown in the following section.
5.2 Module Measurements
The W-band ACL (chip photograph in Figure 3.17(a), results in Figure 5.4) and the W-band
AHL (chip photograph in Figure 3.17(b), results in Figure 5.7), which were integrated in
the SCFE, were packaged into split-block modules (ACL module 1 and AHL module 1).
Furthermore, the 89GHz ACL V 2 depicted in Figure 3.14(b), which is a variant of the
ACL integrated in the ACSS, was packaged (ACL module 3). As stand-alone modules, their
performance can be further investigated. The influence of the wafer-probes is eliminated and
comparative measurements can be performed at other facilities. Furthermore, the suitability
of the active loads as a calibration load for standard noise figure measurements can be
demonstrated and its noise temperature can be compared to the input port noise of standard
LNAs.
Figure 5.8(a) shows the assembled ACL module 1 with a standard UG-387 waveguide
flange and two subminiature type B (SMB) connectors for the gate and drain voltage supply.
In Figure 5.8(b) the open split-block is depicted and the mounting of the ACL MMIC can be
seen. As for the D-band LNA, the transition to waveguide is realized with a feeding patch
and a microstrip line on a quartz substrate. The DC and RF pads are bonded with gold
wires to the bias supply and the microstrip line, respectively.
First, the input return loss of the ACL module was verified at different bias conditions.
Comparing the result in Figure 5.9 to the on-wafer measurement in Figure 5.4(a) shows,
that the assembly into the split block package has negligible influence on the S11. The
minimum of −15 dB is slightly shifted towards 90GHz and above 100GHz the S11 even
improves compared to the bare MMIC. Overall, the input return loss of over 10 dB between
80 and 105GHz is very appropriate for the intended noise temperature measurements. The
70
(a) Assembled ACL module in a
split-block package.
(b) Open module with ACL MMIC and transition to
waveguide.
Figure 5.8: Photographs of the ACL module 125-001.
75 80 85 90 95 100 105 110
-20,0
-17,5
-15,0
-12,5
-10,0
-7,5
-5,0
 0.0 V Gate, 0.6 V Drain
 0.0 V Gate, 0.8 V Drain
 0.0 V Gate, 1.0 V Drain
 0.1 V Gate, 1.0 V Drain
 
 
Frequency [GHz]
S
 - 
P
ar
am
et
er
s 
[d
B
]
Figure 5.9: Measured input return loss (S11) of the ACL module 125-001 at different bias conditions.
71
influence of the S11 on the noise measurement was previously pointed out and will be
further emphasized in the following. The change of S11 due to bias variations, however, has
negligible influence on the noise measurement compared to the effect this bias change has
on the generated noise in the ACL.
75 80 85 90 95 100 105 110
100
150
200
250
300
350
400
450
500
Waveguide Load
 
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
ACL MMIC
ACL Module
ISS
unbiased ACL Module 
Figure 5.10: Noise temperatures of ACL MMIC and module, measured with Setup 1. As a reference, the
measured noise temperatures of calibration standards were added in the graph.
Figure 5.10 compares the measured ACL on-wafer noise temperature (blue curve) with the
measured noise temperature of the ACL module (orange curve). Both measurements were
performed with Setup 1. The bias for the module was Vg = 0.0V and Vd = 0.6V, yielding
an Id = 200mA/mm which was the optimum for minimal noise. Without bias, the ACL
module exhibits an ambient noise temperature (red curve), similar to waveguide calibration
load (green curve) or the on-wafer measurement of an impedance standard substrate (ISS),
which were included in the graph for comparison. As expected, the losses related to chip
bonding and the waveguide transition increase the noise temperature of the ACL module.
However, subsequent measurements with other setups reveal, that the ACL module seems
to exhibit considerable lower noise.
The black curve in Figure 5.11 describes the same measurement of the ACL module 1
with Setup 1, as the orange curve in Figure 5.10. Measured with Setup 2, however, the same
ACL module with identical bias achieves a noise temperature, which is up to 70K lower.
We remember, that the system noise figure of both measurement setups was determined in
Chapter 4 and found to be almost identical - not explaining this difference (see Figure 4.12).
Furthermore, the front-end of both setups is identical up to the LNA2+Mixer stage (see
Figure 4.1). To preclude a fundamental systematic error in the setup, the ACL module was
characterized one more time with the setup at RPG, which has a slightly higher system noise
figure than Setup 1 and Setup 2 up to 100GHz and similar or better for higher frequencies
(see Figure 4.13). The result of the noise temperature measurement with RPG’s setup is
represented by the blue curve in Figure 5.11 which is located right in the middle between
the other two curves. For now, the difference between the measurements is explained by the
different conversion gain of the LNA2+Mixer stage at the IAF setup and a different total
(conversion) gain of the RPG setup. This projects the measured temperature in different
72
80 90 100 110 120 130 140 150
100
150
200
250
300
350
400
450
500
 
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
ACL Modul 001
Setup 2
Setup 1
Setup RPG
D-band Setup
Figure 5.11: Comparison of measurement results of ACL module 125-001 with four different setups.
dynamic ranges of the NFAs power detector which can have non-linearities. Later in this
chapter, further evidence of the setups influence is derived. A fourth measurement was
performed with the modified D-band setup. The setup, including the ACL as a DUT, con-
nected below the reference plane was shown in Figure 4.15(b). The result is the green curve
in Figure 5.11. It is notable, that the D-band result fits very well to the one achieved with
Setup 2. Except for the first W-band isolator and the NFA, totally different components
were used, as described in Chapter 4.
75 80 85 90 95 100
150
160
170
180
190
200
210
 Vg 0.0 Vd +0.6 Id 4mA
 Vg +0.1 Vd +0.5 Id 6mA
 Vg +0.1 Vd +0.8 Id  8mA
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Figure 5.12: Noise temperature of ACL module 125-003 at different bias, measured with Setup 2.
The bias dependency of ACL module 3 was only characterized with Setup 2 and the mod-
ule results then compared with on-wafer results of the MMIC that is mounted in the module.
Figure 5.12 shows the measured noise temperatures at three different bias conditions, yield-
ing a normalized drain current of 133mA/mm (black curve), 200mA/mm (red curve), and
267mA/mm (blue curve). The bias dependent variation of the noise temperature at the
73
designed target frequency 89GHz is less than 5K, which is impressive and of crucial impor-
tance when using an ACL as a reference load. LNAs are typically more sensitive to gain
variations and the advantage of ACLs over LNAs in this respect will be presented in the next
section. But first, the module measurement of module 3 shall be compared to the on-wafer
results.
75 80 85 90 95 100 105 110
50
100
150
200
250
300
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
ACL Module 003 
(ACL89_V2) Setup 2
ACL89_V2 MMIC Setup 1
ACL89_V2 MMIC Setup 2
ACL89_V1 MMIC Setup Millilab
Figure 5.13: Comparison of measurements with setup 1 and setup 2. The black curve is the result of the
exact same chip that was assembled in ACL module 125 − 003 (ACL 89GHz V2). The chip
characterized by the red curve is identical but from a different wafer. As a comparison a
measurement of ACL 89GHz V1 at Millilab is shown.
From the curves depicted in Figure 5.13 several things can be concluded. The red and
the black curve characterize two different dies of the same MMIC, measured with different
setups. The two measurements were performed with a time period of two months in between.
This indicates, that the principal measurement method is fairly reproducible and that both
setups yield comparable results for on-wafer measurements, even with different setups. On
the other hand it would be expected, that the module exhibits a higher noise temperature
than the MMIC, due to the transition losses, as it was seen in Figure 5.10. However, we
have also seen, that module measurement results of ACLs (we will see later, that this is
not the case for LNA modules) vary strongly, depending on the used setup (see Figure
5.11). For example, the packaged ACL module 1 exhibits a comparable temperature when
measured with Setup 2 (red curve in Figure 5.11), as the MMIC when measured with Setup
1 (blue curve in Figure 5.10). Therefore, it is only at the first glance surprising, that the
ACL module 3 (blue curve in Figure 5.13) exhibits a noise temperature which is up to 30K
lower at 89GHz, than the exact same MMIC that was mounted into the module, measured
with the same setup (black curve). Furthermore, we have seen in Figure 5.2(b), that IAF’s
on-wafer measurements tend to yield more conservative noise temperatures. Since the
89GHz ACL V 2 is merely a variation of 89GHz ACL V 1, adapted for packaging but with
the same ACL core (compare Figure 3.14), the temperature of 89GHz ACL V 1 measured
by Millilab is also included into the graph. If now comparing Millilab’s on-wafer result with
the module measurement with Setup 2, the increased noise temperature due to packaging
losses can be explained conclusively.
This comparison is only done to demonstrate some essential findings of this work: Module
74
measurements can not be compared with on-wafer measurements, even when performed
with the same setup. One reason is, that the calibration is done without the wafer probe and
only the scalar loss of the probe is accounted for after the calibration. As was demonstrated
with the previous results in this chapter, the ACL is very sensitive to the impedance it is
connected too. And as explained in Chapter 4, the correct way to account for the wafer
probe’s loss would be to calculate its available gain with Equation 4.7, which is dependent on
its matching and the source reflection. But the return loss of the probe on wafer side (Figure
4.8(a)) is about 10 dB worse, than of the isolator 4.7(a). Therefore, the method to account
for the probe loss has only limited success and a comparison with module measurements is
to be done with caution.
For the same reason, the Millilab on-wafer setup yields different noise temperatures than
IAF’s on-wafer setups (different probe), while Setup 1 and Setup 2 yield comparable results
(same wafer probe).
75 80 85 90 95 100 105 110
275
300
325
1500
2000
2500
3000
3500
4000
4500
5000  AHL module /w bias
 AHL MMIC /w bias
 AHL module w/o bias
 Waveguide Matched Load
 Impedance Standard Substrate
 AHL module @ RPG
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Figure 5.14: Noise temperatures of AHL MMIC and module, measured with Setup 1. As a reference, the
measured noise temperatures of calibration standards were added in the graph.
The AHL module noise temperature was measured with Setup 1 and compared to the on-
wafer measurement with the same setup. This time, for the module a lower noise temperature
would be expected than for the on-wafer result, because losses due to packaging would pull
the noise temperature towards the ambient temperature. But again, this comparison needs
to be done carefully. In Figure 5.14 the measured on-wafer temperature (red curve) and the
module temperature (black curve) yield very similar results. It must be stated, that even
a conservatively assumed loss of 1 dB associated with the packaging would merely yield a
change of 75K in the noise temperature, which would be difficult to identify in relation to the
absolute temperatures of the AHL. Again, as a reference, the measured noise temperatures
of an ISS, a calibration waveguide load and the AHL without bias are added in the graph.
Furthermore, the measurement of the AHL module performed at RPG is included (blue curve).
Not surprisingly anymore, it yields a different result, which can again be explained with the
different dynamic power range in which different setups project the noise temperature, and
the non-linearity associated with that. The dip at 98GHz must be associated with the
75
assembly of the module. Since AHL module 1 was the only module assembled, this effect
could not be investigated further.
5.2.1 Comparison of Active Load Modules to Conventional LNAs
as a Load
Since the design approach for an ACL is very similar to the design of an LNA, it appears
obvious to measure the noise temperature at the input of an LNA in the same way as the
ACL.
75 80 85 90 95 100 105 110
100
150
200
250
300
350
400
450
500
 
