12i

MICROWAVE PHOTONICS: TECHNOLOGICAL
EVOLUTION AND ITS APPLICATIONS

AFSHIN S. DARYOUSH
Drexel University
Philadelphia, Pennsylvania
1. INTRODUCTION
The last two decades of the twentieth century witnessed
significant advances in IC technologies and proliferation of
commercial beroptic communication techniques, and mi-
crowave photonics has emerged as a new discipline. It is
envisioned that microwave photonics systems will be em-
ployed in many civilian and military systems. Although
optical control of microwave devices and circuits was ini-
tially considered as one of the most promising applications
of microwave photonics, now its most widely used appli-
cations are for optical distribution and processing of
information in telecommunication systems, such as radio-
over-ber, as shown for a wireless local-area network in an
ofce environment. In this gure, personal computers (PCs)
are networked together through this hybrid wirelessber-
optic networks in applications such as radio-over-ber
(ROF). The digital data from the PC is upconverted by a
millimeter-wave (MMW) stable carrier and are then radi-
ated by an omnidirectional antenna. The modulated re-
ceived RF signal is then downconverted to the coded digital
signals using the same stabilized MMW local oscillator
(LO). The coded digital signal is then networked to other
users through a high-speed beroptic network at data rates
well above 100Mbps (megabits per second).
An important component of a personal communication
services (PCS) for the portable PC or PDA is the use of the
low-power-consuming frequency translation circuits,
which up- and downconverts the information without any
degradation in its spectral purity. Frequency stability of
the local oscillators used in the MMW wireless communi-
cation, and the clock recovery circuits used in the decision
circuits, are critical in accurate data retrieval Fig. 1.
This evolution could not be realized without technolog-
ical advances in optical sources/ampliers, passive com-
ponents, modulators, and detectors. The presentation in
this article follows the assumption that the reader has a
general understanding of these technologies (please re-
view for fundamentals of optics and device physics aspects
of photonics the books by Saleh and Tiech [1] and Bhatta-
charya [2] respectively) and discussions in here are pri-
marily on subsystem-level implementations of analog
beroptic links for distribution of data and local oscilla-
tor, RF signal processing, analog-to-digital conversion,
and medical imaging.
2. LIGHT INTERACTION WITH DEVICES AND CIRCUITS
From the early beginning of optical control of TRAPPAT
(trapped plasma avalanche triggered transit) and IMPATT
(impact avalanche transit time) in 1970, optical control of
semiconductor devices has provided the promise of remote
control along with isolation of optical source other electrical
sources. Over time a number of techniques have been
developed to control performance of a number of devices
and circuits using optical interactions. Various optically
controlled microwave subsystems have been demonstrated
M
Continued
Floor M
Floor 1
MMW
up/down
conversion
MMW
up/down
conversion
MMW
up/down
conversion
MMW
up/down
conversion
MMW
up/down
conversion
MMW
up/down
conversion
Optical
Tx/Rx
Optical
Tx/Rx
Optical
Tx/Rx
Optical
Tx/Rx
Optical
Tx/Rx
Optical
Tx/Rx
NxN Star
Coupler
Building 2
Building 3
Building 1
Figure 1. Conceptual diagram of hybrid beroptic and wireless local-area distributed network
system ( optical ber distribution; electrical distribution; free-space distribution).
Previous Page
since the early 1980s, such as optically tuned antennas [3],
phase shifters [4], oscillators and lters [5], switches [6],
and ampliers [7]. Although these techniques are intrigu-
ing, they have not yet found significant applications. More
recently, spurred by rapid developments in beroptic-based
networks (MAN and LAN), the chip-level integration of
photonic and microwave components for high-performance
optical receivers have attracted attention. PIN-amplier
congurations, such as PIN-HEMT [8], have been reported,
and more recently high-performance PIN-HBT combina-
tions have been realized by Gutierrez-Aitken et al. [9].
The three-terminal microwave devices are preferred to
perform photodetection and control functions in the receiv-
er front end [10,11]. This conguration enhances receiver
performance by reducing parasitics, requires less pre-
amplication due to intrinsic gain of transistors, and has
lower power consumption and less costly fabrication.
Key to these developments and related applications in
communications and control of microwave systems is un-
derstanding of the optical properties of microwave devices
[12] in transistors. In particular, photoresponse of HBT,
MESFET, and HEMTcould be unied and a comparison is
made in terms of inherent photodetection mechanism,
gain, sensitivity, and bandwidth.
2.1. Static Analysis
Table 1 summarizes various effects that contribute to the
photoresponse of the HBT, MESFET, HEMT, and compar-
ison of these devices to the PIN photodiode. (The perfor-
mance of the PIN photodiode is considered as the
baseline.) For the PIN photodiode, the photoresponse cur-
rent is determined by the photogenerated electronhole
pairs in the intrinsic region and no current gain is expe-
rienced. The light level directly controls the photocurrent
in a linear fashion. On the other hand, in HBT, in addition
to photogenerated electrons at the collector depletion re-
gion, there is an increase in the effective base current due
to the drift of photogenerated holes from the collector de-
pletion region to the base. This source of this term is a
change in the current injection rate due to change in base
voltage and hence it is termed the internal photoconduc-
tive effect; since this increase occurs under constant bias
current, it results in current gain. The photoresponse of
the HBT is linear with optical power. On the other hand,
the photoresponse in both eld-effect transistors (ME-
SFET and HEMT) is attributed to three mechanisms
[13]. The photogenerated carries collected at the gate yield
a photovoltage V
phx
, when passing through an external
resistor. Thus the external photovoltaic increases in gate
bias, which opens the channel and results in a photocur-
rent of I
pvx
=g
m
V
phx
. The internal photovoltaic effect re-
sults in I
pvi
=g
m
V
ph
, where for MESFET [14] V
ph
is a
light-induced modulation of the channel height and in the
case of the HEMT, V
ph
represents a shift in the quasi-Fe-
rmi level [15]. The photoconductive effects I
pc
are very
small and are neglected. The optical responsivity of mi-
crowave semiconductor transistors is compared in Fig. 2.
At low illumination the logarithmic response of FET de-
vices provides large gain. However, photoresponse satu-
rates rapidly, which limits their dynamic range. The HBT,
which has relatively large dark current but low noise [16],
performs best at moderate and high illuminations.
2.2. Dynamic Response
The frequency response of a PIN is limited by the carrier
transit time, where a shorter intrinsic region yields a fast-
er response, but this behavior is at the expense of quantum
efciency for vertical devices. (Traveling-wave photodiodes
are designed to simultaneously meet high power-handling,
speed, and efciency requirements.) The speed of the HBT
is governed by sum of different time constants associated
with the charging time of the baseemitter junction, the
base transmit time, the charging time of the basecollector
junction, and the transit time across the collector depletion
region. The optical response of the HBT is fast since these
time constants are the same as those that determine the
HBT ultra-high-speed response, whereas for FET devices
the external photovoltaic effect is very slow because
of the long charging time of the gate external circuit.
The MESFETs dynamic response is dominated by the
Table 1. Sources of Photodetection Mechanisms in Popular Microwave Semiconductor Transistors and Comparison to PIN
Photodiode as Baseline
I
pvi
I
pvx
I
pc
I
pci
I
pd
Gain Speed Gain Speed Gain Speed Gain Speed Gain Speed
PIN None Very fast
HBT Moderate (linear) Fast None Very fast
FET Large (log) Slow Large (log) Very slow Small Fast
HEMT Large (log) Slow Large (log) Very slow Small Fast

HEMT
HBT
PIN
MESFET
1E04 1E03 1E02 1E01 1E+00 1E+01
100000
10000
1000
100
10
1
0.1
P
h
o
t
o

r
e
s
p
o
n
s
i
v
i
t
y
Optical power (mW)
Figure 2. Comparison of the static light responsivity for micro-
wave transistors and performance comparison to PIN photodiode
as a baseline.
2852 MICROWAVE PHOTONICS: TECHNOLOGICAL EVOLUTION AND ITS APPLICATIONS
dynamics of the internal photovoltaic effect, which is de-
ned by the RC time constant of substrate resistance and
the epitaxiallayersubstrate junction capacitance [16]. For
the HEMT, on the other hand, the RC time constant,
determined by the buffer resistance and the change in
electron concentration in the 2DEG (two-dimensional elec-
tron gas) channel [14], denes the speed. It is important to
note that the speed of the photoresponse of the FETs is
independent of their microwave speed.
The measured and calculated frequency response for
these devices is depicted in Fig. 3. At low frequencies FETs
perform well, but the intrinsically slow photovoltaic effect
yields a small gainbandwidth product. The speeds of the
HBT and the PIN are comparable. Note that increasing
the coupling of efciency of the HBT from 1% to 10%,
which is a feasible task, will reduce the link insertion loss
by 20 dB, thus superseding that of the PIN. Using devices
with larger b (e.g., 250 instead of 25) will also give addi-
tional advantage to the HBT.
A publication by Madjedi et al. on the nonlinear behav-
ior of high-temperature superconducting lm as photo-
mixer has provided an opportunity to generate terahertz
signals using the kinetic inductive photoresponse [17].
These new applications maintain interest in interactions
between light and microwave circuits.
3. FIBEROPTIC LINK DEVELOPMENT
Fiberoptic (FO) links are employed for distribution of fre-
quency reference as well as data communications in dis-
tributed systems, where these links need to provide RF
signals with high dynamic range and low phase noise de-
gradation. The information is distributed by intensity
modulation of light using an either directly or externally
modulated laser (see Fig. 4) and is detected primarily di-
rectly since the evolution of optical ampliers. The FO link
gain and noise performance will impact signal-to-noise
performance, while the nonlinearity of various elements
in the optical system will contribute to a limited dynamic
range. The nonlinear phase and amplitude variation with
input RF power will result in AM-AM and AM-PM con-
version. Moreover, in dense wavelength-division multi-
plexed systems, where channel capacity increases by
multiplexing various colors of light, stimulated Raman
scattering in optical ber could introduce distortion and
channel interference.
First, analytical expressions are presented for gain,
noise gure, and dynamic range, and sources of phase
noise degradation in directly and externally modulated
FO links are presented. Next, ber nonlinearity is intro-
duced in a dense WDMsystem. Finally, the performance of
a multifunction circuit, realized on the basis of a long
FabryPerot laser structure that is monolithically inte-
grated with an electroabsorption modulator, where ef-
cient transmission of data and carrier signals with high
SFDR (spurious-free dynamic range) and low phase noise
degradation are achieved, is reviewed.
3.1. Directly Modulated FO Link
The gain of a beroptic link can be calculated in terms of
microwave scattering parameters using the signal ow
diagram (SFD) technique [18]. The transducer gains of the
optical transmitter and optical receiver are derived sepa-
rately and then combined to yield the gain of the complete
link. The link gain is expressed as
G=
P
out;Tx
P
av;Rx
=G
TX
[H
L
[
2
G
RX
=
[S
21D
[
2
[S
21L
[
2
(1 [G
las
[
2
)(Z
L
K
L
LK
D
Z
D
)
2
[1 G
las
S
22L
[
2
[1 G
SD
S
11D
[
2
When a directly modulated semiconductor laser diode is
employed in the optical transmitter, the SFD is obtained
by considering the forward-bias junction resistance of the
laser diode to be the port 2 termination of a two-port net-
work consisting of the microwave impedance-matching
circuit and driving circuit combined with the other device
parameters of the laser. Whereas a reverse-biased p-i-n
MESFET
&
HEMT
PIN
HBT
HEMT
Photoconductive response
cut-off
Electric characteristics - 1/t
ec
1 10 100 1000 10000 100000
Frequency (MHz)
0
10
20
30
40
50
60
70
80
90
L
i
n
k

