Vous êtes sur la page 1sur 48

MICROWAVE MEASUREMENTS

ANDREA FERRERO
VALERIA TEPPATI
UMBERTO PISANI
Politecnico di Torino
Torino, Italy
1. INTRODUCTION
During World War II, intense research and effort in the
radar development program brought about the radiofre-
quency (RF) spectrum extension beyond the usual appli-
cations in radiocommunications. Shorter wavelengths
induced specific design of reduced dimension laboratory
equipment, in order to generate, convey, transmit, and
detect higher-frequency signals. By convention, the
RF/microwave region ranges between 30 MHz and
300GHz, since shorter than 1 mm wavelengths require
equipment too small to be easily realized.
Voltage, current, and impedance concepts lose their
conventional meanings when operating wavelength is ap-
proximately equal to the dimensions of the structures un-
der test, and the behavior of propagating electromagnetic
waves must be analyzed in terms of electric and magnetic
elds. However, as there is no simple and direct way to
measure these quantities, it is necessary to resort to indi-
rect methods.
Until the 1960s, microwave measurements were car-
ried out by instruments such as
*
microwave cavities, for wavelength measurements
*
power sensors, for power measurements
*
standing-wave-ratio meters (slotted lines) and wave-
guide bridges, for impedance measurement
All the techniques were based on scalar measurements
combined with precision mechanical measurements (e.g.,
probe displacement in a slotted line). Vector quantity, such
as impedance measurements, were indirectly carried out
by making different scalar measurements along a slotted
line.
In the 1960s, another type of microwave power mea-
suring instrument was introduced, based on the principle
of superheterodyne conversion: the spectrum analyzer.
This instrument, which has evolved to have wider bands
and dynamic ranges and to perform math and calibration,
exploits the principle of converting each microwave signal
frequency component to an intermediate frequency (IF),
where the signal can be more easily detected.
At the same time, the rst scattering parameter mea-
surements were performed with the most popular micro-
wave instrument: the vector network analyzer (VNA); this
instrument is based on subsampling or mixing techniques,
to downconvert the microwave signals to an IF frequency,
where they can be detected in both magnitude and phase.
From the rst scalar versions, the instrument evolved into
vector measurements, with faster acquisition times and
higher (up to 130 dB) dynamic ranges. Up-to-date instru-
mentation can include a PC microprocessor plus operating
system and perform automatic measurements as well as
the required calibrations.
In the following sections, power, spectral, and network
analyzer measurements are described. A nal section is
dedicated to advanced microwave measurements.
2. POWER, SPECTRAL AND NOISE MEASUREMENTS
Since the beginning of the microwave years, the main
techniques for microwave measurements were based on
power meters. Those instruments evolved from traditional
bolometer-based systems to diode-based power sensors for
peak power measurements; furthermore, the dynamic
range increased to 70 dB, from the initial 4045dB. Pow-
er measurements are also used as primary microwave
standards with calorimeters, to correlate microwave pow-
er to basic standard units [4,17].
A power meter is typically based on a main unit and
one or more power heads; the heads are generally specied
with respect to the bandwidth and the power range, while
the main unit contains the interface circuits, the head bias
control, and, more recently, math and calibration func-
tionalities.
The core of a power head can be a thermistor, placed on
a low-frequency resistive bridge, as shown in Fig. 1. The
system measures the RF power through a DC substitution
technique by keeping constant the total dissipated power
(RFDC), on the thermistor. Since the value of the
thermistor resistance changes with the temperature (i.e.,
the dissipated power), the bridge becomes unbalanced as
more RF is injected. The feedback network acts on the
bridge DC voltage supply arm to reduce the amount of DC
dissipated power and to keep the bridge balanced. In this
way, the change of DC voltage corresponds to the amount
of RF power injected, and its measurement, usually done
with a digital conversion system, provides an accurate
measurement of the RF power. Since a change of external
temperature may induce the same effect as an RF signal,
an identical (and within the same environment) bridge is
used, so that a temperature compensation is also applied.
Ordinary power heads are calibrated through a refer-
ence chain to the National Power Reference Standard,
generally based on a microcalorimeter [61]. An alternative
way to measure microwave power is given by the spectrum
analyzer, which measures the power of each frequency
component of an input signal and is based on a superhet-
erodyne conversion.
A simplied scheme is shown in Fig. 2; the local volt-
age-controlled oscillator (VCO) sweeps over frequency, so
that the portion of the input signal which has f
in
f
VCO

f
IF
is converted to the IF frequency, then detected and
passed to elaboration or display. The sweep of the local
oscillator is chosen to cover the desired spectrum of the
input signal, while the band of the IF lter determines the
frequency resolution of the spectral measurement and is
thus called resolution bandwidth (RBW). If two adjacent
frequency components of the input signal are within the
RBW, they are combined together at the lter output and
their measurement becomes misleading, as shown in
Fig. 3. A large RBW may introduce unacceptable error
2802 MICROWAVE MEASUREMENTS
Previous Page
especially in measurements such as intermodulation dis-
tortion.
The RBW is also linked with the measurement speed,
since the lter settling time xes the minimum time re-
quired to have a stable output, and thus the maximum
change rate of the mixer output. For this reason if a better
resolution is required, specifically, a narrow IF band, a
slower VCO speed is set.
Another effect of RBW is linked to noise oor. Thermal
noise generally prevails over other noise sources, and its
contribution is constant versus frequency (white noise).
The revealed noise power is thus proportional to RBW. A
reduction of a factor 10 on RBW implies a reduction of
10 dB in noise oor.
Unfortunately, noise oor cannot be reduced by rising
the value of the input attenuator, since noise main contri-
butions are introduced right after the rst converter. Fur-
thermore the RF attenuator reduces the signal-to-noise
ratio (SNR) of the measured signal. Thus, in order to max-
imize the SNR, its preferable to work with the minimum
RF attenuation that gives the minimum distortion of the
mixers.
Spectrum analyzers are widely used for noise measure-
ments. The measurement of network properties in terms
of noise is especially felt in receiving systems, when
noise becomes comparable to signal. To increase the
SNR, a reduction of the noise introduced by the receiver
components is required. For this purpose, reliable noise
measurements are necessary, to characterize the receiver
components.
The most popular quality factor for noise performance
is the noise gure. It is dened for two port devices [34] as
pre-selector or
lowpass
filter
input
signal
RF input
attenuator
mixer
local
oscillator
crystal
reference
IF gain
IF
filter
detector
video
filter
Figure 2. Simplied diagram of a spectrum ana-
lyzer with superheterodyne detection.
RF input
RF bridge
Vrf/2
thermistor
-
+
bridge amplifier
Vrf
Vrf
A/D
compensation
bridge
Vc/2
thermistor
(temperature
compensation)
-
+
bridge amplifier
Vc
Vc
thermistor
mount
P
Figure 1. Simplied diagram of a power
meter thermistor-based power head.
MICROWAVE MEASUREMENTS 2803
the ratio of SNR input to SNR output
F
S
i
=N
i
S
o
=N
o
1
where S
i
, N
i
and S
o
, N
o
are the input and output signal
and noise powers.
Any receiver component, such as an amplier, adds its
own noise to the output. Thus noise gure becomes
F
S
i
=N
i
S
o
=N
o

N
a
G
a
N
i
G
a
N
i
2
where G
a
is the amplier gain S
o
G
a
S
i
and N
a
is the
noise power added by the amplier, and the overall output
noise power becomes N
o
N
a
G
a
N
i
. If the noise can
be represented as thermal noise, then the noise output
power is:
N
o
N
a
G
a
k
b
BT
i
3
where k
b
is Boltzmanns constant, B is the equivalent
bandwidth, and T
i
is the temperature of the noise source
at the input of the amplier. The noise gure is then
F
N
a
G
a
k
b
BT
i
G
a
k
b
BT
i
4
Noise gure meters determine the noise added by a
DUT (device or design under test) by comparing it to the
noise present at the input. Generally, they switch a solid-
state noise source on and off to generate two equivalent
noise temperature points (T
c
and T
h
) and measure the
corresponding two noise power outputs with a tuned
receiver with a xed RBW. As shown in Fig. 4, N
a
is
computed by extrapolating the straight line to the T
i
0
point. The added noise can be then transformed in an
equivalent input noise temperature (see Fig. 4) T
e
with
N
a
G
a
k
b
BT
e
. Finally, with Eq. (4) or
F 1
T
e
T
i
5
the noise gure F is presented to the user.
3. NETWORK ANALYZER MEASUREMENTS
In the 1960s, the basic principle for network analysis was
introduced in an innovative instrument called a vector
voltmeter [15], which is able to measure a microwave sig-
nal in both magnitude and phase.
The idea was the conversion to an intermediate fre-
quency (IF), where the use of ordinary AC voltmeter and
phasemeter is possible, of a test and a reference micro-
wave signal, with a synchronous conversion. This is
achieved through a subsampling process [7], where a
strobe signal, formed by a series of very narrow pulses
spaced by the sampling period t
s
, samples an input signal,
having period Tt
s
cT, as shown in Fig. 5a. In the fre-
quency domain the strobe signal is again a series of pulses,
spaced by f
s
1=t
s
, while the input signal is generally a
sinusoidal one. The process is called subsampling since
the sampling frequency is well below the Nyquist one, and
the technique can be applied if the input is sinusoidal or
with a periodic pattern; thus the microwave signal has an
effective bandwidth well below the sampling frequency.
The microwave signal is acquired when the equation
f
IF
f
in
nf
s
is satised for a certain n value, where nf
s
is
the strobe signal nth harmonic. By a proper choice of f
s
through a PLL (phase-locked loop) technique, the IF fre-
quency can track any change of the input frequency.
By subsampling two microwave signals at the same
frequency with two identical sampling pulses, the IF sig-
nals keep the same magnitude and phase relationships
IF filter
f f
1
f
f f
display spectrum
RBW
RBW
f
2
f
1
f
2
Figure 3. Errors introduced by large RBW in intermodulation
distortion measurements. On the left side of the gure, the input
signals f
1
, f
2
and RBW lter spectra are sketched, while on the
right side the corresponding spectrum analyzer displays are
shown for a correct RBW conguration (top) and in an erroneous
conguration (bottom).
T
i
T
c
T
h
N
o
N
a
N
c
N
h
T
e
slope = k
b
G
a
B
T
i
, Z
i
DUT
S
o
+N
o
T
i
, Z
i
DUT
S
o
+N
o
T
e
, Z
i
noiseless
Figure 4. Noise introduced by an amplier (N
a
) is measured ex-
trapolating two measurement points corresponding to T
c
and T
h
.
The situation can be modeled with a noiseless amplier plus an
additional noise source, with equivalent temperature T
e
.
2804 MICROWAVE MEASUREMENTS
existing between the original waves, but at a much lower
frequency, so that they can be easily measured in ampli-
tude and phase with ordinary instrumentation [7].
The block scheme of the vector voltmeter is shown in
Fig. 6. A PLL technique adjusts the frequency of a voltage-
tuned local oscillator (VTLO) that triggers the samplers as
explained above. The frequencies out of IF band are re-
jected by the narrowband IF lter. Therefore, the system
becomes frequency-selective and reaches great perfor-
mance in terms of dynamic range, reduction of noise,
and resolution. Finally, the two sampled signals are mea-
sured with low-frequency A.C. voltmeter and phasemeter,
or digitally acquired through S/H (sample-and-hold) and
A/D (analog-to-digital) converting circuits.
The phase-locked sampling process is equivalent to a
heterodyne conversion with mixers, but with a basic ad-
vantageit does not require a microwave frequency local
oscillator to downconvert the input signal, but it only
t
strobe
RF
t
s
>>T
T
f
strobe
RF
IF
IF filter
f
in
f
s
2f
s
3f
s
4f
s
IF
RF
t
f
LO
RF
IF IF filter
f
in
f
IF
(a) (b)
(c) (d)
Figure 5. Comparison between sub-
sampling process in time domain (a)
and frequency domain (b) and mixing
process in time domain (c) and frequen-
cy domain (d).
Phasemeter
Bandpass
filter
Voltmeter
Bandpass
filter
(e.g
Down-converted A signal
(e.g. 20 kHz)
Down-converted B signal
(e.g. 20 kHz)
Sampler
Tracking PLL
Sampler
A
B
Microwave frequency
signal A
Microwave frequency
signal B

Figure 6. Block scheme of a vector voltmeter based on sub-sampling techniques.


MICROWAVE MEASUREMENTS 2805
needs a low-frequency pulsed VTLO. As a drawback, the
sampler philosophy requires a PLL to tune the VTLO;
thus the phase lock is reached after a nite amount of time
and it requires a stable reference signal, or it can be lost.
This measurement principle, joined with the introduc-
tion of high-directivity, low-loss broadband directional
couplers, which separate the forward and reverse travel-
ling waves along a transmission line, formed an alterna-
tive system to a direct measurement of reection
coefcients (in amplitude and phase) rather than the clas-
sic slotted-line standing-wave ratio meter. As shown in
Fig. 7, a directional coupler is used to pick up the forward
and reverse traveling waves from the mainline through its
coupled arm where the reference and test inputs of a vec-
tor receiver are connected. In this way a direct reading of
the reection coefcient is easily obtained.
From the basic reectometer, an extension of the S-pa-
rameter measurement system, termed a vector network
analyzer (VNA), follows immediately.
The VNAs were introduced in the late 1960s [5,38,53]
for one-port and two-port device characterization in coax-
ial environment and since the mid-1970s they became a
fundamental testset for all microwave laboratories. While
the lower frequency ranges span from 30 kHz to 100MHz,
the upper frequency of commercial systems extended, at
the beginning, up to 12 GHz, afterward to 18 GHz, and to-
day it reaches 110 GHz.
A complete VNA is a very complex system, which usu-
ally includes
*
A microwave synthesized source
*
Dual reectometer testset based on directional cou-
plers or directional bridges and switches to separate
and select incident and reected waves at the device
ports
*
A multichannel microwave receiver where the sepa-
rated signals are downconverted
*
A central unit for intermediate frequency (IF) detec-
tion, A/D conversion, data processing, and presenta-
tion [18].
A simplied block scheme of a two-port network analyzer
is shown in Fig. 8.
This system allows very broadband measurements
with hundreds of frequency points or limited measure-
ment bands with extremely high-frequency resolution
(useful for testing high selectivity devices).
Normally, VNAs have coaxial (for very broadband ap-
plications) or waveguide measurement ports while
DUTs can have leads on microstrip, on-chip contacts, on-
wafer ports, and so on. Such mechanical and/or electrical

x
Microwave
source
Reflectometer
ka kb
Vector voltmeter
Ref Test
a
b
Figure 7. Reectometer system.
DUT
port 1 port 2
a
1
b
1
a
2
b
2
microwave
source
2-to-1 switch
VTO samplers
phase lock
1 IF
st
a
1
a
2
crystal
reference
oscillator
sample & hold, A/D
display
2 IF
nd
Figure 8. Block scheme of two-port network
analyzer [18].
2806 MICROWAVE MEASUREMENTS
discontinuities from one transmission line to the other re-
quire accurate studies of the interface between the mea-
surement equipment and the device to reduce the
measurement errors, since they can be only partially cor-
rected by a calibration procedure.
3.1. Calibration of Network-Analyzer-Based Systems
Network analyzer measurements are affected by different
kinds of errors and uncertainties, generally classied as
*
Random uncertainties
*
Drifts
*
Systematic errors
Random uncertainties are due mainly to noise and con-
nector repeatability effects, while drifts are typically due
to temperature, but the effects of systematic errors over-
whelm all the others by at least an order of magnitude. For
such reasons the use of an appropriate calibration tech-
nique is mandatory for useful VNA based measurements.
Typical causes of systematic errors are
*
Directional couplers imperfections (directivity, port
mismatch, frequency variation of the coupling coef-
cient)
*
Mismatch errors due to adapters and cables
*
Coaxial conguration switches losses and mismatch
*
Crosstalk
The traditional way to handle such errors was to de-
velop different models based on owgraph representation
of each error effect. The rst studies in this eld began
right after the introduction of vector network analyzers in
the 1960s [5,15,63], but it was only with the introduction
of microprocessor controlled VNA, like the HP8510, that
the so-called calibration procedures became common and
widely used. Here, a brief overview of how an error re-
moval procedure is organized and the limitations of the
today available techniques will be given.
Modern calibration procedures are based on a system-
atic model of errors that overcomes the traditional ap-
proach. The actual VNA is viewed as an ideal system
without systematic error followed by a linear network,
called error box, which models the inuence of the sys-
tematic errors altogether, as shown in Fig. 9.
This error model is the basis of all calibration tech-
niques available today, where the various solutions vary
according to how they handle identication of the error
box parameter. However, the whole approach works if, and
only if, the actual VNA can be seen as a perfect linear re-
ceiver; otherwise the entire process of error coefcient
computation and correction fails, since the error box is
everywhere assumed as a linear network (i.e., the rela-
tionships between raw and corrected parameters are lin-
ear equations). This is a typical case for the current
systems with 4100dB of dynamic range, but for accurate
measurements of a highly isolated device, it is mandatory
to check for proper power levels at each port, in order to
quantify the amount of nonlinearity errors [50].
To compute the error box parameters, calibration tech-
niques use a set of precise components called standards.
Each standard is modeled through electromagnetic simu-
lation or through low-frequency measurement of scaled
devices. A typical model for an open standard is given in
Fig. 10.
The model parameters are stored inside the VNA and
their electrical responses computed during the calibration
procedure. The error box parameters are obtained from
comparison of the modeled responses and the actual stan-
dard measurements.
Different calibrations are normally classied respect to
the degree of the standard knowledge they require:
*
Calibrations based on well-known standards such as
Short (circuit)/open (circuit) load thru (through)
(SOLT)
Offset shorts (short circuits)
*
Self-calibrations such as
Thru reect line (TRL)
Line reect match (LRM)
Short (cirucit)/open (circuit) load reciprocal
(SOLR)
These techniques are only few of the various calibration
algorithms available today; furthermore, although origi-
nally obtained separately, [2,2628,31,32,3941,45,52,64],
they are all special cases of a more general theory derived
for one- and two-port VNAs [56]. Description of the algo-
rithm is beyond the scope of this section; instead, the focus
will be on the differences in terms of accuracy, required
standards, and applicability.
The most widely adopted calibration for the two-port
VNA is the SOLT, an acronym obtained from the required
standards: a short (circuit), an open (circuit), a load, and a
Z
off

off
l
off
C
open
=
=C
0
+C
1
f +C
2
f
2
+C
3
f
3
Figure 10. Typical model of coaxial open standards. The nonlin-
ear capacitance depends on frequency. The line parameters
(length l
off
, characteristic impedance Z
off
, and losses a
off
) are
evidenced in the scheme.
port 1 error
box
a
1
b
1
a
2
b
2
device
under test
port 2 error
box
E
b1
E
b2
S
DUT
a
m1
b
m1
a
m2
b
m2
ideal network analyzer
Figure 9. Error box model of VNA systematic errors.
MICROWAVE MEASUREMENTS 2807
direct port connection, called a thru. This technique has
been applied since the early years of automatic VNAs, and
the corresponding standards are available in almost every
environment (e.g., waveguide, coaxial, on-wafer). All VNA
manufacturers as well as other independent companies
have these standards in their catalogs; this procedure is
also available in every VNA rmware. The main draw-
backs of SOLT are
*
Perfect knowledge of all standards is required.
*
A large number of standard connection and measure-
ments are needed.
*
A direct port connection (i.e., the thru) is mandatory.
The degree of measurement accuracy following the cali-
bration procedure is strongly contingent on the
*
Number of standards implied during the calibration,
*
Degree of their knowledge,
*
Quality of the interconnections.
The SOLT ideally requires a perfect knowledge of all the
standards used, and this condition obviously cannot be
achieved.
Other calibration techniques, as TRL or LRM, do not
require a complete set of fully known calibration stan-
dards, and for this reason they are called self-calibra-
tions. In particular, the TRL technique is based on the use
of one direct connection (i.e., the thru), one reference
transmission line of unknown length (i.e., the line), and
an unknown reection, measured at both VNA ports. Since
the direct port connection cannot be considered standard
and the reection is unknown, this technique requires
only the knowledge of the line reference impedance.
Furthermore, it has been proved [44] that TRL allows
an accurate definition of the scattering parameters in
terms of traveling waves referred to the line reference im-
pedance, which is easy to correlate to the line mechanical
properties. For this reason, TRL is commonly used as a
reference calibration technique at metrology laboratories.
The only signicative drawback of TRL is the working
bandwidth, which is limited below the resonant frequency
of the line. At this frequency, the line does not introduce
any phase shift and is undistinguishable from the thru
connection (zero length, for definition), thus invalidating
the calibration. To overcome this problem, a set of lines of
varying lengths is normally used, and their measure-
ments combine to achieve a broad frequency coverage [45].
Another very popular self-calibration method is the line
reect match (LRM) technique. It uses a fully known one-
port matched load to substitute the line of the TRL algo-
rithm and is particularly useful for on wafer applications,
where probe movement is difcult. Very high-precision la-
ser trimmed resistors are normally used as match stan-
dards, offering constant resistance and low parasitic
effects up to 50 GHz.
To close this brief overview on the most recent calibra-
tion techniques, its worth noting the short/open load re-
ciprocal (SOLR) [31], which requires three fully known
one-port standards and an unknown, but reciprocal, two-
port device. This technique avoids the use of a fully known
two-port device, such as the thru, which is compulsory in
all other calibrations. Thus SOLR is very useful in many
situations where the direct port connection cannot be
achieved or substituted with an accurate two-port device,
such as when the DUT ports have the same gender or
when they are too far apart.
More recently, rather than describing the standards
through electrical models, a precharacterization tech-
nique has been used. The standards are measured in a
certied laboratory with TRL-calibrated VNAs, and data
les substitute the standard electrical models.
This approach is widely used for the so called electronic
calibrator module, generally a pin (positiveintrinsicneg-
ative) diode network that presents a precharacterized se-
ries of electrical states that simulate the different
standard insertions. The user connects only the electron-
ic calibrator to the VNA, while a computer changes the
impedance shown by the module at its ports and measures
them through the VNA, as they would be different stan-
dard connections. The error boxes are computed using a
SOLT-adapted algorithm [25,60,66], where the standard
parameters are substituted by the previously measured
data les.
3.2. Time-Domain Reectometry
The technique of time-domain reectometry (TDR) was
introduced in the early 1960s as a simple way to charac-
terize the position and type of a reection along a line,
using an oscilloscope. In particular, using a step generator
and sampling the reected waveform with an oscilloscope,
the impedance of simple discontinuities could be calculat-
ed [58].
While the traditional TDR was useful as a qualitative
tool, it has some limitations affecting its accuracy and
utility:
*
Limited risetime
*
Sampling scope jitter
*
Poor dynamic range
*
DC path needed
Common TDR systems based on sampling scope and
step or pulse generator, such as the one sketched in
Fig. 11, are still commonly used, especially for digital
bus characterization. After the introduction of VNA and
digital signal processing, it was clear that the reection
coefcient as a function of time could be easily obtained by
the inverse Fourier transform of the network reection
coefcient as a function of frequency, which is normally
measured by the VNA. This technique, called synthetic
TDR, is now widely adopted in every VNA and is used in
many elds, from cable discontinuity detection to parasitic
interconnection modeling.
The use of frequency-domain data, followed by data
processing, has several advantages over the traditional
TDR, in particular
*
Exact knowledge of the step or pulseshape and equiv-
alent bandwidth
2808 MICROWAVE MEASUREMENTS
*
Vectorial error corrected data that avoid spurious
reections from cables and interconnections
*
High dynamic range
Different types of excitation can be easily computed,
and the time-domain network response is normally shown
on the VNA screen; furthermore, the use of sophisticated
signal processing tools, such as time-domain gate function
and frequency-domain windows, allow one to greatly in-
crease the technique capabilities. For example, by gating
the time-domain response of a particular discontinuity
and retransforming the data into the frequency domain,
accurate modeling of interconnections and launchers can
be obtained [11,12,58].
3.3. Multiport and Differential Measurements
The VNA is currently evolving from the basic two-port in-
strument toward a more sophisticated multiport system.
Typical currently available multiport VNA has 4 ports,
while 8 or even 12 ports are available at low-frequency
bands. Applications of such VNAs span from the television
distribution systems up to bus characterization of differ-
ential digital circuits.
In this latter case the traditional definition of scatter-
ing parameters has been extended to the differential
S parameters [14], which are obtained as the difference
between the traditional single-ended lines and allow one
to better highlight the behavior of microwave systems
where the signal ows on coupled lines rather than on a
single-ended ones. This technique is very promising in the
digital marketplace, where the bus speed is reaching the
microwave eld and the crosstalk effects need to be pre-
cisely controlled and simulated.
4. ADVANCED MICROWAVE MEASUREMENTS
The more recent technological advances require the evo-
lution of microwave measurements to very complex setups
that can perform different types of characterization simul-
taneously and can extract more and more information for
the new modeling and designing needs. Nonlinear tech-
niques for active devices will be briefly described in the
following text.
4.1. Nonlinear and Load-Pull Measurements
When the devices work under a strong nonlinear state, the
models and best operating conditions cannot be often pre-
dicted by linear analysis or measurements, since input
and output matching networks (as well as bias point)
inuence the performance of any transistor or power am-
plier. Therefore, a traditional way to characterize a
transistor for power amplier applications consists in
effectively trimming the load G
L
and the source G
S

reection coefcients seen by the device under nonlinear


conditions, to meet the required amplier specications
[21]. In this case of power ampliers, transistor nonlin-
earities play a fundamental role and the optimum loading
conditions may be significantly different from the linear
case, where S parameters are used to compute the best
loading conditions.
Measurement setups that allow one to change the load-
ing conditions are often referred as load- and source-pull
systems [21,24]. Load-pull application examples are high
power amplier, mixer [43], and oscillator design [36],
while source-pull systems have important applications
in low-noise amplier design, where the best noise gure
and, in general, the transistor noise parameters are
found by applying different source impedance values
[3,16,22,42].
A simplied scheme of a generic real-time load source-
pull system is shown in Fig. 12. To obtain a variable load,
passive networks with manual or automatic variable ele-
ments can be used (slug tuners, solid state tuners, etc.)
input
load
port 2
a
m1 m1 m2 m2