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
ACL Module
LNA Module
Figure 5.15: Comparison of an LNA module, measured at its input as a load, versus the ACL module 1.
Measurement performed at RPG.
At RPG a double-LNA with around 40 dB small-signal gain was measured at its input
port with the same setup as the ACL module 1. The LNA module contained an MMIC
manufactured at IAF and packaged at RPG. For the measurement the output port of the
LNA was terminated with a matched waveguide load. The LNA measurement result is
depicted in Figure 5.15 for three bias states: 1.0V (cyan curve), 1.2V (black curve), and
1.5V (green curve). It was also measured with voltages above, but while the increase up
to 1.5V yielded a reduction of noise temperature, it increased again for higher voltages.
Therefore the other results are omitted in the graph. The ACL module was also measured in
two different bias states, with Vd = 0.6V, yielding an Id = 200mA/mm (blue curve), and
Vd = 0.8V, yielding an Id = 266mA/mm (red curve). The ACL module exhibits notable
better noise temperature, than the LNA and more important, the ACL module is negligibly
bias dependent compared to the LNA, whose noise temperature strongly depends on the
bias. Furthermore, the effect of non-optimal input return loss of an LNA at the expense of
good NF can be seen: The LNA noise temperatures are strongly oscillating due to reflected
noise power between LNA and radiometer input. This demonstrates the main advantages
of ACLs compared to LNAs. They are much less sensitive to bias variations and, that they
obviously achieve much lower noise temperatures than standard LNAs typically do.
After this first initial comparison, further IAF-manufactured LNAs were measured with
Setup 1 and the results are shown in Figure 5.16. Since the modules include a bias control
76
75 80 85 90 95 100 105 110
100
150
200
250
300
350
400
450
500
ACL 1
LNA 346
LNA 218
 
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
ACL 3
LNA 355
Figure 5.16: Comparison of three LNA modules, measured at their input as a load, versus the ACL modules
125-001 and 125-003. Measurements performed with Setup 2.
circuit, the exact voltages applied to the MMICs are not known. LNA 218 and LNA 355
posses a small-signal gain of around 20 dB and a NF between 3 and 4 dB. LNA 346 is a
double-LNA with around 45 dB gain and 4 dB NF. Its exact S-Parameter and NF are shown in
Figure 5.17(a) and will be discussed shortly. The orange and blue curves show the measured
noise temperature of ACL module 1 and ACL module 3, respectively. It can be seen, that
standard LNAs exhibit a noise temperature around ambient or above at their input port.
Furthermore, they exhibit a more or less strong ripple over the frequency range. The ACLs,
in contrast, demonstrate a very continuous gradient of the noise temperature. Also, as
we have seen before, they achieve a lower noise temperature. Merely LNA 346 achieves a
comparable noise temperature above 94GHz, but has an notable peak at 95GHz. Therefore,
it is investigated in more detail.
80 85 90 95 100
-25
-20
-15
-10
-5
0
5
40
50
Frequency [GHz]
S
 - 
P
ar
am
et
er
 [d
B
]
S21
S11
S22
NF
(a) Measured S-parameters and noise figure of
LNA module 346.
75 80 85 90 95 100 105
100
150
200
250
300
350
400
450
 
Frequency [GHz]
Te
m
pe
ra
tu
re
 [K
]
One-port noise temperature
Two-port noise temperature
-30
-25
-20
-15
-10
-5
0
5
S
 11 [dB
]
(b) S11 versus one-port noise temperature mea-
surement and standard noise measurement.
Figure 5.17: Investigation of LNA module 346.
Figure 5.17(a) depicts the S-parameters and NF of LNA 346. It can be seen, that for
LNA design it is much simpler to achieve a good output matching (S22), than a good input
matching (S11), because the impedance transformation for optimum noise is different than
77
for optimum power matching and a compromise needs to be chosen. If the LNA is used
as a first stage in a receiver the reduced input matching is not too critical, as long as the
LNA has enough gain. Typically the reflected power at its input does not affect the system
performance. In exchange, prioritization of the optimum noise matching yields a flat NF
below 4 dB (green curve). If this standard two-port measured NF is transformed into a noise
temperature, using Equation 3.1, it can be plotted together with the one-port measured
noise temperature at its input. In Figure 5.17(b), the noise temperature is depicted on the
left scale and the S11 on the right scale. Now, the cause for the noise peak at 95GHz
can clearly be identified. It strongly correlates with the deteriorating S11 in the frequency
range 90 − 97GHz, while the noise temperature is very low for a S11 below −15 dB. The
other notable finding is, that the two-port NF does not correlate with the one-port noise
temperature. This, one more time, emphasizes how important the input matching is for
the one-port measurement of the noise temperature. While for LNA design the goal is to
minimize the noise figure - which characterizes the amount of noise added by the device to
the input signal at the output of the device - the goal of the ACL must be clearly different.
The absolute noise temperature at the input is the relevant quantity and furthermore an
extremely good input matching is mandatory. This clearly distinguishes ACLs from LNAs
(at their input port) and justifies dedicated ACL designs. Only in one respect LNAs seem to
have an advantage over ACL modules, which is the robustness against reflections from the
measurement system. This is presented in the following.
75 80 85 90 95 100 105 110
100
150
200
250
300
350
400
450
500
LNA Modul 346
Setup 1 + Setup 2
 
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
ACL Modul 001 LNA Modul 355
Setup 1 + Setup 2
Setup 2
Setup 1
Setup RPG
Figure 5.18: Comparison of two LNA modules, measured at their input as a load, versus the ACL module
125-001.
Since it was previously shown, that especially module measurements can yield different
results with different setups, two of the LNAs were also measured with Setup 1 and compared
to all the previous measurements of ACL module 1. In Figure 5.18 all the results are
summarized. While the explanation for discrepant results of different setups was based on
the non-linearity of the power detector inside the NFA and different dynamic ranges, this
measurements contradicts the previous argumentation. Obviously, only the ACL modules are
susceptible to changes in the setup, while the LNA modules are very robust in this respect.
We recall, that the only difference between Setup 1 and Setup 2 is the different receiver
78
module. The isolators and LNA 1 (see Figure 4.1), as well as the utilized instruments and
calibration procedure are identical. Therefore, the only explainable source of this discrepancy
is that some reflection at the mixer must leak back to the DUT, all the way thru two isolators
and one LNA. This influence is obviously frequency dependent. The three noise curves of
the ACL intersect at 80GHz at around ambient temperature, while they diverge towards
higher frequencies and lower noise temperatures. The latter leads to believe, that this effect
is only seen at lower noise temperatures and therefore just hidden above ambient. But the
two curves for LNA 346 are very similar up to 100GHz, where they also demonstrate a very
low noise temperature, and only then slightly diverge. A similar divergence above 100GHz
is also observed for LNA 355.
This leads to the conclusion, that the ACL modules are more sensitive to mismatches and
reflections in the setup, than LNAs.
5.2.2 ACL Module as Hot-Cold Load in Noise Measurements
As an alternative to standard noise measurements with a noise diode or with an antenna
pointed at heated and cooled absorber, the ACL module can be used. When noise diode
manufacturers deliver their calibrated diodes, they provide the associated ENR table. Since
noise diodes have a moderate return loss, the manufacturers or suppliers only deliver the
diodes with an isolator at its output and calibrate the combination of both components. The
ENR table is then only valid for this combination. Similar to this procedure it is necessary
to thoroughly characterize the ACL module with an isolator, before using this combination
as a hot-cold load at the input of the DUT. The measured cold noise temperature of ACL
module and isolator is then used to calculate the ENR of this combination. The hot state
for the ENR calculation is when the ACL is switched off, the other way round as for ND. Of
course the isolator reduces the ENR, but it is necessary, because in the off state of the ACL,
its return loss is not defined and what the DUT sees, is the ambient noise of the isolator.
80 85 90 95 100 105
4
5
6
7
8
 Noise Diode
 ACL
 LN
 
 
Frequency [GHz]
N
oi
se
 F
ig
ur
e 
[d
B
]
Figure 5.19: The noise figure of an amplifier was measured with three different methods: Noise diode,
antenna with hot-cold (LN) load, and ACL in on and off state.
To validate this measurement method, an amplifier was chosen as a DUT and its noise
figure measured. Therefore the combination of ACL and isolator was connected as a hot-
79
cold load to the input of the DUT while the output was connected to the NF measurement
setup. With the known ENR of the hot-cold load, the NF of the DUT can be calculated with
the Y-method. The same DUT was then measured with standard methods and the results
are compared in Figure 5.19. When using the ND, one LNA had to be removed from the
measurement setup, to avoid saturation of measurement system. This, again, demonstrates
the advantage of an ACL for calibration of very sensitive receivers with high gain, as they are
used in radiometers. Typically, the noise behavior of an amplifier changes continuously over
frequency. Due to the ripple on the red curve (ND) and the blue curve (ACL), which are
typical for noise power reflections, it is believed that the result achieved with the antenna
and LN is the most accurate of the three. This can be explained with the measured return
losses in Figure 4.7, where the antenna has an even better matching than the isolator in
front of the ACL, thus reducing these reflections. The results of the measurement performed
with the ACL as a hot-cold load and the LN measurement agree very well. Even the results
achieved with the ND are comparable, diverging only in the center of the frequency range
by less than 1 dB. This impressively demonstrates the suitability of an ACL module as a
hot-cold load for noise measurements and even points out its advantages compared to NDs.
5.2.3 Temperature Dependence of ACLs
When characterizing the sensitivity of the ACL to external physical temperature changes, the
entire W-band was measured at each temperature step. For clarity sake only the results at
the center frequency are shown. However, it should be noted that since the performance of
the ACL is frequency dependent, the sensitivity to temperature changes as well.
-40 -20 0 20 40 60 80
150
175
200
225
250
275
300
325
350
 
 
Physical Temperature [°C]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
 Passive Load at Physical Temperature
 Passive Load with Offset
 ACL-Modul
@ 94 GHz
Figure 5.20: Measured noise temperature of ACL module at different physical temperatures compared to
the theoretical value of a passive matched load. Measured with Setup 1.
The solid line in Figure 5.20 shows the noise temperature, which a passive matched load
would exhibit at the accordant physical temperature. The slope of the curve is 1K/◦C. For
a better comparison to the sensitivity of the ACL, an offset was subtracted from this curve
which corresponds to the difference between a passive load and the presented ACL at ambient
temperature. In comparison to the dashed offset curve it can be observed that for physical
80
temperatures below ambient the ACL is less sensitive to temperature changes than a passive
load, while for temperatures above ambient a similar behavior as for a passive load can be
observed. While previous publications only investigated the temperature dependence above
0 ◦C ([34], [65]), this is the first work to analyze the ACL behavior below zero. If in accordance
to previous work, a linear fitting curve is applied only for the measurement points above
0 ◦C, the slope of the curve is 0.9K/◦C. If in a similar fashion the measurement points below
ambient temperature are fitted, an average dependence of 0.5K/◦C can be observed. Finally,
if only measurements below zero are considered, a dependence of 0.36K/◦C can be derived.
In the application scenario of an ACL as a calibration reference, however, such extreme
temperature changes are not expected, since these systems are typically very accurately
temperature stabilized. However, it is noteworthy, that the noise temperature of an ACL,
being an active device, does not increase more than a passive load, when being heated.
5.3 Active Loads with Integrated Switches
5.3.1 Switching Calibration Front-End
One of the critical parameters of the SCFE depicted in Figure 3.20 is the insertion loss
between the internal loads and the output port, which will deteriorate the achieved high
or low temperature of the respective load. Due to the integrated design, it is however
not possible to measure this loss directly. Therefore one of the through paths must be
characterized and the loss added by the additional transmission line, to connect to the ports
at the left edge, must be subtracted. The necessity of this layout was explained in Chapter
3. Since the two input channels of the SFCE are designed symmetrically, only the upper
branch was characterized.
75 80 85 90 95 100 105 110
-35
-30
-25
-20
-15
-10
-5
0
S21, Isolation
 