G
a
i
n

(
d
B
)
Figure 3. Frequency response of microwave devices. Measured
data are represented by discrete points and theory by solid lines.
The optical coupling efciencies were calculated to be less than
1% for the HBT, 4% for the MESFETand HEMT, and 60% for the
PIN diode.
Optical link
RF / Optical
modulator
Optical
source
RF
input
RF
c c c
RF
RF
output
Photo-
Detector
opt
Intensity
modulation
Direct
detection
Optical fiber
(S N)
out (S N)
in
S
in,a
S
out
Figure 4. Overall structure of a beroptic link.
Note that for the directly modulated system the
optical source is internally modulated whereas in
the externally modulated link the electrooptic
property is used to perform intensity modulation
in a MachZehnder modulator.
MICROWAVE PHOTONICS: TECHNOLOGICAL EVOLUTION AND ITS APPLICATIONS 2853
photodiode is employed in the optical receiver, the SFD is
obtained by considering the junction resistance of the di-
ode to be the port 1 terminating load to a two-port network
consisting of the microwave impedance-matching circuit
and the other device parameters of the detector. The link
current transfer function H
L
is dened as the ratio of de-
tector current to RF current across the laser. This is a
measurable quantity that is a function of the electrooptic
device quantum efciencies as well as optical attenuation
and coupling efciency, where ZL is the laser diode exter-
nal quantum efciency, ZD is the detector responsivity, L is
the optical attenuation in the ber, and K
L
, K
D
are the
laser-to-ber and ber-to-detector coupling efciencies, re-
spectively. Nonlinear ber performance due to stimulated
Brillouin and Raman scattering processes for long-ber-
length L is ignored for the moment, but this issue will be
discussed later.
The four contributions to the noise power of the directly
modulated beroptic link. The total noise power at the
output of the detector is the sum of all these individual
noise powers:
N
out
=N
RIN
N
shot
N
th;Tx
N
th;Rx
=N
RIN
N
shot
N
th
The dominant term is the laser RIN (relative intensity
noise) noise power, which is expressed as
N
RIN
=RIN(f )(I
b
I
th
)
2
(Z
L
K
L
LK
D
Z
D
)
2
[S
21D
[
2
[(1 G
SD
S
11D
)[
2
BZ
0
The next dominant noise source is shot noise of the detec-
tor, including dark current noise
N
shot
=2e[(I
b
I
th
)(Z
L
K
L
LK
D
Z
D
) I
d
]
B
[S
21D
[
2
[(1 G
SD
S
11D
)[
2
Z
0
followed with the thermal noise of the transmitter:
N
th;Tx
=4kT
a
B(Z
L
K
L
LK
D
Z
D
)
2
Re{Y
th1
]
[S
21D
[
2
[(1 G
SD
S
11D
)[
2
Z
0
Finally we obtain the thermal noise of the detector:
N
th;Rx
=2kT
a
B(1 [G
in
[
2
)
The only noise source at the input of the link is the ther-
mal noise in the transmitter circuitry: N
in
=kT
a
B. The
noise gure of the beroptic link is dened as
NF
link
=
(SNR)
i
(SNR)
o
=
P
in
+ N
out
P
out
+ N
in
=
1
G
link
+
N
out
N
in
(1)
3.2. Externally Modulated FO Link
The small-signal gain of an externally modulated berop-
tic link is derived using the SFD technique as was applied
to direct modulation. When a MachZehnder interfero-
metric modulator is employed in the optical transmitter to
impress a microwave signal on the optical carrier in a sin-
gle-mode ber, the transmitter SFD is obtained by consid-
ering the capacitance C
M
across the modulator terminals
to be the port 2 termination of a two-port network con-
sisting of the microwave impedance-matching circuit and
other device parameters in the equivalent circuit of the
modulator. The output power of a beroptic link depends
on the amplitude of photocurrent I
det
generated in the de-
tector, which is in turn proportional to the RF voltage V
M
across the capacitor C
M
. This gain is represented as
G=
pLK
D
Z
D
P
in;op
Z
0
2V
p
_ _
2
[S
21M
[
2
[S
21D
[
2
[1 G
M
[
2
[1 S
22m
G
M
[
2
[1 S
11D
G
SD
[
2
where V
p
is bias voltage required for 100% optical modula-
tion and P
in,op
is the optical output of the modulator. Figure
3 compares the achievable gain of directly and externally
modulated beroptic links. Note that a higher gain is
achieved when optical source and detector with efcient
light coupling and responsivity are employed. Finally, gain
of the externally modulated FO link monotonically increas-
es as the optical power squared. Four noise sources con-
tribute to the output noise power of the link. The dominant
term in the case of large input optical power is the shot
noise followed by excess RIN noise of the laser. The shot
noise of the detector, including dark current, is expressed as
N
shot
=2e
Z
op
Z
D
P
in;op
2
_ _
I
d
_ _
B
[S
21D
[
2
[1 G
SD
S
11D
[
2
Z
0
The excess noise of the laser is related to optical source RIN
noise, which could be significantly lower than that for the
semiconductor laser diode:
N
excess
= RIN
Z
op
Z
D
P
in;op
2
_ _
2e
_ _
Z
op
Z
D
P
in;op
2
_ _
B
[S
21D
[
2
[1 G
SD
S
11D
[
2
Z
0
The thermal noise sources of the transmitter and of the
detector are presented respectively as
N
th;Tx
=kT
a
B
pZ
op
Z
D
P
in;op
Z
0
2V
p
_ _
2
[S
21m
[
2
[S
21D
[
2
[1G
M
[
2
[1 G
M
S
22m
[
2
[1 G
SD
S
11D
[
2
N
th;Rx
=2kT
a
B(1 [G
in
[
2
)
The total noise power at the output of the detector is
the sum of all these individual noise powers. As in the
direct-modulation case, the noise power at the input to
the external modulation link is simply kT
a
B. Therefore the
noise gure is given by Eq. (1). Note that the total shot
noise increases as optical power; therefore the signal-to-
noise ratio increases as optical power increases in the
externally modulated links. However, the challenge is to
develop high-speed high-power-handling-capability photo-
detectors.
Figure 5 depicts comparison measured and analytical
calculated gain results of directly and externally modu-
lated beroptic links. The optical power can be high in
externally modulated beroptic links when a solid-state
laser is employed as an optical source. A low-V
p
Mach
Zehnder (MZ) modulator and operation in the quadrature
point of V
b
=V
p
/2 (i.e., f=901) along with a high-power-
handling-capability photodiode allows for the highest re-
ported gain for FO links. Naturally a higher responsivity
(i.e., Z
L
and Z
D
) laser diode and photodiode result in a
lower insertion loss.
3.3. Dynamic Range
Spurious-free and compression dynamic range are directly
related to linearity of the optical modulator (i.e., laser
diode in the case of directly, and MZ modulator in the
externally modulated FO links). The third-order intercept
point and 1dB compression point for the directly modu-
lated are calculated [18] on the basis of the optical
modulation index m:
P
in;int
=
m
2
int
(I
b
I
th
)
2
[1 S
22L
G
Las
[
2
Z
0
[S
21L
[
2
(1 [G
Las
[
2
)
P
in;1 dBCP
=
m
2
1dBCP
(I
b
I
th
)
2
[1 S
22L
G
Las
[
2
Z
0
[S
21L
[
2
(1 [G
Las
[
2
)
In a similar fashion, the third-order intercept point and
1 dB compression point are derived for externally modu-
lated beroptic links as follows:
P
in;int
=
8V
2
p
[1 S
22m
G
M
[
2
p
2
Z
0
[S
21m
[
2
[1 G
M
[
2
P
in;1 dBCP
=
0:950454
2
V
2
p
[1 S
22m
G
M
[
2
p
2
Z
0
[S
21m
[
2
[1 G
M
[
2
Using the derived relationship for intercept and compres-
sion points, spurious-free and compression dynamic range
are calculated using the following expressions:
SFDR=
2
3
10 log
10
P
out;int
kT
a
GNF
_ _
dB Hz
2=3
CFDR=10 log
10
P
out;1dBCP
kT
a
GNF
_ _
dB Hz
3.4. Fiber Nonlinearity
WDM optical systems are employed in radio-over-ber
and optically controlled phased-array antenna architec-
tures. In pushing these systems to the limits of transmis-
sion capability, aspects such as ber nonlinearities need to
be understood. These aspects include the nonlinear ber
phenomena of SRS (stimulated Raman scattering) [20],
SBS (stimulated Brillouin scattering) [21], and XPM
(cross-phase modulation) [22]. These nonlinear effects be-
come noticeable particularly in WDM subcarrier-multi-
plexed (SCM) systems that may cover many closely packed
video channels cover long distances. These nonlinearities
result in optical power transfer to higher and lower optical
frequencies. Each nonlinear effect creates a distortion lev-
el that becomes intolerable above the acceptable threshold
level. Threshold requirements are based on electrical non-
linear distortion requirements, composite second-order
(CSO) and composite triple (or third-order) beat (CTB)
(particularly CTB), which have been determined in pub-
lished works such as Refs. 23 and 24.
Figure 6 shows the SRS crosstalk level for a two-chan-
nel WDM system, while the ber length is varied and for
Fiber length (km)
S
R
S

c
r
o
s
s
t
a
l
k

(
d
B
c
)
Model 1
Model 1
Measured
2.7 ps/nm/km
17 ps/nm/km
55
60
65
70
75
80
85
0 20 40 60 80 100
Figure 6. Measured and simulated SRS-induced crosstalk ver-
sus ber length for an input optical power level of 4dBm, ber
dispersion of 17ps mm
1
km
1
RF frequency of 151.85MHz, and a
channel spacing of 9.4 nm.