S
port 1 port 3
microwave
source
b
output
load
a
1
b
1
a
2
b
2
a
b

in
a
t
b
t

out

t
VNA
Figure 12. Real-time load/source-pull setup.
step
generator
display
(a)
t
r
reflection
discontinuity
(b)
Figure 11. TDR system based on sampling scope and pulse gen-
erator (a) and typical time-domain pulse response (b).
MICROWAVE MEASUREMENTS 2809
[1,46,55,62]. The alternatives are active loads, which elec-
tronically synthesize the required reection coefcients by
properly amplifying, phase shifting, and combining micro-
wave signals [10,59].
In the simplest load-pull implementation, the device is
driven by the microwave source at a single frequency, and
its performance is measured while physically changing G
L
or G
S
. Vector measurement techniques with VNA allow
real-time determination of the load and source reection
coefcients, as well as input and output power, through
the dual directional couplers, shown in Fig. 12.
A very popular and less expensive technique exploits
power meters, connected to the systemthrough directional
couplers, at the input and output of the DUT. Since no
vectorial real-time correction is possible in this case, the
measurements rely on the accuracy of directional coupler
and tuner precalibration. For this reason, VNA-based sys-
tems are faster and have lower uncertainties than do pow-
er-meter-based systems, as has been demonstrated in
Ref. 33.
4.1.1. Harmonic Measurements. Load-pull measure-
ments are inherently for single-frequency or small-band-
width applications; nevertheless, nonlinear devices
produce harmonics, and the loading conditions at harmon-
ic frequencies may significantly affect device performance,
as proved by theory [57] and experiments [30]. Harmonic
source- and load-pull systems allow one to change G
S
and
G
L
values at a discrete set of frequencies (typically two or
three), while measuring device performance over the en-
tire spectrum of interest. A typical application is the
design of high-efciency solid-state ampliers and trans-
ceiver output stages [6,54]. The introduction of harmonic
tuners [47,48] has provided a compact and economic solu-
tion for harmonic source load-pull testsets. However, like
any other passive tuner, harmonic tuners cannot offer
highly reective terminations at higher frequencies, due
to losses; a good solution is, again, the active load, realized
by combining, with power splitters and combiners, more
single-frequency active loads [37,48,51].
4.1.2. Intermodulation Set-ups. By driving the device
with two tones or with digitally modulated signals, inter-
modulation or adjacent-channel power ratio (ACPR) are
traditionally measured to evaluate the linearity charac-
teristics of the transistor. A typical application is the pow-
er amplier design for CDMA- and WCDMA-based
telecommunication systems, where the phone and base
station performance is more sensitive by amplitude dis-
tortion than in previous access technologies based on con-
stant-envelope modulation schemes.
Like all the other device characteristics, intermodula-
tion distortion depends on the loading conditions; there-
fore, load-pull systems have evolved to integrated load-
pull and intermodulation measurement capabilities, as
shown in Fig. 13 [23].
4.2. Time-Domain Waveform Measurements
The growth of strongly nonlinear devices for wireless ap-
plications (mobile units and base stations), has compelled
the development of another measurement philosophy:
time-domain waveform measurements. This technique
can provide important information for both device model-
ing [35,49] and design [9,13]. For example, if the wave-
forms at the input and output of the device are directly
accessible, the effects of the harmonic terminations [8] or
of the envelope impedance [65] can be studied directly in
the time domain.
Time-domain waveforms are commonly measured by
sampling oscilloscopes, which are based on microwave sig-
nal subsampling, as sketched in Fig. 5. The IF lter is not
a narrow-, bandpass lter one as in the VNA, but a low-
pass lter, and IF signal spectra become exact copies
of the microwave signals. The magnitude and phase
f
1
microwave
auxiliary
source
port 2 port 1 port 3
a
1
b
1
a
2
b
2
a
t
b
t
isolators
spectrum
analyzer
f
0
microwave
auxiliary
source
output
load
a
m1
b
m1
a
m2
b
m2
VNA
Figure 13. Load-pull system for intermodulation mea-
surements [23].
switchable
lowpass
filter
microwave
sampler
CH1
CH2
synt
IF
step gain
amplifier
jP
ADC MEM DSP
display
ADC MEM DSP
trigger
circuitry
Figure 14. Simplied block scheme of
a time-domain waveform receiver.
2810 MICROWAVE MEASUREMENTS
relationships between the spectral components are main-
tained and are measured by A/D conversion and FFT of
the IF signals, as shown in Fig. 14 [19,20].
Calibrations can be performed in the frequency do-
main, after fast Fourier transform (FFT); then, corrected
measurements are displayed again in the time domain,
after an inverse fast Fourier transform (IFFT) [29]. The
main drawbacks of waveform measurements is the low
dynamic range and speed.
BIBLIOGRAPHY
1. V. Adamian, 2-26.5GHz on-wafer noise and S-parameter mea-
surements using a solid state tuner, 34th ARFTG Conf. Di-
gest, Dec. 1989, pp. 3340.
2. V. Adamian, A novel procedure for network analyzer calibra-
tion and verication, 41st ARFTG Conf. Digest, June 1993,
pp. 817.
3. V. Adamian and A. Uhlir, A novel procedure for receiver noise
characterization, IEEE Trans. Instrum. Meas. IM-22:181182
(June 1973).
4. J. W. Allen, F. R. Clague, N. T. Larsen, and M. P. Weidman,
NIST Microwave Power Standards in Waveguide, NIST Tech-
nical Note 1511, Boulder, CO, 1999.
5. R. Anderson and O. Dennison, An advanced new network an-
alyzer for sweep-measuring amplitude and phase from 0.1 to
12.4GHz, Hewlett-Packard J. 18:29 (Feb. 1967).
6. I. Bahl, E. Grifn, A. Geissberger, C. Andricos, and T.
Brukiewa, Class-B power MMIC ampliers with 70 percent
power-added efciency, IEEE Trans. Microwave Theory Tech.
MTT-37:13151320 (Sept. 1989).
7. A. Bailey, Microwave Measurements, 2nd ed., Peter Pere-
grinus, London, 1989, Chaps. 1516.
8. D. Barataud, C. Arnaud, B. Thibaud, M. Campovecchio, J. M.
Nebus, and J. P. Villotte, Measurements of time-domain volt-
age/current waveforms at RF and microwave frequencies
based on the use of a vector network analyzer for character-
ization of nonlinear devicesapplication to high-efciency
power ampliers and frequency-multipliers optimization,
IEEE Trans. Instrum. Meas. IM-47:12591264 (Oct. 1998).
9. D. Barataud, M. Campovecchio, and J.-M. Nebus, Optimum
design of very high-efciency microwave power ampliers
based on time-domain harmonic load-pull measurements,
IEEE Trans. Microwave Theory Tech. 49(6):11071112 (June
2001).
10. G. P. Bava, U. Pisani, and V. Pozzolo, Active load technique for
load-pull characterization at microwave frequencies, Elec-
tron. Lett. 18(4):178179 (Feb. 1982).
11. C. Beccari, A. Ferrero, and U. Pisani, In-xture calibration of
an s-parameter measuring system by means of time domain
reectometry, 32nd ARFTG Conf. Digest, Phoenix, AZ, Dec.
1988, pp. 8997.
12. C. Beccari, A. Ferrero, and U. Pisani, Time domain reect-
ometry applied to mmic passive component modeling, 33rd
ARFTG Conf. Digest, Long Beach, CA, June 1989, pp. 110.
13. J. Benedikt, R. Gaddi, P. J. Tasker, and M. Goss, High-power
time-domain measurement systemwith active harmonic load-
pull for high-efciency base-stantion amplier design, IEEE
Trans. Microwave Theory Tech. 48(12):26172624 (Dec. 2000).
14. D. Bockelman and W. Eisenstadt, Combined differential
and common-mode scattering parameters: Theory and simu-
lation, IEEE Microwave Theory Tech. MTT-43:15301539
(July 1995).
15. R. Carlson and F. Weinert, The RF vector voltmeteran im-
portant new instrument for amplitude and phase measure-
ments from 1 MHz to 1000MHz, Hewlett-Packard J. 17:212
(May 1966).
16. G. Caruso and M. Sannino, Determination of microwave two-
port noise parameters through computer-aided frequency-
conversion techniques, IEEE Trans. Microwave Theory Tech.
MTT-27:779783 (Sept. 1979).
17. F. R. Clague, A Calibration Service for Coaxial Reference
Standards for Microwave Power, NIST Technical Note 1374,
Boulder, CO, 1995.
18. Hewlett-Packard Co., HP8510 Network Analyzer System Op-
erating and Programming Manual, HP 08510-90005, Santa
Rosa, CA, 1985.
19. Hewlett-Packard Co., The Microwave Transition Analyzer: A
Versatile Measurement Set for Bench and Test, HP Product
Note 70820-1, Rohnert Park, CA, 1991.
20. Hewlett-Packard Co., The Microwave Transition Analyzer:
Measure 25 ps Transitions in Switched and Pulsed Microwave
Components, HP Product Note 70820-2, Rohnert Park, CA,
1991.
21. S. C. Cripps, RF Power Ampliers for Wireless Communica-
tions, Artech House, Boston, 1999.
22. A. Davidson, B. Leake, and E. Strid, Accuracy improve-
ments in microwave noise parameter measurements, IEEE
Trans. Microwave Theory Tech. MTT-37:19731978 (Dec.
1979).
23. M. Demmler, B. Hughes, and A. Cognata, A 0.550GHz on-
wafer, intermodulation, load-pull and power measurement
system, IEEE MTT-S Int. Microwave Symp. Digest, Orlando,
FL, May 1995, pp. 10411044.
24. R. Soares, ed., GaAs MESFET Circuit Design, Artech House,
Norwood, MA, 1988, Chap. 6.3.
25. G. Engen, R. Judish, and J. Juroshek, The multistate two
port: An alternative transfer standard, 41st ARFTG Conf. Di-
gest, June 1994, pp. 1118.
26. H. J. Eul and B. Schieck, Thru-match-reect: One result of a
rigorous theory for de-embedding and network analyzer cal-
ibration, Proc. 18th European Microwave Conf., Stockholm,
1988, pp. 909914.
27. H. J. Eul and B. Schieck, A generalized theory and new cal-
ibration procedures for network analyzer self-calibration,
IEEE Trans. Microwave Theory Tech. MTT-39:724731 (April
1991).
28. H. J. Eul and B. Schieck, Reducing the number of calibration
standards for network analyzer calibration, IEEE Trans.
Instrum. Meas. IM-40:732735 (Aug. 1991).
29. P. Ferrari, G. Angenieux, and B. Flechet, A complete calibra-
tion procedure for time domain network analyzers, IEEE
MTT-S Int. Microwave Symp. Digest, Albuquerque, NM,
June 1992, pp. 14511454.
30. A. Ferrero and U. Pisani, Large signal 2nd harmonic on wafer
MESFET characterization, 36th ARFTG Conf. Digest, Mon-
terrey, CA, Dec. 1990, pp. 101106.
31. A. Ferrero and U. Pisani, Two-port network analyzer calibra-
tion using an unknown thru, IEEE Microwave Guided
Waves Lett. MGWL-2:505507 (Dec. 1992).
32. A. Ferrero and F. Sanpietro, A simplied algorithm for leaky
network analyzer calibration, IEEE Microwave Guided Wave
Lett. MGWL-5:119121 (April 1995).
33. A. Ferrero, V. Teppati, and A. Carullo, Accuracy evaluation of
on-wafer load-pull measurements, IEEE Trans. Microwave
Theory Tech. MTT-49(1):3943 (Jan. 2001).
MICROWAVE MEASUREMENTS 2811
34. H. T. Friis, Noise gures of radio receivers, Proc. IRE, July
1944, pp. 419422.
35. R. Gaddi, J. A. Pla, J. Benedikt, and P. J. Tasker, Ldmos
electro-thermal model validation from large-signal time-do-
main measurements, IEEE MTT-S Symp. Digest, 2001, Vol. 1,
pp. 399402.
36. F. M. Ghannouchi and R. Bosisio, Source-pull/load-pull oscil-
lator measurements at microwave/mm wave frequencies,
IEEE Trans. Microwave Theory Tech. MTT-41:3235 (Feb.
1992).
37. F. M. Ghannouchi, R. Larose, and R. Bosisio, A new multi-
harmonic loading method for large-signal microwave transis-
tor characterization, IEEE Trans. Microwave Theory Tech.
MTT-39:986992 (June 1991).
38. R. A. Hackborn, An automatic network analyzer system,
Microwave J. 11:4552 (May 1968).
39. H. Van Hamme and M. Van Den Bossche, Flexible vector net-
work analyzer calibration with accuracy bounds using an
8-term or a 16-term error correction model, IEEE Microwave
Theory Tech. MTT-42:976987 (Aug. 1991).
40. J. Helton and R. Speciale, A complete and unambiguous so-
lution to the super-TSD multiport-calibration problem, IEEE
MTT-S Int. Microwave Symp. Digest, Boston, May 1983,
pp. 251252.
41. J. Jargon, R. Marks, and D. Rytting, Robust SOLT and alter-
native calibrations for four-sampler vector network analyz-
ers, IEEE Trans. Microwave Theory Tech. MTT-47:20082013
(Oct. 1999).
42. D. Le and F. Ghannouchi, Noise measurements of microwave
transistor using and uncalibrated mechanical stub tuner and
a built-in reverse six-port reectometer, IEEE Trans. In-
strum. Meas. IM-44:847852 (Aug. 1995).
43. D. Le and F. Ghannouchi, Multitone characterization and de-
sign of FETresistive mixers based on combined active source-
pull/load-pull techniques, IEEE Trans. Microwave Theory
Tech. MTT-46:12011208 (Sept. 1998).
44. R. Marks and D. Williams, A general waveguide circuit the-
ory, J. Res. NIST 97:533561 (Sept. 1992).
45. R. B. Marks, A multiline method of network analyzer cali-
bration, IEEE Trans. Microwave Theory Tech. MTT-39:1205
1215 (July 1991).
46. C. McIntosh, R. Pollard, and R. Miles, Novel MMIC source-
impedance tuners for on-wafer microwave noise-parameter
measurements, IEEE Trans. Microwave Theory Tech. MTT-
47:125131 (Feb. 1999).
47. ATN Microwaves, A load-pull system with harmonic tuning,
Microwave J. 128132 (March 1996).
48. Focus Microwaves, An affordable harmonic load pull setup,
Microwave J. 180182 (Oct. 1998).
49. D. G. Morgan, G. D. Edwards, A. Phillips, and P. J. Tasker,
Full extraction of phemt state functions using time domain
measurements, IEEE MTT-S Symp. Digest, 2001, Vol. 2, pp.
823826.
50. HP8510C Network Analyzer System Specication and Perfor-
mance Verication Program Manual. HP8510 Software Sup-
plement.
51. D. B. Poulin and R. B. Stancliff, Harmonic load-pull. IEEE
MTT-S Int. Microwave Symp. Digest, 79(1):185187 (April
1979).
52. M. Roos and V. Sotoudeh, A measurement and calibration
technique for accurate measurement of amplier S parame-
ters, IEEE MTT-S Int. Microwave Symp. Digest, Las Vegas,
NV, June 1987, pp. 449451.
53. D. Rytting and S. Sanders, A system for automatic network
analysis, Hewlett-Packard J. 21:210 (Feb. 1970).
54. F. Sechi, High efciency microwave FET power amplier, Mi-
crowave J. 5963 (Nov. 1981).
55. F. Sechi, R. Paglione, B. Perlman, and J. Brown, A computer
controlled microwave tuner for automated load pull, RCA Rev.
44:566572 (Dec. 1983).
56. K. J. Silvonen, A general approach to network analyzer cal-
ibration, IEEE Microwave Theory Tech. MTT-40:754759
(April 1992).
57. D. Snider, A theoretical analysis and experimental conrma-
tion of the optimally loaded and overdriven rf power amplier,
IEEE Trans. Electron. Dev. ED-14:851857 (Dec. 1967).
58. H. E. Stinehelfer, Sr., Discussion of de-embedding techniques
using time-domain analysis, IEEE Proc. 74(1):9094 (Jan.
1986).
59. Y. Takayama, A new load-pull characterization method for
microwave power transistor, IEEE MTT-S Int. Microwave
Symp. Digest, Cherry Hill, NJ, June 1976, pp. 218220.
60. Agilent Technologies, Agilent Electronic Calibration (ECal)
Modules for Vector Network Analyzers, 5963-3743E, Santa
Rosa, CA, 2003.
61. Agilent Technologies, Fundamental of RF and Microwave
Power Measurements, 5988-9213/6E, Santa Rosa, CA, 2003.
62. C. Tsironis, A novel design method of wideband power ampli-
er, Microwave J. 303304 (May 1992).
63. V. Pozzolo and U. Pisani, Errors caused by directional cou-
plers and impedance mismatches in a test set for s-parameter
measurements, Alta Frequenza 37(10):911915 (April 1968).
64. R. L. Vaitkus, Wide-band de-embedding with a short, an open
and a through line, Proc. IEEE 74:7174 (Jan. 1986).
65. D. J. Williams, J. Leckey, and P. J. Tasker, A study of the effect
of envelope impedance on intermodulation asymmetry using
a two-tone time domain measurement system, IEEE MTT-S
Int. Microwave Symp. Digest, Seattle, WA, June 2002, pp.
18411844.
66. K. Wong and R. Grewal, Microwave electronic calibration:
Transferring standards lab accuracy to the production oor,
Microwave J. 94105 (Sept. 1994).
MICROWAVE MIXERS
R. S. TAHIM
RST Scientific Research, Inc.
New York
1. INTRODUCTION
Mixer as a component is found in almost all kinds of mi-
crowave and millimeter-wave systems that include vari-
ous communication systems, interferometer systems,
phased arrays, sensors, and radar systems. Microwave
mixers form a highly mature technology [1]. However,
there are still interesting and new circuits that continue to
be developed in order to make the microwave/millimeter-
wave systems more efcient. A number of microwave
systems are extremely vital to the defense applications
involving the collection and distribution of information
2812 MICROWAVE MIXERS
both to and from the forward-deployed troops and between
the command centers and the troops in the eld. Wide-
band radar systems that incorporate wideband mixers are
needed in large-scale surveillance, tracking targets, iden-
tication, and reconnaissance applications. Therefore, the
U.S. military has placed significant emphasis on the de-
velopment of wideband microwave and millimeter-wave
transceiver systems for both radar and data transmission
systems at high frequencies. Such wideband systems are
needed for transferring high-value data such as videos,
high-resolution images, and radar, optical, and computer
data that military personnel need to carry out the military
operations more efciently and effectively.
Radar and the wireless telecommunications industry
have also established itself as one of the strongest growth
areas in the commercial sector side. Although vigorous
expansion is expected to continue in the established wire-
less applications, the next major growth areas in the wire-
less revolution have already begun to emerge. As a result,
variety of new applications are envisioned in such areas as
homeland security, intelligent highway systems, autono-
mous landing systems, mobile satellite communications,
intersatellite links, tracking and data relay, direct conver-
sion receivers, ultra-wide-band (UWB) systems, advanced
security systems, and local multipoint distribution sys-
tems. In most of these applications, low-noise receivers
that include high-performance microwave mixers are
desirable.
In all RF, microwave, and millimeter-wave systems
there are two major parts: the transmitter and the receiv-
er. In a microwave communication system, the transmit-
ter and receiver systems are installed at different
locations. The message is sent from the transmitter to
the receiver using the antennas. In a radar system, the
transmitter and the receiver are located very close togeth-
er. The transmitted signal is reected by a target or the
targets to the receiver; thus the information about the
target (range and velocity) is obtained by the radar. Along
with the development of communication and radar sys-
tems for military applications, another highly interesting
area called electronic warfare (EW) has also emerged over
the years. In EW applications, the receivers are used to
intercept signals of limited information from a hostile
transmitter, while a jamming transmitter is used to gen-
erate false information or noise to mask the true signal
received by the possible hostile radar receiver.
2. MIXER OPERATION
In the microwave systems for both radar and communica-
tions, the incoming signal at microwave frequency (f
s
) is
downconverted in a mixer to generate a low intermediate
frequency (f
IF
), which is further processed to generate the
baseband signal in order to retrieve the required informa-
tion or data. Frequency down-conversion is accomplished
by utilizing the properties of a nonlinear impedance ele-
ment of the mixer, which is simultaneously interacting
with the receive signal (f
s
) and a usually a much stronger
signal at f
LO
generated by a local oscillator. In the mixer
conversion process, information about the frequency,
amplitude, and phase of the original signal is preserved.
Therefore, we nd the mixer as a component in almost all
kinds of microwave systems that include various commu-
nication systems, interferometer systems, phased arrays,
radars, and all kinds of instruments such as frequency
upconverters, modulators, frequency synthesizers, spec-
trum analyzers, and network analyzers.
3. MIXERS IN MICROWAVE/MM-WAVE SYSTEMS
Microwave mixers in a communication receiver and radar
system play a critical role in the data transfer and the
identication of the targets. Figure 1 is a basic block di-
agram of a wireless communication system, which gener-
ally consists of two transceivers, each transmitting and
receiving the microwave signals. In the modern telecom-
munication systems, a receiver consists of a low-noise-am-
plier (LNA) followed by a mixer circuit.
A monopulse receiver topology consisting of low-noise
amplier and microwave/millimeter-wave mixers is
shown in Fig. 2 [2]. In the receiver system, the signal
f
s
from the antennas is rst amplied in the low-noise
amplier before being downconverted using a balanced
Transceiver Transceiver
TX
RX TX
RX
Modulator/demodulator Modulator/demodulator
Data interface Data interface
Figure 1. Schematic of a microwave communication system.
LO
W-Band MMIC
amplifier
Subharmonic
mixer
IF 2
IF 3
CPW
Transmission
lines
LO
CPW-fed
slot-ring
antenna
1st Stage
IF amp
IF 1
IF 4
Extended
hemopherical
silicon lens
Figure 2. A monopulse receiver topology with four slot-ring
antennas.
MICROWAVE MIXERS 2813
mixer to the IF frequency band. A low noise and a highly
stable local oscillator source or a dielectric resonator os-
cillator (DRO) is generally used as the local-oscillator (LO)
source to drive the mixer (frequency downconverter).
A system block diagram of a typical radar system il-
lustrating the use of a microwave mixer in the system is
shown in Fig. 3. In the radar receiver, the returned echo is
mixed with the stable oscillator frequency to obtain the
Doppler shift frequency [3]. The Doppler frequency is
passed through a set of Doppler lters to determine the
target velocity. The return signal is also compared with
the reference time in a time comparator to determine the
target range. In the system, the exciter provides the local
oscillator power to the downconverter (mixer) to generate
the desired intermediate frequency (IF).
In the receiver, as long as the amplier gain before the
mixer is sufciently high to mask the contribution of the
mixer noise gure, the overall noise gure of the receiver