 
Frequency [GHz]
S
 - 
P
ar
am
et
er
 [d
B
] S22, Isolation
S21, Through
S11, Through
S22, Through
Figure 5.21: S-parameters between input port 1 and output port for switch position to input port 1 (through)
and to internal matched load (isolation).
In Figure 5.21 S-parameter measurement results are shown, where port 1 is the input port
and port 2 is the output port (compare with Figure 3.20). In through position all branches
are closed (Vsw = +0.1V), except for the upper input signal path (Vsw = −0.5V). An
81
insertion loss between 4 and 5 dB is measured within the bandwidth of 80− 100GHz. The
line length difference of 1700µm between the path from the internal loads and the path
from input to the switch core yields a simulated added loss of −0.7 dB. This results in an
estimated insertion loss of 3.3 dB at the center frequency for the two internal paths to the
active loads. To measure the isolation of the SCFE, the output port is routed to the internal
matched load and S-parameters are measured at the same ports as above. An isolation of
more than −25 dB is measured within the entire W-band. Furthermore, the input return
loss is very good in all switching states of the SCFE. Even though the S11 in the through
position is measured at the input port 1 (black curve), it gives a good indication of the
internal matching towards the internal loads. As previously shown, the very good matching
of less than −15 dB will allow a good performance of the internal loads.
75 80 85 90 95 100 105 110
175
200
225
250
275
300
325
350
375
400
 
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
 internal matched passive load
 impedance standard substrate
 ACL w/o bias
 ACL  Vg +0.1 Vd +0.7
Figure 5.22: Noise temperatures in switch positions ACL and passive load, compared to standard substrate
load and ACL without bias.
For the characterization of the cold reference on the SCFE, the SP5T switch was set to
the accordant position. The ACL was biased with Vg = +0.1V and Vd = +0.7V, yielding
a drain current of 173mA/mm at which the lowest noise temperature was achieved. The
result is shown in Figure 5.22 (blue curve), compared to measurement of an ISS (green
curve). Furthermore, the noise temperature of the SCFE was measured when the SP5T is
switched to the internal matched load (red curve) and switched to the ACL at zero bias. For
the last two states it must be noted, that it can not be differentiated if it is really the noise
temperature of the passive load, of the ACL without bias or the switch itself, since all are
at ambient temperature. For frequencies above 90GHz, the SCFE achieves a mean noise
temperature of 220K in the cold position. Compared with the measured temperature of
150K of the stand-alone ACL, the increase of 70K is due to the losses of the SP5T switch.
The temperature change, due to a resistive loss is described by Equation 5.1.
T2 = T1·Llin + T0· (1− Llin) (5.1)
For example, the cold temperature of the ACL before the switch would be T1 and Llin
the linear loss of the switch at an ambient temperature T0, then the cold temperature after
the switch would increase to T2. This of course also applies for a temperature T1 above T0,
82
e.g. for a hot load, then the resulting temperature is pulled down by the switch loss and T2
decreases.
LdB = 10· log10(Llin) = 10· log10
(
T2 − T0
T1 − T0
)
(5.2)
If the measured noise temperatures at 95GHz of the stand-alone ACL MMIC of 140K
(see Figure 5.4(b)) and of the SCFE of 225K (see Figure 5.22) are inserted into Equation
5.2 as T1 and T2, respectively, the resulting loss yields −3.35 dB. This is very close to the
insertion loss, derived above from the through measurement by de-embedding the simulated
line loss.
Unlike the AHL, the ACL is very insensitive to bias variations. The difference of the highest
and the lowest achievable noise temperature with extreme bias conditions is less than 20K
within the entire operation bandwidth. Figure 5.23 shows the achieved noise temperatures
at different bias.
85 90 95 100 105 110
200
210
220
230
240
250
260
270
280
 Vg +0.1 Vd +0.6 Id 154 mA/mm
 Vg +0.1 Vd +0.8 Id 193 mA/mm
 Vg +0.1 Vd +0.7 Id 173 mA/mm
 Vg +0.1 Vd +1.0 Id 233 mA/mm
 Vg 0.0 Vd +1.0 Id 140 mA/mm
 Vg 0.0 Vd +0.8 Id 109 mA/mm
 Vg 0.0 Vd +0.5 Id 71 mA/mm
 
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Figure 5.23: Noise temperatures in switch positions ACL at different ACL bias.
The hot reference on the SCFE was characterized accordingly, by routing the signal from
the AHL to the output. In contrast to the ACL, however, it is possible to achieve different
levels of noise temperature as a function of transistor biasing (see Figure 5.24). Again, the
originally higher noise levels of the stand-alone AHL are reduced by losses of the SP5T,
according to Equation 5.1. Nevertheless, noise temperatures as high as 1750K at 94GHz
were achieved, measured at the output of the SCFE. Again measurements of an ISS and the
SCFE switched to the internal load are included in the graph, as well as the temperature
with zero bias AHL.
If the achievable ENR of the SCFE is calculated applying Equation 4.4, with 1700K for the
hot and 225K for the cold state at 95GHz, this yields an ENR of 7 dB. This is much higher,
than an ENR achieved only with a cold and an ambient load and confirms the advantage of
a SCFE for hot-cold calibration. A novel application of such a device could be to use it as
a wafer-probe reference to calibrate a noise measurement setup. As described in Chapter 4,
the state-of-the-art is to calibrate such a setup without the wafer-probe and then to account
83
75 80 85 90 95 100 105 110
290
300
310
750
1000
1250
1500
1750
2000
2250
 
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
 Vg +0.25 Vd +1.0 Id 393 mA/mm
 Vg +0.2 Vd +1.0 Id 326 mA/mm
 Vg +0.25 Vd +0.8 Id 350 mA/mm
 Vg +0.15 Vd +0.8 Id 224 mA/mm
 Vg +0.1 Vd +0.8 Id 172 mA/mm
 Vg +0.05 Vd +0.8 Id 129 mA/mm
 Vg +0.0 Vd +0.8 Id 95 mA/mm
 AHL w/o bias
 impedance standard substrate
 internal matched passive load
Figure 5.24: Noise temperatures in switch positions AHL at different AHL bias, compared to standard
substrate load and AHL without bias.
for the loss afterward. The error introduced with this approach was discussed earlier in this
chapter. With the SCFE presented in this work an on-wafer calibration of the setup could be
performed. Reflections due the non-ideal matching of the wafer-probe would be accounted
for during the calibration, yielding more reliable results. Furthermore, the variable hot noise
temperature of the SCFE at different bias states of the AHL allows to extend the number of
reference loads above three and to improve the calibration accuracy even more. However, it
remains the necessity to analyze the influence of biasing and of temperature changes on the
stability of the AHL and to calibrate the SCFE in the first place to calculate an ENR table.
5.3.2 Active Calibration Sub-System
The two versions of the ACSS presented in Chapter 3 were characterized similar as the
SCFE. The S-parameters were measured at Millilab and on-wafer noise measurement were
measured with Setup 2 at IAF and with the Millilab setup. Figures 5.25 and 5.26 depict the
S-parameters in different switching states for ACSS V 1 and ACSS V 2, respectively.
Since the SP3T of both versions is identical, the measured S11 and S21 for the through
position are very similar, as well as the S21 for isolation, when the switch is set to the internal
matched load. The S22 describes the matching at the output of the ACSS, which will be
connected to the radiometer. It changes, depending on the switch position of the SP3T. The
better matching of 89GHz ACL V 4 over 89GHz ACL V 1 at the design frequency (compare
Figures 5.1(a) and 5.2(a)) also reflects in the ACSS results. In both cases the matching
when switched to the ACL (orange curve) is better at 89GHz, than when switched to the
matched load (green curve) or the through path (blue curve). For ACSS V 2, an ideal match
of −25 dB was achieved at design frequency for the ACL switch position.
Figure 5.27 compares the achieved noise temperatures of ACSS V 1 and V 2, measured
with the Millilab setup. As for the stand-alone MMICs, the first version achieved a lower
noise temperature, while the prioritized design goal for the second version was the improved
return loss. It can be observed, that the difference in noise temperature between the two
ACSS versions is less, than between the respective stand-alone ACLs. This is explained by
84
75 80 85 90 95 100 105 110
-50
-40
-30
-20
-10
0
S21, Isolation (ML)
 
Frequency [GHz]
S
 - 
P
ar
am
et
er
 [d
B
]
S22, ML
S21, Through
S11, Through
S22, Through
S22, ACL
Figure 5.25: S-parameters of ACSS V 1 in different switch positions.
75 80 85 90 95 100 105 110
-50
-40
-30
-20
-10
0
S21, Isolation (ML)
 