L
i
n
k

R
F

g
a
i
n

(
d
B
)
Average detector current (mA)
Gain predicted
By model
Measured gain
Externally modulated
solid state laser
(V
= 0.65 V, [ = 90)
Directly modulated high-
slope-efficiency laser
Directly modulated low-
slope-efficiency laser
0.001 0.1 1 10 100 0.01
40
30
20
10
0
10
20
30
40
Figure 5. Comparison of experimental and simulation results of
directly and externally modulated beroptic links at HF [17].
(Courtesy of Ed Ackerman of Photonics Inc.)
9.4 nm channel spacing, and input power of 4 dBm, a dis-
persion of 17 ps nm
1
km
1
, and a subcarrier frequency of
152MHz. The two curves in the gure represent the high-
and low-dispersion bers. The more dispersive ber gen-
erates lower XT due to the walkoff effect. As is expected,
the crosstalk level begins to increase as the ber length
increases. However, the XT level reaches a peak value and
then decreases as the length is increased further. As the
length is increased beyond 70km, eventually the SRS lev-
el reaches a steady-state value. This behavior shows a si-
nusoidal dependence of crosstalk on ber length for
dispersive ber but a monotonic increase for the disper-
sion-shifted ber. The importance of this result is that,
due to the walkoff effect, the crosstalk level for a length of
ber may not increase as the ber length increases.
Hence, the crosstalk level can be reduced or increased de-
pending on the ber length. In comparison of the mea-
surement and simulation (Model 1: Wang et al. model [24]
and Model 2: PhillipsOtt model [25]), the maximum error
is less than 5%. Measurement results of the SRS crosstalk
while increasing the RF frequency from 50 to 725MHz in
steps of 50 MHz is depicted in Fig. 7. Also included in the
gure is a simulation of PhillipsOtt model [25]. As de-
picted in Fig. 7, the SRS crosstalk level changes as the
frequency is increased, following a behavior similar to a
sinc function (i.e., sin x/x) squared fashion.
3.5. Phase Noise Degradation
The FO distribution link contributes residual phase noise
to the reference signal, which is a function of operation
frequency. This impact is significant mostly in the directly
modulated beroptic links where the RIN of the optical
source is far stronger than the externally modulated FO
links. Moreover, directly modulated beroptic links are
preferable to externally modulated links because of their
lower cost than. The phase noise of the reference signal
could be degraded if residual phase noise were too close to
the signal noise oor level. Therefore, an appropriate se-
lection of reference frequency is necessary to avoid signif-
icant degradation after passing through the FO link. For
example, as shown in Fig. 8, to generate a 12-GHz local
oscillator (LO) at front end, a reference signal at frequency
of 100MHz (UHF), 4 GHz (C band), and 12 GHz (X band)
can be sent through FO link. Since the phase noise con-
tributions for FO links are different at these frequencies,
an optimum frequency for reference signal can be found to
have the least phase noise degradation due to the FO link.
53
56
59
62
65
68
71
74
77
80
S
R
S

c
r
o
s
s
t
a
l
k

(
d
B
c
)
Frequency (MHz)
0 50 100 150 200 250 300 350 400 450 500 550 600 650 700 750
Model 2 (20 km)
Measured (20 km)
Model 1 (20 km)
Model 2 (40 km)
Figure 7. Measured and simulated SRS crosstalk
versus RF frequency with a ber length of 20km, -
ber dispersion of 17ps nm
1
km
1
, channel spacing
of 9.4nm, and a 4dBm (2.5 mW) input power per
channel; also, a simulation of 40km using only the
PhillipsOtt model [25].
Reference
Fiber optic link
Tx Rx
100 MHz
UHF
X40
Microwave or
MMW generation
Reference frequency
alternatives
Reference frequency
generation
Microwave or
MMW generation
Microwave or
MMW generation
4 GHz
C-Band
12 GHz
X-Band
Figure 8. Distribution of frequency reference and
generation of MMW signal using various frequency
references used in a FO-based local-oscillator (LO) syn-
chronization network.
The optical spectra of a modulate optical signal are ex-
pressed as P
opt
{1mcos(o
m
t df
m
(t)) n
RIN
(t)} cos (o
opt
t
f
opt
(t) df
opt
(t)), where P
opt
is the averaged optical
power and m is the optical modulation index at modulat-
ing microwave carrier o
m
. The focus of present work is
df
m
, the residual phase noise added to the microwave
carrier from the laser diode noise source. n
RIN
is the rel-
ative intensity noise; o
opt
is the optical frequency; f
opt
is the optical phase signal due to side modes and modu-
lation, and df
opt
is the phase noise of the optical signal.
Since most beroptic links for antenna remoting appli-
cations use intensity detection, only the noise signals in
optical intensity affect the microwave carrier signal,
namely, df
m
and n
RIN
. The n
RIN
could contribute to the
FM noise of the reference signal through nonlinear AM/
PM conversion [27].
The laser diode SSB phase noise of the nth harmonic of
the modulating signal '
out
has contributions from three
noise terms: (1) the input reference signal phase noise,
'
in
, (2) the low-frequency noise of the laser diode upcon-
verted to the carrier frequency '
up
; and (3) the RIN noise
at the offset microwave carrier '
RIN
. This behavior is
quite analogous to microwave systems [28]. Therefore at
angular offset carrier frequency of O, '
out
can be approx-
imately expressed as [29]:
'
out;no
(O) =n
2
'
in;o
(O) n
2
'
up;o
(O) '
RIN;no
(O) (2)
The factor of n is the harmonic order of the modulation
signal, if any nonlinearity of the laser diode is exploited to
generate the nth harmonic [28]. (If the fundamental fre-
quency is employed, then n=1.) The subscript o indicates
the modulation frequency. The upconversion factor of low-
frequency RIN to phase noise is the dominant noise
source. Calculation of this upconversion factor depends
on the derivative of the RF phase with respect to the RF
derive amplitude and is C
up;o
=
1
2
(@y(o)=@P
o
)
2
P
2
o
, which y is
phase of optical signal at the modulating frequency o. The
dependence of phase on the optical output power in a di-
rectly modulated system is a bit more complicated than in
an externally modulated system since the nonlinear be-
havior is dependent on modulation index of laser diode
and the relationship of its operation frequency compared
to the relaxation oscillation frequency. This process is
nonlinear and at certain frequencies results in AM-PM
compression. The results are related to modulation index
through the a parameter, which is a function of modula-
tion frequency and averaged optical power [18].
Since RIN noise in semiconductor laser diodes is strong
up to 100MHz because of mode partition noise, it is pre-
dicted that the spectral purity of the UHF reference signal
is greatly degraded, resulting in a higher FM noise. More-
over, the X-band modulating signal is close to the relax-
ation oscillation frequency where RIN is peaked. The best
frequency for reference signal distribution through DMFO
link is the C-band signal as depicted in Fig. 9, where the
phase noise of the 12-GHz LO signal is generated from the
reference signal through the abovementioned DMFO link.
Clearly, the signal generated from a C-band signal has the
best phase noise performance. The signal from UHF ref-
erence degrades greatly because the residual phase noise
of the FO link is higher than the reference phase noise at
offset frequency higher than 100 Hz.
3.6. Monolithically Integrated Mode-Locked Laser
A monolithic version of a laser with external cavity is re-
alizable using semiconductor fabrication process and re-
ported by a number of researchers. Figure 10a shows a
schematic drawing of the monolithic laser with an inte-
grated EA (electroabsorption) modulator. Stacked struc-
ture consisting of two MQW layers, a MQW
(multiquantum well) for laser diode (MQW-LD) and a
MQW for EA modulator (MQW-MD) are employed. The
FabryPerot (FP) cavity length for our experiment is
cleaved approximately for a length of 2170mm. This total
length is composed of a 1970-mm-long gain section, a 150-
mm-long modulator, and a 50-mm-long separation region.
The facet of the modulator section is coated with high-re-
ectivity lm (RE85%). The facet of the gain section is as
cleaved. The laser is mounted in a high-frequency pack-
age. The schematic diagram of the long FP laser with in-
tegrated EA modulator is shown in Fig. 10b.
10
6
10
5
10
4
10
3
10
2
10
1
10
0
10
1
180
160
140
120
100
80
60
40
20
From 4 GHz ref.
From 12 GHz ref.
From 100 MHz ref.
Offset frequency (Hz)
FO Noise@12 GHz
FO Noise@4 GHz
FO Noise@100 MHz
P
h
a
s
e

n
o
i
s
e

(
d
B
c
/
H
z
) Ref. @100 MHz
Ref. @4 GHz
Ref. @12 GHz
Figure 9. The simulated phase noise of a LO signal at
12 GHz, is generated from different reference signals
through a directly modulated FO link as depicted in Fig. 8.
The gain section of the laser diode is forward-biased at
different bias currents, and the EA section is reverse-
biased by different voltage levels. A resonance peak is
observed that is associated with the longitudinal mode
separation in the long FP laser. The longitudinal-mode
separation is calculated as Df =c/2nLE19.3 GHz, where c
=300 mm GHz is the speed of light in free space, nE3.5
is the index of refraction of the waveguide, and
LE2.17mm is the FP cavity length. This resonant fre-
quency has a frequency tuning sensitivity of E1 MHz/mA.
The optical oscillations are stabilized using an injection-
locking process [27], where a single oscillation peak that
appears as the gain section is modulated by a frequency
reference of P
m
Z1 dBm at f
m
=19.258 GHz. The familiar
one-sided injection-locking spectra are observed outside
the injection-locking range and the close-in to carrier
phase noise is significantly reduced within the locking
range. The measured close-in to carrier phase noise de-
gradation at 100Hz offset carrier is depicted in Fig. 11,
where 31 and 6dB degradation are measured for the
injected power of P
m
= 0.5dBm in the resistively and
reactively-matched modules respectively. However for
injected power level of 4.5 dBm, a close-in to carrier
phase noise identical to the reference source is measured
for the reactively matched case Fig. 11.
The resonance frequency could also be stabilized using
fundamental mode locking by modulating the EA section
by a synthesized source. The close-in to carrier phase
noise are measured and comparison is made against the
reference signal from an HP83640A source. The results
are summarized in Table 2 for different laser operation
points. As indicated, a very small phase noise degradation
is observed; however, the results for V
m
= 1Vare better
Frequency reference
Cross over point
> 0 dBm
+70
+60
+50
+40
+30
+20
+10
0
0
20
20 15 5 5 10 15 20
10
10
30
Reactively-Matched
Resistively-Matched
Cross Over Point
> +12 dBm
Input power (dBm)
P
h
a
s
e