is determined mainly by the amplier noise gure. How-
ever, at millimeter-wave range, the way to obtain high
gain for the amplier is to cascade several ampliers. In
such a case, either the cost or the complexity of the circuit
is increased. In order to achieve a low receiver noise gure
in the system, it is more desirable to use the mixer in the
receiver front end as a rst element and to minimize the
mixer conversion loss to achieve the low receiver noise
gure.
The receiver and mixer parameters in the microwave
communication system, are related as [4]
S=NG
T
P
T
G
R
=T
S
e
p
=kDf l
2
=4pR
2
1
where
G
T
transmit antenna gain
P
T
input power to antenna
G
R
receive antenna gain
T
S
effective noise temperature at receiver input
k Boltzmann constant 1.37910
23
228.6dB
(W/Hz K)
Df Bbandwidth of receiver
The receiver noise gure in the system in which mixer
is used in the front end is given by [5]
F L
c
N
r
F
IF
1 2
where L
c
is the total conversion loss (including frequency
conversion loss, diode series resistance loss, and the mis-
match losses) in the mixer, then F
IF
is the noise gure of
the IF amplier followed by the mixer and N
r
is the diode
noise ratio.
4. MIXER CONFIGURATIONS
Seven basic types of mixers have been developed for var-
ious applications:
1. Single-ended mixers
2. Single-balanced mixers
3. Double-balanced mixers
4. Subharmonically pumped mixers
Time Pulse
modulator
Stable
oscillator
Up-
converter
Amplifier Circulator
Antenna
Mixer
IF
amplifier
Coherent
oscillator
Doppler
mixer
Time
comparator
Doppler
filters
Range Velocity
Figure 3. Microwave mixer in Doppler radar
system.
2814 MICROWAVE MIXERS
5. Image-rejection mixers
6. Image-enhanced mixers
7. Quadrature mixers
The single-ended mixer is the simplest mixer congura-
tion using one mixer diode. However, microwave mixer
congurations using two or more Schottky diodes are
quite common and have proved to provide substantial ad-
vantages in terms of operating bandwidth, intermodula-
tion, spurs, LO noise cancellation, high dynamic range,
low intermodulation, and so on [6,7]. At high microwave
and millimeter-wave frequency bands, subharmonically
pumped mixer designs are preferred because the LO fre-
quency requirement in subharmonic mixers is roughly
half that required in the fundamental mixer design. The
two-diode mixer also provides substantial AM local oscil-
lator noise suppression [7]. The noise associated with the
LO sources (used in the frequency downconverter), when
translated at the IF frequency band during the conversion
process, can be a source of problems in the receiver. An-
other important property of subharmonically pumped
mixer is the strong attenuation of downconverted local
oscillator noise available at the IF output.
5. MIXER PARAMETERS
High-performance mixers in the microwave and millime-
ter-wave receivers form an important part in the current
telecommunications, meteorology, and electronic warfare
systems. Important mixer parameters that determine its
usefulness in such microwave and millimeter-wave sys-
tems are
*
Conversion loss
*
Operating bandwidth
*
Local-oscillator power requirements
*
Isolation between the RF, LO, and IF ports
*
RF and IF impedance match
*
Intermodulation
*
LO noise cancellation
*
Dynamic range
The conversion loss of a microwave mixer is dened as the
ratio of the IF output power to the signal input power.
Conversion loss measures the efciency with which mixer
converts RF energy to the IF energy and is one of the most
important mixer parameters. Conversion loss L
c
is ex-
pressed in the form
L
c
L
o
L
p
3
where L
o
includes the intrinsic conversion loss arising
from the conversion process within the nonlinear resis-
tance of the diode plus the signal loss associated with the
impedance mismatch losses at the RF and IF ports. L
p
is
the parasitic loss associated with the parasitic elements of
the diode, such as junction capacitance C
j
, the spreading
resistance R
s
, and series inductance L
s
.
The junction capacitance C
j
, series resistance R
s
, and
other parasitic reactance L
s
are shown in Fig. 4 for the
diode equivalent circuit, and they are frequency-depen-
dent. Their effect on controlling the mixer performance
such as operating bandwidth and the conversion efciency
becomes much stronger at high microwave and millime-
ter-wave frequencies [8].
Since C
j
pd
2
and R
s
pd
1
, where d is the diameter of
the junction, d should be reduced as the operating fre-
quency is increased. Reducing the size of the junction area
of the diode increases the current density through the di-
ode and consequently may increase the intrinsic conver-
sion loss of the mixer [9].
In addition to the parasitic losses, the conversion loss
depends on the losses in the RF circuitry surrounding the
mixer, the quality of the nonlinear device, the mismatch
between the RF circuit and the diodes and the bias level to
the diodes. When no energy is dissipated at the idler fre-
quencies, the fundamental limit on the conversion loss of a
mixer is 3dB [10]. The discrepancy between the measured
mixer loss and the predicted loss is shown to have three
main causes and is generally divided into the following
three parts:
L
1
mismatch loss due to impedance mismatch at RF
and IF ports
L
2
diode parasitic losses due to junction capacitance
and series resistance
L
3
intrinsic junction loss of the diode
The overall conversion loss L is the sum of three losses:
LL
1
L
2
L
3
all indB 4
The conversion loss is a very important mixer parameter,
especially in the receivers without RF amplier in front of
the mixer, since the rst component at the input of a re-
ceiver has the most inuence on the noise gure of the
receiver. In a mixer circuit, in which the 2f
LO
f
s
frequency
product lies close to the RF signal, the conversion loss can
be further improved by extracting the power associated
with 2f
LO
f
s
signal [11] by remixing this signal with LO to
generate more IF power.
L
p
R
s
C
j
C
p
G
Figure 4. Equivalent circuit of a mixer diode.
MICROWAVE MIXERS 2815
When the signal power approaches the same power
level as the local oscillator, the mixer becomes saturated;
specifically, the output IF power no longer increases lin-
early with the input power. Since the mixer is a three-port
device, isolation between RF and LO ports, isolation be-
tween RF and IF ports, and isolation between LO and IF
ports is very important. Isolation is the measure of circuit
balance within the mixer. Poor isolation between the RF
and LO means that a LO signal can reach the RF port and
power can radiate through the receiving antenna [12]. In
receivers, where LNA forms the front end, the poor isola-
tion between the LO and RF can result in signal distortion
because the FET devices in the amplier also exhibit non-
linear characteristics. Isolation between the mixer ports
becomes a problem in a multichannel receiver in which a
single LO is used to feed a number of mixers [13].
In the resistive mixer operation, the signal level in a
mixer is significantly lower than that of LO power and
does not therefore perturb the LO-pumped conductance
waveform. Under such conditions, no harmonics of
signal are generated. However, if the RF and LO signal
powers are at sufciently high level, the harmonics and
inter-modulation products are generated according to
7mf
RF
7nf
LO
, where m0,1,2,y and n0,1,2,y, al-
though a lter is used at the mixer output to allow only
f
IF
to pass. In case the intermodulation frequency product
falls within the IF frequency range, it must be 40 dBc
below the actual IF signal level.
In a wireless system, which must operate in severe in-
terference environments, microwave receivers are espe-
cially susceptible to intermodulation frequency products
generated in mixers. Spurious signals or distortion in mix-
ers continue to be problems in virtually all microwave
systems. Mixers are frequently the dominant components
in establishing the systems distortion performance. Inter-
modulation (IM) products are undesirable mixer generat-
ed output products exiting the mixer from any port. Any
RF frequency that satises the following equation can
generate spurious responses, spurs, in a mixer
mf
RF
-- nf
LO
f
IF
5
where f
IF
is the desired IF frequency.
6. MIXER DESIGN TOPOLOGIES
Although single-ended mixers occasionally are used, most
practical mixers are balanced mixers incorporating more
than one Schottky mixer diode. Balanced mixers require
baluns or hybrids, which largely determine the bandwidth
and overall performance of the mixer. Thus, they have
been the subject of considerable research interest. As a
result, a number of balanced mixer congurations have
been developed for various applications. A few of these
design congurations are:
Planar balanced mixer
Crossbar balanced mixer
Finline balanced mixer
Subharmonically pumped mixer
In spite of the maturity of FET circuits, diode mixers are
still widely used in microwave circuits, because diode mix-
ers have an important advantage over FETs and bipolar
devices. The Schottky barrier diode is inherently a resis-
tive device and as such has very wide bandwidth. The
bandwidths of diode mixers are generally limited primar-
ily by the bandwidths of baluns, not the diodes [14]. FETs
have high-Q gate input impedance, causing difculties in
achieving at wide bandwidth as compared to the mixer
designs using Schottky mixer diodes. Diode mixers usual-
ly have 58 dB conversion loss, while active mixers usually
can achieve at least a few decibels of gain. More recently, a
number of design topologies for the balanced mixers
have been developed for various applications. Many of
these design topologies provide state-of-the-art perfor-
mance in terms of bandwidth, conversion loss, and other
parameters.
Marchand baluns have been found to be quite useful for
planar mixer designs and can achieve a 101 bandwidth.
Figure 5 shows the mixer design topology using Marchand
balun [14]. Planar mixers using microstrip transmission-
line couplers operating up to W-band frequency ranges
have also been developed [15].
Figure 6 shows the conguration of broadband micro-
wave integrated circuit crossbar mixer. In the crossbar
mixer RF and LO signals are coupled (using waveguides)
to two Schottky mixer diodes orthogonally. Such congu-
ration provides the necessary isolation between the RF
and LO ports. Such design congurations also provide
very broadband performance. Crossbar mixers have been
developed in high-microwave and millimeter-wave fre-
quency spectra [16,17].
The two-diode subharmonically pumped mixer design
topology on suspended substrate strip transmission line is
shown in Fig. 7. Two Schottky mixer diodes are shunt-
mounted with opposite polarities. Two lowpass lters are
used to separate the signal frequency at o
s
2o
p
7o
IF
, the
LO frequency o
p
, and the IF frequency o
IF
.
Quadrature mixers are important in many communi-
cation systems and in direct-conversion receivers. A quad-
rature mixer typically consists of a pair of balanced mixers
with the input signal split into in phase and quadrature
Figure 5. Planar mixer using Marchand balun.
2816 MICROWAVE MIXERS
IF
Output
Matching
stub
Mixer
Substrate
Sliding
short
RF input
waveguide
IF filter
LO
Figure 6. Crossbar mixer design using sus-
pended stripline conguration.
o
S
WR
10
o
P
o
IF
WR
19
A
A
Quartz substrate
Schottky barrier
diodes
Figure 7. Millimeter-wave subharmonically
pumped mixer.
S.C.
I II
IV
III
Pump
input
r.f. input
i.f. output
O.C.
O.C.
Figure 8. Microstrip layout of subharmoni-
cally pumped mixer.
MICROWAVE MIXERS 2817
parts, resulting in the in-phase and quadrature signals
translated to the intermediate frequency.
A planar microstrip subharmonically pumped mixer
circuit layout is shown in Fig. 8 [18]. In this design, the
short-circuited stub (SC) I does not impede the pump sig-
nal but shorts the RF signal to the ground. The open stub
(OC) II, on the other hand, short circuits the pump signal
without affecting the RF signal. The SC stub and OC stubs
are l/4 in length at LO frequency. Two antiparallel pair of
Schottky mixer diodes are connected between the LO and
RF ports as shown in Fig. 8. The conversion loss of har-
monic mixers is about 0.51 dB higher than that of a con-
ventional balanced mixer [18].
7. CONCLUSIONS
Although mixer design is often viewed as a mature tech-
nology, there seems to be no end to the supply of new ideas
for mixer circuits. Mixer parameters are generally deter-
mined by the system designers who are constantly
researching for high performance, low-cost, manufactur-
ability and integratable designs. This is particularly true
of mixers for wireless and other modern communication
systems and radar systems where distortion is a major
concern. As a result, there are continual developments in
mixer technology. These include new circuits, new devices
and device technologies.
BIBLIOGRAPHY
1. S. A. Mass, Microwave Mixers.
2. S. Raman, S. Barker, and G. M. Rebeiz, A W-band dielectric-
lens-based integrated monopulse radar receiver, IEEE Trans.
Microwave Theory Tech. 46:22832288 (Dec. 1998).
3. G. W. Stimson, Introduction to Airborne Radar, 2nd ed.,
Mendham, NJ, 1998.
4. R. J. Mailloux, Phased Array Antenna Handbook, Artech
House, Boston, 1994.
5. B. L. Sharma, Metal-Semiconductor Schottky Barrier Junc-
tions and Their Applications, Plenum Press, New York, 1984.
6. E. R. Carlson, M. V. Schneider, and T. F. McMaster, Sub-har-
monically pumped millimeter-wave mixers, IEEE Trans. Mi-
crowave Theory Tech. MTT-26:706715 (Oct. 1978).
7. P. S. Henry, B. S. Glance, and M. V. Schneider, Local-oscillator
noise cancellation in sub-harmonically pumped down-con-
verter, IEEE Trans. Microwave Theory Tech. MTT-24:254
257 (May 1976).
8. L. Mania and G. B. Stracca, Effects of the diode junction ca-
pacitance on the conversion loss of microwave mixers, IEEE
Trans. Commun. COM-22:14281435 (Sept. 1974).
9. M. McColl, Conversion loss limitations on Schottky-barrier
mixers, IEEE Trans. Microwave Theory Tech. MTT-25:5459
(Jan. 1977).
10. A. J. Kelly, Fundamental limits on conversion loss of double
side-band resistive mixers, IEEE. Trans. Microwave Theory
Tech. MTT-25:867869 (Nov. 1977).
11. T. H. Oxley et al., Image recovery mixers, Paper presented at
European Microwave Conf., Sweden, 1971.
12. K. Chang, RF and Microwave Wireless Systems, Wiley-Inter-
science, New York, 2000.
13. J. B. Tsui, Microwave Receivers and Related Components,
Avionics Laboratory, U.S. Air Force, Wright Aeronautical Lab-
oratories, 1983.
14. S. A. Maas and K. W. Chang, Broadband, planar, doubly bal-
anced monolithic Ka-band diode mixer, IEEE Microwave and
Millimeter-wave Monolithic Circuits Symp. Digest, 1993, p.
53.
15. K. Chang, D. M. English, R. S. Tahim, A. J. Grote, T. Pham, C.
Sun, G. M. Hayashiabara, P. Yen, and W. Piotrowski, W-band
(75110GHz) microstrip components, IEEETrans. Microwave
Theory Tech. MTT-31:13751382 (Dec. 1985).
16. L. T. Yuan, Design and performance analysis of an octave
bandwidth waveguide mixer, IEEE Trans. Microwave Theory
Tech. MTT-25:10481054 (Dec. 1977).
17. R. S. Tahim, G. M. Hayashibara, and K. Chang, Design and
performance of W-band broad-band integrated circuit mixers,
IEEE Trans. Microwave Theory Tech. MTT-31:277283
(March 1983).
18. J. D. Buchs and G. Begemann, Frequency conversion using
harmonic mixers with resistive diodes, IEE J. Microwave Opt.
Acoust. 2:7176 (May 1978).
MICROWAVE OSCILLATORS
JONGHOON CHOI
AMIR MORTAZAWI
University of Michigan
Ann Arbor, Michigan
Oscillators are an integral part of receiving and transmit-
ting systems used for communication, radar, and other
applications. Any oscillator circuit consists of an active
device and a resonant circuit that determines the frequen-
cy of oscillation. The active device can be either a two-ter-
minal negative-resistance device such as Gunn diode,
IMPATT (impact avalanche transit time) diode, resonant
tunneling diode (RTD), and so on, or a transistor with ap-
propriate feedback to cause instability. In general, an os-
cillator can operate at a xed frequency, or its frequency of
operation can be tunable. The frequency tuning can be
achieved either mechanically or electronically. An oscilla-
tor is usually characterized by its frequency of operation,
output power, noise, long-term frequency stability, and
DC-to-RF efciency.
1. MODELING OF MICROWAVE OSCILLATORS
An oscillator consists of linear circuits and active devices,
and its operation relies on the nonlinear behavior of the
active devices. In order to accurately characterize an os-
cillator, it has to be analyzed using linear and nonlinear
circuit analysis techniques. An oscillator can be analyzed
using either the feedback model or the negative-resistance
model. These two analysis models are identical; however,
the negative-resistance model is the most commonly used
approach in the design of microwave oscillators. The pop-
ularity of the negative-resistance model is due to its sim-
plicity and the fact that it can be related to device and
2818 MICROWAVE OSCILLATORS
circuit reection coefcients since they can be measured
accurately at microwave frequencies. In the negative-re-
sistance model, the oscillator is divided into two parts, the
active device and the embedding passive circuit. In a sim-
plied oscillator analysis, the active device is assumed to
operate under small-signal conditions. This type of anal-
ysis can provide some information about startup condi-
tions for oscillation and predict the approximate frequency
of oscillation. In order to accurately determine the oscil-
lation frequency, output power, stability, and spectral pu-
rity of an oscillator, large-signal analysis of the oscillator
circuit must be performed. Usually simplied large-signal
models for active devices can provide valuable insight into
the operation of oscillators.
1.1. Oscillator Design Using One-Port Negative-Resistance
Devices
Figure 1 shows the simplied block diagram of an one port
oscillator circuit. The small-signal impedance of the neg-
ative-resistance device and the embedding circuit imped-
ance can be denoted as Z
d
R
d
jX
d
and Z
c
R
c
jX
c
.
For a series-resonant circuit, assuming that current I is
owing through the circuit, we obtain
IZ
d
Z
c
0 1
which results in
Z
d
Z
c
0 2
The startup conditions for oscillation for a series resonant
circuit are given as follows:
R
d
R
c
o0 3
X
d
X
c
0 4
This means that the magnitude of the device negative re-
sistance must be larger than the overall circuit losses in
order for oscillation to build up. The startup of oscillation
is initiated either from noise present in the circuit or from
transient introduced during power turnon. As the oscilla-
tion amplitude builds up, the device enters into its non-
linear region, where the device negative resistance drops
and eventually becomes equal to the overall circuit loses.
At this point the oscillation amplitude reaches its steady
state.
For a parallel-resonant circuit, the steady-state oscil-
lation condition is given in terms of active device and cir-
cuit admittances Y
d
and Y
c
, respectively as
Y
d
Y
c
0 5
The startup conditions for a parallel-resonant circuit are
given by
G
d
G
c
o0 6
B
d
B
c
0 7
where Y
d
G
d
jB
d
and Y
c
G
c
jB
c
.
The steady-state oscillation condition can be written
in terms of device and circuit reection coefcients G
d
and
G
c
as
G
d
G
c
1 8
Voltage-controlled devices such as the Gunn diode should
be connected to a parallel-resonant circuit, and the appro-
priate startup conditions for oscillation should be satis-
ed. Current-controlled devices such as IMPATT diode are
placed in a series-resonant circuit for proper operation. In
this case startup conditions for oscillation for a series-res-
onant circuit should be met. The startup oscillation fre-
quency can be different than the steady-state oscillation
frequency. This is because the large-signal device imped-
ance is a function of the voltage (or current) amplitude
across (through) the device. As the oscillation amplitude
builds up, the device impedance varies until it reaches its
nal value. However, for high-Q circuits, the steady-state
oscillation frequencies are very close to the startup fre-
quency of oscillation. It should also be mentioned that the
startup condition for oscillation is a necessary but not
sufcient condition for having an unstable circuit [13].
(Nyquist or root locus analysis can always be used to
determine the circuit instability.)
1.2. Oscillator Design Using Two-Port Devices
The block diagram of a two-port oscillator circuit is shown
in Fig. 2. Most microwave transistors are conditionally
stable within a limited range of frequencies. By plotting
the input and output stability circles on a Smith chart, one
can graphically determine the range of impedances for
unstable operation. When designing microwave oscilla-
tors, the input port of the two-terminal device can be ter-
minated with a purely reactive load that lies in the
unstable region of the Smith chart. This is shown in
Fig. 2 as Z
T
. This will result in the output impedance
(Z
out
) looking into the device to render negative real parts
(|G
out
|41). The load impedance Z
L
must be chosen to
satisfy the startup conditions for oscillation. In practice,
I
Z
d
Z
c

d
Figure 1. Simplied block diagram of a one-port oscillator.
MICROWAVE OSCILLATORS 2819
the load impedance is determined by Eqs. (9) and (10) in
order to achieve maximum power
X
L
o
o
X
out
o
o
9
R
L
o
o

jR
out
A; o
o
j
A0
j
3
10
where A is the amplitude of i(t), o
o
is the oscillation
frequency, R
L
and X
L
are the real and imaginary parts
of the load impedance (Z
L
), and R
out
and X
out
are the
real and imaginary parts of the output impedance
(Z
out
).
When a microwave oscillator is designed according to
this procedure, the measured oscillation frequency will
usually deviate from the design frequency by several per-
centages. This is because the design procedure is based on
the small-signal, linear S parameters that do not consider
the fact that R
out
and X
out
are functions of the oscillation
amplitude.
The steady-state oscillation condition for a two-port os-
cillator in terms of the transistors S-parameter load and
source reection coefcients is given by
1
G
T
S
11