Frequency [GHz]
S
 - 
P
ar
am
et
er
 [d
B
]
S22, ML
S21, Through
S11, Through
S22, Through
S22, ACL2
Figure 5.26: S-parameters of ACSS V 2 in different switch positions.
85
75 80 85 90 95 100
150
175
200
225
250
275
300
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
ACSS V1
ACSS V2
Figure 5.27: Achieved noise temperatures of ACSS V 1 and ACSS V 2 in switch position ACL, measured at
Millilab.
the loss of the SP3T, which has a relatively higher impact on the lower noise temperature,
than on the higher noise temperature ACL (compare Equation 5.1), pulling both towards
ambient temperature.
75 80 85 90 95 100
175
200
225
250
275
300
325
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Passive matched load
Switch voltage 0.6 V
Switch voltages 0.8 V and 1.0 V
ACL
Figure 5.28: Noise temperatures of ACSS V 1 in switch positions ACL and passive matched load, measured
with Setup 2 at IAF. If the switch is not supplied with the appropriate bias, the isolation
breaks down and noise from the matched load leaks into the ACSS output.
Since the performance of the SP3T is crucial for the application of the ACSS, its influence
on the achieved noise temperature in different switch positions was further analyzed. The
bias threshold of the switch was tested and found to be around +0.7V. For any voltage
below this threshold, the FET switches does not shorten the signal to ground and the basic
functionality of the SP3T is corrupted. The result is, that noise power from all three input
channels is superposed at the star point and routed to the output of the ACSS. This effect
is shown in Figure 5.28 for ACSS V 1. The switches are set to route the ACL signal to the
output and thus isolating the other two channels. The ACL is biased with Vg = −0.2V and
86
Vd = +0.6V. For a switch bias of +0.6V the noise power of internal matched load and the
input port leak into the output , increasing the measured noise temperature (black curve).
For any switch voltage above the threshold, the isolation of the switch is identical, which
is demonstrated in Figure 5.28 with +0.8V and +1.0V (green curves). This confirms the
very important property of the designed SPxT switches, that their performance is not bias
sensitive and thus not introducing an additional error source for the temperature references.
The good isolation of the SP3T is also confirmed, when the switches are set to route the
signal to the internal matched load. In this state, the noise temperature from the ACL must
be isolated from the output. In Figure 5.28, the measured noise temperature at the output
in this state is shown, while during one measurement the ACL was biased (orange curve)
and during another measurement not biased (red curve). Both curves are very identical,
proving that the internal low noise temperature of the ACL around 150K does not influence
the output noise temperature, due to good isolation.
It is notable, that the on-wafer measurements of ACSS show almost no difference between
results achieved with the Millilab setup (Figure 5.27) and with the IAF setup (Figure 5.28).
This indicates, that the loss of the switch in front of the ACL isolates the ACL from the
output, making it less sensitive to the impedance of the setup. Furthermore, as for the
characterization of the stand-alone ACL in 5.2(b), the comparison shows that the measures
taken to reduce noise reflections in the setup, yield a smaller ripple on the temperatures
measured with Setup 2.
In the course of the Calload project, ACSS V 1 was selected for packaging due to the
slightly better noise performance. Some tuning with bond wires during the assembly was
done to improve the matching at the center frequency. The performance was good and a
noise temperature of 227K was achieved for the module, which is again equal to the value
in on-wafer measurements. This is the same effect that was discussed in this chapter. The
long-term stability of the ACSS was characterized with the highly temperature stabilized
test-bed at DA-Design and was also very good.
5.3.3 Active Hot-Cold Load
The noise temperature measurements of the AHCL in cold and hot load switch biasing are
shown in Figure 5.29. As for the two previously presented MMICs with integrated ACLs
and switches, the loss of the SP3T increases the cold temperature and decreases the hot
temperature, compared to the stand-alone circuit. For the simulation results shown in Figure
5.29, this loss was accounted for, by inserting the simulated SP3T loss and the measured noise
temperature of the stand-alone circuit into Equation 5.1. Since the stand-alone ACL test-
structure is measured with RF-pad, which has an insertion loss of about 0.3 dB, and which is
not employed in the AHCL, this value can be subtracted from the switch loss. This yields an
effective insertion loss of 1.7 to 2.7 dB. Thus, at 93GHz where the stand-alone ACL provides
a temperature of about 190K, an AHCL cold temperature of 225 to 240K can be expected.
This is in agreement with the measured AHCL cold temperature, as can be seen in 5.29. In
the frequency range of 91 to 104GHz the AHCL cold temperature is lower than 250K. The
measured AHCL hot temperature has its maximum at 91GHz with a temperature of 860K.
When comparing measurement and simulation, it can be stated that the simulation was
able to predict the AHCL temperatures at its center frequency. The measured noise power
87
bandwidth is not as large as the simulated, which can be explained by small frequency-
dependent impedance mismatches between the ACL, SP3T and the additional switching
periphery, which in total lead to a noticeable deviation of measurement and simulation. The
AHCL has a power dissipation of 3.2mW, which is the same value as obtained with the ACL
test structure. This very low power consumption can only be accomplished due to the fact
that only one noise source is employed and that the switches have a negligible DC power
consumption.
75 80 85 90 95 100 105 110
200
250
300
600
700
800
900
N
oi
se
 T
em
pe
ra
tu
re
 (K
)
Frequency (GHz)
Ts1,simp.hot
hot meas.
cold meas.Ts1,simp.cold
ambient
Figure 5.29: Measurement versus simulation of the AHCL’s hot and cold noise temperatures. The dotted
line indicates ambient temperature.
Summary and Conclusion
Designing the different active load circuits based on different simulation approaches, allowed
to verify the suitability of each simulation method. It was shown, that the noise temperature
discrepancies between simulation and measurement are also due to utilized measurement
setup. Nevertheless, none of the simulation methods was found to be particularly superior or
inferior to the others in terms of accuracy. However, the novel method presented in this work,
has a significant advantage in terms of shorter design flow and faster numerical optimizer
possibilities.
Furthermore, ambiguous noise temperature results were presented, but by cross-checking
the measurements with many different setups, a fairly good understanding of the effects
influencing one-port noise measurements was gained. It was found, that it is not sufficient
to rely on one particular setup and that especially for short-term on-wafer and module mea-
surements only an achieved temperature range can be derived from the actual measurement
results. Overall, state-of-the art noise temperatures were achieved for all presented active
loads, even if the measurement setup yielding the higher result was considered. Also, the
W-band ACLs presented in this chapter are the first active loads at such high frequencies.
A comparison to previously published stand-alone ACLs will be given at the end of the next
chapter, after the D-band ACL results were presented.
After the difference between LNA and ACL designs was pointed out, packaged ACLs were
characterized and compared to LNAs as a load. With the result, that not only the suitability
88
of an ACL as a calibration reference was demonstrated, but also its advantages compared
the conventional loads were emphasized. It was shown that an ACL module can be used to
perform noise measurements on LNAs, yielding comparable results as with noise diodes. The
behavior of an ACL for below zero ambient temperatures was investigated for the first time
and the results compared to previous work which only analyzed the temperature dependency
for above zero.
Since this work is the first one to present an active hot load based on a FET, it was possible
to integrate cold, ambient and hot loads into integrated switching calibration front-ends.
Different implementations of such front-ends were developed and characterized. If the noise
performance due to the switch losses and the herewith associated temperature increase is
suitable for future space mission instrumentation has to be investigated in future work. But
no similar highly integrated calibration front-ends based on MMIC technology are known up
to date and a patent was granted [78].
It will be demonstrated in the following chapter, that the state-of-the-art noise performance
of the presented ACLs in this chapter and the concept of integrating FET switches, is not
only limited up to W-band frequencies, but also outstanding results can be achieved even up
to 150GHz.
89

6 Characterization of D-Band Active
Load MMICs
Similar to the presented results in the previous chapter, the designed D-band active load
circuits were characterized with the on-wafer measurement setup described in Section 4.4.
None of the D-band active load MMICs were assembled in a package, therefore the results are
limited to on-wafer measurements. The designed D-band LNA necessary for the measurement
setup, however, was of course mounted into a module, as depicted in Figure 3.27. Therefore,
S-parameter and noise figure measurements of the LNA module 667 are presented, in addition
to on-wafer results of the mounted MMIC which are compared to the simulations.
6.1 LNA Module for Measurement Setup
Just as all the developed active loads, the LNA was designed using IAF’s proprietary model
library, which, at the time of the LNA design, was based on extractions from measurements
up to 110GHz. The results of the S-parameters and the noise figure simulated in ADS
are shown in Figure 6.1 (solid lines) and compared to on-wafer S-parameter measurements
(dashed lines). The applied voltages for the measurement were −0.1V at the gate and 0.8V
at the drain, yielding a drain current density of 430mA/mm. This results in a total DC
power consumption of 41.3 mW. The simulation shows a very flat gain of around 23 dB
over the entire D-band, whereas the measurement result exhibits a slowly decreasing gradient
from about 26 dB at 110GHz to 16 dB at 170GHz. Also, the input and output return losses
show some discrepancy between simulation and measurement. These inconsistencies might
be due to the limitation of the modeling data up to 110GHz. However a good matching of
below -10 dB was achieved over the bigger part of the D-band.
More important for the purpose of this work, however, is the performance of the LNA mod-
ule, since especially the noise figure and the input matching of this component are ultimately
responsible for the evaluated measurement setup described in Section 4.4. Remember, that
the chip layout and packaging explained in Section 3.5 allows to bias the first two transis-
tor stages and the subsequent two stages with different drain voltages. The programmable
DC-control board was set to supply +0.8V at stage 1 and 2 and +1.0V at stage 3 and 4,
while all gates are biased with −0.1V. Figure 6.2 shows the measured S-parameters and
noise figure of LNA module 667. It exhibits a small-signal gain (S21) of far above 20 dB in
a frequency range of 50GHz and covers the entire D-band with a linear gain of more than
15 dB. Input- and output return losses (S11 and S22) of the LNA are better than −5 dB
in the entire band and mostly below −10 dB. This makes it ideal for the purpose of any
kind of broadband measurement setup. The gradient in the gain can be observed again and
is identical to the on-wafer measurement. Above 150GHz, the performance of the LNA in
terms of small-signal gain and S11 deteriorates, which is not critical, because the frequency
91
110 120 130 140 150 160 170
-30
-20
-10
0
10
20
30
 
Frequency [GHz]
S
 - 
P
ar
am
et
er
 / 
N
oi
se
 F
ig
ur
e 
[d
B
]
Simulated Noise Figure
 
S21
 
S11
S22
Figure 6.1: S-Parameter of D-band LNA MMIC.
110 120 130 140 150 160 170
-30
-20
-10
0
10
20
30
 
Frequency [GHz]
S
 - 
P
ar
am
et
er
 / 
N
oi
se
 F
ig
ur
e 
[d
B
]
Measured Noise Figure
 
S21
 
S11
S22
Figure 6.2: Measured S-Parameters and Noise Figure of D-band LNA module 667.
92
range of the measurement setup is limited by the used mixer. Losses due to the packaging
and the waveguide transitions are compensated by the modified operating voltages. The
noise figure measurement was performed with a re-calibrated ISSN-06 noise diode from Elva.
The setup allowed measurements in the frequency range from 113− 150GHz, again limited
by the mixer, where the LNA module achieves a state of the art noise figure of below 5 dB
and even below 4 dB between 115 and 135GHz. This validates the usability of this LNA
module for the dedicated one-port noise measurement setup even in a much wider frequency
range, than was finally achieved due to the limitations described in Section 4.4. In the
relevant frequency range for the following ACL characterization, a slightly lower noise figure
than 5 dB is measured, which is consistent with the achieved system noise figure of the
measurement setup around 6 dB (see Figure 4.16), that includes the additional loss of the
isolator at the first stage. Furthermore, the input return loss (S11) is best in the frequency
range 140− 150GHz, which is the range where the one-port noise measurement setup has
the smallest ripple and therefore the results presented in the following are the most reliable.
Even though the expected input power at the first stage LNA module is extremely low, it
should be considered, that radiometers or one-port noise measurement systems are cascading
several high gain LNA stages and that a linear behavior of all stages must be assured. There-
fore, it is important to know the large signal characteristics of the LNA when dimensioning
the measurement system to provide enough dynamic range. Linearity measurements were
performed at the center frequency of 140GHz. Figure 6.3 shows the measured output power
and the calculated gain for input power from −25 dBm to −11 dBm. The 1-dB compression
point is located at −22 dBm input power, which for the expected noise power to be measured
gives enough headroom even for the second stage LNA.
-26 -24 -22 -20 -18 -16 -14 -12 -10
0
5
10
15
20
25
30
35
 
Input Power [dBm] @ 140 GHz
G
ai
n 
[d
B
] @
 1
40
 G
H
z
-2
-1
0
1
2
3
4
5
O
utput P
ow
er [dB
m
] @
 140 G
H
z 
Figure 6.3: Measured output power and gain of the D-band LNA module 667.
Compared to other published work on LNAs in various technologies, this work achieves
some outstanding figures of merit (FOM). Figure 6.4(a) shows the gain-bandwidth product
of several published LNAs at their center frequencies. The gain-bandwidth product of the
LNA of this work is marked with the red star. Since this figure presents the absolute
bandwidth, similar FOM as in this work were achieved at higher frequencies. In D-band,
however, the LNA of this work has a very prominent position in the FOM, especially because
93
it covers the entire band with a gain of around 20 dB. Figure 6.4(b) shows the published
noise figures of LNAs in several semiconductor technologies. Note that accurate noise figure
measurements at these frequencies are challenging and not all published work can present
their noise figures. Also, most of the noise figures in Figure 6.4(b) were measured on-wafer,
neglecting the degradation associated with waveguide packaging. But most important, the
other LNAs in this second FOM have typically a very narrow bandwidth, because their low-
noise performance is optimized for a certain frequency region, depending on the intended
application.
50 100 150 200 250 300 350 400 450
0
100
200
300
400
500
600
700
800
900
1000
1100
1200
1300
 