n
o
i
s
e

d
e
g
r
a
d
a
t
i
o
n
@
1
0
0

H
z

(
d
B
)
Figure 11. Measured FM noise degradation at
100-Hz-offset carrier for the injection locked inter-
modal oscillation. The measured injection-locking
power of the long FP laser diode is also depicted at
the crossover point for the resistively ( ) and re-
actively matched ( ) lasers.
Gain
section
EA modulator
Data signal
150m 50m
Gain section
1970m
Reference signal
(Mode-Locked)
Reference signal
(Injection-Locked)
EA
Modulator
section
HR
p-InGaAsP
MQW-LD
MQW-MD
InP InP
n-InP
P-InP
Isolation
section
(a) (b)
Bias current
Figure 10. (a) Conceptual representation of the long FP laser integrated with an electroabsorp-
tion modulator; (b) mechanical xture used for mounting of the laser diode for injection-locking,
mode-locking, and optoelectronic conversion evaluations.
Table 2. Phase Noise Degradation of LO Signal as
Function of Various Offset Frequencies and Laser
Operation Points
L(O)(dBc=Hz) D(dB)
f
m
(GHz) V
m
(V) O=100Hz O=1kHz O=100Hz O=1 kHz
19.13400 4.5 75.4 81 2.6 2.4
19.29469 2.5 77.0 82 1.0 1.1
19.29406 1.0 77.6 82 0.4 0.4
than the results for V
m
=4.5V for the same laser current
of I
b
=140mA.
Since this stabilized signal has much cleaner close-in to
carrier phase noise than does the free-running oscillation,
it could be employed as the LO signal. The laser diodes
gain section is forward-biased (I
b
=140 mA), and the EA
modulator section is reverse-biased (V
m
= 1 V). This op-
erating point is selected because of the efcient mode-lock-
ing process while maintaining the least amount of phase
noise degradation of the carrier signal. The laser diode
optical output is collimated to a single-mode optical ber
using a polarizing collimator with an overall ber coupling
efciency of 12%. The optical ber output is connected to
an optical receiver integrated with RF analyzer, which
automatically dis-embeds the calibrated optical receiver
response from the measurement and displays it in the op-
tical domain. The electrical domain power levels, in dBr,
are easily calculated by multiplying the depicted optical
domain results, displayed in dBm, by a factor of 2.
Next the gain section of this laser is modulated by
S-band signals (2.2 GHz750 MHz). Strong nonlinearity
of the mode-locked laser at the LO signal of 19.3 GHz
upconverts the S-band signals to 17.1 and 22.5GHz as
shown in Fig. 12. The data modulation power level is
changed over a wide range. An optical conversion loss is
dened as the ratio of the generated mixed RF signal
(19.372.2GHz) to the IF signal (2.2 GHz). The optical
conversion loss is as low as 1.4 dB, resulting in an electri-
cal conversion loss of 2.8dB. The optoelectronic conversion
loss for the lower sideband (LSB) at 17.1GHz is higher
than the upper sideband (USB) of 22.5 GHz by 1.3dB
(i.e., 2.6dB electrical) [31].
On the other hand, a modulation loss greater than
51 dB is measured when the gain section is directly
modulated by the RF signal at 17.1GHz. The spurious-
free dynamic range (SFDR) of this optoelectronic
mixer is also evaluated. The intermodulation distortion
(IMD) measurements are conducted for two modulating
tones that are 5 MHz apart (e.g., f
1
=2.200 GHz and
f
2
=2.205 GHz). Both tones are upconverted by a stable
LO signal of 19.360GHz and IMDs of the upconverted
RF signals are measured at LSB and USB frequencies.
Based on the mode-locked laser IMD and RIN noise mea-
surement results for the upconverted RF tones, SFDR
LSB and USB RF signals are E88 dB
.
Hz
2/3
and
89 dB
.
Hz
2/3
, respectively.
4. FIBEROPTIC DISTRIBUTION FOR PHASED-ARRAY
ANTENNAS
One of the simplest methods antenna remoting is based on
the concept of direct replacement of the electrical inter-
connects by FO links. However, there are challenges as-
sociated with reliability of optical components, cost of
system integration, and the architecture employed for
achieving the best attributes possible.
4.1. Device Innovations and Reliability
A vast number of research work reported in the literature
have focused on the device improvements to meet the per-
formance requirement of commercial beroptic communi-
cation. Performance of sampled directly and externally
modulated beroptic links operating at S band is rendered
in Table 3 (circa 1995), where the best performance is
achieved for DFB laser monolithically integrated with the
EA modulator. The performance of the mode-locked laser
is quite acceptable for many telecommunication systems.
(The results of the MZ modulator may appear worse than
what is reported in the literature, but in this case a semi-
conductor laser with optical power of only a few mW is
used as a source.) Monolithically integrated EA modulator
with sampled grating DBR laser (SGDBR) [32] has been
developed where SFDR of 120 dB
.
Hz
2/3
is achieved over
a large tuning bandwidth. However, the harsh military
and space environment imposes additional burdens on the
performance of lightwave technology components.
Frequency (GHz)
N
o
r
m
a
l
i
z
e
d

l
i
g
h
t

o
u
t
p
u
t

p
o
w
e
r

(
d
B
)
5
40 10 20 30 50
5
15
20
25
30
35
fs
fm2fs fm+2fs
fmfs
fm+fs
2fmfs
2fm+fs
fm
2fm
(b) (a)
RL 0.00 dBm MKR #1 FRQ 2.20 GHz
ATTEN 10 dB LW OPT
5.00 dB/DIV
AVG PWR 1.4 dBm OPT
MARKER
2.20 GHz
19.82 dBm
19.82 dBm
1
Carrier
RF signal
IF signal
CENTER 11.00 GHZ
RB 3.00 MHz VB 10.0 KHz
SPAN 20.00 GHz
ST 2.000 sec
Figure 12. Optoelectronic mixing of data signal with local-oscillator signal in the mode-locked
long FP laser diode: (a) experimental results for operation conditions I
b
=140mA, V
m
= 1 V, f
m
=
19.360GHz, P
m
= 15dBm, f
data
=2.200GHz, and P
data
= 25dBmdata, lower sideband RF,
and local-oscillator signals (center frequency 11.0 GHz and frequency span 20.0GHz); (b) simula-
tion results based on traveling-wave rate equation [30].
4.2. Packaging Requirements
As indicated in the gain expressions of directly and exter-
nally modulated beroptic links, a dB improvement in the
optical coupling improves insertion gain by 2dB. However,
mechanical tolerances of optical bers and sources are in
submicrometer range, hence making the low-cost integra-
tion of optical components with optical bers challenging
at the least. Moreover, this process has to be done in a cost-
effective manner. Another important aspect of the light
coupling is that reection has to be minimized since any
optical feedback introduces modulation of the dynamic
response, which resembles the transmission characteris-
tics of FP resonators. Therefore, optical isolators combined
with angle polished bers are required to reduce the light
feedback level below50dB in certain applications. Another
important aspect is temperature control of semiconductor
devices to avoid any sensitivity to temperature in harsh
environments of space. Finally, directly modulated ber-
optic links or externally modulated systems using EA
modulators experience input impedance that corresponds
to high-reectivity coefcients (i.e., approximately short
for a laser diode and open for an EA modulator). To avoid
high reection loss, impedance matching circuits are
needed to be developed, which is not easy to accomplish
over a large fractional bandwidth. Figure 13 depicts a
designed structure of monolithically integrated optical
source with an EA modulator [33], which is used for dis-
tribution of both LO signal and data signal. This structure
is also based on cascading a number of lasers with a mono-
lithically long FP cavity in series, hence increasing the
forward PN junction resistance, while maintaining the
same RF current modulating all the gain sections. In es-
sence, since the input impedance of laser diode (i.e., a for-
ward-biased p-n junction) is about 4O, by series
combination of the impedances, a level closer to 50O is
achieved.
1
Moreover, a lower Q
ex
factor is obtained, which
simplies the matching circuit design. Finally, the ber
coupling is achieved cost-effectively by combining a num-
ber of lensed bers mounted on a silicon V groove. This
process will enable packaging of a large number of laser
diode sources.
4.3. Architecture Innovations
In large-aperture phased-array antennas or ROF systems,
the RF signal could be downconverted to the IF signals for
further processing at the centralized receiver. This archi-
tecture, shown in Fig. 14, is the conventional one. The
challenges for implementation of optically controlled
Integrated Optical
feedback cavity
Feedback mirror
Laser
Laser chip
Metal carrier
Laser chip
Metal carrier
Optical module
fixture
Ground
metalization
Reference and
data signal
Silicon V-groove
fiber alignment
platform
Fibers
Bond wires
Fiber lens
To antennas
Microstrip line
Figure 13. Conceptual representation of an optimized optical transmitter using eight series-
mounted laser diodes with monolithically integrated external optical cavity coupled to the lensed
bers using a Si V-groove ber alignment system.
Table 3. Comparison of Various Components Off the Shelf (COTS) Fiberoptic Links at S-Band
a
Directly Modulated FO Links Externally Modulated FO Links
Mode-Locked Laser Ortel DFB Laser DFB/EA Modulator Sumitomo MZ Modulator
Frequency 2.2GHz 2.5 GHz 2.2 GHz 2.5 GHz
Gain(dB) 8 44 12 40
IP3 (dBm) 17 27 14 23
Noise oor (dBm/Hz) 142 151 151 151
SFDR (dB
.
Hz
2/3
) 101 86 103 90
a
Note the MZ modulator is based on DFB laser as optical source.
1
A similar approach is currently pursued in quantum cascade la-
sers, although the latter is a unipolar device. For more informa-
tion, see, for example, the paper by Faist et al. [34].
phased array using this architecture are (1) a high-dy-
namic-range beroptic links are required at ultrahigh fre-
quencies, (2) phase and frequency control must be
maintained in the distribution network all the way to
the central processor, and (3) as will be shown later, a
higher resolution for a true time-delay device is required.
On the other hand, the T/R-level data mixing architecture,
shown in Fig. 15, provides a great opportunity to perform
downconversion of the RF signals to IF and avoid the lim-
itations encountered in the CPU-level data mixing. Addi-
tional requirements are: (1) the need for stabilized LO at
each element to coherently down-or upconvert the re-
ceived RF or IF signals, (2) increase in the number of op-
tical links, and (3) the requirement of phase control in
addition to TTD to obtain a squint-free radiated beam.
Nonetheless, experimental comparison of a 24 MMIC-
based C-band phased-array antenna was conducted,
where a superior dynamic range was measured for T/R-
level data mixing architecture over CPU level 1 [35].
These apparent limitations were avoided using a cascad-
ed ILPLL oscillator [36], a self-oscillating mixer [37], and
an optoelectronic mixer using MLL [38]. The most impor-
tant advantage of T/R-level data mixing is its reduction in
the number of resolution bits required in real-time-delay
devices to generate a squint-free-beam. This issue is high-
lighted next.
Figure 16 depicts radiation pattern of a 25-element lin-
ear phased array (with l/2 separation) designed for oper-
ation at center frequency of 33 GHz with bandwidth of
3 GHz (i.e., each graph is composite of three simulated
graphs at frequencies of 31.5, 33, and 34.5 GHz). The sim-
ulation results are for CPU-level data mixing. The re-
quired time delay is achieved using a switched delay line
TTD (real-time-delay device) with minimum time resolu-
tion of 10 ps. As this simulation result indicates as the
beam is pointed away from broadside, sidelobe levels in-
crease to only 6dBc, and the mainbeam decreases by
2 dB. On the other hand, Fig. 17 depicts the simulated
performance of the same phased array when it is con-
structed on the basis of T/R-level data mixing. This struc-
ture employs a 2p analog phase shifter based on the
concept of cascaded oscillators [36] along with a TTD
with a time resolution of 30 ps (i.e., decreasing the time
delay number of bits by factor of B4). As this gure clearly
indicates, no reduction in mainbeam peak level or increase
in sidelobe levels is observed for any scan angles. In fact,
the sidelobe levels are compatible with the expected the-
oretical level of 13.6 dB for a uniform array.
5. MICROWAVE PHOTONIC SIGNAL PROCESSORS
One of the most significant advantages of microwave pho-
tonics is not the antenna remoting concepts, but rather
the opportunity to perform signal processing in the optical
domain. The primary gure of merit is the timeband-
width product, which could exceed 10
4
, hence leading to
significant rejection and ltering using various delay
lines. A few realizations of signal processors using micro-
wave photonic techniques are discussed next.
5.1. Memory Loop
Fiberoptic-based memory loops are used for recirculation
of the incoming RF pulses. The simplied schematic dia-
Demod.
CPU MODULE T/R MODULE OBFN
Transmit
Oscillator
Receive
Oscillator
Synthesizer
Mod.
SSPA
Phase
Shifter
Optical
Receiver
Optical
Receiver
Optical
Receiver
Optical
Transmitter
Optical
Transmitter
Optical
Transmitter
n-bit
TTD
n-bit
TTD
LNA
Figure 15. T/R-level data mixing architecture
for transmit/receive-mode operation, where dis-
tributed local oscillators need to be synchro-
nized to a frequency reference. Note that both
real-time-delay and phase shifter devices are
required for beam-squint-free operation in
broadband systems.
Demod.
Synthesizer
Mod.
Optical
Transmitter
Optical
Transmitter
Transmit
Oscillator
Receive
Oscillator
Optical
Receiver
Optical
Receiver
n-bit
TTD
n-bit
TTD
SSPA
LNA
CPU MODULE
T/R MODULE OBFN
Figure 14. CPU-level data mixing architecture
for transmit/receive-mode operation. Note that
real-time-delay are required for broadband op-
eration without beam squint.
gram of a beroptic-based recirculating memory loop is
shown conceptually in Fig. 18. This system consists of four
basic elements: a switch, an electrical amplier, a ber-
optic time-delay element, and a gain equalizer. The gain
equalizer is composed of a YIG tunable lter and an at-
tenuator. The RF input pulse is routed through the switch
to the time-delay device. The switch closes the loop and
thus controls the recirculation. As the signal reenters the
microwave circuit, it is amplied and rerouted through
the ber. As a result, a pulsetrain is obtained that has a
pulse repetition interval corresponding to one recircula-
tion time through the loop.
The number or recirculations is not limited by disper-
sion, and for higher recirculation the following steps are
established (1) reducing the insertion loss and noise gure
of the beroptic link and (2) attening the frequency
response of the closed-loop system. Using a 1-km optical
delay line over 24 GHz with an insertion loss of 11 dB
(atness of 4dB) and a spurious-free dynamic range of
87 dB Hz
2/3
, a short electrical pulse has recirculated for as
long as a millisecond. The spectral purity of the recircu-
lated signal is evaluated, and the spectral purity of the
output pulses is as shown in Fig. 19 after 10, 20, and 35
recirculations. The phase noise degradation is measured
for offset carrier frequencies of 10, 50 and 100 Hz.
Since the frequency response of the open loop is not, in
practice, at over the bandwidth, to enhance the perfor-
mance of the memory loop, a gain equalizer is required.
The amplication of the recirculating signal can be real-
ized in either the electrical [39] or the optical domain. For
broadband microwave signal processing, however, where
the incoming signals in the frequency range of 218 GHz
are analyzed, pulse recirculation in the optical domain is
preferable to that in the electrical domain.
The maximum number of recirculations in the loop in
terms of the characteristics of the system components can
then be numerically evaluated as a function of gain at-
ness C. In particular, the maximum number of recircula-
tions n
max
is limited to the maximum number allowable by
NF
Tmax
and the open-loop noise gure NF
B
as follows [39]:
NF
Tmax
-
1
2
(NF
B
1)
1
C
n
1 C
n
1 C
n
_ _
The implication of nonat frequency response of the delay
unit is that the noise will increase at a faster rate at fre-
quencies where open-loop gain is greater than unity.
Therefore, the nonat frequency response and high noise
gure of the delay element will restrict the maximum time
delay attainable by the memory loop. Naturally, to over-
come the n
max
limitation, while achieving long total time
delays of nt, one could increase the unit time delay t, but
the long unit delay will produce a void in the time domain
for the short input pulses.
5.2. Advanced Optical Signal Processing Techniques
The use of passive optical components such as optical
isolators, array waveguide grating, superposed array
20
35
50
65
90 70 50 30 10 10 30 50 70 90
5
12dBc
R
a
d
i
a
t
i
o
n