S
12
S
21
G
L
1 S
22
G
L
11
1
G
L
S
22

S
12
S
21
G
T
1 S
11
G
T
12
If the oscillation conditions are satised at one port, they
are satised at all other ports [4,5]. If the oscillator is sta-
ble at the desired frequency of operation, it can be made
unstable by using series (Z
F
in Fig. 2) or shunt feedback.
There are a total of six congurations: three for series
feedback and three for shunt feedback as shown in
Fig. 3 [6].
1.3. Oscillator Design Using Large-Signal Analysis
The large-signal design approach, called device-line mea-
surement, can be applied to one-port oscillator design in
order to achieve better accuracy compared to the oscillator
design using small-signal S parameter [7]. For the large-
signal oscillator design, a nonlinear model for the active
device and a nonlinear simulation technique such as
the harmonic balance are needed to predict the large-
signal S parameters and the output power.
Starting with a two-port network (Fig. 2), one rst ter-
minates one of the ports with a proper impedance (Z
T
)
such that the impedance of the resulting one-port network
(Z
out
) has a negative real part, or equivalently, the mag-
nitude of its reection coefcient (G
out
) is larger than uni-
ty. A test source is then connected to the aforementioned
one-port negative-resistance network, as shown in Fig. 4,
to determine its large-signal reection coefcient. It
should be mentioned that the device-line measurement
system requires the internal impedance (Z
s
) to prevent the
measurement circuit from oscillating. Therefore, R
s
,
which has a larger value than the magnitude of the real
Z
L
Z
T
Z
F

out

in

L
i(t)
Figure 2. Block diagram of a transistor oscillator.
x
1
x
2
x
3
R
L
s
x
2
x
1
x
3
R
L
s
x
1
x
2
x
3
R
L
s
b
3
b
2
b
1
G
L
s
b
1
b
3
b
2
G
L
s
b
2
b
1
b
3
G
L
s
(a)
(b)
Figure 3. Circuit congurations for (a) series
feedback oscillators and (b) shunt feedback os-
cillators.
2820 MICROWAVE OSCILLATORS
part of the output impedance (|R
out
|), should be inserted
next to the test source. By varying the amplitude of the RF
signal at the design frequency, the large-signal depen-
dence of G
out
can be measured. Since |G
out
| is larger than
unity, due to the negative resistance R
out
(A, o
o
), the add-
ed power (P
add
), as given in the following equation, is
therefore positive
P
add
P
av
jG
out
j
2
1 13
where P
av
is the available power of the source.
When the load impedance (Z
L
) is selected to satisfy the
steady-state oscillation condition, specifically, G
out
G
L
1,
the designed oscillator can generate the same power as the
added power (P
add
) that the one-port negative resistance
device generated as determined by Eq. (13). On the basis
of the large-signal output reection coefcient, the load
impedance Z
L
is determined as follows:
X
L
o
o
X
out
A
o
; o
o
14
R
L
o
o
R
out
A
o
; o
o
15
The device-line measurement technique can be used to
maximize the output power of an oscillator. By varying RF
source power at the design frequency, the point where the
added power is maximized can be found. At this power
level, one can measure the large-signal output reection
coefcient and design the load impedance on the basis of
Eqs. (14) and (15). Furthermore, a more accurate estimate
of oscillator frequency is obtained using this technique.
1.4. Injection Locking of Microwave Oscillators
Injection locking of a microwave oscillator is accomplished
by applying a small signal to the free-running microwave
oscillator, provided the frequency of the small signal is
close enough to the free-running frequency of the oscilla-
tor. Through injection-locking a free-running microwave
oscillator to a low-noise low-power source, the free-run-
ning frequency is locked to the injection signal frequency,
thereby improving its frequency stability and phase
noise. It has been shown that the phase noise of the
injection-locked oscillator is reduced to that of the exter-
nal locking source near the carrier frequency; however, it
approaches that of the free-running oscillator far from the
carrier frequency [12].
The effect of the injection signal can be studied by add-
ing a small current source in parallel with the equivalent
circuit for a free-running microwave oscillator as shown in
Fig. 5. Suppose that the injected and the free-running os-
cillator signals are represented by i
i
(t) I
i
(t) exp j[o
i
t
f
i
(t)] and v(t) V(t) exp j[o
o
t f(t)], respectively.
The circuit equation for this injection-locked oscillator
is given by
C
dvt
dt
GG
d
vt
1
L
Z
vt dt i
i
t 16
where G
d
is the negative conductance of a nonlinear
active device. Using the assumption that the amplitudes
of v(t) and i
i
(t) are very slowly time-varying signals, and
dividing the real and imaginary parts of Eq. (16), the am-
plitude and phase equations for the injection-locked oscil-
lator can be obtained by
dVt
dt

GG
d
2C
Vt
1
2C
I
i
t cosft f
i
t 17
dft
dt
o
i
o
o

1
2C
I
i
t
Vt
sinft f
i
t 18
When this system reaches the steady state, the rst term
in Eq. (17) with the time derivative goes to zero and can be
further simplied:
GG
d

I
i
t
Vt
cosft f
i
t 19
Assuming that the device negative conductance is not a
function of frequency, the left side of Eq. (19) goes to zero
because GG
d
E0. Also, the right side of Eq. (19) is close
to zero because the amplitude of the injected signal is
negligibly small compared to the amplitude of the free-
running oscillator (|I
i
(t)|5|V(t)|). Ultimately, the left
and right sides of the amplitude equation [Eq. (17)] be-
come balanced in the steady state. On the other hand,
the phase equation [Eq. (18)] has a more significant mean-
ing in the injection-locking process. In the steady-state
G
Free running oscillator equivalent circuit
G
d C
L i
i
(t)
Injected
signal
Figure 5. Equivalent circuit of an injection-locked oscillator.
R
S
Measurement
system
One-port oscillator
equivalent circuit
X
out
(A,o
o
)
R
out
(A,o
o
)
Z
L
Z
out
(P
in
, f
o
)
Figure 4. Device-line measurement setup.
MICROWAVE OSCILLATORS 2821
condition, Eq. (18) can be simplied to
o
i
o
o

o
o
2Q
I
i
t
It
sinft f
i
t o
m
sinf 20
where Q is the external quality factor (Qo
o
C/G).
The following condition must be satised for the phase
difference f in Eq. (20) to be real:
o
m
o
i
o
o
o
m
21
This condition indicates that when the injection signal
frequency remains within the locking range, the steady-
state solution for the phase equation can be obtained and
the injection-locking behavior is realized.
The locking bandwidth can be expressed as
Do
lock
2o
m

o
o
2Q

P
i
P
o
s
22
where P
i
and P
o
are the injection signal power and the
free-running oscillator output power, respectively. Equa-
tion (22) shows that one can decrease the oscillator circuit
Q and increase the injection signal power in order to in-
crease the locking bandwidth.
Through injection locking, the oscillators free-running
frequency is locked to the injection frequency and the
phase difference between the free-running signal and the
injection signal is determined by Eq. (20).
2. NOISE IN OSCILLATORS
The spectral purity of an oscillator is degraded by the
random uctuations of its amplitude, frequency, and
phase. Noise generated in the active device and passive
components modulates the signal produced by the oscilla-
tor. The sources of random sideband noise in an oscillator
include thermal noise, shot noise, and icker (1/f) noise.
These noise sources result in amplitude and phase noise in
oscillators. In general, the output of a noisy oscillator can
be represented by
vt Aat coso
o
t ft 23
where A and o
o
are deterministic amplitude and frequen-
cy and a(t) and f(t) are random amplitude and phase
noise, respectively. In practice, amplitude noise can be
dramatically attenuated by an amplitude-limiting mech-
anism of oscillators. In case an oscillator has large ampli-
tude noise, amplitude noise can be easily eliminated by
placing a limiter at the output of the oscillator. Therefore
phase noise plays a dominant role in the spectral purity
performance of oscillators. As shown in Fig. 6b, the exis-
tence of phase noise broadens the frequency spectrum of a
noisy oscillator output around the carrier frequency,
whereas the frequency spectrum of an ideal oscillator out-
put is represented by a delta function (Fig. 6a).
Phase noise of the oscillators in radio receivers and
transmitters limits the performance of the communi-
cation system. Phase noise of the local oscillator in a ra-
dio receiver downconverts the adjacent channels into
intermediate frequency (IF), thereby limiting the receiv-
ers immunity to nearby interference. In a radio transmit-
ter phase noise can overwhelm the nearby weak
channels.
Normally, phase noise is measured as the ratio of noise
power (P
n
) in one sideband contained in a specied band-
width (B) at an offset frequency (Do) compared to the car-
rier output power, referred to here as P
s
. Such a power
ratio is usually expressed as decibels below carrier per
hertz (dBc/Hz).
To explain the behavior of phase noise, several models
for phase noise can be considered. The most general model
for phase noise in oscillators was described by Leeson [13].
This model is constructed using an oscillator consisting of
a feedback amplier and a resonator, under the assump-
tion that the oscillator is a linear and time-invariant (LTI)
system. When we consider only white noises such as ther-
mal and shot noise, the noises decrease with frequency
offset as a result of the ltering effect of the resonator in
the oscillator. Therefore, phase noise is given by
LDo 10log
2kT
P
s
o
o
2Q
L
Do

2
" #
24
where k is Boltzmanns constant, T is the absolute tem-
perature, and Q
L
is the loaded quality factor of the reso-
nator. However, since icker (1/f) noise in active devices is
upconverted to the oscillation frequency by means of
Frequency
+o
B
Power
P
s
P
n
(a) (b)
o
o o
o
o
o
Figure 6. Output spectrum in (a) noiseless and
(b) noisy oscillators.
2822 MICROWAVE OSCILLATORS
nonlinear modulation mechanism in practical oscillators,
Eq. (24) is modied into the nal form of Leesons
formula as
LDo 10 log
2kTF
P
s
1
o
o
2Q
L
Do

2
( )
1
Do
1=f
3
jDoj

" #
25
where F is the excess noise measure (empirical constant)
and Do
1=f
3 is the corner frequency between 1/f
2
and 1/f
3
regions. Leesons formula [Eq. (25)] can be graphically de-
picted as seen in Fig. 7. The effect of icker noise is dom-
inant near the carrier frequency, leading to the decrease of
phase noise with offset frequency at 9dB/octave up to the
1/f
3
corner frequency. From the 1/f
3
corner frequency to o
o
/
2Q
L
, phase noise shows a decreasing slope of 6dB/octave.
Also, Leesons model suggests the following methods to
reduce phase noise of an oscillator:
1. Use a high-Q resonator because phase noise is in-
versely proportional to Q
2
.
2. Increase the signal output power.
3. Select an active device with low icker noise and low
noise gure.
Even though Leesons LTI model provides valuable in-
sights into microwave oscillator designs in an engineering
perspective, it cannot explain several phase noise phe-
nomena such as icker noise upconversion. To overcome
this limitation, many approaches based on linear time-
varying (LTV) models have been investigated. More
recently a LTV model in conjunction with the concept of
impulse-sensitive function based on the fact that phase
uctuation varies according to where the impulse noise is
injected into a free-running oscillator has been introduced
[14]. For microwave oscillators with distributed compo-
nents, a perturbation theory for phase noise behavior was
proposed, based on the solution of the Langevin equations
describing the stochastic behavior [15]. A unifying theory
for phase noise derived an exact nonlinear equation for
oscillators in the presence of deterministic and random
perturbations, which led to a general mathematical solu-
tion for phase noise representation [16].
3. OSCILLATOR TYPES
3.1. Gunn Oscillators
The Gunn diode [17] is a negative-resistance device that
operates on the basis of transfer of electrons between two
valleys in the conduction band of a semiconductor mate-
rial. For this reason it is also referred to as a transferred-
electron device (TED). A Gunn device may be described as
a bulk of GaAs that exhibits negative resistance when DC
bias is applied. In this respect the Gunn diode is not pre-
cisely a diode since it does contain any junctions. Gunn
diodes are capable of producing from a few milliwatts to a
few watts with efciencies up to 1520%. The operation
frequency of Gunn diodes reaches 100GHz. Gunn diodes
have a good phase noise performance. The equivalent of
the Gunn diode chip can be described by a parallel RC
circuit where the value of R is negative. Gunn diodes ex-
hibit a negative resistance at low frequencies as well as
RF frequencies. Precautions must be taken in order to
avoid parasitic oscillations due to resonances introduced
by the bias circuit. This can be achieved by rendering the
bias circuit lossy at low frequencies in order to stabilize it.
The Gunn diode oscillator can be made in waveguide,
coaxial, and microstrip lines. The most widely used Gunn
oscillator circuit is an iris-coupled waveguide cavity oscil-
lator. This circuit has high stability and low phase noise
because of its high Q. Furthermore one can mechanically
or electronically tune the oscillation frequency. Figure 8
shows the iris-coupled waveguide cavity Gunn oscillator.
The resonance frequency of the waveguide cavity is deter-
mined by the distance between the iris to the effective
backwall and should be a half-wavelength. The effective
backwall is in between the diodes post and the physical
backwall. If the diodes post is near the cavitys sidewall,
then the effective backwall is very close to the physical
backwall. By adjusting the diameter and the position of
the diodes post with respect to sidewalls, optimum im-
pedance match to the cavity resonator can be achieved. In
order to mechanically tune the frequency of the cavity, a
dielectric rod can be inserted into the cavity resonator.
The electronic tuning of oscillation frequency can be ac-
complished by mounting a varactor diode inside the cavity
using another post. By adjusting the varactor bias, the
Diode post
Gunn diode
z/2
Effective back wall
Waveguide cavity
Iris
Figure 8. An iris-coupled waveguide cavity Gunn diode
oscillator.
Offset frequency (o)
o
o
/2Q
Power
1/f
3
1/f
2
P
s
2FkT
10log
1/f
3
o
Figure 7. Phase noise spectrum of Leesons model.
MICROWAVE OSCILLATORS 2823
oscillation frequency is tuned. This is a simple method for
frequency-modulating such an oscillator.
3.2. IMPATT Oscillators
The term IMPATT is an acronym for impact avalanche
transit time. IMPATT diode is used in the design of solid-
state microwave and millimeter-wave oscillators with an
operating range of frequencies from several gigahertz to
4200GHz. The IMPATT diode exhibits negative resis-
tance due to 1801 phase delay of the current with respect
to the voltage. This phase delay is due to (1) the nite de-
lay between the applied RF voltage and the current due to
avalanche breakdown and (2) the subsequent transit of
carriers through a drift region. One version of IMPATT
operation was rst proposed by Read in 1958 [18]. The
rst demonstration of an IMPATT oscillator was reported
by DeLoach [19]. The two most commonly used versions of
IMPATT diodes are the single-drift region and double-drift
region. IMPATT diodes produce from 0.54 W with ef-
ciencies usually greater than 10% (r30 GHz) and 4%
(r100GHz). IMPATT diodes differ from Gunn diodes in
several respects: (1) the output powers of IMPATTs are up
to 10 times higher than Gunn diodes, (2) IMPATT diodes
are more efcient than Gunn diodes, (3) the operating
voltage for the IMPATT is higher than that of the Gunn
diode (20100V for an X-band IMPATT as compared to
r10 V for an X-band Gunn), (4) IMPATT diodes are nois-
ier than Gunn diodes, and (5) the circuit design using
IMPATT diodes is more difcult because their negative
resistance is about one order of magnitude lower than that
of Gunn diodes.
The IMPATT diode is a current-controlled device and
therefore requires a current source for bias. The IMPATT
diode equivalent circuit consists of a series RC circuit,
where R is a negative number. Since IMPATT diode is a
current-controlled device, it is usually placed in a series-
resonant circuit for stable oscillations. Usually a radial
line (Fig. 9a) or a coaxial line (Fig. 9b) quarter-wave trans-
former is used to match the low negative resistance of the
IMPATT to a waveguide cavity.
3.3. Dielectric Resonator Oscillators
Microstrip resonators have a limited Qon the order of 100.
In order to reduce the phase noise of oscillators, it is nec-
essary to increase the resonant circuits Q. A low-cost
technique to increase the resonators Q is to couple a
dielectric resonator to a microstrip circuit. Typical dielec-
tric resonators have a cylindrical geometry with a dielec-
tric constant between 10 and 100. The TE
01d
of the
cylindrical dielectric resonator can be coupled to a micro-
strip line by placing it on top of the substrate close to the
microstrip line. The distance between the resonator and
the microstrip line determines the coupling factor. The Q
of dielectric resonators is on the order of several thousand.
The resonance frequency of a dielectric resonator can be
adjusted by placing a movable metal plate on top of the
dielectric resonator. Figure 10 shows different congura-
tions of dielectric resonator oscillators.
3.4. Electronically Tunable Oscillators
In order to construct electronically tunable oscillators, ei-
ther a varactor diode is connected to the resonator, or an
yttrium iron garnet (YIG) sphere is used to construct the
Bias filter
Radial line
transformer
z/4
Reduced height
waveguide
Bias filter
Radial line
transformer
z/4
Reduced height
waveguide
Quarter wave
transformer
Gunn diode
(b)
(a)
Figure 9. IMPATT diode oscillators in waveguide using (a) a
radial-line transformer and (b) a coaxial-line transformer.
Z
F
Z
o
Matching circuit
Feedback
Z
o
Dielectric resonator
(a)
(b)
Z
o
Z
o
Dielectric resonator
Figure 10. Dielectric resonator oscillator circuits using (a) series
feedback and (b) shunt feedback.
2824 MICROWAVE OSCILLATORS
resonator. In varactor-tuned oscillators, by changing the
reverse bias voltage across the varactor, the varactor ca-
pacitance and thereby the resonance frequency can be
tuned. The tuning range is dependent on the varactors
capacitance ratio C
max
/C
min
. This ratio is largest for
hyperabrupt varactor diodes. The frequency tuning limit
of a varactor tuned oscillator is determined by [4].
o
max
o
min