 
 CMOS
 InP / mHEMT
 SiGe HBT
 IAF mHEMT
 This work
G
ai
n*
B
an
dw
id
th
  [
dB
*G
H
z]
Frequency [GHz]
(a) Product of gain and bandwidth of published LNAs.
50 100 150 200 250
0
2
4
6
8
10
12
 
 
 CMOS
 InP / mHEMT
 SiGe HBT
 IAF mHEMT
 This work
N
oi
se
 F
ig
ur
e 
[d
B
]
Frequency [GHz]
(b) Noise figures of published LNAs.
Figure 6.4: State-of-the art for LNAs.
6.2 Active Cold Load MMIC in D-Band
Having the LNA presented in the previous section available, the D-band ACL has been
characterized using the setup described in Section 4.4. Due to the limitations thoroughly
94
elaborated in that section, no reliable results are expected in the lower part of the measured
bandwidth. Therefore, this area in the following result figures is shaded, but still shown, to
illustrate the effects discussed in Chapter 4. First, the noise temperature of the stand-alone
ACL, as illustrated in the chip photograph in Figure 3.24(a), is shown. In Figure 6.5, the
noise temperature at different bias conditions is plotted versus the frequency. An outstanding
noise temperature of around 150K is achieved, which is in fact similar to the results of the
presented W-band ACLs. Also, as for the W-band loads, the resulting noise temperature
is not very sensitive to different bias conditions. Merely at a high drain current density of
283mA/mm (red curve), the noise temperature is noticeably increased.
110 115 120 125 130 135 140 145 150
0
50
100
150
200
250
300
350
 
 
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
 Vg +0.0 Vd +0.6 Id 214 mA/mm
 Vg +0.0 Vd +1.0 Id 283 mA/mm
 Vg -0.1 Vd +0.6 Id 122 mA/mm
 Vg -0.1 Vd +0.8 Id 148 mA/mm
 Vg -0.1 Vd +1.0 Id 178 mA/mm
Figure 6.5: Measured noise temperatures of D-band ACL at different bias.
To better evaluate this bias dependency, Figure 6.6 depicts the noise temperature at
145GHz versus the drain current density. At all measured points below 214mA/mm, the
ACL achieves a noise temperature below 150K and exhibits a very flat curve, proving a
behavior not too sensitive to bias changes. As was repeatedly pointed out in the previous
chapters, this is a crucial property of an ACL when it is intended to be used as a reference
load.
For a deeper analysis of the influence of the bias conditions, Figure 6.7 shows the noise
temperature for different drain voltages versus the gate voltage. Obviously, a higher drain
voltage favors a noise temperature increase, even if the gate voltage controls the drain current
to be low. The left measurement point at the red 1V curve represents the temperature at
a drain current density, regulated by the gate voltage, of 178mA/mm. For a drain voltage
of 0.6V, however, even at a higher drain current density of 214mA/mm, a lower noise
temperature is achieved (right measurement point at the blue curve). This is an important
finding, because it confirms that the noise temperature exhibited by an ACL is not mainly
dependent on the drain current, but that the appropriate bias setting of the transistor needs
to chosen to assure the reflection coefficients on that the ACL design was based in the
simulations. Less surprisingly, it can be seen, that within one fixed drain voltage setting, the
noise temperature can be increased by increasing the drain current (controlled by the gate
voltage).
95
100 125 150 175 200 225 250 275 300
60
80
100
120
140
160
180
200
220
V
g 
-0
.1
 V
d 
+0
.6
Drain Current Density [mA/mm]
 
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
@145 GHz
V
g 
-0
.1
 V
d 
+0
.8
V
g 
-0
.1
 V
d 
+1
.0
V
g 
0.
0 
V
d 
+0
.6
V
g 
0.
0 
V
d 
+1
.0
Figure 6.6: Measured noise temperatures of D-band ACL at 145GHz versus drain current density.
-0.2 -0.1 0.0 0.1
100
110
120
130
140
150
160
170
180
190
200
Gate Voltage [V]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Drain Voltage +1.0 V
122 mA / mm
283 mA / mm
Drain Voltage +0.6 V
214 mA / mm
178 mA / mm
@145 GHz
Figure 6.7: Measured noise temperatures of D-band ACL at 145GHz versus gate voltage.
96
Similar to the SCFE or the ACSS presented in Section 5.3, the stand-alone D-band ACL
was integrated on one MMIC together with a SP2T and a matched load (ML), as illustrated
in Figure 3.24(b). Figure 6.8 shows as a result the noise temperatures measured at the
MMIC output during the two possible internal switching states. For comparison, the noise
temperature of the stand-alone ACL is included into the graph (black curve). When switched
to the internal matched load, the resulting noise temperature corresponds as expected to the
ambient temperature, but only for frequencies above 140GHz. This is explained by the fact,
that the matching of the SP2T towards the two loads and towards the output is most critical.
Reflected noise power will superimpose and cause noise peaks as seen at 133GHz in both
curves (red and blue). When switched to the internal ACL, the output noise temperature of
the MMIC is reduced about 100K. Due to the very low noise of the ACL core at 137GHz
(see black curve), the peak seen in the red curve at this frequency is even compensated, so
that the noise temperature of the device when switched to the ACL (blue curve) is around
200K and the noise temperature curve very flat from 135GHz to 150GHz. If now the noise
temperature of the stand-alone ACL is compared to the ACL integrated with the SP2T, a
mean difference of approximately 50K can be observed. This increase is due to the losses of
the switch and the internal lines for signal routing. To calculate the loss associated with this
temperature increase, Equation 5.2 is applied again with the temperatures T2 = 200K and
T1 = 150K, which roughly correspond to mean values of the blue and black curves. This
results in a calculated loss of 1.8 dB. At 140GHz, where the SP2T was originally centered
and hence the difference in temperature between stand-alone ACL and the integrated version
is the smallest (T2 = 180K and T1 = 145K), the calculated loss is as low as 1.1 dB. These
are remarkable values for a D-band SP2T based on FETs and are also due to the very
compact design of the integrated switch and ACL and the short internal transmission lines.
110 115 120 125 130 135 140 145 150
100
150
200
250
300
350
400
switch to ML
switch to ACL
Frequency [GHz]
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
ACL only
Figure 6.8: SP2T switch with integrated matched load and D-band ACL.
Summary and Conclusion
As mentioned in Section 3, ACL noise temperatures of 180K on MMIC level were predicted
at 100GHz in [26] and assessed to be feasible for the use as reference loads in Post-EPS
97
context. The presented ACLs in this chapter were characterized up to 150GHz, yielding a
state-of-the art noise temperature of below 150K on MMIC level. Figure 6.9 compares the
noise temperatures of all stand-alone ACLs, designed and characterized in the course of this
work to all previously published ACLs in different technologies. It is notable, that a similarly
low noise temperature in D-band was achieved, as with the best of the designed W-band
loads.
0 20 40 60 80 100 120 140 160
25
50
75
100
125
150
175
200
 SiGe HBT
 GaAs MESFET
 InP HEMT
 GaAs pHEMT
 GaAs pHEMT
 GaAs mHEMT
 GaAs mHEMT
 This work 
         (GaAs mHEMT)
N
oi
se
 T
em
pe
ra
tu
re
 [K
]
Frequency [GHz]
Figure 6.9: State of the Art of previously published ACLs compared to the ACLs of this work.
98
7 Conclusion
In summary, two goals are of particular importance for the calibration of radiometer sys-
tems: For the more scientific application of remote sensing it is to increase the radiometric
accuracy, while for the more mass-market oriented application of PMMW imaging, e.g. in
security scanners, it is the goal to increase the repetition rate of the calibration procedure.
The first goal can be achieved by using references with noise temperatures in the vicinity of
the expected target or measurement range. This not only avoids extrapolation errors and
non-linear behavior of the receiver, but also improves the radiometric resolution, being an
equally crucial parameter in radiometry. The second goal can be achieved by placing fast
and reliably operating switches before the receiver, to switch between the references and the
receiving antenna (or multiple receiving channels).
Consequently, this work has presented active electronic loads, that are covering a wide range
of noise temperatures, encompassing the dominant measurement range in radiometric appli-
cations. In addition, for the first time such loads were integrated together with front-end
switches on the same MMIC technology at frequencies as high as 150GHz.
Even though the effect of a cold appearing transistor was described more than 30 years
ago, only few such circuits were published since then. As a result, no established design
guidelines were available at the beginning of this work, but only few publications - all with
different simulation approaches. Therefore, part of this work was to compare and evaluate
the known noise temperature calculation from previous work. It has been demonstrated with
a 89GHz ACL design, that several simplifications of the general calculation presented in [71]
are allowed, if a deviation of 5% from the original temperature calculation is acceptable. A
mandatory requirement for these simplifications is to design very good matching networks, to
reduce the impact of reflections at the external terminals of the load. However, it was found,
that even with the best realistically achievable matching networks, a remaining influence on
the backward gain and thus on the generated noise at the transistor gate can not be avoided.
This observation is of course based on the simulations with the available transistor model
and passive components library.
No matter which of the previously known simulation methods are used, they require to itera-
tively design the output and input matching networks, before the resulting noise temperature
can be simulated. If applicable, several re-design loops might be necessary to optimize the
noise temperature result. The required design-flow is therefore very time-consuming. To
overcome this, a novel one port noise temperature simulation method was presented in this
work, which can directly calculate the resulting noise temperature, while modifications at
the matching and supply networks are done. This allows to benefit from the possibilities of
optimization algorithms which are part of modern numerical simulation tools.
More important, than the relatively small differences in simulated noise temperature between
the different simulation methods, is the variation found during characterization of the active
load circuits. Even more than regular noise measurements of amplifying two-port devices,
the characterization of one-port low-noise loads was found out to be particularly challenging.
99
The results of the measurements were very sensitive to the utilized setup and a thorough
analysis and selection of the crucial components was necessary to assure accurate results. It
can be stated, that even if all measures are taken, to minimize the influence of the measure-
ment system itself and to reduce all inherent errors, it is not safe to rely on the absolute
accuracy of the measurement result. This was found out by cross-checking some selected
results with independently designed measurement systems of other research facilities in that
field. Also, for one of the two alternative measurement setups, a different calibration method
(heated waveguide) was used. A comparison and discussion of the discrepant results lead to
the conclusion, that none of utilized setups can be considered as superior to another. And
in consequence, that a stated noise temperature of an active cold load is only accurate for
the utilized setup, or, that it is only valid within a certain accuracy range. This was a major
finding of this work.
Knowing this, the designed circuits were characterized and state-of-the-art results for active
cold loads in W-band and D-band were presented. The lowest up to now published noise
temperature of 50K at a frequency of 1.4GHz is only 100K lower, than the noise tem-
perature presented in this work at a frequency one hundred times higher (140GHz)! For
the highest previously reported frequency of an ACL (52GHz), a temperature was achieved,
which is only less than 10K lower than the result of this work! It is remarkable, that the
lowest achieved temperatures for the presented D-band loads are similar to the achieved
temperatures for the W-band loads. Since the same technology, the same design-flow, and
the same characterization method was applied, this might lead to the conclusion, that a
similar noise performances might be reached even at higher frequencies. This might be one
focus of future work.
In addition to the realized cold loads, an active hot load was presented for the first time. It
does not achieve as high noise temperatures as noise diodes, which is in fact even a disadvan-
tage of noise diodes in the context of radiometer calibration. The dynamic range of the by
nature very sensitive radiometers is not suited for the calibration with noise diodes in general.
Instead, the diode’s relatively high noise power is reduced by implementing attenuators be-
fore feeding it to the radiometer input for calibration. The active hot load, however, provides
a convenient noise temperature, which advantageously expands the hot-cold range for the
two-point calibration, especially when combined with an active cold load. Furthermore, the
AHLs can be manufactured with the same mHEMT technology as the cold loads, and, as
a matter of fact, as existing multiple-throw FET switches. Consequently, integrated active
hot-cold loads were developed, that allow to switch internally between the references. For
this novel approach, a patent was granted. It should be noted, that the long-term stability
and the bias dependency of active hot loads were not extensively investigated in this or
previous work. This might be another topic for future studies.
To add a third reference to such a switching calibration front-end, an impedance-matched
NiCr resistor was included, that exhibits a noise temperature equivalent to the physical tem-
perature of the MMIC. By adding a third calibration reference, the error due to a non-linear
behavior of the radiometer can be further reduced. By including a temperature sensor on
the MMIC, not only the exact temperature of the passive load could be monitored, but also
the possible influence on the active loads could be accounted for. This idea was also part
of the granted patent. The sensitivity of the ACL to its physical temperature, however, was
investigated as part of this work and found to be less than for a passive load.
100
If even more internal references are desired, another idea shall be presented as an outlook:
Covering a certain temperature range with the active cold and hot load, respectively, it is
possible to generate any arbitrary temperature in between. For that, several delay lines with
different losses could be placed between the load and the output towards the radiometer
and switches would select which delay line to loop into the reference path. By that, the
noise temperature of the active load would be pulled towards the physical temperature of
the lines and reduce (for an AHL) or increase (for an ACL) the temperature at the output by
an amount equivalent to the loss of the correspondent line. Since the loss is only dependent
on the physical length of the lines, the difference between the multiple generated references
will always be stable, assuming all the lines are at the same physical temperature.
Another promising application of the developed calibration MMICs in this work, is the use as
a reference in on-wafer noise-measurement systems. Once such a hot-ambient-cold MMIC is
thoroughly characterized, it could be used similar to an impedance standard substrate. Typ-
ically the receiver is calibrated at waveguide level before the wafer-probe is connected, just
as it was done in this work. The disadvantage is, that the probe is not part of the calibration
loop and as it was clearly demonstrated in this work, especially on-wafer measurements are
afflicted with large errors, because the poor matching on probe-side is greatly influencing
the noise temperature measurement. With the proposed calibration on wafer-level, using a
hot-cold reference MMIC, this error could be effectively reduced.
101