p
a
t
t
e
r
n

(
d
B
)
Scan angle (degrees)
Figure 17. Simulated radiation pattern of 25radiating-element
linear multibeamphased-array antenna based on a T/R-level data
mixing architecture where a real-time-delay line with 30 ps res-
olution along with analog phase shifter for LO is employed to
generate beams at different angles (f
LO
=24GHz, f
data
=
7.510.5GHz, f
RF
=31.534.5GHz).
RF IN
RF OUT
SPDT
Switch
Amplifier
Coupler
Gain Equalizer
Optical
Receiver
Optical
Transmitter
Delay Unit
Figure 18. Conceptual drawing of a beroptic-based recirculat-
ing delay line. It is composed of a SPDT switch, electronic ampli-
er, coupler, optical delay element, and gain equalizer.
Scan angle (degrees)
5
20
35
50
65
90 70 50 30 10 10 30 50 70 90
2dB
6dBc
R
a
d
i
a
t
i
o
n

p
a
t
t
e
r
n

(
d
B
)
Figure 16. Simulated radiation pattern of 25-radiating-element
linear multibeam phased-array antenna based on a CPU-level
data mixing architecture where a real-time-delay line with 10ps
resolution is employed to generate beams at different angles (f
LO
=24 GHz, f
data
=7.510.5GHz, f
RF
=31.534.5GHz).
grating, and spatial light modulators provides for a num-
ber of signal processing techniques, such as interference
mitigation [4042] and adaptive waveform generation
[43]. The basic principle of these techniques is based on
translating spectrum to time using dispersive bers or
delay lines. As the number of taps increases, increased
resolution in frequency domain can be observed. On the
other hand, using mode-locked pulses with short sampling
periods will increase the time resolution. As the tap
weights and unit time delay are adjusted, an arbitrary
waveform in time domain is generated that corresponds to
the desired transfer function. Figure 20 depicts the struc-
ture of a tunable lter, where a tunable lter with a Q as
high as 800 is demonstrated [44]. Although discrete grat-
ing arrays are simpler for design implementation, but su-
perposed arrays are quite practical for sampling
bandwidth in the terahertz range [44].
Moreover, tapped delay lines are employed in combina-
tion with positive and negative optical amplitude to adjust
transfer function and shape of the transversal lter. High-
birefringence materials combined with a polarizer are em-
ployed to create all-optical transversal lters. Figure 21
depicts the shape of lter impulse response. The desired
impulse response is converted to the desired bandpass l-
ter. Moreover, notch lters could be developed using RF
interference in the optical ber while the other frequen-
cies are transmitted through without much attenuation.
Interference mitigation by 50 dB is experimentally
demonstrated at 75 MHz [44].
6. OPTICAL ADC TECHNIQUES
Another application of microwave photonics is in the de-
velopment of the analog-to-digital converter (ADC). Opto-
electronic devices have demonstrated ultrafast switching
speed, and mode-locked solid-state lasers have achieved
high-speed and accurate optical pulses as low as 10 fs.
With capability of sampling electrical signals of resolu-
tions of subpicoseconds, implementation of optical ADC
has long intrigued many researchers because of the
following advantages: (1) optical sampling has time jitter
that is about two orders of magnitude lower than that of
an electronic clock, (2) optical sampling decouples the
electrical clock signal used for optical sampling from the
sampled electrical signal, and (3) an optical sampled or
quantized signal is easy to distribute by ber and to re-
motely control. Also, many photonic ADC approaches also
produce output as Gray codes directly, eliminating the
need for additional encoding circuits. However optical
Circulator
Passive filter
Active filter
L
1
L
2
L
3
3dB couplers
980 nm
Pump laser
EDF
WDM
Grating Grating L
act
(a)
(b)
40
50
60
A
m
p
l
i
t
u
d
e

(
d
B
)
600 550 650 700 750 800
Frequency (MHz)
Figure 20. Optical signal processing using a tunable lter: (a)
experimental setup; (b) transfer function in terms of modulated
frequency. (Courtesy of Prof. Robert Minasian of University of
Sydney.)
Input signal
10 Recirculation
24 Recirculation
35 Recirculation
10
20
30
40
50
P
h
a
s
e

n
o
i
s
e

d
e
g
r
a
d
a
t
i
o
n

(
d
B
)
10 20 30 40 50 60 70 80 90 100 110
Frequency offset carrier (Hz)
Figure 19. Spectral purity and phase noise de-
gradation compared to the input electrical pulse at
various offset carriers as a function of number of
recirculations.
quantization is generally limited to resolution of a few
bits, and this remains a major challenge. There are two
approaches to quantize the information after sampling by
optical signal: (1) hybrid opticelectronic ADC, where elec-
trical quantizers are employed; and (2) all-optical ADC,
where optical quantization is implemented.
6.1. Hybrid Opticelectronic ADC
Hybrid Opticelectronic ADC, also called optically assis-
tant ADC, employs optical sampling followed by electron-
ic quantization. It attempts to combine advantages of both
optical and electronic ADC technologies: high-speed rate
of optical sampling and high resolution of electronic quan-
tization. But the speed of electronic devices is much lower
than the optical sampling rate. So the sampled optical
signal has to be downconverted to a lower speed and chan-
neled to parallel electronic quantizers. For example,
100Gsps sampled pulses can be split into eight channels
in time domain, and the pulse rate in every channel is
12.5 Gsps. Most optical sampling transducers are imple-
mented with MachZehnder modulators [45], where the
output intensity of the MZ interferometer is a function of
the applied voltage
I
out
=I
i
cos
2
j
0
2
_ _
pV
2V
p
_ _ _ _
where j
0
=2pnL=l
0
, the optical distance of a branch, and
V
p
is the half-wave voltage, dened as the applied voltage
at which the phase shift changes by p. Optical sampl-
ing operates in a small range around V
p
, the output is
approximately linearly proportional to applied voltage V.
To extend the linear range of MZ modulator, a number
of linearization approaches were demonstrated. In pho-
tonic ADC, the linearization can be implemented in
digital domain by directly inverting the transfer function
in DSP.
The key issue in hybrid optoelectronic ADC is to chan-
nelize high-speed sampled optical pulses and ensure that
channels match in both amplitude and time. Based on ap-
proaches of channelization, three major schemes are pro-
posed: (1) time interleaving [46], (2) WDM channelization
[47], and (3) time stretching [48]. The optical sampling
and time interleaving is employed by Juodawlkis et al.
[46] to implement an ADC with bandwidth up to 505Msps
by using 18 optical demultiplexers and 14-bit electrical
quantizers. A dual-output LiNbO
3
MZ modulator is used
for linearization processing and a 65 dB SFDR and 47 dB
SNR, corresponding to an effective resolution of 7.5 bits.
The sampled optical pulses are split into eight channels by
optical time-division demultiplexers, which are composed
of three stages of 1 2 switches controlled by 505-MHz
driving signals. Therefore the 16 parallel high-resolution
electronic quantizers are operating at 63 Msps [47]. To
achieve an interleaving SFDR of 80 dB, the converter-to-
converter gains must be matched to B0.01%, the offset
must be matched to B0.01% of the signal amplitude, and
the converter-to-converter crosstalk must be less than one
part in 10
4
.
Time (ns)
0 5 10 15 20
0
1
set 1 set 2 set N
1
2
M
T
1
2
M
O
p
t
i
c
a
l