C
max
C
min
s
26
A typical varactor-tuned oscillator circuit is shown in
Fig. 11.
Another technique for construction of an electronically
tunable oscillator is to make the resonator by using a
magnetic material. Single-crystal yttrium iron garnet is a
magnetic material that resonates at microwave frequen-
cies when subjected to a DC magnetic eld. The resonance
frequency is directly proportional to the applied magnetic
eld; hence a linear frequency tuning over wideband can
be achieved by adjusting its biasing DC magnetic eld.
The YIG resonator consists of a RF coupling loop, an elec-
tromagnet, and a YIG sphere. YIG is a low-loss material,
and YIG resonators can provide Q factors as high as sev-
eral thousand. YIG tuned oscillators are commonly used
in sweep generators. The frequency tuning range of the
YIG oscillators is determined by the bandwidth over
which the active device provides negative resistance. Fig-
ure 12 shows a diagram of a YIG tuned oscillator. YIG
tuned oscillators with oscillation frequencies as high as
60 GHz have been reported [20].
3.5. MMIC Oscillators
Because of their low cost, improved reproducibility, and
small size, many oscillators have been implemented using
monolithic microwave integrated circuit (MMIC) tech-
nique. Table 1 compares the different types of RF/micro-
wave oscillators reported in more recently published
scientific papers in terms of active device, resonator
type, operation frequency, output power, and phase noise.
In the design of MMIC oscillators, important gures of
merit are phase noise, DC-to-RF efciency, and the tuning
range for phase-locked loop applications. To satisfy
these goals, the choice of active devices is crucial. Each
device has different characteristics such as the 1/f corner
frequency, forward gain, and cutoff frequency (f
T
), which
can produce oscillators with varying degrees of perfor-
mance.
In most cases, practical MMIC oscillators have been
implemented with GaAs MESFET or silicon bipolar tran-
sistor. The GaAs MESFETcan operate at high frequencies
(up to millimeter-wave frequency) because of its high cut-
off frequency. In addition, it has higher gain and higher
output power than does the silicon bipolar transistor.
However, the high 1/f noise corner frequency of the
GaAs MESFET (around 10100 MHz) degrades the noise
performance of oscillators significantly. Therefore, many
designs for low-noise oscillators are based on the silicon
bipolar transistor because of its low 1/f noise corner.
Since the high-electron-mobility transistor (HEMT)
has the advantages of high electron mobility and high f
T
,
it presents high power capability at high frequencies and
low-noise characteristics. Therefore, HEMT MMIC oscil-
lators show good performance at extremely high frequen-
cies of 70200 GHz.
The heterojunction bipolar transistor (HBT) is a bipolar
transistor with a structure similar to that of a silicon bi-
polar transistor; however, HBT consists of different semi-
conductor materials, while the BJT contains only silicon.
The heterojunction structure enables the HBT to operate
at higher frequencies than the silicon BJT. In addition, the
Table 1. Recently Reported Integrated Microwave Oscillators
Year [Ref.] Active Device Resonator Type Frequency (GHz) Output Power (dBm) Phase Noise (dBc/Hz)
2000 32 SiGe HBT LC 4.634.90 13 100 at 100kHz
2000 32 Si BJT LC 5.595.94 9 90 at 100kHz
2000 33 GaAs MESFET LC 11.2511.8 11.5 91 at 100kHz
2000 34 CMOS LC 1.11.45 4 119 at 600kHz
2002 35 SiGe HBT Sapphire resonator 4.85 133 at 1KHz
2002 36 SiGe BJT LC 43.647.3 5.6 108 at 1MHz
2003 37 CMOS LC 5.866.02 4 110 at 1 MHz
2004 38 InGaP-GaAs HBT LC 12.7113.54 0 113.8 at 1 MHz
YIG
Z
F
R
L
Matching circuit
Feedback
Figure 12. Diagram of a YIG-tuned oscillator.
Z
F
Z
o
Matching circuit
Feedback
Figure 11. A varactor-tuned oscillator.
MICROWAVE OSCILLATORS 2825
HBT presents intrinsic characteristics of low 1/f noise cor-
ner, which makes it a good candidate for low-noise oscil-
lators at microwave frequencies. Additionally, the
capability of generating high negative resistances over a
wide frequency range makes this device suitable for the
VCO design. Currently, SiGe HBTs have been actively re-
searched because of reduced fabrication costs, higher in-
tegration levels, and the compatibility with lower
frequency circuits on the same chip.
The advances in CMOS processing technology begin to
challenge GaAs devices and bipolar transistors in com-
mercial circuits with the benet of cost-effective mass pro-
duction. Reducing the minimum channel length leads to
the increase in f
T
, which makes CMOS an attractive can-
didate for microwave oscillators. However, the major chal-
lenges in the oscillator design using silicon technology
including silicon BJT, SiGe HBTs, and CMOS, lie in the
difculty in implementing good passive elements, such as
on-chip inductors and high-Q resonators. This is due to
the fact that silicon substrates exhibit frequency-depen-
dent losses. Therefore, the development of high-Q on-chip
inductors is one of the active research topics in the oscil-
lator design. Many topologies such as the optimized one-
layer inductor, the seriesparallel stacked inductor, the
differential inductor, and the 3D micromachined solenoid
inductor have been proposed in order to increase circuit Q.
BIBLIOGRAPHY
1. N. M. Nguyen and R. G. Meyer, Start-up and frequency sta-
bility in high-frequency oscillators, IEEE J. Solid-State Circ.
27(5):8108120 (May 1992).
2. R. W. Jackson, Criteria for onset of oscillations in microwave
circuits, IEEE Trans. Microwave Theory Tech. 40(3):566569
(March 1992).
3. R. D. Martinez and R. C. Compton, A general approach for the
S-parameter design of oscillators with 1 and 2-port devices,
IEEE Trans. Microwave Theory Tech. 40(3): 569574 (March
1992).
4. I. Bahl and P. Bhartia, Microwave Solid State Circuit Design,
Wiley-Interscience, New York, 1988, Chap. 9.
5. G. D. Vandelin, A. M. Pavio, and U. L. Rohde, Microwave Cir-
cuit Design Using Linear and Nonlinear Techniques, Wiley-
Interscience, New York, 1990, Chap. 6.
6. K. L. Kotzebue and W. J. Parrish, The use of large signal S-
parameters in microwave oscillator design, Proc. Int. IEEE
Microwave Symp., 1975.
7. W. Wagner, Oscillator design by device line measurement,
Microwave J. (Feb. 1979).
8. S. A. Maas, Nonlinear Microwave and RF Circuits, 2nd ed.,
Artech House, Norwood, MA, 2003.
9. G. Gonzalez, Microwave Transistor Ampliers: Analysis and
Design, Englewood Cliffs, NJ, Prentice-Hall, 1997.
10. A. E. Siegman, Lasers, Univ. Scientific Books, Mill Valey, CA,
1986.
11. R. A. York and T. Itoh, Injection- and phase-locking tech-
niques for beam control, IEEE Trans. Microwave Theory Tech.
46:19201929 (Nov. 1998).
12. H. Chang, X. Cao, M. J. Vaughan, U. K. Mishra, and R. A.
York, Phase noise in externally injection-locked oscillator
arrays, IEEE Trans. Microwave Theory Tech. 45:20352042
(Nov. 1997).
13. D. B. Leeson, A simple model of feedback oscillator noise
spectrum, Proc. IEEE 54(2):329 (Feb. 1966).
14. A. Hajimiri and T. H. Lee, A general theory of phase noise in
electrical oscillators, IEEE J. Solid-State Circ. 33:179194
(Feb. 1998).
15. F. X. Kaertner, Determination of the correlation spectrum of
oscillators with low noise, IEEE Trans. Microwave Theory
Tech. 37:90101 (Jan. 1989).
16. A. Demir, A. Mehrota, and J. Roychowdhury, Phase noise in
oscillators: A unifying theory and numerical methods for
characterization, IEEE Trans. Circ. Syst. 47:655674 (May
2000).
17. J. B. Gunn, Microwave oscillation in IIIV semiconductors,
Solid State Commun. 1:88 (1963).
18. W. T. Read, A proposed high efciency negative resistance di-
ode, Bell Syst. Tech. J. 37:401 (1958).
19. B. C. DeLoach, Jr., The IMPATT story, IEEE Trans. Electron.
Devices ED-23:657 (1976).
20. D. Zenous et al., GaAs FET YIG oscillator tunes from 26 to
40GHz, Microwaves RF 22:120139 (Oct. 1983).
21. S. Yngvesson, Microwave Semiconductor Devices, Kluwer Ac-
ademic Publishers, 1991, Chap. 6.
22. K. K. Kurokawa, Some basic characteristics of broad band
negative resistance oscillator circuit, Bell Syst. Tech. J.
48(6):19371955 (July 1969).
23. M. E. Hines, J. R. Collinet, and J. G. Ondria, FM noise re-
duction of an phase locked oscillator, IEEE Trans. Microwave
Theory Tech. 16(9):738742 (Sept. 1968).
24. K. Kurokawa, An Introduction to the Theory of Microwave
Circuits, Academic Press, New York, 1969, Chap. 9.
25. F. Schwierz and J. J. Liou, Modern Microwave Transistors
Theory, Design, and, Performance, Wiley, Hoboken, NJ,
2003.
26. I. D. Robertson and S. Lucyszyn, RFIC and MMICDesign and
Technology, IEE, London, 2001.
27. K. Chang, Microwave Solid-State Circuits and Applications,
Wiley, New York, 1994.
28. E. C. Niehenke, R. A. Pucel, and I. J. Bahl, Microwave and
millimeter-wave integrated circuits, IEEE Trans. Microwave
Theory Tech. 50(3):846857 (2002).
29. L. M. Burns, Application for GaAs and silicon integrated cir-
cuits in next generation wireless communication systems,
IEEE Trans. Microwave Theory Tech. 30(10):10881095
(1995).
30. R. Gotzfried, F. Beisswanger, and S. Gerlach, Design of RF
integrated circuits using SiGe bipolar technology, IEEE J.
Solid-State Circ. 33(9):14171422 (1998).
31. V. Radisic et al., 80GHz MMIC HEMT VCO, IEEE Microwave
Wireless Compon. Lett. 11(8):325327 (2001).
32. H. Jacobsson et al., Low-phase-noise low-power IC VCOs for
5-8-GHz wireless applications, IEEE Trans. Microwave The-
ory Tech. 48(12):25332539 (2000).
33. C. H. Lee, S. Han, B. Matinpour, and J. Laskar, A low phase
noise X-band MMIC GaAs MESFET VCO, IEEE Microwave
Guided Wave Lett. 10(8):325327 (2000).
34. F. Svelto, S. Deantoni, and R. Castello, A 1.3GHz low-phase
noise fully tunable CMOS LC VCO, IEEE J. Solid-State Circ.
35(3):356361 (2000).
35. O. Llopis, G. Cibiel, Y. Kersale, M. Regis, M. Chaubet, and
V. Giordano, Ultra low phase noise sapphire-SiGe HBT
2826 MICROWAVE OSCILLATORS
oscillator, IEEE Microwave Wireless Compon. Lett. 12(5):157
159 (2002).
36. H. Li, H.-M. Rein, R. Kreienkamp, and W. Klein, 47GHz VCO
with low phase noise fabricated in a SiGe bipolar production
technology, IEEE Microwave Wireless Compon. Lett. 12(3):79
81 (2002).
37. Y. Chu and H. Chuang, A fully integrated 5.8 GHz U-NII
Band 0.18-mm CMOS VCO, IEEE Microwave Wireless
Compon. Lett. 13(7):287289 (2003).
38. D. Baek, S. Ko, J. Kim, D. Kim, and S. Hong, Ku-Band InGap-
GaAs HBT MMIC VCOs with balanced and differential top-
ologies, IEEE Trans. Microwave Theory Tech. 52(4):1353
1359 (2004).
MICROWAVE PARAMETRIC AMPLIFIERS
MAREK T. FABER
Warsaw University of
Technology
Warsaw, Poland
1. INTRODUCTION
It has been known since nineteenth century that a me-
chanical system or an electric circuit in which there is a
parameter that varies periodically with time may oscillate
under certain conditions. The rst use of this periodic
variation to amplication, frequency changing, or har-
monic generation was investigated in the 1940s, but real
progress in the parametric amplication dated from the
development in the mid-1950s of semiconductor diodes in
which the barrier capacitance could be modulated at mi-
crowave frequencies. Since then, there has been a contin-
ual and tremendous increase in the amount of theoretical
and experimental work devoted to parametric ampliers
[111]. This is because of their ability to maintain a low
level of noise that is inevitably introduced at each stage of
analog signal processing. A typical microwave receiver of
the early 1960s consisted of a silicon point-contact diode
mixer followed by a vacuum-tube amplier. The addition
of a parametric amplier (or paramp, as it was popularly
called) ahead of the mixer gave an order of magnitude im-
provement in sensitivity of the receiver.
The question arose at that time as to why the para-
metric amplication allows noise levels lower than with
the usual type of amplier. To answer this fundamental
question, it should be remembered that an amplier con-
sists of both passive elements and active energy sources
and that the whole is terminated by two ports, one receiv-
ing the signal to be amplied and the other delivering the
amplied signal. In the ordinary type of amplier, the en-
ergy sources consist of DC sources that are incorporated
into a network of passive elements that may be linear or
nonlinear. The essential condition for such a system to act
efciently as an amplier is the inclusion of a nonlinear
and/or electrically controlled resistive element to provide
an energy transfer from the source to the signal [11].
Unfortunately, its operation is accompanied by a back-
ground noise that is an inevitable result of its dissipative
character. In order to obtain an amplier with a low noise
level, the energy must no longer be supplied to the net-
work directly from DC sources, and so other methods of
energy transfer must be envisioned. One of these methods
is to employ AC sources. It has been shown [1,4,6,8] that if
a nonlinear reactance is incorporated in the network, a
system of this kind may be used efciently as an amplier.
In such a parametric amplier a nonlinear reactance does
not generate any noise, provided it is free from loss.
The rst practical realization of the principle of trans-
ferring energy from a pump source at a high frequency to
a signal at a lower frequency was based on a nonlinear
inductance [2,3,7], with the nonlinearity depending on the
properties of ferrite materials. Unfortunately, large levels
of pump power were required to provide the necessary
nonlinearity. With pump levels in the kilowatt region and
relatively high loss in the reactance, this type of ampliers
had only very limited applications.
Capacitive parametric ampliers depend on the non-
linear capacitancevoltage characteristic of semiconductor
diodes. The diodes, which are specially made for the pur-
pose, are called varactor diodes or more simply varactors.
When used in a parametric amplier, the varactor junc-
tion is biased in the reverse direction to prevent current
ow across the junction and thus suppress shot noise
associated with the current, leaving only thermal noise
present in series resistance of a real diode. The effective
noise level of the amplier depends on both the signal and
pump frequencies as well as on the properties of the diode,
and, as it is thermal in origin, the noise level also depends
on the temperature of the diode. Thus, it can be lowered by
cooling the amplier.
Varactor parametric ampliers could be operated at
convenient temperatures from room temperature, through
liquid nitrogen temperature, down to liquid helium tem-
perature, and offered versatility of system design. These
ampliers gained popularity because the improvement in
sensitivity they offered outweighed the complications they
introduced (the requirement for a low-loss ferrite circula-
tor and a high-power, high-frequency pump oscillator
coupled with a reputation for being touchy to operate) [12].
Among uncooled ampliers, the parametric amplier was
the most sensitive amplier in existence in the 1960s. The
noise level of a cooled paramp was not as low as that of a
maser, but the paramp was much smaller and much less
costly and did not require a liquid helium cryostat to op-
erate. The wideband ultra-low-noise paramps [5,6,9] had
been the key devices that had brought commercial and
military communication satellite systems into existence.
They also dominated in radio astronomy receivers
throughout the world, opening new horizons in explora-
tion of our universe. The growth in technology of the wide-
band paramps in the late 1960s had been exceptionally
rapid. The microwave parametric amplier evolved from a
relatively high noise temperature device (250300 K) with
narrow bandwidth of 2550 MHz to one capable of opera-
tional noise temperature as low as 12K in cryogenically
cooled mode and as low as 50120 K in an uncooled mode
and 1dB bandwidth in excess of 500MHz. In the early
MICROWAVE PARAMETRIC AMPLIFIERS 2827
1970s, Peltier cooled paramps were the conguration of
the day.
In the early 1980s parametric ampliers were chal-
lenged by gallium arsenide eld-effect transistor
(MESFET) ampliers. With their simplicity and cost
advantages, the MESFETs were encroaching on the pre-
viously exclusive property of the paramps, eroding their
monopoly. Rapid improvement in GaAs MESFET technol-
ogy and development of the high-electron-mobility tran-
sistor (HEMT) resulted in lower intrinsic noise, and thus
less noisy ampliers, leaving paramps behind in the race
for the highest sensitivity [13] and nally displacing them
at microwave frequencies. But the elegance and rene-
ment of the parametric amplication are still fascinating,
and the technique might emerge in its classical form at the
terahertz region of the electromagnetic spectrum where
there are no active semiconductor devices but there are
many powerful far-infrared lasers to pump contemporary
submillimeter-wave varactors [14]. More recently optical
control in optoelectronic devices and nonlinear interaction
between optical and microwave signals in semiconductor
devices have gained much interest because of potential
application in signal switching, mixing, and frequency
modulation. Parametric amplication plays an important
role in increasing sensitivity of such devices employing
semiconductor photodetectors [15,16].
2. PRINCIPLE OF OPERATION
In a parametric amplier, the energy at one (signal) fre-
quency is increased by supplying energy at a second
(pump) frequency. The basic idea is illustrated in Fig. 1.
Considering the simple resonant circuit of Fig. 1, it is as-
sumed for the purpose of illustration that the plates of the
capacitor can be separated mechanically [2,17]. Let us as-
sume that prior to a time t 0, the circuit has been in-
duced to oscillate at its resonant frequency. Suppose that
when the charge in the capacitor is reaching its rst max-
imum after t 0, separation of the plates is suddenly in-
creased, thereby decreasing the capacitance. Because of
the electric eld, this separation requires mechanical
work to be supplied. The relation between charge q, ca-
pacitance c, and voltage v on a capacitor is, of course,
q cv. Thus, if the plate separation is increased and the
capacitance thereby decreased, in a time short enough for
q to be considered essentially constant, the voltage and
also the stored energy qv/2 must both be suddenly in-
creased. If, at the next zero of voltage across the plates, the
original separation is suddenly restored, this does not
change the stored energy of the circuit since the eld is
then zero. If the whole sequence is then repeated every
half-cycle, the voltage and energy may also be increased
every half-cycle, giving a buildup of the voltage across the
capacitor and the energy stored in the circuit (thus also
the charge in the capacitor). This buildup will continue
until the energy added by separation of the plates exactly
equals the energy dissipated in the circuit. It should be
noted that the plates are moved (or pumped) at twice the
resonant frequency and that the phase of the pumping is
important. A variable-capacitance amplier operating
exactly on this principle, in which a signal is supplied at
the resonant frequency f
s
with the pump frequency f
p

2f
s
, is called a degenerate amplier.
Nondegenerate parametric ampliers, which do not im-
pose this restriction on pump frequency or phase, can be
constructed by connecting a second tuned circuit across
the variable capacitance. This circuit is known as the
idler, and its frequency f
i
is tuned to f
p
f
s
. Operation of
such an amplier can be explained in much the same way
as above for the single resonant circuit; interested readers
may refer to Ref. 17.
In practical circuits a semiconductor junction is used as
an electronically variable capacitance instead of a me-
chanically controlled parallel-plate capacitor. The wave-
form changing such a capacitance is sinusoidal (rather
than rectangular) and is produced by a special generator
C
a
p
a
c
i
t
a
n
c
e
C
h
a
r
g
e
V
o
l
t
a
g
e
E
n
e
r
g
y
s
u
p
p
l
i
e
d
Time
Figure 1. Illustration of basic parametric amplier principle.
The separation of the plates forming the capacitor is increased
(thus the capacitance is decreased) so fast that the charge q re-
mains constant, giving buildup in voltage v q/c and stored en-
ergy qv/2. Original separation is then restored at the moment
when q 0. The whole sequence is repeated every half-cycle.
2828 MICROWAVE PARAMETRIC AMPLIFIERS
(called a pump) pumping energy to the nonlinear capaci-
tance. The theoretical power ow into and out of an ide-
alized lossless nonlinear reactance is described in terms of
two generalized equations known as the ManleyRowe
relations, which can be written as
X
1
m0
X
1
n1
mP
mn
mf
1
nf
2
0
X
1
n0
X
1
m1
nP
mn
mf
1
nf
2
0
1
where P
mn
represents, algebraically, the power ow into
the nonlinear reactance at the frequencies mf
1
nf
2
.
These equations are a result of only the nonlinear varia-
tion of the reactance and are independent of the shape of
its characteristic and of the driving power levels. The
spectrum of signals at the nonlinear reactance is illustrat-
ed at Fig. 2. If we consider a typical case of three-frequen-
cy ampliers (however, more frequencies may be used in
some specific applications)f
1
, f
2
, f
3
, where f
3
f
1
f
2

then, provided f
1
5f
1
f
2
, the general ManleyRowe equa-
tions can be simplied to
P
1
f
1

P
3
f
3

P
2
f
2

P
3
f
3
0 2
This equation and Fig. 2 can be used to understand the
operation of some of the different types of parametric am-
pliers (nonlinear capacitance is assumed here to be loss-
less).
Let f
1
f
s
be the signal frequency and power be sup-
plied from an external source at frequency f
2
f
p
. Since
pump power P
2
P
p
is supplied to the capacitor (P
p
40),
then P
3
o0 represents power leaving the reactance at the
frequency f
3
f
p
f
s
; hence P
1
P
s
40. It follows that the
device is absolutely stable and has the maximum power
gain equal to f
3
/f
1
1f
p
/f
s
. This type of amplier is called
an upper-sideband (noninverting) upconverter.
Now, let f
1
f
s
be the signal frequency but let power be
supplied from the pump at the frequency f
3
f
p
. Hence, P
3
P
p
40 and both P
1
P
s
o0 and P
2
o0; so the reactance
can deliver energy at frequencies f
1
f
s
and f
2
f
p
f
s
. If
the power is extracted at the frequency f
2
, the device is
called a lower-sideband (inverting) upconverter. The term
inverting is used because the input signal spectrum is in-
verted at the output in this mode of operation.
If, as stated, the nonlinear capacitance is pumped at
the frequency f
3
f
p
, that is, P
3
P
p
40 and P
1
P
s
o0
and P
2
o0, but the output is at the frequency f
1
f
s
, then
the negative sign of P
1
indicates that the capacitor emits
more power than that fed from the generator at f
1
f
s
. It
should be noted that the power supplied from the varactor
to the signal source is independent of that supplied by the
source itself; thus innite gain is possible and the device is
able to oscillate. It indicates that the pumped varactor
presents a negative resistance to the signal source. Hence,
with this frequency arrangement the signal power can be
amplied at the same frequency, in contrast to the previ-
ous cases. The powers P
1
and P
2
are strongly dependent on
the pump power and the external impedances. When the
input and output frequencies are the same at f
1
f
s
, power
at f
2
f
p
f
s
is simply dissipated in the circuit and is un-
used. It justies the name idler used traditionally for this
frequency. The idler signal is an inevitable byproduct of
this type of amplication, and suppressing it would also
suppress the desired amplication of the signal. The sep-
aration of the idler and signal frequency is an important
factor determining design and properties of this single-
port (or reection type) amplierthe closer the idler to
the signal, the more difcult it is to separate them by l-
tering. If the signal and idler frequencies are separated far
enough so that the signal circuit does not pass the idler,
the amplier is called a nondegenerate amplier. In the
opposite case, if the signal circuit passes both the signal
and idler bands and the input termination is common to
both, that is, f
i
f
p
f
s
% f
s
, the amplier is said to be de-
generate.
3. VARACTOR DIODES
The most convenient nonlinear reactance element is a
semiconductor diode specially designed to provide large
variation of diode junction capacitance as a function of the
applied (reverse) pumping voltage. Varactors may be clas-
sied into two broad groups, depending on the method of
fabrication: the junction varactors, widely used at micro-
wave frequencies, and the Schottky barrier varactors, gen-
erally used at millimeter waves and in high-performance
ampliers. Both groups are extensively discussed else-
where in this encyclopedia, so only brief descriptions are
given here and only problems specific for parametric am-
pliers applications are discussed.
A typical junction varactor is made on an n-type silicon
or gallium arsenide wafer, with a highly doped conducting
substrate on which is grown a lower doped epitaxial layer.
A suitable p-type dopant is then diffused into the epilayer
to the obtain p

region and to form the p

-n junction.
Ohmic contacts are made to a small circular area on the
top of the wafer for the anode and to the bottom of the
wafer for the cathode. Most of the epitaxial layer is then
etched away, except the portion that is under the top con-
tact. In this way a mesa of the desired diameter is formed.
Diodes fabricated in this manner are called diffused epit-
axial varactors.
A Schottky barrier varactor diode consists of a circular
metallic contact pad (usually platinum) deposited on a
lightly doped n-type layer epitaxially grown on a heavily
doped GaAs substrate. The epitaxial layer is conductive
A
m
p
l
i
t
d
e
0 f
s
f
p
2f
p
f
p
f
s
f
p
+f
s
2f
p
f
s
2f
p
+f
s
Frequency
Figure 2. Spectrum of signals at a nonlinear reactance. Input
signal spectrum (f
s
and its vicinity) appears at both sides of the
harmonics nf
p
of the pumping signal.
MICROWAVE PARAMETRIC AMPLIFIERS 2829
except in the vicinity of the metalsemiconductor inter-
face, where an insulating depletion zone is formed. Deple-
tion-layer thickness, and thus the capacitance, varies with
the biasing voltage applied to the metalsemiconductor
junction.
Simple, but justied by relatively low operating fre-
quencies, varactor models are used in the analysis and
design of parametric ampliers; the reader is referred to
Ref. 14 for advanced modeling of varactor diodes. A typical
equivalent circuit of a microwave varactor is shown in
Fig. 3. The reverse-bias junction is modeled in this simple
circuit by a voltage dependent capacitor C
j
(v) and series
resistance R
s
. The terms L
p
, C
p
, and C
s
are linear parasitic
inductance, capacitance, and stray capacitance, respec-
tively. The dependence of the junction capacitance on the
applied reverse voltage v is given by
C
j
v C
j0
1
v
f

W
3
where f is the barrier potential and C
j0
is the zero-bias
capacitance. The exponent W depends on the doping prole
of the epitaxial layer of the varactor. For Schottky diode
varactors with uniform epitaxial layer doping, it is close to
1
2
. For p

-n-junction varactors the W factor varies from


1
3
for
a linearly graded diffused junction up to
1
2
for an ideal
abrupt junction. Other doping proles (e.g., hyperbolic)
have also been used in hyperabrupt varactors to obtain
larger capacitance variation and thus higher values of the
W factor (W in excess of 1 have been reported). However, the
performance of the parametric amplier depends not only
on the capacitance change DC
j
but also on the series re-
sistance R
s
of the varactor. Unfortunately, hyperabrupt
varactors have high series resistance (DC
j
/R
s
is lower than
for other varactors), which excludes their use in high-per-
formance ampliers.
In Schottky barrier varactors the series resistance is
dominated by the contribution from the undepleted epit-
axial layer. For parametric ampliers, a breakdown volt-
age of a few volts is sufcient and this allows relatively
high doping and a thin epitaxial layer, thereby reducing
the series resistance. In diffused epitaxial varactors there
are additional sources of parasitic resistance, namely, the
resistance of the diffused p