References
[1] R. H. Dicke, “The measurement of thermal radiation at microwave frequencies,” Review
of Scientific Instruments, vol. 17, pp. 268 –275, july 1946.
[2] M. Tiuri, “Radio astronomy receivers,” Antennas and Propagation, IEEE Transactions
on, vol. 12, pp. 930 – 938, dec 1964.
[3] N. Skou, Microwave radiometer systems: design and analysis. Artech House, 1989.
[4] B. Vowinkel, Passive Mikrowellenradiometrie. Friedr. Vieweg und Sohn Verlagsge-
sellschaft, 1988.
[5] F. T. Ulaby, R. K. Moore, and A. Fung, Microwave Remote Sensing - active and passive.
Artech House, Norwood MA, 1981.
[6] S. Kazama, T. Rose, R. Zimmermann, and R. Zimmermann, “A precision autocalibrat-
ing 7 channel radiometer for environmental research applications,” Journal of Remote
Sensing Society of Japan, vol. 19, no. 3, pp. 265 – 273, 1999.
[7] T. Rose, S. Crewell, U. Löhnert, and C. Simmer, “A network suitable microwave ra-
diometer for operational monitoring of the cloudy atmosphere,” Atmospheric Research,
vol. 75, pp. 183 – 200, May 2005.
[8] T. L. Wilson, K. Rohlfs, and S. Hüttemeister, Tools of Radio Astronomy. Springer-
Verlag, fifth ed., 2009.
[9] R. Wilson, “The cosmic microwave background radiation,” Nobel Lecture, December
1978.
[10] A. R. Thomson, J. M. Moran, and G. W. Swenson, Interferometry and Synthesis in
Radio Astronomy. Wiley, second ed., 2001.
[11] A. Hulsmann, A. Tessmann, A. Leuther, I. Kallfass, E. Weissbrodt, M. Knelangen,
M. Schlechtweg, and O. Ambacher, “Integrated circuits beyond 100 ghz for stand-off
detection of concealed weapons,” in Fraunhofer VVS Future Security 2010 - 5th Security
Research Conference, pp. 74–77, 2010.
[12] M. Peichl, S. Dill, M. Jirousek, and H. Suess, “Passive microwave remote sensing for
security applications,” in Radar Conference, 2007. EuRAD 2007. European, pp. 32–35,
2007.
[13] A. Luukanen, L. Gronberg, T. Haarnoja, P. Helisto, M. Leivo, A. Rautiainen, J. Pent-
tila, J. Bjarnason, C. Dietlein, and E. Grossman, “Passive broadband terahertz camera
103
for stand-off concealed threat identification using superconducting antenna-coupled mi-
crobolometers,” in Microwave Conference, 2008. EuMC 2008. 38th European, pp. 943–
946, 2008.
[14] L. Yujiri, M. Shoucri, and P. Moffa, “Passive millimeter wave imaging,” Microwave
Magazine, IEEE, vol. 4, pp. 39 – 50, sept. 2003.
[15] L. Gilreath, V. Jain, H.-C. Yao, L. Zheng, and P. Heydari, “A 94-ghz passive imag-
ing receiver using a balanced lna with embedded dicke switch,” in Radio Frequency
Integrated Circuits Symposium (RFIC), 2010 IEEE, pp. 79 –82, may 2010.
[16] A. Luukanen, J. Ala-Laurinaho, M. Leivo, D. Gomes-Martins, M. Gronholm, J. Hakli,
P. Koivisto, S. Makela, P. Pursula, P. Rantakari, M. Sipila, J. Saily, A. Tamminen,
H. Toivanen, R. Tuovinen, A. Rautiainen, and A. Ra?isa?nen, “Developments towards
real-time active and passive submillimetre-wave imaging for security applications,” in
Microwave Symposium Digest (MTT), 2012 IEEE MTT-S International, pp. 1–3, 2012.
[17] J. Richter, D. Notel, F. Kloppel, J. Huck, H. Essen, and L. Schmidt, “A multi-channel
radiometer with focal plane array antenna for w-band passive millimeterwave imaging,”
in Microwave Symposium Digest, 2006. IEEE MTT-S International, pp. 1592–1595,
2006.
[18] W. G. . of the Joint Committee for Guides in Metrology (JCGM/WG 2), “International
vocabulary of metrology - basic and general concepts and associated terms (vim),” tech.
rep., Joint Committee for Guides in Metrology, 2008.
[19] ESA, “Calibration loads for radiometers,” statement of work, European Space Agency,
August 2009.
[20] P. Schlüssel, P. Phillips, C. Accadia, and R. Munro, “Post-eps mission requirements
document,” tech. rep., Eumetsat, June 2010.
[21] J. Randa, J. Lahtinen, A. Camps, A. Gasiewski, M. Hallikainen, D. Le Vine, M. Martin-
Neira, J. Piepmeier, P. Rosenkranz, C. Ruf, J. Shiue, and N. Skou, “Recommended
terminology for microwave radiometry,” tech. rep., National Institute of Standards and
Technology, 2008.
[22] J. Randa, “Traceability for microwave remote-sensing radiometry,” tech. rep., National
Institute of Standards and Technology, 2004.
[23] K. R. Lang, A Companion to Astronomy and Astrophysics: Chronology and Glossary
with Data Tables. Springer, 2006.
[24] M. Schneebeli and C. Matzler, “A calibration scheme for microwave radiometers using
tipping curves and kalman filtering,” Geoscience and Remote Sensing, IEEE Transac-
tions on, vol. 47, no. 12, pp. 4201–4209, 2009.
[25] M. Janssen, Atmospheric Remote Sensing by Microwave Radiometry. Wiley, 1993.
[26] J. Uusitalo, J. Charlton, and J. Lahtinen, “Advanced internal calibration techniques for
post eps radiometers,” tech. rep., Harp Technologies Ltd, July 2010.
104
[27] R. Frater and D. Williams, “An active "cold" noise source,” Microwave Theory and
Techniques, IEEE Transactions on, vol. 29, pp. 344 – 347, apr 1981.
[28] E. Weissbrodt, A. Tessmann, M. Schlechtweg, I. Kallfass, and O. Ambacher, “Active
load modules for w-band radiometer calibration,” in Geoscience and Remote Sensing
Symposium (IGARSS), 2012 IEEE International, pp. 2945–2948, 2012.
[29] M. Kantanen, E. Weissbrodt, J. Varis, A. Leuther, M. Seelmann-Eggebert, M. Ro?sch,
M. Schlecthweg, T. Poutanen, I. Sundberg, M. Kaisti, M. Altti, P. Jukkala, and P. Pi-
ironen, “Active cold load mmics for ka-, v-, and w-bands,” Microwaves, Antennas
Propagation, IET, vol. 9, no. 8, pp. 742–747, 2015.
[30] L. Dunleavy, M. Smith, S. Lardizabal, A. Fejzuli, and R. Roeder, “Design and charac-
terization of fet based cold/hot noise sources,” in Microwave Symposium Digest, 1997.,
IEEE MTT-S International, vol. 3, pp. 1293 –1296 vol.3, jun 1997.
[31] P. Buhles and S. Lardizabal, “Design and characterization of mmic active cold loads,”
in Microwave Symposium Digest., 2000 IEEE MTT-S International, vol. 1, pp. 29 –32
vol.1, 2000.
[32] W. Ciccognani, F. Giannini, E. Limiti, and P. Longhi, “Analysis, design and measure-
ment of active low-noise terminations,” in Microwave Techniques, 2008. COMITE 2008.
14th Conference on, pp. 1 –4, april 2008.
[33] N. Skou, S. Sobjaerg, and J. Balling, “Measurements on active cold loads for radiometer
calibration,” in Microwave Radiometry and Remote Sensing of the Environment, 2008.
MICRORAD 2008, pp. 1 –4, march 2008.
[34] E. Leynia de la Jarrige, L. Escotte, J. Goutoule, E. Gonneau, and J. Rayssac, “Sige hbt-
based active cold load for radiometer calibration,” Microwave and Wireless Components
Letters, IEEE, vol. 20, pp. 238 –240, april 2010.
[35] C. Bredin, D. Sanson, N. Mohamed, J. Orlhac, J. Goutoule, L. Escotte, and P. Piironen,
“23.8 ghz and 36.5 ghz active cold loads for radiometer calibration,” in Microwave
Technologies and Techniques Workshop, 2010.
[36] J. Piironen, “personal communication,” 2011.
[37] Agilent, “Understanding rf/microwave solid state switches and their applications.” Ap-
plication Note, 2010.
[38] J. Lemmetyinen, J. Uusitalo, J. Kainulainen, K. Rautiainen, N. Fabritius, M. Levander,
V. Kangas, H. Greus, J. Pihlflyckt, A. Kontu, S. Kemppainen, A. Colliander, M. Hal-
likainen, and J. Lahtinen, “Smos calibration subsystem,” Geoscience and Remote Sens-
ing, IEEE Transactions on, vol. 45, no. 11, pp. 3691–3700, 2007.
[39] A. Colliander, L. Ruokokoski, J. Suomela, K. Veijola, J. Kettunen, V. Kangas, A. Aalto,
M. Levander, H. Greus, M. Hallikainen, and J. Lahtinen, “Development and calibration
of smos reference radiometer,” Geoscience and Remote Sensing, IEEE Transactions on,
vol. 45, no. 7, pp. 1967–1977, 2007.
105
[40] B. Lambrigtsen, “Calibration of the airs microwave instruments,” Geoscience and Re-
mote Sensing, IEEE Transactions on, vol. 41, no. 2, pp. 369–378, 2003.
[41] T. Misra, A. M. Jha, D. Putrevu, J. Rao, D. B. Dave, and S. S. Rana, “Ground
calibration of multifrequency scanning microwave radiometer (msmr),” Geoscience and
Remote Sensing, IEEE Transactions on, vol. 40, no. 2, pp. 504–508, 2002.
[42] E. Njoku, J. M. Stacey, and F. Barath, “The seasat scanning multichannel microwave
radiometer (smmr): Instrument description and performance,” Oceanic Engineering,
IEEE Journal of, vol. 5, no. 2, pp. 100–115, 1980.
[43] C. Ruf, S. Keihm, and M. Janssen, “Topex/poseidon microwave radiometer (tmr). i.
instrument description and antenna temperature calibration,” Geoscience and Remote
Sensing, IEEE Transactions on, vol. 33, no. 1, pp. 125–137, 1995.
[44] S. Brown, S. Desai, W. Lu, and A. Tanner, “On the long-term stability of microwave
radiometers using noise diodes for calibration,” Geoscience and Remote Sensing, IEEE
Transactions on, vol. 45, no. 7, pp. 1908–1920, 2007.
[45] P. Pingree, M. Janssen, J. Oswald, S. Brown, J. Chen, K. Hurst, A. Kitiyakara, F. Mai-
wald, and S. Smith, “Microwave radiometers from 0.6 to 22 ghz for juno, a polar orbiter
around jupiter,” in Aerospace Conference, 2008 IEEE, pp. 1–15, 2008.
[46] S. Brown, B. Lambrigtsen, R. Denning, T. Gaier, P. Kangaslahti, B. Lim, J. Tan-
abe, and A. Tanner, “The high-altitude mmic sounding radiometer for the global hawk