p
u
l
s
e

a
m
p
l
i
t
u
d
e
(
a
.
u
.
)
positive
section
negative
section
set 1 set 2 set N
1
2
M
T
1
2
M
(a)
(b)
Spectral Filters
i M
Negative coefficient section
Grating
Power
splitter
Multiple
Wavel ength
Source
Photodetector
Positive coefficient section
Circulator
Output
Photodetector
Time slot
delay
z
1
z
2
z
z
z
z
z z
z
z
t
t
t
Figure 21. Filter impulse response (FIR) us-
ing spectral time mapping: (a) desired signal in
the domain; (b) experimental set-up. (Courtesy
of Prof. Robert Minasian of University of
Sydney.)
On the other hand, using optical dispersive compo-
nents, a multiwavelength optical wave is smeared (for
continuous spectrum) or split (for discrete wavelengths) in
time, and borrowing a concept from WDM communication,
the sampled optical signal can be channelized in both
wavelength and time domains. The RF signal is sampled
by the WDM pulses and then channelized by a WDM
demultiplexer. Clark et al. demonstrated a 100-Gsps pho-
tonic ADC based on the time- and wavelength-interleaved
scheme [47] using a mode-locked ber laser (MLFL) that
generates 12.5 GHz pulsetrain, and a 100-GHz sampling
optical pulse is obtained by 8 multiplexer. The pulses
propagate through different ber delays and attenuators,
which then are recombined in the WDM. The delay bers
and attenuators can be adjusted for time matching and
amplitude equalizing, respectively. The time-interweaved
pulsetrain is then demultiplexed into eight channels ac-
cording to wavelengths. The resulting parallel pulses are
then quantized by 412.5-Gsps quantizers, but an 8-bit
ADC operating at only 781 Msps has been achieved with
SNR 2226 dB, which corresponds to about 4 bits.
The resolution of wavelength-channelized ADC is still
limited since it faces difculties similar to those encoun-
tered by the time interleaving ADC. The time uncertainty
and amplitude uniformity between channels are difcult
to control.
Bhushan et al. demonstrated a record ultrafast sam-
pling rate of 130 Gsps using the time stretching approach
[48]. The idea behind this ADC is similar to wavelength-
channelized ADC, but in the time-stretched approach an
RF input signal modulates a broadband optical continuous
wave (CW) other than that sampled by optical pulses with
discrete wavelengths. The detected analog optical signal is
sampled and quantized by electronic ADC. Figure 22
shows a limited time application of time-stretch prepro-
cessing, where a passively mode-locked ber laser with
20 MHz repetition rate followed by a 17-nm lter is used to
generate broadband short pulses. The optical pulse prop-
agates through a dispersive ber of length L
1
and is dis-
persed in time to yield a time aperture of 0.8ns. The wave
is then modulated by the RF signal to be converted. The
modulated wave is sent to another piece of dispersive ber
of L
2
to be stretched in time domain. The ADC obtains a
stretch ratio of M=16.2 by correct choice of either L
1
or
L
2
, where the stretched signal is detected and digitized by
a single 8-Gsps electronic ADC of an oscilloscope. So the
effective sampling rate is about 130Gsps (8Gsps 16.2)
with SNR B45 dB, corresponding to 7.5 bits of resolution.
On the other hand, for continuous signals in time, a
parallel architecture must be used in order to preserve the
information. In this process an arrayed waveguide grating
(AWG) is employed to sample a portion of the optical spec-
trum, and since each optical wavelength corresponds to a
different propagation time delay, the lter performs sam-
pling in time. Then each segment is time-stretched by the
same factor M prior to entering a slow electronic ADC;
however, both time alignment and amplitude in balance
between various arms of are crucial importance.
6.2. All-Optical ADC
There are quite a few ways to implement optical quanti-
zation, but probably the best-known photonic ADC is
based on a patent from 1977 [49]; a revised design called
optical folding-ash ADC was patented in 1995 [50].
Figure 23 shows a block diagram of a 4-bit optical folding-
ash ADC. The geometric scaling of V
p
or electrode length
is eliminated by a parallelserial combined conguration.
This scheme uses identical electrode length but sets DC
bias at different points on the interferometer transfer
characteristic curve. The resulting transfer function of
MSB-2 and LSB branch is obtained by multiplying the
functions of all stages at different bias, showing doubled
frequency compared to the previous bit. However, this
scheme also presents some additional challenges: (1) the
MSB is susceptible to high noise levels because of
the slowly changing slope at the digital edges, and (2)
the transit-time limitation is still not eliminated; more-
over, the hardware complexity increases exponentially in
terms of interferometer number as 2
(b1)
1 and strongly
relies on accurate bias.
Another way to quantize an optical signal is to exploit
variable electroabsorption semiconductor modulators
demonstrated by Hayduk [51]. The quantization is
achieved using an architecture that relates the received
analog voltage to an optical intensity, which is based on
the same concept as the electronic ash ADC, in which
2
N
1 comparators with different threshold values are
used. The use of passive materials in this ash photonic
ADC architecture with no external voltage requirement
gives this module very low power dissipation. The authors
claim that the ADC has the potential to operate at more
PD
Stretched signal
Digitizer
Input electrical signal
Modulator
Time
stretch
x n
SC
source
Figure 22. Time-limited signal converted by time
stretching ADC [45]. (Courtesy of Prof. B. Jalali
from UCLA.)
than 100Gsps combined with a resolution as high as
1214 bits. But this scheme is susceptible to amplitude
uctuation and unbalanced energy splitting. A similar
idea is also employed in all-optical ADC quantization us-
ing photodetectors with different sensitivities [52].
In the two ADC schemes mentioned above, the RF signal
is sampled and quantized by optical amplitude. However,
the amplitude errors strongly depend on source uctuation,
device linearity, and loss along the optical link. So it is dif-
cult to build a construct high-resolution all-optical ADC
based on optical amplitude. An all-optical ADC operating in
spectrum domain is demonstrated by Zmuda et al. [53], as
shown in Fig. 24. The input signal is sampled by a tunable
laser and quantized by processor lters with binary behav-
ior. The output wavelength of the tunable laser is modulat-
ed by the applied electrical eld, so the electrical amplitude
is represented as wavelength in spectrum domain. The
postsampled light is processed by a parallel optical lter
array. If the spectrum line falls into the passband, the out-
put is represented as 1; otherwise a 0 will be read. Each
lter has periodic equally spaced passband and stopband
and organized in a Gray code manner.
The major challenges for this scheme are the lter de-
sign of sharp transition and tunable laser. A 4-bit optical
ADC using Bragg grating lters and a ring cavity tunable
laser is proposed and analyzed. Because of the conver-
gence time limitation of the FabryPerot cavity, only 4 bits
of resolution can be achieved at 10 Gsps. Moreover, the
performance of Bragg grating lters (see Fig. 25) show
limited resolution. The authors proposed a folding circuit
to enhance its resolution, where a MachZehnder inter-
ferometer performs the optical folding circuit and a rst-
or second-order linearization circuit corrects the nonlinear
folded signal. Thus, two low-resolution ADCs are coupled
with the folding circuit to achieve MN bits of resolution.
The author claimed that the ADC would be able to operate
at conversion speeds in excess of 10 GHz with up to 1012
bits of resolution. However, it is not clear how the upper
M-bit ADC achieved the additional 68 bits. But it indi-
cates a promising way to perform analog-to-digital con-
version in spectrum domain.
7. NEAR-IR OPTICAL SPECTROSCOPY
Near-infrared (NIR) spectroscopy is a new, noninvasive
technique employed to analyze living tissue. In NIR spec-
troscopy, the main aim is to extract the optical properties
(absorption and scattering) of the living tissue. Absorption
information is used to characterize the concentration of
biological chromophores, such as hemoglobin (in oxy and
deoxy forms), which in turn indicates the physiological
changes in blood. Scattering data provide information on
(a)
0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1
0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0
0 0 1 1 1 1 0 0 0 0 1 1 1 1 0 0
0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 0
E C' D'
B
A
D
B'
E
C
MSB
MSB-1
MSB-2
A
C
B B
LSB
(b)
(c)
MSB
MSB-1
MSB-2
LSB
V
x
0
Va
V
x
V
x
V
x
0 V
x
V
in
Input buffer
v
x
B
I
0
pulsed
laser
Figure 23. Optical folding-ash ADC [47]. a) various bias points
of all optical ADC, b) realization of MZ modulators for quantiza-
tion, c) the received signal.
Tunable
wavelength
source S
p
l
i
t
t
e
r
MACH-ZEHNDER
PROCESSOR FILTERS
L
PD
PD
PD
PD
PROCESSOR FILTERS
Delay equalizer
c
V
REF
c
V
REF
c
V
REF
LSB
c
V
REF
MSB
Photodetector
Sign bit
V
in
(t)
Comparator
Figure 24. Optical ADC using tunable laser and
lters [53].
composition, density, and organization of tissue structures,
such as cells and subcellular organelles [54,55]. Therefore,
NIR techniques provide information about disease-related
functional and structural changes. More specifically, it has
been shown that physiological changes such as ischemia,
necrosis, and malignant transformation can produce im-
portant perturbations in tissue optical properties [55]. The
importance of the NIR spectrum lies in the fact that in this
region tissue absorption is much lower than in other parts
of the spectra (see Fig. 26). Apart from tissue information
content, this region is attractive since NIRinstruments are
inexpensive to construct and are easily portable. These
features render NIR instruments as an attractive alterna-
tive to other techniques, such as MRI (magnetic resonance
imaging). Moreover, NIR light is not an ionizing radiation;
therefore, it can be used as a usual clinical monitoring of
patients in radiation therapy.
The modulated NIR could be employed for greater spa-
tial and temporal information and could be explained on the
basis of the following physical principle. When photons en-
ter a turbid (multiply scattering) medium, the photons scat-
ter randomly in all directions, diffuse through the medium,
and are absorbed during this diffusion process. When source
detector separation is large enough and scattering domi-
nates the absorption, the diffusion theory provides a very
suitable approximation for photon transport
1
c
@F(r; t)
@t
DV
2
F(r; t) m
a
F(r; t) =S(r; t)
where F is the uence rate (W/cm
2
), c is the speed of light in
the tissue, S is the source term, m
a
is the absorption, and D
is the diffusion constant, which is related to the reduced
scattering constant m
/
S
by D=1=(3m
/
S
). In an innite medium
for a point source, the photon diffusion wave (PDW) can be
Transmission
MSB
13.2
10.0
5.0
0.0
15.8
10.0
0.0
5.0
12.6
10.0
5.0
0.0
18.3
15.0
10.0
5.0
0.0
1.5436E6 1.5460E6
1.5460E6
1.5480E6
1.5480E6
1.5500E6
1.5500E6
1.5520E6
1.5520E 6
1.5540E6
1.5540E6
1.555
1.555
1.555
1.555
1.5436E6
1.5460E6 1.5480E6 1.5500E6 1.5520E6 1.5540E6 1.5436E6
1.5460E6 1.5480E6 1.5500E6 1.5520E6 1.5540E6 1.5436E6
LSB+2
LSB+1
LSB
Figure 25. Performance of Bragg grating lters.
Hb
Therapeutical
window
Water
HbO
2
600 700 800 900 1000
0.0
0.1
0.2
0.3
0.4
0.5
A
b
s
o
r
p
t
i
o
n