-region and the resistance of


the ohmic top contact. The p

-region resistance becomes


large at cryogenic temperatures, causing serious degrada-
tion of diode performance. Similarly, silicon diodes cannot
be used in cooled ampliers because of series resistance
increase and carrier freezeout below 40 K [14]. Hence gal-
lium arsenide Schottky varactors have been the best
choice for cooled parametric amplier applications
because of their low, relatively temperature-insensitive
resistance.
3.1. Varactor Figures of Merit
Efcient varactor operations require the reactance of the
junction capacitance to be much larger than the diode se-
ries resistance. It places an upper frequency limit on the
usefulness of a given varactor, and gures of merit quan-
tify this limit.
The static cutoff frequency f
c
gives an indication of loss
and is dened as that frequency at which the capacitive
reactance of the junction at zero bias becomes equal to the
series resistance:
f
c

1
2pR
s
C
j0
4
For varactor applications the diode is characterized by the
dynamic cutoff frequency dened as
f
cd

1
2pR
s
1
C
j;min

1
C
j;max

5
where C
j,min
and C
j,max
are the values of C
j
at the reverse
breakdown voltage and at the zero bias voltage (or at 1 mA
forward current), respectively.
Correspondingly, the quality factors at a specied fre-
quency f
0
are given as Qf
c
/f
0
and for a pumped abrupt
junction as Q
d
0.25 f
cd
/f
0
.
In an amplier, the capacitance is modulated by the
application of microwave power at the pump frequency f
p
and thus varies periodically in time. Hence C
j
(t) can be
expanded into Fourier series
C
j
t C
0
1
X
1
n1
2g
n
cos 2pnf
p
t
!
6
where C
0
and g
n
are the Fourier coefcients. The values of
g
n
, in particular that of g
1
, determine parametric amplier
performance. In most practical cases higher terms may be
ignored and only C
0
(the average value of the pumped
capacitance) and g
1
need to be considered. The diode
capacitance modulation coefcient g
1
is, of course, a func-
tion of the voltage developed across the diode at pump
frequency, but will have a maximum attainable value,
termed nonlinearity factor, given by
g
C
j;max
C
j;min
2C
j;max
C
j;min

7
When g is referred to, it is usually this maximum value
that is intended.
For parametric amplier applications the most useful
quantity to characterize the diode is the pumped gure of
merit, dened as
Mgf
c0
8
C
p
C
s
L
p R
s
C
j
(v)
Figure 3. Equivalent circuit of a varactor. The reversed-bias
junction is represented by a voltage-dependent capacitor C
j
(v) and
a series resistance R
s
. L
p
, C
p
, and C
s
are parasitic inductance,
capacitance, and stray capacitance, respectively.
2830 MICROWAVE PARAMETRIC AMPLIFIERS
where
f
c0

1
2pR
s
C
0
9
is the pumped cutoff frequency. A high pumped gure of
merit indicates a low attainable noise of the amplier.
4. NONDEGENERATE PARAMETRIC AMPLIFIER
As mentioned above, several three-frequency amplier
congurations are feasible. Of these, the negative-resis-
tance nondegenerate parametric amplier with a varactor
diode providing a suitable nonlinear capacitance has
achieved the most success as a practical low-noise pa-
ramp. A lumped-circuit analog of the microwave amplier
is used to obtain some insight into the operation of this
type of amplier. The model, shown in Fig. 4, consists of
three resonant circuits coupled by a varactor that may be
represented by the equivalent circuit of Fig. 3. So that the
properties of the nonlinear element, rather than these of
the resonant circuits, can be emphasized, it is assumed
that the resonant circuits are ideal, and this is taken to
mean that at certain frequencies they are pure resistanc-
es, while at all other frequencies they are very high im-
pedances and, therefore, virtually open circuits.
With these assumptions, the behavior of the amplier
at the frequency f
1
f
s
can be described in terms of a neg-
ative resistance < (voltage developed across < is out of
phase of the current owing through it), which appears in
the equivalent circuit of the amplier at this frequency. Its
value is given by [1,8]
<
g
2
o
s
o
i
R
Ti
C
2
0
10
where R
Ti
R
s
R
2
is the total series resistance at the
idler frequency f
i
f
2
.
The transducer power gain is taken as the ratio of pow-
er dissipated in the load resistance R
L
to the available
power from the signal source (internal resistance R
g
) and
is given by
G
p

4R
g
R
L
R
g
R
L
R
s
<
2
11
clearly, for high gain R
g
R
L
R
s
E|<|.
In common with all other negative-resistance ampli-
ers, the negative-resistance parametric amplier is ex-
tremely sensitive to small changes in the value of < and
changes in source and/or load resistance when it is oper-
ated under high-gain conditions. In practice < may change
as a result of uctuations in the pump power and frequen-
cy, and control circuits have often been employed in con-
junction with klystron pump sources in an effort to
minimize such uctuations. The popular approach was
to control the attenuator in the pump line to maintain a
constant varactor bias. Frequency uctuations were han-
dled in critical applications by locking the pump to a stable
reference. The introduction of Gunn oscillators as pump
sources resulted in a considerable improvement in stabil-
ity, making stabilization schemes unnecessary in most
cases [12].
4.1. Operation with a Circulator
Nonreciprocal devices such as circulators are used to sep-
arate the amplied output from the input and to protect
the amplier against changes in impedance at the input or
at the following stage of the receiver. An ideal circulator is
a circuit element that directs energy from one port to the
next port without loss and prevents transmission in the
opposite direction. This passive component obtains its
nonreciprocity from the presence of a central ferrite struc-
ture placed in a DC magnetic eld.
Early paramps used three-port circulators, but most
recent designs incorporated ve-port circulators to in-
crease immunity from the effect of source impedance vari-
ation and to provide greater isolation between adjacent
stages in multistage ampliers. (For stability, the gain was
usually limited to 2025 dB for a single-stage device. If
higher gain was required, a number of 1015-dB stages
were connected in cascade.)
Knowing that the variation of the varactors junction
capacitance is completely controlled by the pump, it is
then possible to concentrate entirely on that variation,
and omit details of the source that produced it. The circuit
of Fig. 4 can therefore be redrawn as in Fig. 5, leaving out
the pump circuit; the capacitance is then specied as C
j
(t),
emphasizing in this way the time variation. The signal
source and the load are connected to the amplier via a
three-port circulator.
4.1.1. Power Gain. The power gain of the ampliercir-
culator combination can be seen to depend on the ratio of
the powers entering and leaving the amplier. Since
the circulator has a characteristic impedance Z
0
to which
it should be matched, source and load resistances will
have the same value, and the circuit available gain will
equal the transducer gain. The required power gain is
therefore equal to the square of the magnitude of the
voltage reection coefcient G and, since R
g
R
L
Z
0
, is
Tuned to f
i
= f
2
Tuned to f
s
= f
1
Tuned to f
p
= f
3
C
2
C
1
L
2
L
1
R
L
V
L
3
C
3
Input
Idler
circuit
Output
Signal
source
Pump
Figure 4. Lumped-element model of a microwave nondegenerate
parametric amplier. Three ideal resonant circuits tuned to sig-
nal, f
s
f
1
, idler, f
i
f
2
, and pump, f
p
f
3
, frequencies are coupled
by a varactor represented by the equivalent circuit of Fig. 3.
MICROWAVE PARAMETRIC AMPLIFIERS 2831
given by
G
p
GG

R
s
R
L
<
R
s
R
L
<

2
12
With the use of a circulator, the input loop of the amplier
effectively contains only one of two resistors, R
g
or R
L
. For
high gain, that is, R
L
R
s
E|<|, Eq. (12) can be rewritten
as
G
p

4R
2
L
R
L
R
s
<
2
13
Comparing the expressions for power gain with and with-
out the circulator (setting R
g
R
L
and remembering that
R
s
5R
L
), it can be seen that the use of a circulator has in-
creased the power gain 4 times for the same gain stability,
where instability of gain results from changes in <.
4.1.2. Noise Temperature. In low-noise ampliers, it is
more convenient to express noise performance in terms of
effective noise temperature (or, in short, noise tempera-
ture), which is related to the noise gure F by the equation
T
e
T
0
F 1 14
where T
0
290K is a standard temperature.
The effective noise temperature of the parametric am-
plier depends on the thermal noise contributions from all
the resistances in the circuit (shot noise is not present in
the reverse-biased junction). Neglecting contributions oth-
er than that from the varactor itself, the noise tempera-
ture of a negative-resistance reection-type parametric
amplier can be calculated in this limiting case from the
following equation [5,7,8]
T
ea
T 1
1
G
p

f
c0
f
i

2
g
2
1
f
2
c0
f
s
f
i
g
2
1
15
where T is the diode physical temperature and G
p
is the
amplier power gain. It is possible to calculate the pump
frequency f
p,opt
at which the parametric amplier noise
temperature has its minimum value. This is given by
f
p;opt
f
s
1
f
c0
f
s

2
g
2
" #
1=2
% gf
c0
M 16
and the optimum noise temperature for large gain be-
comes
T
ea;opt
T
2f
s
gf
c0

2f
s
T
M
17
Thus, it is apparent that for good amplier performance a
diode should combine high cutoff frequency with a marked
capacitance variation, even at low temperatures, in order
to achieve low noise temperature.
If we now add contributions from losses in the signal
and idler circuits, the circuit behavior of the circulator-
type amplier (where R
L
R
g
), is then described by the
noise temperature
T
ea

T
1
R
1
R
g

T
d
R
s
R
g

f
s
f
i
j<j
R
g
T
d
R
s
T
2
R
2
R
2
R
s
!
18
where < is the negative resistance given by Eq. (10), R
g
is
the generator resistance in the signal circuit and R
1
and R
2
represent losses (other than R
s
) in signal, f
1
f
s
, and idler,
f
i
f
2
, circuits. T
1
, T
2
, and T
d
are the temperatures of R
1
,
R
2
, and R
s
, respectively. In practical ampliers tempera-
tures are usually equal, that is, T
1
T
2
T
d
T and the
amplier is designed to have, as far as possible, R
1
and R
2
negligibly small in comparison with R
s
. Then, at high gain,
that is, j<j % R
g
R
s
, the noise temperature is given by
T
ea
T
R
g
R
s
R
g
f
p
f
i
1

19
For many applications cooling is not necessary. Adequate
low-noise performance can be achieved using simple cir-
cuits with diode of moderate cutoff frequency. If lower noise
temperatures are required, it is necessary to use higher
idler frequencies, and this is possible only if a diode of suf-
ciently high cutoff frequency is used. When cooling is nec-
essary to lower the noise temperature, it must be
remembered that any loss in the path between the signal
source, such as an antenna, and the parametric amplier
may considerably degrade the overall noise temperature.
If the loss of the circulator is A10log(L), then the noise
temperature of the circulatoramplier cascade is calcu-
lated from
T
e;ca
T
e1

T
e2
G
1
L 1T LT
ea
20
If, for example, a circulator with A0.41dB transmission
loss is ahead of a low-noise, T
e
100-K amplier, then at
room temperature, T290K, the receiver has noise
E
g
R
g
R
s
R
i
C
1
C
j
(t )
L
1
R
L
= R
g
= Z
0
L
2
C
2
R
1
R
2
f
s
f
1
= f
s
f
2
= f
i
Z
0
f
s
Input
Output
Circulator
Figure 5. Equivalent circuit of the nondegenerate parametric
amplier employing a circulator to separate the amplied output
from the input. The pumped varactor is represented by a capac-
itance C
j
(t) periodically varying in time and series resistance R
s
.
2832 MICROWAVE PARAMETRIC AMPLIFIERS
temperature T
e,ca
0.1T1.1 T
ea
139K. When the am-
plier is cooled down to temperature 20K, its noise tem-
perature lowers to, say, T
ea
10K. Leaving the circulator
uncooled will increase the overall noise temperature 4
times to T
e,ca
40K. It is thus obvious that both the cir-
culator and the amplier must be cooled down to obtain
low-noise performance of the receiver (T
e,ca
13K in this
illustrative example).
4.1.3. Bandwidth. Much effort has been spent in devis-
ing ways of obtaining the broad bandwidth essential in
some applicationssee Ref. 9 for a thorough review of ad-
vanced techniques used for the purpose. Because the
bandwidth of an amplier depends on its gain (decreases
with increase of gain), the gainbandwidth product is used
to characterize the amplier performance. It can be shown
[8] that for a negative-resistance amplier operated in
conjunction with a circulator, the gainbandwidth product
may be derived as
G
1=2
p
B2
1
B
s

1
B
i

1
21
where B
s
and B
i
are the unpumped signal and idler circuit
bandwidths, respectively. Both signal and idler circuits
should therefore be as broadband as possible to give the
amplier a good gainbandwidth product.
Since the signal circuit is loaded externally by the
source resistance, whereas the idler has no external load-
ing, the latter will tend to be a high-Q circuit limiting the
bandwidth. The resistance present in the idler has already
been xed at the diode resistance R
s
, and any increase in
this will degrade the amplier performance. Therefore, in
order to optimize the bandwidth of the idler circuit, it is
necessary to keep the reactance of the idler circuit as low
as possible. (The bandwidth of a series-tuned circuit is
given by the R/L ratio.) This can be achieved by conning
the idler power to the varactor encapsulation. From Fig. 3
it can be seen that there is the possibility of a series
resonance, associated with L
p
, C
0
(the average value
of the pumped junctions capacitance), C
s
, and R
s
at the
frequency given by
o
2
si
%
1
L
p
C
0
C
s

22
neglecting the effect of R
s
. The resonant frequency f
si
can
be arranged to be the idler frequency of the amplier, but
in order to use the series resonance in this way, a return
path for the current must be provided (e.g., by a lumped
circuit or a length of transmission line). An elegant solu-
tion to this problem uses two antiparallel connected di-
odes. When sufciently excited, the idler current will
circulate around this structure and will not propagate to
any other part of the amplier. A further development of
this idea has seen the production of suitable diodes in one
encapsulation. Another approach is to mount the diodes
back to back across the pump waveguide with signal line
entering through the sidewall and contacting the junction
between them (crossbar conguration).
For parametric ampliers with a high idler frequency it
is more convenient to use the parallel resonance of the
encapsulated diode (which is actually a series resonance
for the idler currents) to form the idler circuit. Such an
amplier uses a single diode and can be further rened by
modifying the series resonance to support the signal
frequency.
Careful attention to the design of the idler circuit
leaves the bandwidth of the signal circuit as the main
limitation on the overall bandwidth of the parametric am-
plier. An estimate of the maximum attainable bandwidth
under these conditions may be made putting B
i
5B
s
in Eq.
(21) to yield
G
1=2
p
B 2B
s
23
If the signal circuit is now designed for the minimum pos-
sible Q-factor associated with the source resistance, then
the added inductance must just resonate the diode capac-
itance at the signal frequency (i.e., single-tuned circuit).
Then under the high-gain condition, it is found that
G
1=2
p
B 2gf
s
gf
c0
f
p
f
s