unmanned aerial vehicle: Instrument description and performance,” Geoscience and
Remote Sensing, IEEE Transactions on, vol. 49, no. 9, pp. 3291–3301, 2011.
[47] I. Corbella, A. Gasiewski, M. Klein, V. Leuski, A. Francavilla, and J. Piepmeier, “On-
board accurate calibration of dual-channel radiometers using internal and external ref-
erences,” Microwave Theory and Techniques, IEEE Transactions on, vol. 50, no. 7,
pp. 1816–1820, 2002.
[48] A. Tanner, “A high stability ka-band radiometer for tropospheric water vapor measure-
ments,” in Aerospace Conference, 2001, IEEE Proceedings., vol. 4, pp. 4/1849–4/1863
vol.4, 2001.
[49] M. Lapinoja, V.-H. Kilpiä, V. Kangas, and P. Jukkala, “Active cold load technology
trade-off study,” tech. rep., Elektrobit Microwave, July 2005.
[50] D. Sanson and J.-M. Goutoule, “Active calibration for radiometer - final report,” tech.
rep., EADS Astrium, June 2010.
[51] X. Mei, C. Lin, L. J. Lee, Y. M. Kim, P. H. Liu, M. Lange, A. Cavus, R. To, M. Nishi-
moto, and R. Lai, “A w-band ingaas/inalas/inp hemt low-noise amplifier mmic with
2.5db noise figure and 19.4 db gain at 94ghz,” in Indium Phosphide and Related Mate-
rials, 2008. IPRM 2008. 20th International Conference on, pp. 1–3, 2008.
[52] K. Elgaid, H. McLelland, C. Stanley, and I. Thayne, “Low noise w-band mmmic amplifier
using 50nm inp technology for millimeterwave receivers applications,” in Indium Phos-
phide and Related Materials, 2005. International Conference on, pp. 523–525, 2005.
106
[53] P. Riemer, B. Buhrow, J. Hacker, J. Bergman, B. Brar, B. Gilbert, and E. Daniel,
“Low-power w-band cpwg inas/alsb hemt low-noise amplifier,” Microwave and Wireless
Components Letters, IEEE, vol. 16, no. 1, pp. 40–42, 2006.
[54] P. Kangaslahti, B. Lim, T. Gaier, A. Tanner, M. Varonen, L. Samoska, S. Brown,
B. Lambrigtsen, S. Reising, J. Tanabe, O. Montes, D. Dawson, and C. Parashare, “Low
noise amplifier receivers for millimeter wave atmospheric remote sensing,” in Microwave
Symposium Digest (MTT), 2012 IEEE MTT-S International, pp. 1–3, 2012.
[55] A. Leuther, A. Tessmann, I. Kallfass, R. Losch, M. Seelmann-Eggebert, N. Wadefalk,
F. Schafer, J. Gallego Puyol, M. Schlechtweg, M. Mikulla, and O. Ambacher, “Meta-
morphic hemt technology for low-noise applications,” in Indium Phosphide Related Ma-
terials, 2009. IPRM ’09. IEEE International Conference on, pp. 188 –191, May 2009.
[56] A. Leuther, A. Tessmann, I. Kallfass, H. Massler, R. Loesch, M. Schlechtweg,
M. Mikulla, and O. Ambacher, “Metamorphic hemt technology for submillimeter-wave
mmic applications,” in Indium Phosphide Related Materials (IPRM), 2010 International
Conference on, pp. 1–6, 2010.
[57] A. Leuther, A. Tessmann, H. Massler, R. Losch, M. Schlechtweg, M. Mikulla, and
O. Ambacher, “35 nm metamorphic hemt mmic technology,” in Indium Phosphide and
Related Materials, 2008. IPRM 2008. 20th International Conference on, pp. 1 –4, May
2008.
[58] A. Leuther, S. Koch, A. Tessmann, I. Kallfass, T. Merkle, H. Massler, R. Loesch,
M. Schlechtweg, S. Saito, and O. Ambacher, “20 nm metamorphic hemt with 660 ghz
ft,” in Compound Semiconductor Week (CSW/IPRM), 2011 and 23rd International
Conference on Indium Phosphide and Related Materials, pp. 1–4, 2011.
[59] M. Seelmann-Eggebert, F. Schäfer, A. Leuther, and H. Massler, “A versatile and cryo-
genic mhemt-model including noise,” in Microwave Symposium Digest (MTT), 2010
IEEE MTT-S International, pp. 501 –504, may 2010.
[60] E. Leynia de la Jarrige, L. Escotte, E. Gonneau, and J.-M. Goutoule, “Long-term
stability of an sige hbt-based active cold load,” in Geoscience and Remote Sensing
Symposium (IGARSS), 2011 IEEE International, pp. 3839 –3842, july 2011.
[61] E. de la Jarrige, L. Escotte, E. Gonneau, and J. Goutoule, “Sige hbt-based active cold
load: Design, characterization and stability measurements,” in Noise and Fluctuations
(ICNF), 2011 21st International Conference on, pp. 332 –335, june 2011.
[62] E. de la Jarrige, L. Escotte, E. Gonneau, and J. Goutoule, “Stability analysis of an sige
hbt-based active cold load,” Microwave Theory and Techniques, IEEE Transactions on,
vol. 59, pp. 354 –359, feb. 2011.
[63] E. de la Jarrige, Sources de bruit actives ultra stables a faible temperature de bruit pour
la radiometrie micro-onde. PhD thesis, Universite Paul Sabatier Toulouse III, 2011.
107
[64] J. Randa, L. Dunleavy, and L. Terrell, “Stability measurements on noise sources,” In-
strumentation and Measurement, IEEE Transactions on, vol. 50, pp. 368 –372, apr
2001.
[65] S. Sobjaerg, N. Skou, and J. Balling, “Measurements on active cold loads for radiometer
calibration,” Geoscience and Remote Sensing, IEEE Transactions on, vol. 47, pp. 3134
–3139, sept. 2009.
[66] V.-H. Kilpiä and P. Jukkala, “Active cold load for radiometer calibration,” tech. rep.,
DA Design Electronics, March 2010.
[67] H.-J. Michel, Zweitor-Analyse mit Leistungswellen. B.G. Teubner, 1981.
[68] F. Ellinger, Radio Frequency Integrated Circuits and Technologies. Springer-Verlag,
2007.
[69] U. Tietze, C. Schenk, and E. Gamm, Halbleiter-Schaltungstechnik. Springer-Verlag,
thirteenth ed., 2008.
[70] H. Rothe and W. Dahlke, “Theory of noisy fourpoles,” Proceedings of the IRE, vol. 44,
no. 6, pp. 811–818, 1956.
[71] M. Weatherspoon and L. Dunleavy, “Experimental validation of generalized equations
for fet cold noise source design,” Microwave Theory and Techniques, IEEE Transactions
on, vol. 54, pp. 608 – 614, feb. 2006.
[72] D. Wait and G. F. Engen, “Application of radiometry to the accurate measurement
of amplifier noise,” Instrumentation and Measurement, IEEE Transactions on, vol. 40,
no. 2, pp. 433–437, 1991.
[73] S. Diebold, E. Weissbrodt, H. Massler, A. Leuther, A. Tessmann, and I. Kallfass, “A w
-band monolithic integrated active hot and cold noise source,” Microwave Theory and
Techniques, IEEE Transactions on, vol. 62, pp. 623–630, March 2014.
[74] E. Weissbrodt, M. Schlechtweg, O. Ambacher, and I. Kallfass, “W-band active loads
and switching front-end mmics for radiometer calibration,” International Journal of
Microwave and Wireless Technologies, vol. 5, pp. 293–299, 6 2013.
[75] M. Thumm, W. Wiesbeck, and S. Kern, Hochfrequenzmesstechnik. B.G. Teubner,
1998.
[76] M. Edwards and J. Sinsky, “A new criterion for linear 2-port stability using a single ge-
ometrically derived parameter,” Microwave Theory and Techniques, IEEE Transactions
on, vol. 40, no. 12, pp. 2303–2311, 1992.
[77] I. Kallfass, S. Diebold, H. Massler, S. Koch, M. Seelmann-Eggebert, and A. Leuther,
“Multiple-throw millimeter-wave fet switches for frequencies from 60 up to 120 ghz,” in
Microwave Conference, 2008. EuMC 2008. 38th European, pp. 1453–1456, 2008.
[78] E. Weissbrodt, I. Kallfass, and S. Diebold, “Radiometrische kalibrationseinrichtung mit
monolithisch integriertem mehrfachschalter.” patent, april 2011. 102011016732.
108
[79] E. Weissbrodt, I. Kallfass, A. Tessmann, H. Massler, A. Leuther, M. Schlechtweg, and
O. Ambacher, “A broadband low-noise d-band amplifier module in 35 nm mhemt tech-
nology,” in 6th ESA Workshop on Millimetre-Wave Technology and Applications, may
2011.
[80] E. Weissbrodt, A. Hulsmann, A. Massler, H.and Leuther, I. Kallfass, M. Schlechtweg,
and O. Ambacher, “Integrated d-band mmics for receiver front-ends,” in Microwave
Technology and Techniques Workshop, pp. 1–6, 2012.
[81] E. Weissbrodt, I. Kallfass, R. Weber, A. Tessmann, H. Massler, and A. Leuther, “Low-
noise amplifiers in d-band using 100 nm and 50 nm mhemt technology,” in German
Microwave Conference, 2010, pp. 55–58, 2010.
[82] H. Friis, “Noise figures of radio receivers,” Proceedings of the IRE, vol. 32, pp. 419 –
422, july 1944.
[83] Agilent, “Fundamentals of rf and microwave noise figure measurements.” Application
Note 57-1, 2006.
[84] Agilent, “Noise figure measurement accuracy - the y-factor method.” Application Note
57-2, 2004.
[85] L. Belostotski, “A calibration method for rf and microwave noise sources,” Microwave
Theory and Techniques, IEEE Transactions on, vol. 59, no. 1, pp. 178–187, 2011.
[86] Agilent, “Practical noise figure measurement and analysis for low-noise amplifier de-
signs.” Application Note 1354, 2000.
[87] Agilent, “10 hints for making successful noise figure measurements.” Application Note
57-3, 2009.
[88] Elva, “Precision calibrated noise sources.” Data Sheet, 2015.
[89] Noisecom, “Precision calibrated millimeter-wave wave guide noise sources.” Data Sheet,
2015.
[90] “Answer from semic rf electronic to support request.” email, november 2010.
[91] J. L. Cano, E. Villa, D. Ortiz, and E. Artal, “Transición en guía wr-28 de acceso al
criostato para la medida de sistemas enfriados,” in Actas del XXIV Simposium Nacional
de la Unión Científica Internacional de Radio (URSI 2009), sept 2009.
[92] “Cuming microwave technical bulletin 390-1.” C-RAM SFC Data Sheet, 2011.
[93] J. Saily and A. V. Raisanen, “Studies on specular and non-specular reflectivities of radar
absorbing materials (ram) at submillimetre wavelengths,” February 2003.
109