c
o
e
f
f
i
c
i
e
n
t

(
c
m
1
)
Wavelength (nm)
Figure 26. Absorption spectrum of Hb (deoxy-hemoglobin),
HbO2 (oxy-hemoglobin), and water in the NIR region.
expressed analytically as
F(r; t) =
S
4pD
exp(ikr)
r
where k is the complex wavevector (i.e., k=k
real
ik
imag
) and is described as combination of modulating
frequency, diffusion, and absorption coefcients as k =
m
a
=Dio=(c
.
D)
_
. The backscattered PDW has phase
lag and amplitude attenuation relative to the source as
Y
lag
(r; o) =k
imag
:r
A
att
(r; o) = exp(k
real
r)=(4pDr)
Figure 27 shows the solution of amplitude attenuation
and phase shift of photon diffuse waves with respect to
frequency (up to 1GHz) for breast tissue with optical
absorption and scattering coefcients properties of m
a
=
0:05 cm
1
; m
/
S
=10 cm
1
. NIR techniques are also being
used for brain imaging as seen in Fig. 28 [57,58]. The
image on the left is obtained from MRI. The image on the
right is obtained from diffuse optical tomography.
The NIR technique uses continuous-wave (CW), time-
domain, and frequency-domain instruments according to
their applications and information content. CW systems
are very inexpensive but suffer from limited resolution.
Frequency-domain (FD) instruments are more compact
and cheaper, and FD algorithms are easier to handle than
those of the time-domain techniques; hence FD instru-
ments are more attractive. In FD photon migration con-
cept is also easier. When light is modulated by frequencies
in the megahertz region, diffuse photon density waves
(PDW) are generated, propagating with a wavelength of
several centimeters [5962]. At the detector one measures
the amplitude decay and phase shift data of these waves
(see Fig. 27). Amplitude and phase data are used to map
the optical absorption and scattering properties of the me-
dium. Optical constants in turn are used to obtain hemo-
globin concentration, blood volume, and oxygen
saturation. Greater accuracy of the extracted results is
achieved when a frequency-swept mode is employed, and
as the modulation frequency increases to the microwave
region, a higher spatial resolution is attained.
8. CONCLUSIONS
This article provides a personal perspective of microwave
photonics and its evolution toward a mature eld. From its
fundamental beginning of light interaction with semicon-
ductor devices and circuits, the goal has been to take ad-
vantage of isolation between optical and electrical
systems. Moreover, the injection photogenerated elec-
tronhole pair inuences the dynamics of microwave de-
vices, leading to the development of novel devices. Light
interaction with microwave devices and circuits has seen a
new resurgence, particularly in applications dealing with
integrated optical detectors with microwave functions.
HBT (heterojunction phototransistor)-based devices
seem to provide high gain and speed performance.
Figure 28. Brain activation image: MRI image is shown on the
left, and the image from diffuse optical tomography is depicted on
right. (Courtesy of Prof. Britton Chance of University of Pennsyl-
vania.)
0.95
1
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0 100 200 300 400 500 600 700 800 9001000
N
o
m
a
l
i
z
e
d

m
a
g
n
i
t
u
d
e
Frequency(MHz)
(a)
120
100
80
60
40
20
0
100 200 300 400 500 600 700 800 9001000 0
Frequency (MHz)
(b)
P
h
a
s
e