2g
2
f
c0
f
s
f
i
24
The result suggests that high-quality diodes are impor-
tant in securing large gainbandwidth products, and that
there must be a tradeoff between effective noise tempera-
ture and bandwidth in selecting the pump and, therefore,
the idler frequency.
Considerable improvement in the overall gainband-
width product can be achieved by introducing a lter
structure in place of a simple single resonant circuit. Im-
provements of over 5 times have been reported with little
increase of noise temperature. For a maximally at
design, the limiting bandwidth should be given by
Blog(G
p
) const, but in practice the use of more than
two or three compensating elements leads to practical dif-
culties in tuning the device.
5. OTHER PARAMETRIC AMPLIFIER CONFIGURATIONS
5.1. Degenerate Parametric Ampliers
In the degenerate parametric amplier, the pump fre-
quency is approximately twice the signal frequency. The
signal and idler passbands overlap, and instead of idler
circuit being terminated inside the amplier, it is effec-
tively terminated in the input of the amplier. The ab-
sence of a separate idler circuit makes construction of such
an amplier much simpler than the corresponding non-
degenerate version. The additional advantages are a low
pump frequency and a wider bandwidth. Therefore,
degenerate ampliers were nding applications in some
early broadband radiometers and more recently in milli-
meter-wave paramps.
The phase-coherent degenerate amplier in which the
pump frequency is exactly twice the signal frequency is by
its nature a single-frequency device since no departure
from coherence with the pump is allowed [1,4,8]. To
MICROWAVE PARAMETRIC AMPLIFIERS 2833
achieve the required frequency relationship, in practice it
would be necessary to synchronize the pump frequency to
the second harmonic of the signal frequency with special
phase-locked loop (PLL) circuitry. It can be shown [1] that
the gain of the amplier changes with changes in phase of
the pump signal, rending the use of this amplier imprac-
ticable in the majority of potential applications.
Even if phase relations are loosened and the signal fre-
quency is only approximately equal to the idler frequency,
care must be taken in using the amplier. It should be
noted that the degenerate amplier is not suitable for di-
rect use with frequency-modulated signals, since when f
s
increases in frequency, then f
i
f
p
f
s
, which is present at
the same terminals, decreases in frequency. The signal fed
into a degenerate amplier may be amplitude-modulated,
although if the amplier has the idler frequency only ap-
proximately equal to the signal frequency, beats between
the two waveforms can cause interference. A cascade
of the degenerate paramp followed by a parametric con-
verter pumped synchronously at 0.5f
p
was used to over-
come both difculties (1).
The major disadvantage of degenerate ampliers be-
comes apparent when we consider that a signal entering
receiver appears in both idler and signal responses of the
amplier and the output contains both the signal and its
image (also noise from both responses adds at the ampli-
er output). When coherent communication signals are
involved, this double response is unacceptable. In radiom-
eter applications, where the signal takes the form of
broadband noise, this type of amplier has good sensitiv-
ity, since the signal is received equally in both the signal
and idler bands. In general, the use of degenerate ampli-
er must be judged carefully considering the nature of the
signal, single- or double-sideband operation at the input
and the output of the amplier, and the nature of the de-
tector employed in the receiver. A detailed discussion of
this subject can be found in Refs. 4, 10, and 11.
5.2. Multiple-Idler Parametric Ampliers
To circumvent some of the disadvantages of the three-fre-
quency paramps, many other frequency combinations had
been proposed for parametric ampliers in the hope that
an improved performance would outweigh the disadvan-
tages of (usually) increased complexity. If more than one
idler frequency is used, it is possible to use a pump fre-
quency lower than the signal frequency. In the 2-idler
case (the so called four-frequency paramp), the usual
restriction
f
i
f
p
f
s
> 0 25
is replaced by
f
i1
f
s
f
p
> 0 26
for the rst idler, and
f
i2
f
p
f
i1
2f
p
f
s
> 0 27
for the second idler. Therefore
1
2
f
s
of
p
of
s
is required
for efcient operation (negative resistance at signal
frequencies f
p
of
s
o2f
p
). Note that if f
p
is chosen as
2
3
f
s
,
then the two idler frequencies will both be equal to
1
3
f
s
.
The lack of adequate pump generators was the main rea-
son to use four-frequency paramps at millimeter waves.
However, more complex microwave construction usually
resulted in higher losses and hence poor noise perfor-
mance and, therefore, multiple-idler paramps found only
limited applications [9].
5.3. Traveling-Wave Parametric Amplier
Up to now parametric devices utilizing essentially reso-
nant structures have been considered. Such circuits suffer
from previously discussed drawbacks; some of them can be
minimized by resorting to nonresonant propagating cir-
cuits. Avariety of congurations are possible, each with its
own characteristics. Perhaps the simplest is the case
where all the three traveling wavessignal, idler, and
pumphave positive phase and group velocities. This was
exploited in ampliers taking the form of a transmission
line periodically loaded with varactor diodes. It was dif-
cult in practice to satisfy phase requirements for all the
signals, and some experimental ampliers provided sepa-
rate pump feeds for each diode [10]. For frequencies below
1.5 GHz, experimental ampliers had usually the form of
stripline with diodes between the central and outer con-
ductors [1,7]. For higher frequencies, the diodes were
mounted across rectangular waveguide or a series of cou-
pled cavities [10].
In practice, because of the difculties of providing a
considerable number of identical diodes and the experi-
mental difculties of providing the correct phase charac-
teristics at the three frequencies, the performance of
traveling-wave ampliers was worse than that achieved
with much more simple and requiring much less pump
power single-diode devices. Therefore, traveling-wave
parametric ampliers have never been developed beyond
the experimental stage and have never left the laboratory.
A distributed varactor was needed for success in this eld.
However, it appeared at the end of the 1970s in a form of
nonlinear transmission line employing distributed
Schottky barrier varactors [14]. It was too lateparamet-
ric ampliers were just giving way to new technology of
GaAs eld-effect transistor ampliers.
6. ILLUSTRATIVE EXAMPLE
There were many species of microwave parametric ampli-
ers, and a wide range of ampliers design were available
(see Refs. 46, and 911 for detailed and complete design
theories and practical considerations). The choice was de-
pendent on the particular application. In many applica-
tions, the lower limit of background noise in the system
was determined by thermal noise entering the antenna
from the terrestrial surroundings, and for such applica-
tions relatively simple uncooled parametric ampliers
were commercially available. For reception of the weak
signals from communication satellites or interplanetary
probes, however, extremely low system noise temperature
was essential and a whole range of specialized cooled am-
pliers was developed offering noise temperatures as low
as 1215K (comparable with masers).
2834 MICROWAVE PARAMETRIC AMPLIFIERS
Bandwidth and operating frequency requirements
were widely diversied. Paramps were made to amplify
signals up to the millimeter-wave frequencies (60-GHz
paramps are reported in Ref. 9). For some terrestrial com-
munication and radar requirements, with several MHz
bandwidth were routinely manufactured. Satellite com-
munication systems required 500 MHz bandwidth, and
specially designed paramps were developed for such
systems.
A 3.25-MHz nondegenerate parametric amplier [12]
has been selected as a representative example to illustrate
the elegant design and renement of microwave construc-
tion. The amplier designed for spectral line radio astron-
omy covers the frequency range 3.13.4GHz with an
instantaneous bandwidth of 40 MHz. A section drawing
of the amplier is shown in Fig. 6. Description of the am-
plier is rewritten here with permission of John Wiley &
Sons, Inc.:
The varactor diode is mounted in the E-plane of a reduced-
height waveguide which couples pump power from a 22 GHz
reex klystron to the varactor. A short length of high-imped-
ance coaxial line series resonates the diode mean capacitance
at the signal frequency. A three-element low-pass lter iso-
lates the pump and idler from the input line while the pump
waveguide is cut off at the idler frequency, conning the idler
to the vicinity of the varactor and the idler cavity. The idler
circuit contains a tunable cavity coupled to the varactor by an
iris. The position of this iris was chosen to optimize the pump
coupling to the varactor. The idler frequency is determined by
the combination of the package parasitic reactances, the cou-
pling, and the tunable cavity, which consists of a micrometer-
adjustable noncontacting short circuit in a cylindrical tube.
The pump and idler blocking lter forms part of a quarter-
wave.
transformer in the input coaxial line, which is used to
transform the characteristic impedance of an external cir-
culator to the value required to obtain a desired gain
[R
L
in Eq. (13)].
No external bias is provided, the varactor being pumped until
self-bias is developed.
The amplier gain was set to 20 dB and, from noise mea-
surements of uncooled receiver, its effective noise temper-
ature was estimated to be 60K.
BIBLIOGRAPHY
1. L. A. Blackwell and K. L. Kotzebue, Semiconductor-Diode
Parametric Ampliers, Prentice-Hall, Englewood Cliffs, NJ,
1961.
2. B. W. Siergowancew, Microwave Parametric Ampliers (in
Russian), Sowietskoje Radio, Moscow, 1961.
3. A. P. Bielousow, Parametric Ampliers with Varactor Diodes
(in Russian), Oborongiz, Moscow, 1961.
4. P. Peneld, Jr. and R. P. Rafuse, Varactor Applications, MIT
Press, Cambridge, MA, 1962.
5. M. Uenohara, Cooled varactor parametric ampliers, in
L. Young, ed., Advances in Microwaves, Academic Press,
New York, 1967, Vol. 2.
6. M. Uenohara and J. W. Gewartowski, Varactor applications,
in H. A. Watson, ed., Microwave Semiconductor Devices and
Their Circuit Applications, McGraw-Hill, New York, 1969.
7. H. N. Daglish, J. G. Armstrong, J. C. Walling, and C. A. P.
Foxell, Low-Noise Microwave Ampliers, Cambridge Univ.
Press, Cambridge, UK, 1968.
8. D. P. Howson and R. B. Smith, Parametric Ampliers,
McGraw-Hill, London, 1970.
9. C. L. Cuccia, Ultralow-noise parametric ampliers in com-
munication satellite earth terminals, in L. Young, ed., Ad-
vances in Microwaves, Academic Press, New York, 1971,
Vol. 7.
Varactor
Quarter-wave transformer
Pump and idler filter
Micrometer
Varactor
Idler tuning cavity
Coupling iris Signal tuning
Reduced height and
width waveguide
Figure 6. Representative example of the mi-
crowave construction of a parametric amplier:
section diagram of a 3.25-GHz nondegenerate
parametric amplier. (Source: J. W. Archer and
R. Batchelor, Multipliers and parametric devic-
es, in K. Chang, ed., Handbook of Microwave
and Optical Components, Vol. 2, p. 187. r1990
John Wiley & Sons. Reprinted by permission of
John Wiley & Sons, Inc.)
MICROWAVE PARAMETRIC AMPLIFIERS 2835
10. K. St. Grabowski, Parametric Ampliers and Mixers with
Varactor Diodes (in Polish), Wydawnictwa Naukowo-Tech-
nicz-ne, Warsaw, 1968.
11. J. C. Decroly, L. Laurent, J. C. Lienard, G. Marechal, and J.
Vorobeitchik, Parametric Ampliers, Macmillan, London,
1973.
12. J. W. Archer and R. A. Batchelor, Multipliers and parametric
devices, in K. Chang, ed., Handbook of Microwave and Opti-
cal Components, Wiley, New York, 1990, Vol. 2.
13. S. Weinreb, M. W. Pospieszalski, and R. Norrod, Cryogenic,
HEMT, low-noise receivers for 1.3 to 43GHz range, IEEE
MTT-S Int. Microwave Symp. Digest, 1988, pp. 945948.
14. M. T. Faber, J. Chramiec, and M. E. Adamski, Microwave and
Millimeter-Wave Diode Frequency Multipliers, Artech House,
BostonLondon, 1995.
15. A. Khanifar, R. J. Green, A. Khosrowbeygi, N. T. Ali, and M.
Milovanovic, The analysis of photoparametric amplifying de-
vices and characteristics, IEEE MTT-S Int. Microwave Symp.
Digest, 1995, pp. 15031506.
16. J. -C. Lee, H. F. Taylor, and K. Chang, Degenerate parametric
amplication in an optoelectronic GaAs CPW-to-slotline ring
resonator, IEEE Microwave Guided Wave Lett. 7(9):267269
(1997).
17. H. V. Shurmer, Microwave Semiconductor Devices, Pitman
Publishing, London, 1971.
FURTHER READING
S. Okwit, An historical view of the evolution of low-noise concepts
and techniques, IEEE Trans. Microwave Theory Tech. MTT-
32(9):10681082 (1984).
M. E. Hines, The virtues of nonlinearitydetection, frequency
conversion, parametric amplication and harmonic genera-
tion, IEEE Trans. Microwave Theory Tech. MTT-32(9):1097
1104 (1984).
D. G. Tucker, Circuits with Periodically-Varying Parameters, Mac-
donald, London, 1964.
S. A. Maas, Nonlinear Microwave and RF Circuits, 2nd ed.,
Artech House, BostonLondon, 2003.
S. Yngvesson, Microwave Semiconductor Devices, Kluwer, Boston,
1991.
S. A. Maas, Microwave Mixers. 2nd ed., Artech House, Boston
London, 1993.
MICROWAVE PHASE SHIFTERS
JOSEPH F. WHITE
JFW Technology, Inc.
1. SCANNING ARRAY ANTENNA APPLICATIONS
Mechanical motion necessary for antenna scanning was
perceived to be slow and unreliable. For this reason the
microwave industry developed an intense interest in
phased-array antennas, primarily for military but also
for commercial applications. The antennas radiated wave-
front would be steered by thousands of individual radia-
tors, roughly one for each half-wavelength square area of
the radiating aperture. Each radiator would be controlled
by a solid-state (either semiconductor or ferrite) phase
shifter, having low insertion loss and 03601 of phase shift.
For computer control the phase shift would be accom-
plished in binary bits. Thus a 3-bit phase shifter would
have a 1801, a 901, and a 451 section and these could be
used to provide 03151 control in 451 steps (the last step to
3601 is not needed, being equivalent to 01 in the steady
state). The antennas would be more expensive, both be-
cause of their need for numerous control elements (a cir-
cular aperture 30 wavelengths in diameter requires about
2500 elements) and the fact that, since a phased array
provides only about 7451 of steering, four separate aper-
tures are needed for 3601 azimuthal coverage. But they
would be fast and nearly failsafe, since a failure of a few
elements would result in but graceful degradation of the
system.
Ideally time delay (Fig. 1) is used to steer an array of
antenna elements, and a two-dimensional array uses total
time delay equal to that required for both azimuth and
elevation steering. When time delay is used, the steering
is frequency-independent, very desirable for a broadband
antenna. However, the time delay required, equivalent to
70% of the antenna width for 451 beam steering along ei-
ther of the antennas steering axes, can amount to thou-
sands of degrees of control. Instead, phase control is used.
The requisite time delay is rst calculated by the beam-
steering computer and then all integer wavelengths
dropped. The residue in degrees is then provided (to with-
in one-half of the least significant bit) as a command to the
binary bit phase shifter. In some cases, groups of adjacent
phase shifters (subarrays) may employ time-delay steer-
ing to enhance antenna bandwidth performance.
Over the last four decades there has been a keen com-
petition between the rival technologies, semiconductor
and ferrite, to achieve the phase control. The semiconduc-
tor devices and their circuitry are generally faster switch-
ing and inherently reciprocal (having the same phase shift
on transmit as receive), a useful antenna property. Being
switches, their phase control is essentially temperature
invariant, and the circuits more easily reproduced. Driver
circuits, which interface the microwave control circuit to
the array antenna beam steering computer, are very sim-
ple for the semiconductor phase shifter. However, semi-
conductors are discrete and small switching ele-
ments, and thereby limited in their peak power-handling
(N + 1) (N 1) (2) (1) (N)
0
0
d
N
S S
. . . . . .
. . . S
Figure 1. A linear phased array steered with time delay.
2836 MICROWAVE PHASE SHIFTERS
capacity. Furthermore, their insertion losses increase with
frequency.
Ferrites are a controllable propagation medium for mi-
crowaves and thus have more volume and a higher power-
handling capacity. Properly designed, they have low losses
at higher microwave frequencies and, for certain circuit
congurations, can be made reciprocal, even though prop-
agation through the medium itself is nonreciprocal. Con-
siderable attention must be given to the ferrites ux
driver circuitry to achieve reproducible binary phase con-
trol from unit to unit and over temperature and bias sup-
ply voltage changes. In fact, taking its driver circuit into
account, a ferrite phase shifter typically includes more
semiconductors than does a semiconductor phase shifter.
This section treats the semiconductor phase shifter.
While they could be built using a variety of semiconduc-
tors, diodes, and bipolar and eld-effect transistors, the
principal development was with silicon pin diodes because
of their relatively low cost, high microwave Q, and inher-
ent inertia to changes in characteristics with applied mi-
crowave (RF) excitation. To appreciate this requires some
description of the pin diode.
2. THE PIN DIODE
Generally, semiconductor pn junctions have rapid re-
sponse to an applied voltage and even can be used to rec-
tify an RF signal for detection purposes. However the pin
has a high resistivity (intrinsic), undoped region between
its p and n zones. The result is that holes and electrons
which are injected from the p and n zones under forward
bias move by diffusion into the i region, where they serve
as mobile charge not unlike electrons in copper, rendering
the i region conductive to an applied RF signal. Electrons
and holes can combine with one another, resulting in car-
rier death, but to do so they must give up energy equal to
the energy difference between the valence and conduction
bands (the bandgap). For silicon this is 1.1 electronvolts,
and such a drop in energy requires an energy emission, if
performed in one step, of a photon of visible light. We do
not observe silicon to be glowing with such light emission,
because such a transition is very unlikely. Put another
way, the lifetime of an electronhole pair is long, tens of
microseconds for the resistivities obtained in practical di-
odes. The staircase of energy steps resulting from impu-
rities and stresses in an otherwise ideal silicon crystal
produce a far more likely energy transition between
bands, consequently lower carrier lifetime.
The charge storage in a pins i region is equal to the
product of the lifetime and the forward bias (Fig. 2). Thus,
for example, a 1000V breakdown pin diode might have a
5 ms lifetime and be biased with a current of 100mA, re-
sulting in a stored charge of 0.5 mC. When a 1GHz sinu-
soid having a peak current of 50 A is applied to the diode,
it causes a peak-to-peak charge movement of less than
0.025 mC, less than 5% of the charge stored by the bias.
The result is that the diode appears to be a low value of
resistance throughout the entire RF sinusoid.
The same diode, when operated at a reverse bias of
100V, is able to sustain, without conduction, an applied
RF sinusoid of 1000V peak. This is because the diode re-
quires a microsecond or more to establish a conducting
state in the i region, much longer than the 0.5ns forward-
going voltage duration of a 1 GHz sinusoid.
Well-made pin diodes enjoy a bulk breakdown voltage
of about 10 V/mm (250 V/mil) of i-region width. It is this
bulk breakdown that determines the pins ability to sus-
tain RF voltage. Conduction due to impact ionization in
the i region can occur rapidly, even within an RF half-
cycle.
Ryder [1] has likened the bias on a pin diode to the
large signal and the RFas the small ac component, the
truth of which is evident from the relative magnitudes of
the charges related to each. The result of this remarkable
behavior is that the pin can control tens of kilowatts of RF
power, using only fractions of a watt of bias power.
Using the charge control approach for determining the
RF properties the RF resistance, R
i
, of the pin under for-
ward bias is found to be [2, p. 62]
R
i
W
2
=2m
AP
ti
0
1
where, in the pin diodes i region, Wthe i region thick-
ness; m
AP
the ambipolar mobility (the effective average
velocity per unit applied electric eld of the holes and
electrons); t the average lifetime of holes and electrons;
and i
0
the forward bias current.
For the pin used in the example of Fig. 1, W100 mm
(4 mils), m
AP
610cm
2
/V (in silicon), t 5ms, and a suit-
able bias current is 0.1 A, resulting in an RF resistance of
only 0.16 O. To R
i
must be added the ohmic contributions
of the p and n regions of the diode, as well as the contact
resistances of the diode package. Even so, the total for-
ward-biased resistance R
F
is usually 0.5 O or less. Inter-
estingly, R
i
is not dependent on i-region diameter D,
directly. However, indirectly it is, since smaller diameter
diodes have lower t, because carriers, on average, are
0.5 ns
f = 1000 MHz
Diode bias current
Stored charge = Q = 0.1 A 5 10
6
s
= 0.5 jC
> (charge) <
=
50 A 0.5 10
9
s
0.025 jC
V
A
V
B
V
F
V
100 mA
I
50A
1000 V
Figure 2. Example comparing charge stored in a pin diode by the
bias to the charge movement due to a high-level RF signal.
MICROWAVE PHASE SHIFTERS 2837
closer to the i-region boundaries at which recombination
can more readily take place.
PIN diode area A does relate to the junction capaci-
tance, C
J
, which follows the parallel plate formula fairly
closely.
C
J
e
R
e
0
A=W 2
At low frequencies, say, 1 MHz, a C change between zero
and reverse bias voltage is observable; however, at RF, it is
the minimum capacitance that is experienced due to the
dielectric relaxation of the i region [2]. With the high di-
electric constant of silicon (e
R
11.8), there is little fring-
ing of the electric eld.
In series with this capacitance is a resistance R
R
(not
necessarily equal to R
F
), which is determined by measure-
ment. An RF gure of merit for the pin is the switching
cutoff frequency F
CS
, given by
F
CS
1=2pC
J

R
F
R
F
p
3
As will be described later, the F
CS
value permits a predic-
tion of the minimum insertion loss to be obtained in a
phase shifter, switch, or duplexer circuit, even before the
circuit conguration has been specied [2, Chap. 5]. The
RF equivalent circuit of the pin chip in its two bias states
is shown in Fig. 3. Package capacitance and inductance
must be added for a complete packaged diode. For our
sample diode, having a junction diameter of 0.49mm
(19 mils), C
J
0.2 pF.
When suitably soldered either into a package or onto a
good heat sink, it is found that the junction temperature
rise of our sample diode is about 151C per watt of power
dissipated in the junction. Equivalently stated, its ther-
mal resistance, y 151C/W. In pulsed RF power applica-
tions there may be insufcient time during the pulse for
thermal equilibrium to be reached. The junction temper-
ature may rise linearly and the diode i region, small as it
may seem, must sink the heat dissipated until it can ow
out through the thermal resistance path (Fig. 4). The heat
capacity HC of the i region is given by
HCspecific heat density volume 4
which, for a silicon pin(0.74 J/g-1C) (2.43g/cm
3
)
(pD
2
W/4).
For our sample diode, having D49mm (19mils) and
W100mm (100 mils), HC34 mJ/1C. This is indeed
a small heat capacity, yet it implies that, for a given
temperature rise, the pin could dissipate nearly 7 times
as much power for 1 ms as it could sustain with continuous
dissipation. Furthermore, this HC calculation is conser-
vative, because it ignores the heatsinking capacity of the
bondwires and the p and n portions of the diode that are in
intimate contact with the i region. The product of HC and
y gives the thermal time constant, t
T
, from which the tem-
perature rise, DT
J
of the i region can be estimated for any
pulselength, t, of power dissipation, P
D
. Thus
t
T
HCy 5
DT
J
P
D
y1 e
t=t
T
6
The temperature rise is shown graphically in Fig. 5. For
the sample diode the minimum t
T
is 500ms. If a safe tem-
perature rise is considered to be 100 1C, then the diode
could dissipate 6.6 W continuously, 16 W for 500ms, 66 W
for 50ms, and so forth. Following the pulse, the diode cools
during the interpulse periods with the same thermal time
constant (Fig. 6).
This same reasoning could be applied to develop a more
complete thermal model of the diode, which includes its
thermal surroundings. Figure 7 shows a more represen-
tative model of the diode, with its thermal elements and
(High power
loss element)
Forward
bias
Reverse
bias
R
F
R
R
C
J
G
R
Figure 3. pin-diode chip equivalent circuit.
T
W
P+
Minimum
heat sink
model
N+
W
A
D
D
A
Figure 4. pin model used for heatsinking calculation.
0
0 1 2
^
T

=

P
D

t
/
H

C
^T = P
D

3 4
0.2
0.4
0.6
0.8
1.0
t /t
T
, Ratio of pulse length
to thermal time constant
^
T
m
/
P
D
0
,

N
o
r
m
a
l
i
z
e
d

j
u
n
c
t
i
o
n
t
e
m
p
e
r
a
t
u
r
e

r
i
s
e

Figure 5. General pulsed temperature-rise prole of a pin diode,
using the minimum-time-constant model.
2838 MICROWAVE PHASE SHIFTERS
that of the packaging materials. Generally, however, the
simple conservative model is sufcient to estimate the
maximum temperature rise to be expected from a given
pulsed power dissipation.
A representative listing of a wide range of pin diodes is
shown in Table 1. Given these parameters, it is possible to
estimate most of the performance of a variety of RF phase
shifter, switch, and duplexer circuits, even before the cir-
cuits themselves are specied.
3. LOADED LINE PHASE SHIFTER
A diode phase shifter is a device whose primary function is
to change, by means of a control bias, the propagation
phase of a microwave signal. Most switches, attenuators,
limiters, and duplexers introduce phase shift, although
seldom by design. Moreover, since any reactance placed in
series or shunt with a transmission line introduces phase
shift, the possibilities for phase shift networks are unlim-
ited. However, adding the requirement that the device has
minimum insertion and reective losses reduces the se-
lection of practical circuits.
Most think of switching between circuit paths as a di-
rect means of phase shift (Fig. 8). Actually this is a
switched time-delay circuit, producing phase shift that is
linearly proportional to frequency. This might seem all the
more desirable, since it could lead to broadband array an-
tenna steering. However, in practice, the switching be-
tween paths is accomplished with limited isolation of the
nonselected path. Figure 9 shows how the loss can in-
crease dramatically when the OFF arm resonates.
While time delay circuits have a place, they are not ef-
cient. All the RF power must be switched between paths
and four diodes minimally are required to do this. The in-
sertion loss is the same for all bits, whereas in a phase
shifter circuit, only two diodes are required per bit and the
diode loss is much less for small phase shift bits. Accord-
ingly, we shall omit further discussion of time-delay cir-
cuits and proceed to phase shifters (which, generally, do
not have linearly increasing phase change with frequency).
Initially it was thought that very high power phase
shifters would be required. In fact, the author conducted a
Navy-sponsored project, whose objective was a 100kW
peak power phase shifteran objective that was met and
applied to a high power array! The key to the development
was the recognition that, since numerous diodes would be
required for very high power, each contributing a small
amount of the total phase shift required, a circuit lightly
coupling the diodes to the propagating wave was needed.
The solution was the transmission phase shifter
(Fig. 10), in which pairs of diode-switched susceptances
load the transmission line. The spacing of the suscep-
tances, about 901, is selected to cause mutual cancellation
of their reections. The magnitude of the susceptances is
made small (less than 0.4 Y
o
), and so the diodes are sub-
jected to relatively small RF currents and voltages, pro-
ducing low dissipation in each diode. This enhances both
power-handling and insertion-loss performance.
This loaded line section has an equivalent circuit [2, p.
410] consisting of a uniform line section of new character-
istic admittance Y
E
and electrical length y
E
, related to the
loaded lines admittance Y
o
and electrical length y by
cos y
E
cos y B=Y
O
sin y 7
Y
E
Y
O
1 B=Y
O

2
2B=Y
O
cot y
1=2
8
Consider Eq. (7) rst. If the line loading susceptance is
somehow switched by the pin diode between equal-mag-
nitude and opposite-sign susceptors, B
1
and B
2
, then the
electrical length of the loaded line is described by the vec-
tor diagram in Fig. 11.
Notice from Fig. 11 that when y 901 the sine of the
phase shift Dj/2, produced by each of the equal suscep-
tances B
i
, is equal to the normalized susceptance term
T
J
,

M
a
x
.

j
u
n
c
t
i
o
n

t
e
m
p
.
(

C
)
T
3
= 40
60
80
100
120
P
o
0
t /T
T
4
5T
T
r
T
3 2 1 7 6 5
D
i
s
s
i
p
a
t
i
o
n
Figure 6. Sample estimate of pin junction temperature during a
train of power dissipating pulses.
Heat sink
electrode
Heat sink
Solder
Peripheral silicon 3
Solder
Top weight 1
Active I region
(a)
(b)
HC
3
P
D
t
4
2
5
6

3
0

1
0
2
0
4
0
5
0
6
0
HC
2
HC
4
HC
5
HC
6
HC
1
Figure 7. Construction pin detail and its thermal model.
MICROWAVE PHASE SHIFTERS 2839
Table 1. Typical Parameters of Available PIN Diodes
Offered by the M/A-COM Division of AMP Inc. in Burlington, MA.
2
8
4
0
M
I
C
R
O
W
A
V
E
P
H
A
S
E
S
H
I
F
T
E
R
S
BZ
o
(Z
o
1/Y
o
). Then, approximating the sine by its angle,
the total phase shift in radians obtained by switching be-
tween B
1
and B
2
is given by
Dj B
2
B
1
Z
O
9
For example, if the normalized susceptances switch be-
tween plus and minus 0.2, then the phase shift is 0.4 ra-
dians, near 22.51, a 16th of a wavelength. The respective 4
bits of a phase shifter can be made up of one, two, four, and
eight such sections in cascade.
But, the reader may ask, suppose that the individual
reections from each section, although small in them-
selves, combine when 15 such sections are cascaded to
produce very large reection and with it high mismatch
loss.
Such is not the case. Referring to Eq. (8) and applying
the values y 901 and |B
i
Z
o
|0.2, gives Y
E
0.98Y
o
.
This is true for either positive (capacitive) or negative
(inductive) line loading. Thus, even as the phase length of
the section changes, its characteristic admittance does
not. Nor is its value very different from Y
o
. Accordingly,
an arbitrarily long cascade combinations of such sections
would not result in a VSWR larger than 1.04, provided
that there are no line sections intervening between the
phase shift sections. Even this small mismatch could be
further compensated by installing a quarter-wave line of
admittance 1.02Y
o
at each end of the phase shifter cas-
cade. This inherent match of the loaded line phase shifter
is one of its most useful attributes.
It now remains to design the line loading circuits, such
that a two-state diode can yield the 70.2 normalized
susceptance switching. The rst circuit approach used
shunt stubs, whose length was varied by pin-diode switch-
es (Fig. 12). The diodes were similar to the 0.2-pF, 4-mil I
region model described in Table 1. The linelengths a
1
and
a
2
were adjustable. The phase shift was proportional to a
1
while the average of the two lengths was adjusted to con-
trol the transmission match. With 51 of phase shift per
stub pair, a level of 140 kW peak power was sustained with
0.001 duty cycle, 5m-sec-long pulses at 1300 MHz. There
were eight sections in the experimental model and the
results are shown in Fig. 13.
The maximum powers listed are those that cause or
nearly cause burnout, usually occasioned by voltage
breakdown of the pin diodes in the reverse biased state.
For pins, as well as all semiconductors, reliable operation
requires a rating that impose only 50% of this maximum
voltage stress on the semiconductor. Since power is related
0
2
4
6
8
20
40
60
80
1.8 1.7 1.6
Long path, L
2
switched in
Z
0
= 50 , loss = 0.1 dB/
2V
O
+
0.2 pH
50
1