Patents and Personal Publications
[P-1] E. Weissbrodt, I. Kallfass, S. Diebold, “Radiometrische Kalibrationseinrichtung mit
monolithisch integriertem Mehrfachschalter," German Patent, 102011016732 , April
2011.
[P-2] E. Weissbrodt, R. Lenz, W. Wiesbeck, W. Albright, J.Nicoll, “The role of point tar-
gets in spaceborne L-Band polarimetric SAR calibration," EUSAR 2006 - 6th European
Conference on Synthetic Aperture Radar, VDE Verlag GmbH, 2006.
[P-3] E. Weissbrodt, I. Kallfass, R. Weber, A. Tessmann, H. Massler, A. Leuther, “Low-
noise amplifiers in D-band using 100 nm and 50 nm mHEMT technology," German
Microwave Conference, pp. 55 – 58, 2010.
[P-4] E. Weissbrodt, I. Kallfass, A. Tessmann, H. Massler, A. Leuther, M. Schlechtweg,
O. Ambacher, “A Broadband Low-Noise D-Band Amplifier Module in 35 nm mHEMT
Technology," 6th ESA Workshop on Millimetre-Wave Technology and Applications,
2011.
[P-5] E. Weissbrodt, I. Kallfass, A. Hulsmann, A. Tessmann, A. Leuther, H. Massler, O.
Ambacher, “W-band radiometer system with switching front-end for multi-load calibra-
tion," Geoscience and Remote Sensing Symposium (IGARSS), 2011 IEEE International,
pp. 3843 – 3846, July 2011.
[P-6] E. Weissbrodt, A. Hulsmann, H. Massler, A. Leuther, I. Kallfass, M. Schlechtweg,
O. Ambacher “Integrated D-Band MMICs for Receiver Front-Ends," in Microwave
Technology and Techniques Workshop, Mar. 2012, pp. 1 – 6.
[P-7] E. Weissbrodt, A. Tessmann, M. Schlechtweg, I. Kallfass, O. Ambacher, “Active
load modules for W-band radiometer calibration," Geoscience and Remote Sensing
Symposium (IGARSS), 2012 IEEE International, pp. 2945 – 2948, 2012.
[P-8] E. Weissbrodt, A. Leuther, M. Schlechtweg, I. Kallfass, O. Ambacher, “Highly in-
tegrated switching calibration front-end MMIC with active loads for w-band radiome-
ters," inMicrowave Integrated Circuits Conference (EuMIC), 2012 7th European, 2012,
pp. 203 – 206.
[P-9] E. Weissbrodt, M. Schlechtweg, O. Ambacher, I. Kallfass, “W-band active loads
and switching front-end MMICs for radiometer calibration," International Journal of
Microwave and Wireless Technologies, vol.5, pp. 293 – 299, 6 2013.
[P-10] R. Weber, V. Hurm, H. Massler, E. Weissbrodt, A. Tessmann, A. Leuther, T. Narhi,
I. Kallfass, “An H-band low-noise amplifier MMIC in 35 nm metamorphic HEMT tech-
nology,"in Microwave Integrated Circuits Conference (EuMIC), 2012 7th European,
2012, pp. 187 – 190.
111
[P-11] A. Hulsmann, A. Tessmann, A. Leuther, I. Kallfass, E. Weissbrodt, M. Knelangen,
M. Schlechtweg, O. Ambacher, “Integrated circuits beyond 100 GHz for stand-off de-
tection of concealed weapons," in Fraunhofer VVS Future Security 2010 - 5th Security
Research Conference, 2010, pp. 74 – 77.
[P-12] I. Kallfass, A. Hulsmann, A. Tessmann, A. Leuther, E. Weissbrodt, M. Schlechtweg,
O. Ambacher, “W-band direct detection radiometers using metamorphic HEMT
technology," in SPIE - The International Society for Optical Engineering: Passive
Millimeter-Wave Imaging Technology XIV, vol. 8022, 2011, pp. 80220O – 80220O-5.
[P-13] S. Diebold, E. Weissbrodt, H. Massler, A. Leuther, A. Tessmann, I. Kallfass, “A
W-Band Monolithic Integrated Active Hot and Cold Noise Source," Microwave Theory
and Techniques, IEEE Transactions on, vol. 62, no. 3, pp.623-630, March 2014.
[P-14] M. Kantanen, E. Weissbrodt, J. Varis, A. Leuther, M. Seelmann-Eggebert, M.
Rösch, M. Schlecthweg, T. Poutanen, I. Sundberg, M. Kaisti, M. Altti, P. Jukkala,
P. Piironen “Active Cold Load MMICs for Ka-, V-, and W-Bands," IET Microwaves,
Antennas and Propagation, vol.9, issue 8, June 2015, pp. 742 – 747.
112
Acknowledgments
This doctoral thesis is a result of my work as a scientist at the Fraunhofer Institute for
Applied Solid-State Physics (IAF) in Freiburg, Germany, where I not only got the support
from an excellent scientific and technical staff, but also had access to IAF’s worldwide
leading III/V semiconductor technology process and its outstanding measurement facilities.
Therefore, I would like to thank Professor Dr. Oliver Ambacher, head of the institute and
Dr. Michael Schlechtweg, head of the department for High Frequency Devices and Circuits,
who gave me this unique opportunity.
I am deeply grateful to my advisor Professor Dr. Ingmar Kallfass, head of the Institute
of Robust Power Semiconductor Systems at the University of Stuttgart for his enduring
scientific and personal guidance. Moreover, for his confidence in my capabilities from the
first day on and his encouragement and unlimited patience throughout the years of this
research.
I also want to thank Professor Dr. Gilles Dambrine, deputy director of the Institute of
Electronics, Microelectronics and Nanotechnology (IEMN) at the University of Lille, and a
distinguished scientist in the field of noise theory, for taking the position of second reviewer.
Furthermore, I would like to acknowledge the basic funding of my general research by the
German Federal Ministry of Defence (BMVg) and the Bundeswehr Technical Center for
Information Technology and Electronics (WTD 81) and in particular the support by the
European Space Agency (ESA) for funding and supporting the very successful Calload
project. In this context I would like to express my gratitude to my colleague Dr. Matthias
Seelmann-Eggebert who acquired this project, introduced me into noise theory and modeling
and provided accurate simulation models - as well as to Dr. Petri Piironen, ESA’s project
coordinator, for his feedback and suggestions for improvement. During the course of the
project, I gained valuable knowledge in the field of calibration thru fruitful discussions with
the Finnish project partners, namely: Dr. Jussi Varis and Mikko Kantanen from Millilab and
Torsti Poutanen from DA-Design. The good atmosphere in the project team and the always
constructive cooperation during the project facilitated its success.
As mentioned above, the support of many colleagues facilitated my daily work. Therefore,
I want to thank Hermann Massler for sharing his profound measurement experience and
Sandrine Wagner and Roger Lozar for supporting my measurement activities. Furthermore,
I thank Dr. Arnulf Leuther, in representation of all the staff members of the technology and
processing group for providing excellent mHEMT MMICs , Dr. Axel Tessmann for constant
consulting in almost every field and always thoroughly reviewing my papers, and Rainer We-
ber for the constant support in ADS libraries and introduction to LNA design. Moreover, I
thank Dr. Beatriz Aja, for sharing her extensive knowledge on radiometer instruments, noise
modeling and network analysis, and my fellow (former) PhD students Dr. Daniel Bruch, Dr.
Daniel Lopez-Diaz, Daniel Müller, and Dr. Sebastian Diebold for many fruitful discussions.
The constant support from the mechanical hardware shop and the IT department was also
much appreciated.
113
For the always good atmosphere in the office and for the continuous motivation special
thanks go to Sibylle Hämmerle and to my roommates Dr. Jutta Kühn, Dr. Dirk Schwan-
tuschke, Dr. Robert Makon, and Dr. Vincenzo Carruba.
On many occasions, I had the chance to exchange my recent research progress with
experienced scientists in that field, from whom I gained valuable knowledge and got new
impulses for my work. For that I am very grateful to:
Professor Dr. Steven C. Reising, professor at the Colorado State University and Dr. Axel
Murk from the University of Bern for the feedback to my presented work and discussions on
several conferences.
Dr. Frank Schaefer and Sener Türk from the Max-Planck-Institute for Radioastronomy for
suggestions concerning the Hot-Cold measurement-setup, low-noise measurements and for
feedback on radioastronomy requirements.
Achim Walber and Martin Philipp from Radiometer Physics GmbH for their interest in my
work and for the opportunity to cross-check the ACL and AHL module measurements in
their laboratories.
Dr. Markus Peichl and Dr. Matthias Jirousek from the National Aeronautics and Space
Research Centre (DLR) for their advice on system requirements in passive millimeter wave
imaging, security applications, and radiometry in general.
The scientists from the meteorology institute, campus nord, of the Karlsruhe Institute of
Technology, for demonstrating their radiometer and calibration procedure.
I also would like to thank my friend Dr. Oliver Pink, who always stands by my side if I
need advice - not only limited to scientific questions.
All the necessary energy to stay the course thru the years of research I gained from the people
closest to me. Therefore I am deeply grateful to Julia and my whole family, especially to
Maria and Waldemar who always support me and believe in me.
114
Für das Können gibt es nur einen Beweis:
Das Tun.
(Marie von Ebner-Eschenbach)