1
+
Short path, L
1
switched in
L
2
= 205 @ 1.5 GHz
V
L
= ARG (V
O
/V
L
) [
L
1
= 160 @ 1.5 GHz
1.5
Frequency (GHz)
1.4 1.3 1.2
I
n
s
e
r
t
i
o
n

l
o
s
s

(
d
B
)
z

[
,

p
h
a
s
e

s
h
i
f
t

(

)
Figure 9. Switched path circuit example demonstrating loss
resonances.
Z
0
jB
1
jB
2
jB
2
jB
1
0
Figure 10. Switched transmission phase shifter section, also
called the loaded line phase shifter.
Unit radius
circle
+1 1
B
2
Z
0
B
1
Z
0

e
1
e
2
[
[
[
1 [
2
[
[
Figure 11. Graphical representation of the loaded line phase
shifters change in electrical length caused by switching the load-
ing susceptances.
(a) (b)
Figure 8. Schematic for switched delay line phase shifter.
MICROWAVE PHASE SHIFTERS 2841
to the square of voltage, this means devices need be rated
at one-fourth the power level, which would cause imme-
diate failure.
For microwave phase shifters this is especially useful in
the event of a short-circuited output, which could nearly
double the voltage stress on its diodes. Such a short can
result from an arc over or damaged radiating element,
even a disconnected output. Indeed, a customary accep-
tance test for a high-power-control device is operation
into a short-circuited load, which is varied through all
phases. Given this derating, very high reliability of pin
phase shifters is experienced, as is necessary in an array
antenna.
While it is true that no practical phased array could
radiate such levels (a 2500-element array using 35 kW
phase shifters would radiate 87 MW peak power), this
result is significant, because single-pole double-throw
switches can be constructed by installing such phase shift-
ers between 3 dB hybrid couplers, allowing, for example,
the full output power of a radar to be switched between
alternate antennas.
The loaded line approach was extended to 3GHz in a
circuit in which the diodes own capacitance terminates
the quarterwave shunt stub. Switching between forward
and reverse bias changed between j50 O (the diode has
about 3 pF capacitance) and its forward resistance of 0.5 O.
Adjusting the shunt stub impedance produced as much as
451 phase shift per pair and a maximum RF peak power of
70 kW [2, p. 429]. This phaser was used in the U.S. Safe-
guard system, of which only a prototype system was built.
0
0 2 4
Number of reverse biased sections
(other sections forward biased)
6
140 kW
10 kW Loss
55 kW
32 kW
Burnout Peak Power = 16 kW
Measured
8
22
1
/2
0.5
0
20
60
100
140
180
Input
VSWR
a
2
a
1
22
1
/2
15

10
5

1.16
1.12
1.14
1.08
S
t
e
p

p
h
a
s
e

s
h
i
f
t

(

)
I
n
s
e
r
t
i
o
n

l
o
s
s

(
d
B
)
Figure 13. L-band measurements of phase shift, insertion loss,
and ultimate peak power capability of the switched stub, loaded
line phase shifter. Varying switched stub lengths (a
2
and a
1
) pro-
duced the different phase shift values and adjusted transmission
match.
V
O 1
Y
S
Y
S
a
2
a
2
V
2
+
To
bias
1
RF
choke
Y
C
= 1

a
1
a
1
V
1
+

0
Figure 12. Equivalent circuit for one section of the switched stub, loaded line phase shifter tested
at 1300MHz under high peak power.
2V
O
V
1
jb jb
jb jb jx
V
2
1
(1)
For a match, A + B
2
= 1
and under this condition,
=
1.2
A
C
B
D
1
0
jx
1
1
jb
0
1

1
jb
0
1
+

(2)
=
x =
= arg (V
2
/V
O
) = arg (A + B )
= tan
1
2b
1 +b
2

2b
b
2
1
(1 bx
j b (2 bx)
jx
(1 bx)
[
[
. .
Figure 14. The lumped-element p-conguration phase shifter.
2842 MICROWAVE PHASE SHIFTERS
4. LUMPED-ELEMENT PHASE SHIFTERS
If the quarterwave section of line in the loaded line section
is replaced with a diode switchable reactance (Fig. 14), the
lumped-element, highpasslowpass p circuit is obtained,
which can yield up to 1801 of phase shift. A similar tee
conguration [2] is also practical. These circuits are ad-
vantageous in integrated circuit applications, because
they employ a minimum of switching elements and no
space-consuming distributed elements.
For modest power levels, the most common diode phase
shifter conguration is the reection circuit, employing a
3-dB, 901 coupler (Fig. 15). The coupler can be realized in
numerous ways, three of which are shown in Fig. 16. The
operation of the coupler is to convert the pair of variable
phase reection circuits containing pin diodes into a
matched two-port network, having the reection angle
change of the terminations.
This operation can be explained based on the couplers
operation. Consider the backward-wave (hybrid coupler)
circuit at the bottomof Fig. 16. Power enters the coupler at
port 1 and divides evenly to exit at ports 2 and 4. The wave
exiting port 4 has an additional 901. On encountering the
reective circuits at ports 2 and 4, all energy reenters the
coupler, but due to the second 901 phase difference in the
signals on this second pass, they cancel at the input
(port 1), but add perfectly at the normally decoupled port 3.
This operation requires perfectly even power split and
901 phase difference. The backward coupler (but not the
other types shown) has the remarkable property that the
901 phase difference prevails at all frequencies [2, p. 194].
Of course, the power split varies with frequency, being
equal at only one frequency (or two frequencies if the de-
sign is overcoupled at the center frequency). Nevertheless,
more than octave bandwidth (Fig. 17) with modest VSWR
Symmetric, switchable,
reflective terminations
(3)
(1)
(2)
(4)
3 dB Hybrid (90)

1
2
1
2
Figure 15. The reection phase shifter circuit employing a cou-
pler to achieve matched two-port transmission.
1.0
1.5
2.0
2.5
3.0
3.5
1.6 1.4 1.2 1.0
4.0 dB
3.6
3.3
3.0
2.7
2.4
2.0
f /f
0
.8
Coupling = 20 log
10
k =
(3) (2)
Z
0
Z
OE,
Z
OO
(1) (4)
T

T
.6 .4
I
n
p
u
t

V
S
W
R
0
Figure 17. VSWR performance with frequency for a coupled-line
hybrid coupler terminated in symmetric reections for various
coupling values. Note that the 2.4-dB coupler would operate with
a maximum VSWR of 1.35 over a band of 0.51.5 f
0
, a 31 fre-
quency range (f
0
is the frequency at which the coupling section is
901 long).
Z
o
Z
o
Z
o
Z
o
/ 2
/4
/4
/4
/4
/4
/4
2
2
4
4
Branch line
Hybrid phase
Shifter bit
Backward-wave
coupler bit
Rat race bit
1
1
1
4
2
3
3
3
2 Z
o
Z
o
/ 2
3 /4

z
z
z
z
z z
z
Figure 16. Methods of realizing 3-dB, 901 cou-
plers.
MICROWAVE PHASE SHIFTERS 2843
can be obtained with a single coupled line section, even
more bandwidth with multistage couplers.
As was true of the loaded line circuit, there are numer-
ous ways to congure the pin in a reection circuit to yield
any desired phase shift. However, regardless of what con-
guration is used, if the circuit is designed to present to
the pin its maximum sustainable RF voltage V
M
in the
reverse biased state and the maximum sustainable RF
current I
M
in the forward-biased state, Hines [3] showed
that the maximum power P
M
sustainable when the circuit
yields a phase shift Dj is as shown in Table 2. Similarly, if
the circuit is designed such that the fraction of incident
power dissipated (P
D
/P
A
) is the same in both forward bias
and reverse bias, then the minimum for this ratio is that
shown in Table 2.
Generally, the choice of circuit that would provide the
maximum power stresses to the diode is not the same
choice that would produce equal power dissipation in its
two states, but the two limits are very useful for estimat-
ing what performance limits a practical circuit might in-
cur. Furthermore, since phase shifter bits are usually
designed for low loss, the average of the losses in the
two bias states is about equal to the minimum value spec-
ied in Table 2. The loss so calculated is for pin dissipation
only. Circuit losses add to this value; but, as will be shown,
a practical 3-bit, L-band phase shifter can be made with
less than 1dB of total insertion loss.
If the series resistance of the pin is about the same in
both bias states (R
F
R
R
), an equal loss phase shifter can
be made by installing the pin at the 3dB outputs of the
coupler with a series inductance whose reactance magni-
tude is half that of the pins capacitive reactance. By in-
stalling a quarterwave transformer between coupler and
diode termination (Fig. 18), the phase shift can be adjust-
ed to any desired value, allowing use of the same pin and
series inductance for all bits (Fig. 19).
The method for constructing the backward-wave cou-
pler [2] reection phase shifter is shown in Figs. 20 and
21. A three-layer dielectric stripline sandwich is em-
ployed, wherein the center dielectric is used for the cou-
pled lines. Using this approach, a 3-bit phase shifter was
designed for use in the Cobra Dane radar built for the U.S.
Air Force by the Raytheon Company on Shemya Island,
near the western tip of the Aleutian Island chain. To min-
imize losses, the outer dielectrics were removed, resulting
in air stripline in the transformer and diode regions. The
operating bandwidth was approximately 12001400 MHz,
To Z
O
coupler
Socket points
X
S
= X
C
= X
L
/ 2
R
S
= R
F
= R
R
Z
T
R
F
R
R
L
Z
S
= R
S
jX
S jX
C
j X
L
F
R
C
0
Figure 18. The loss-equalized phase shifter termination with
series inductor and quarter-wave transformer to adjust phase
shift value.
Figure 19. The reection coefcients seen at the
coupler for loss-equalized 1801, 901, and 451
phase shift bits.
2844 MICROWAVE PHASE SHIFTERS
and the required power-handling capacity was to be 1 kW
peak, with evenly spaced pulses of up to 2000ms and 0.05
duty cycle.
The schematic diagram for the unit is shown in Fig. 20.
Four separate phase shifters and a Wilkinson equal-phase
power divider were housed in a single assembly to reduce
costs and interface connections. Figure 21 shows the diode
mount detail, and measured performance is shown in Fig.
22. A photograph of the completed assembly is shown in
Fig. 23.
The diode used for all bits was the 3-pF, 1800-V bulk
breakdown pin listed in the rst column of Table 1. The
individual phase shifter section, tested with 200V bias on
all diodes, sustained 4.14.8 kW peak power before burn-
out, and therefore could be rated for 1 kW operating level.
The insertion loss of each phase shifter, including both pin
diode and circuit losses, was 0.7dB. About 16,000 phase
shifters were installed in the Cobra Dane antenna array,
which radiates approximately 16 MW peak and 1 MW of
average power. At the time of installation the array was
operated 20 h per day, resulting in nearly 2 million device
hours daily.
In separate projects, pin phase shifters of 3- and 4-bit
designs were implemented at S, C, and X bands [2, Chap.
6]. C-band phasers nd use in the scanning-beam micro-
wave landing system (MLS), for which a worldwide stan-
dard exists. Generally, higher-frequency diode phasers
have progressively higher insertion loss and lower peak
power capacity. At X band, the 4-bit design had 2dB of
insertion loss and a burnout power of about 1000 W with
1 ms pulselengths and 0.001 duty cycle.
5. CONSTANT PHASE SHIFT WITH FREQUENCY
Frequently there is a need for a phaser whose phase shift
is constant over a considerable bandwidth. Schiffman [2,4]
observed that when the backward wave coupler has ports
2 and 4 connected to each other (Fig. 24), an allpass net-
work results, having a dispersion characteristic (an elec-
trical length that does not increase linearly with
frequency) that can be adjusted with the coupling coef-
cient (Fig. 25).
Input
4 kW
Outputs
(1)
(2)
(3)
(4)
3 Bit phase shifter
& driver
4-TO-1 Wilkinson
power divider
100
100
100
Figure 20. The 4-to-1 power divider and phaser assembly.
40
50
80
90
100
160
170
180
190
0
20
16
14
12
10
0.5
1.0
1500 1400 1300
Frequency (MHz)
1200 1100
P
h
a
s
e

s
h
i
f
t

(

)
R
e
t
u
r
n

l
o
s
s

(
d
B
)

I
n
s
e
r
t
i
o
n

l
o
s
s

(
d
B
)
Design bandwidth
Max. return loss envelope (all 8 states)
Min. loss (all 8 states)
Max. loss (all 8 states)
90 bit
45

bit
180

bit
Figure 22. Measured performance for the L-band, 3-bit stripline
phase shifter.
Detail view
of diode
(to scale)
Fired glass passivation of
I region is applied to diodes
in slice form
Mounted
diode
chip
Threaded copper
heat sink
Flexible copper strap
Tungsten stand
off contactor
Diode chip with
150 jm (6 mil)
I region width
3.6 mm
(90 mils)
1.2 mm
(48 mils)
Figure 21. The high-voltage pin chip mounted on a copper heat-
sink.
Table 2. Limits of Power Handling and Insertion Loss (P
D
/
P
A
) for Transmission and Reection Phase Shifters
Power Loss
Reection circuit P
M

V
M
I
M
4 sinDf=2
P
D
P
A
% f
f
f
cs

sin
Df
2

Transmission circuit P
M

V
M
I
M
2 tanDf
P
D
P
A
% 2
f
f
cs

tanDf
10fof
cs
f
cs

1
2pc
J

R
F
R
R
p :
MICROWAVE PHASE SHIFTERS 2845
A switched-path phase shifter (Fig. 26), which alter-
nates between a Schiffman section and a uniform trans-
mission line of appropriate length, can be made to have
a nearly constant phase shift over an octave bandwidth
(Fig. 27).
6. VARACTOR DIODE, CONTINUOUS PHASE SHIFTER
The varactor diode, lacking a wide I region, consists only
of a pn junction, whose capacitance at RF frequencies is
variable over a 51 or wider range with the application of a
reverse bias. Such a variable reactance can be used in
place of the pin in the reection phase shifter circuit, to
provide phase control that varies continuously with ap-
plied bias voltage. Of necessity, the continuous phase
Figure 23. Photograph of the Cobra Dane 4-to-1 divider and
highpower phase shifter assembly.
Z
OE
, Z
OO
Z
O
= Z
OO
Z
OE

+
Z
O
Z
O
2V
O
V
2
V
1
0

Figure 24. Schiffman phase shift section.


[
(

)
0 ()
0
0 45 90 135 180
j = 1.00 (no coupling)
3.01 (6 dB)
5.83 (3 dB)
90
180
270
360
Figure 25. Dispersion characteristic of the Schiffman section.
RF Output
Bias
Bias
Bias
return
z
/
4
z
/
4
z
/
4
RF
Input
SPDT
switch
Dispersive
path
Nondispersive
path
Figure 26. A stripline Schiffman phase shifter using diode path
switching (after Grauling and Geller [5]).
80
120
160
200
240
280
320
360
400
50 70 90
[= [
O
[
1
k

=

3

[
O
[
()
110 130
O

a
n
d
1

(

)
= 90
[ = 90

1
[
+c
= 4.8 c
c
c
= +4.8

=

3
.
0
1
j
[ = 90

[
[
0
Figure 27. The octave bandwidth Schiffman phase shifter
formed by a r 3.01 [2] coupled line pair y long a uniform line
path 3 y long. Phase shift is 90174.81 for 551 ryr1251, a 2.24:1
frequency ratio.
2846 MICROWAVE PHASE SHIFTERS
shifter is limited to low power, below 1 W, since the var-
actor capacitance can change at the RF rate. In fact, var-
actors are used as the nonlinear element in frequency
multipliers.
For varactor phase shifters, a gure of merit F applies
[2, p. 486], relating the number of degrees of phase shift
per decibel of loss to the cutoff frequency of the varactor, f
c
1/2pRC
MIN
, where C
MIN
is the minimum capacitance
obtained at a reverse bias just before the breakdown volt-
age, V
B
; M is the ratio of the capacitance at zero volts to
that at V
B
; and f is the frequency of operation.
F fc=f 1 1=M6:6

=dB 10
This equation applies when the loss is small, below 1 dB.
Thus a varactor having a junction capacitance which var-
ies from 10 to 2pF in series with a 2.6 O resistance
(Fig. 28) has a cutoff frequency of 159GHz and could yield
3231/dB at 1 GHz. Circuit losses must be added to this
value.
Generally, higher loss is obtained as a result of the
tuning effect of the varactors series inductance and the
circuit reactance employed to transform the reection co-
efcient into a range which covers both the upper (induc-
tive) and lower (capacitive) halves of the Smith chart
(Figs. 29 and 30).
The varactor phase shifter experiences little variation
with temperature, typically only a 1% change in total
phase shift over a 501C temperature change. Even this
small change may be attributable to circuit changes and
possibly could be reduced further. However, variations in
RF power beyond one watt produce significant changes,
particularly near zero bias, at which the applied RF volt-
age swings into the forward conducting region (Fig. 31).
7. THE FET AS A SWITCHING DEVICE
Actually, any electronically switched device can be used as
the control element in a phase shifter as long as it has
sufcient Q to provide acceptably low insertion loss. The
considerable strides made in the development of eld effect
90 80 70 60 50 40
Reverse bias
45 @ 1 GHx
C
1.9 nH
Z
O
= 50
2.6
30 20 10 0
0
2.0
4.0
6.0
8.0
10.0
C
,

T
o
t
a
l

d
i
o
d
e

c
a
p
a
c
i
t
y

(
p
F
)
Figure 28. Representative varactor diode capacitance and cir-
cuit termination for a varactor continuous phase shifter.
Figure 29. Phase shift (change in reection co-
efcient angle) and loss (departure from unity
reection coefcient magnitude) of varactor cir-
cuit in Fig. 28.
MICROWAVE PHASE SHIFTERS 2847
transistors (FETs) both for microwave power generation
and switching [1218] have two important implications for
semi-conductor phase shifting.
First, phase shifting for array antennas can be per-
formed at low power levels with subsequent amplication
by FETs and other solid-state devices to the required out-
put level at the array antenna radiating element. With
this approach the achievement of lowest insertion loss is
less critical. Of course, the amplication generally will be
nonreciprocal, requiring that switching be performed be-
tween transmission and reception modes.
Second, the phase shifter can be realized using FET
elements for switching instead of pin diodes. FETswitches
have a inherent advantage when compared to pin diodes
in that the FET has a third (gate) terminal to which bias is
applied, simplifying the design of the microwave circuitry
in which bias blocking otherwise is needed for two termi-
nal control elements.
FETs can be modeled as a lossy capacitor in the non-
conducting state and a resistor in the conducting state,
essentially the same equivalent circuit format used for the
PIN diode. As a result, most of the formation developed for
the pin can be used directly with FETs. The FET switched
phase shifter can be reciprocal, as is the pin phase shifter.
Acknowledgment
The contents of this article have been excerpted, with per-
mission, from the authors text, Microwave Semiconductor
Engineering [2]. The author thanks Randy Rhea and Gary
Breed for this opportunity. The phase shifter development
described took place at M/A-COM, formerly Microwave
Associates, and now a division of AMP, Inc., located in
Burlington, Massachusetts. Much of the development of
both the pin and varactor diodes, as well as the circuits to
exploit them, was sponsored by the U.S. Government
through its Air Force, Navy and Army research agencies.
BIBLIOGRAPHY
1. R. Ryder, Bell Telephone Labs, Murray Hill, NJ, in a talk giv-
en at the NEREM Conference in Boston, MA, circa 1970.
2. J. F. White, Microwave Semiconductor Engineering, Noble
Publishing, Tucker, GA (originally published under the title
Semiconductor Control, Artech House, Norwood, MA, 1977;
later republished under the current title by Van Nostrand
Reinhold Co., New York, 1982, and translated with permis-
sion into Japanese).
3. M. E. Hines, Fundamental limitations in RF switching and
phase shifting using semiconductor diodes, Proc. IEEE
52:697708 (1964).
4. B. M. Schiffman, A new class of broadband microwave 901
phase shifters, IEEE Trans. Microwave Theory Tech. MTT-
6:232237 (1958).
5. C. H. Grauling and B. D. Geller, A broadband frequency
translator with 30 dB suppression of spurious sidebands,
IEEE Trans. Microwave Theory Tech. MTT-18:651652
(1970).
6. W. J. Ince and D. H. Temme, Phasers and time delay elements,
in L. Young, ed., Advances in Microwaves, Vol. 4, Academic
Press, 1969, pp. 1189.
0
0
.2
.4
.6
.8
1.0
1.2
1.4
1.6
1.8
2.0
20
40
60
80
100
120
140
160
180
200
220
240
90 80 70 60 50 40
Bias (
.
V)
30 20 10 0
0
90 80 70 60
Calculated
Measured
50 40
Reverse bias (
.
V)
30 20 10
P
h
a
s
e

s
h
i
f
t

(

)
I
n
s
e
r
t
i
o
n

l
o
s
s

(
d
B
)
Calculated
Measured
Figure 30. Calculated and measured phase
shift and insertion loss at 1 GHz for the var-
actor termination of Fig. 27.
20
80 70
15 W PK
0.1 mW
5 W PK
1 = 1.2 GHz
60 50 40
Bias voltage
30 20 10
10
0
10
20
30
40
50
60
70
80
90
P
h
a
s
e

s
h
i
f
t

(

)
Figure 31. Typical variation of phase shift with RF input power.
2848 MICROWAVE PHASE SHIFTERS
7. I. Bahl and P. Bhartia, Microwave Solid State Circuit Design,
Wiley, New York, 1988, Chaps. 8, 13.
8. J. F. White, Semiconductor Control Devices: PIN Diodes, in K.
Chang, ed., Handbook of Microwave and Optical Components,
Vol. 2, Wiley, New York, 1990, Chap. 4.
9. A. I. Sreenivas and R. Stockton, Semiconductor control devic-
es: Phase shifters and switches, in K. Chang, ed., Handbook of
Microwave and Optical Components, Vol. 2, Wiley, New York,
1990, Chap. 5.
10. S. Yngvensson, Microwave Semiconductor Devices, Kluwer,
Norwell, MA, 1991, Chap. 9.
11. K. Chang, Microwave Solid State Circuits and Applications,
Wiley, New York, 1994, Chap. 8.
12. A. Mallet-Guy et al., Modeling and performance of a sub-
nanosecond high isolation DC-18GHz monolithic SPST with
driver, IEEE MTT-SInt. Microwave Symp. Digest, 1991, Vol.
1, pp. 193196.
13. H. Takasu et al., GaAs FET switch model for X-band MMIC
phase shifter design, 1994 IEEE MTT-S Int. Microwave
Symp. Digest, 1994, Vol. 3, pp. 14131416.
14. M. J. Schindler and T. E. Kazior, High power 218GHz
MMIC TR switch, Appl. Microwave Mag. 9094 (summer
1991).
15. F. McGrath et al., Multi gate FET power switches, Appl. Mi-
crowave Mag. 7786 (summer 1991).
16. T. Tokumitsu, I. Toyoda, and M. Aikawa, Low voltage, high
power T/R switch MMIC using LC resonators, IEEE Micro-
wave Millimeter-Wave Monolithic Circuits Symp. Digest, June
1993, pp. 2730.
17. A. Ehoud et al., Extraction techniques for FET switch mod-
eling, IEEE MTT-S Trans. Aug. 1995, pp. 18631867.
18. K. Purnell et al., GaAs MESFET, passive element, MMIC
phase shifter, IEEE Int. Microwave Symp. Digest, 1996, Vol.
2, pp. 11971200.
MICROWAVE PHASE SHIFTERS 2849
Next Page

Vous aimerez peut-être aussi