s
h
i
f
t

(
D
e
g
r
e
e
)
Figure 27. Normalized amplitude attenuation (a) and phase shift (b) of photon diffuse waves as a
function of frequency.
Fiber-fed phased-array antennas and wireless commu-
nications have been the second driver for the microwave
photonics. The performance of analog beroptic links is
analyzed in terms of link gain and dynamic range for data
signal distribution; performance of FO links are also an-
alyzed in terms of AM/PM conversion employed in remot-
ing of the frequency reference to stabilize the distributed
local oscillators. Moreover, distortion induced by SRS in
dense WDM systems is discussed. Fiberoptic links based
on direct modulation could meet costperformance re-
quirements for many applications, even though monolith-
ically integrated EA modulators are becoming very
attractive. Another innovation is in the design of multi-
function circuits, such as the monolithically, integrated FP
laser with EA modulator, where simultaneous transmis-
sion of frequency reference of data signals is accomplished.
The optical oscillation in this novel device could be stabi-
lized as a LO signal using either injection locking or mode
locking. The achieved close-in to carrier phase noise of the
stabilized LO signal is lower in the case of injection lock-
ing than mode locking for the same modulating power
level. However, the nonlinear behavior of the mode-locked
laser provides opportunity for efcient optoelectronic mix-
ing of LO and data signals, while maintaining a very high
SFDR. This multifunction circuit creates the possibility of
generating a RF signal from the frequency reference and
data signals, thus bypassing the need for integration with
electrical mixers in up/downconversion.
Among the technologies that are unique to microwave
photonics is the issue of optical signal processing, which
could lead to very large timebandwidth products, hence
resulting in high frequency selectivity. Memory loop de-
vices, transversal lters, and tapped delay lines are at-
tractive solutions that command unique advantages over
the electrical signal processing techniques. Moreover, the
optical ADC is considered another important advantage
over electrical systems because of its lower timing jitter,
hence achieving a high resolution ADC at the Gsps level.
This technology is critical in the development of digital
receivers for software radiocommunication systems. Fi-
nally, new applications for microwave photonics are
emerging in the medical imaging using the photon densi-
ty wave. The RF modulated light at microwave frequency
at wavelengths associated with absorption peaks of oxy-
and deoxyhemoglobin provides a higher spatial resolution
with function imaging of the biological tissues. This tech-
nique is currently being pushed for clinical applications to
medical imaging of brain, breast, and skin tissues.
There many innovations in the eld of microwave pho-
tonics that, because of space limitations in this article,
impossible to cover here. However, the author wishes to
recommend as further reading a number of special issues
of IEEE Transactions on Microwave Theory and Tech-
niques that are dedicated to reviews of the latest innova-
tions in microwave photonics as further reading.
Acknowledgment
The author wishes to acknowledge the contribution of
many of his students, particularly Dr. Murilo Romero,
Dr. Edward Ackerman, Dr. Reza Saedi, Mr. Adam McIn-
vale, Dr. Tsang-der Ni, Dr. Xiang-dong Zhang, Dr. Man-
ouchehr Ghanevati, Dr. Joong-Hee Lee, and Dr. Xiaobo
Hou. Moreover, technical discussions with his colleagues
Prof. Asher Madjar, Technion, Tel Aviv, Israel; Dr. Kenji
Sato, NTT, Yokosuka, Japan; Dr. Hiroyo Ogawa, CRL,
Yokosuka, Japan; Prof. Robert Minasian, University of
Sydney, Australia; Prof. Peter Herczfeld, Drexel Univer-
sity, Philadelphia; Prof. Bahram Jalali, UCLA, Los Ange-
les; Prof. Britton Chance, University of Pennsylvania,
Philadelphia; and Prof. Tibor Berceli, Budapest Universi-
ty of Science and Technology, Budapest, Hungary are
greatly appreciated.
BIBLIOGRAPHY
1. B. Saleh and M. Tiech, Fundamentals of Optics, Wiley, New
York, 1991.
2. P. Bhattacharya, Semiconductor Optoelectronic Devices,
Prentice-Hall, Englewood Cliffs, NJ, 1994.
3. A. S. Daryoush, K. Bontzos, and P. R. Herczfeld, Optically
tuned patch antenna for phased array applications, IEEE Int.
APS Symp. Digest, Philadelphia, PA, 1986.
4. P. Cheung, D. P. Neikirk, and T. Itoh, Optically controlled co-
planar waveguide phase shifters, IEEE Trans. Microwave
Theory Tech. 38(5):586595 (1990).
5. M. Belaga, Application of an Optically Controlled PIN Diode
in Microwave Circuits, M.S. thesis, Drexel Univ., Philadel-
phia, PA, 1987.
6. A. Rosen and F. Zutavern, High-Power Optically Activated
Solid-state Switches, Artech House, 1994.
7. R. Simons, Microwave performance of an optically controlled
A1GaAs/GaAs high electron mobility transistor and GaAs
MESFET, IEEE Trans. Microwave Theory Tech.
35(12):14441455 (1987).
8. Y. Akatsu, High-speed monolithically integrated pin-HEMT
photoreceivers, Proc. 1995 IEEE/LEOS Summer Topical
Meeting, ICs for New Age Lightwave Communications, pp.
3334.
9. A. L. Guiterrez-Aitken, K. Yang, X. Zhang, G. I. Haddad, P.
Bhattacharya, and L. M. Lunardi, 16-GHz bandwidth In-
A1As-InGaAg monolithically integrated p-i-n/HBT photore-
ceiver, IEEE Photon. Technol. Lett. 7(11):13391341 (Nov.
1995).
10. S. Chandrasekhar, L. Lunardi, A. H. Ganuck, R. A. Hamm,
and G. J. Qua, High-speed monolithic p-i-n/HBT and HPT/
HBT photoreceivers implemented with simple phototransis-
tor structure, IEEE Photon Technol. Lett. 5(11):13161318
(1993).
11. J. Lasri, D. Dahan, A. Bilenca, G. Eisentein, and D. Ritter,
Clock recovery at multiple bit rates using direct optical injec-
tion locking of self oscillating InGaAs/InP heterojunction bi-
polar photo-transistor, IEEE Photon. Technol. Lett. 13:1355
1357 (2001).
12. A. J. Seeds and A. A. A. de Salles, Optical control of micro-
wave semiconductor devices, IEEE Trans. Microwave Theory
Tech. 38(5):577585 (1990).
13. A. Madjar, P. Herczfeld, and A. Paolella, A novel analytical
model for optically generated currents in GaAs MESFETs,
IEEE Trans. Microwave Theory Tech. 40(8) (1992).
14. M. A. Romero, Modulation Doped Field Effect Photodetectors,
Ph.D. thesis, Drexel Univ., Philadelphia, PA, 1995.
15. H. Ogawa, S. Banba, E. Suematsu, H. Kamitsuna, and D.
Polifko, A comparison of noise performance between a pin di-
ode and MMIC HEMT and HBT optical receivers, IEEE Int.
Microwave Symp. Digest, Atlanta, GA, 1993, pp. 225228.
16. A. Paolella, A. Madjar, and P. Herczfeld, Modeling the GaAs
MESFETs response to modulated light at radio and micro-
wave frequencies, IEEE Trans. Microwave Theory Tech.
42(7):11221130 (1994).
17. A. H. Madjedi, S. K. Chaudhuri, and S. Safavi-Naeini, Opti-
cal-microwave interaction modeling in high-temperature su-
perconducting lm, IEEE Trans. Microwave Theory Tech.
49(10):18731881 (2001).
18. A. S. Daryoush, E. Ackerman, N. Samant, D. Kasemsat, and
S. Wanuga, Interfaces for high-speed ber optic links: Anal-
ysis and experiment, IEEE Trans. Microwave Theory Tech.
39(12) (Dec. 1991).
19. E. Ackerman, Photonics Inc., private communications.
20. A. R. Chraplyvy and P. S. Henry, Performance degradation
due to stimulated Raman scattering in wavelength-division-
multiplexed optical-ber systems, Electron. Lett. 19(16) (Aug.
1983).
21. A. R. Chraplyvy, Limitations on lightwave communications
imposed by optical-ber nonlinearities, J. Lightwave Technol.
8(10) (Oct. 1990).
22. Z. Wang et al., Effects of cross phase modulation in wave-
length SCM video transmission systems, Electron. Lett.
31(18) (Aug. 1995).
23. T. E. Darcie et al., Lightwave multi-channel analog AM video
distribution systems, IEEE Int. Conf. Communications, Bos-
ton, MA, June 1989.
24. Z. Wang, A. Li, C. Mahon, G. Jacobsen, and E. Bodtker Per-
formance limitations imposed by stimulated Raman scatter-
ing in optical WDM SCM video distribution systems, IEEE
Photon. Technol. Lett. 7(12) (Dec. 1995).
25. M. R. Phillips and D. M. Ott, Crosstalk due to optical ber
nonlinearities in WDM CATV lightwave systems, J. Light-
wave Technol. 17(10) (Oct. 1999).
26. X. Zhang, and T. D. Ni, and A. S. Daryoush, Laser induced
phase noise in optically injection locked oscillator, IEEE MTT
Symp. Digest, 1992, pp. 765768.
27. X. Zhang and A. S. Daryoush, Bias dependent low frequency
noise up-conversion in HBT oscillators , IEEE Microwave
Guided Wave Lett. 4(12):423425 (Dec. 1994).
28. T. D. Ni, X. Zhang, and A. S. Daryoush, Experimental study
on close-in carrier phase noise of laser diode with coherent
feedback, IEEE Trans. Microwave Theory Tech. 43(9):2277
2283 (Sept. 1995).
29. A. S. Daryoush, K. Sato, K. Horikawa, and H. Ogawa, Ef-
cient opto-electronic mixing at Ka-band using a mode-locked
laser, IEEE Microwave Guided Wave Lett. 9(8):317319 (Aug.
1999).
30. A. S. Daryoush et al., Applications of opto-electronic mixing in
distributed systems, 2001 Topical Symp. Millimeter Wave Digest,
(TSMMW2001), Yokosuka, Japan, March 2001, pp. 175178.
31. A. S. Daryoush, K. Sato, K. Horikawa, and H. Ogawa, Efcient
optoelectronic mixing at Ka-band using a mode-locked laser,
IEEE Microwave Guided Wave Lett. 9(8):317319 (Aug. 1999).
32. H. X. Shi, D. A. Cohen, J. Barton, M. Majewski, L. A. Coldren,
M. C. Larson, and G. A. Fish, Dynamics range of widely tun-
able sampled grating DBR lasers, Electron. Lett. 38:180181
(Feb. 2002).
33. A. S. Daryoush, K. Sato, K. Horikawa, and H. Ogawa, Dy-
namic response of long optical cavity laser diode for Ka-band
communication satellites, IEEE Trans. Microwave Theory
Tech. 45(8):12881295 (Aug. 1997).
34. J. Faist, F. Capasso, D.L. Sivco, C. Sirtori, A.L. Hutchinson,
and A.Y. Cho, Science 264: 553 (1994).
35. R. Saedi, W. Jen, N. Samant, A. S. Daryoush, D. Sturzebecher,
and P. Herczfeld, Comparison of CPU level data mixing to T/R
level data mixing architectures in optically controlled phased
arrays, Proc. 1993 Int. Microwave Symp., Atlanta, GA, June
1993.
36. X. Zhang and A. S. Daryoush, Full 3601 phase shifting of in-
jection locked oscillators, IEEE Microwave Guided Wave Lett.
3(1):1416 (1993).
37. X. Zhou and A. S. Daryoush, An efcient self-oscillating mixer
for communications, IEEE Trans. Microwave Theory Tech.
42(10):18581862 (Oct. 1994).
38. A. S. Daryoush, K. Sato, K. Horikawa, and H. Ogawa, Ef-
cient optoelectronic mixing at Ka-band using a mode-locked
laser, IEEE Microwave Guided Wave Lett. 9(8):317319 (Aug.
1999).
39. R. Saedi, A. S. Daryoush, and P. Herczfeld, Fiberoptic based
recirculating memory loop, IEEE Int. Microwave MTT-S
Symp. Digest, Albuquerque, NM, June 1992, Vol. II, pp.
677680.
40. D. B. Hunter, R. A. Minasian, and P. A. Krug, Tunable optical
transversal lter based on chirped gratings, Electron. Lett.
31:22052207 (1995).
41. J. Campany, D. Pastor, and B. Ortega, New and exible ber-
optic delay line lters using chirped Bragg gratings and laser
arrays, IEEE Trans. Microwave Theory Tech. 47:13211326
(1999).
42. X. Wang and K. T. Chan, Tunable all-optical incoherent bi-
polar delay-line lter using injection-locked FP laser and ber
Bragg gratings, Electron. Lett. 36(24) (2000).
43. A. M. Weiner et al., IEEE J. Quantum Electron. 28:908
(1992).
44. R. A. Minasian, K. E. Alameh, and E. H. W. Cahan, Photonics-
based interference mitigation lters, IEEE Trans. Microwave
Theory Tech. (special issue on microwave and millimeter-
wave photonics) 49(10):18941899 (2001).
45. B. L. Shoop, Photonic Analog-to-Digital Conversion, Springer,
2001.
46. P. W. Juodawlkis, J. C. Twichell, G. E. Betts, J. J. Hargreaves,
R. D. Younger, J. L. Wasserman, F. J. ODonnell, K. G. Ray,
and R. C. Williamson, Optically sampled analog-to-digital
converters, IEEE Trans. Microwave Theory Tech. 49(10):
18401853 (Oct. 2001).
47. T. R. Clark, J. U. Kang, and R. D. Esman, Performance of a
time- and wavelength-interleaved photonic sampler for ana-
log-digital conversion, IEEE Photon. Technol. Lett.
11(9):11681170 (Sept. 1999).
48. A. S. Bhushan, P. V. Kelkar, B. Jalali, O. Boyraz, and M. Is-
lam, 130-Gsa/s photonic analog-to-digital converter with time
stretch preprocessor, IEEE Photon. Technol. Lett. 14(5):684
686 (May 2002).
49. H. F. Taylor, Electro-Optic A/D Converter, U.S. Patent
4,058,722 (1957).
50. B. Jalali and Y. M. Xie, Optical folding-ash analog-to-digital
converter with analog encoding, Opt. Lett. 20(18):19011903
(1995).
51. M. J. Hayduk, Photonic Analog-to-Digital Conversion Using
Light Absorption, U.S. Patent 6,326,910.
52. J. T. Gallo, Photonic A/D Converter Using Parallel Synchro-
nous Quantization of Optical Signals, U.S. Patent 6,188,342.
53. H. Zmuda, E. N. Toughlian, G. Li, and P. Li Kam Wa, A pho-
tonic wideband analog-to digital converter, Proc. 2001 IEEE
Aerospace Conf., 2001, Vol. 3, pp. 14611472.
54. B. Chance, M. Cope, E. Gratton, N. Ramanujam, and B.
Tromberg, Phase measurement of light absorption and scat-
ter in human tissue, Rev. Sci. Instrum. 69:34573481 (1998).
55. B. Tromberg, N. Shah, R. Lanning, A. Cerussi, J. Espionza, T.
Pham, L. Svaasand, and J. Butler, Non-Invasive in vivo char-
acterization of breast tumors using photon migration spec-
troscopy, Neoplasia 2:2640 (2000).
56. J. B. Fishkin, O. Coquoz, E. Anderson, M. Brenner, and B.
Tromberg, Frequency-domain photon migration measure-
ments of normal and malignant tissue optical properties in
a human subject, Appl. Opt. 36:1020 (1997).
57. B. Chance, E. Anday, S. Nioka, S. Zhou, L. Hong, K. Worden,
C. Li, T. Murray, Y. Ovetsky, D. Pidikiti, and R. Thomas, A
novel method for fast imaging a brain function, non-invasive-
ly, with light, Opt. Express 2:411423 (1998).
58. V. Ntziachristos, A. G. Yodh, M. Schnall, and B. Chance, MRI-
guided diffuse optical spectroscopy of malignant and benign
breast lesions, Neoplasia 4(4):347354 (2002).
59. J. B. Fishkin and E. Gratton, Propagation of photon density
waves in strongly scattering media containing an absorbing
semi-innite plane bounded by a straight edge, J. Opt. Soc.
Am. A 10:127140 (1993).
60. S. Fantini, M. Franceshini, and E. Gratton, Semi-innite-ge-
ometry boundary problem for light migration in highly scat-
tering media: A frequency-domain study in the diffusion
approximation, J. Opt. Soc. Am. B 11:21282138 (1994).
61. J. B. Fishkin, S. Fantini, M. J. vandeVen, and E. Gratton,
Gigahertz photon density waves in a turbid medium: Theory
and experiments, Phys. Rev. E 53:2307 (1996).
62. T. H. Pham, O. Coquoz, J. B. Fishkin, E. Anderson, and B.
Tromberg, Broad bandwidth frequency domain instrument
for quantitative tissue optical spectroscopy, Rev. Sci. Instrum.
71:25002513 (2000).
MICROWAVE POWER AMPLIFIERS
PAOLO COLANTONIO
FRANCO GIANNINI
ERNESTO LIMITI
Universita` degli Studi di Roma
Tor Vergata
Rome, Italy
1. INTRODUCTION
A power amplier (PA) is a system component whose main
task is to increase the power level of the signal at its input
up to a predened level. As contrasted to low-level ampli-
ers, often specied in terms of small-signal gain, the ab-
solute output power level is the PA main feature. This
characteristic on one hand forces the selection of the active
devices composing the PA on the basis of their output pow-
er capabilities; on the other hand, in order to fully exploit
such capabilities, the devices are typically operated under
large-signal regimes, forcing the exploration of nonlinear
operating regions. A PA is therefore an intrinsically non-
linear system component, whose large-signal operating
conditions often lead to detrimental effects on the output
signal, resulting in a distorted replica of the input. More-
over, the linear approximation underlying small-signal
amplier design techniques is no longer strictly valid and
hence does not allow their direct application to PA design.
Dedicated methodologies therefore have to be adopted,
even if preliminary and rst-guess simplied approaches
are often employed. The present contribution is aimed at
addressing the microwave power amplier design and
performance. Radiofrequency (RF) or other approaches at
lower operating frequencies (below 1GHz) are therefore
not covered in detail, except where explicitly stated.
Microwave PA system applications span a broad range
of areas [1], including telecommunications, radar [24],
electronic warfare, heating [5,6], and medical microwave
imaging [712], which represent only a few examples.
Given such extremely diversied elds, PA specications
differ greatly on operating, technological, and design
requirements. Examples of such differences range from
traveling-wave-tube (TWT) ampliers in satellite pay-
loads to solid-state ampliers for personal wireless com-
munication handsets, from microwave heating tubes to
ampliers composing hyperthermia apparatuses.
The large differences in system applications are reected
back into the technologies adopted for realization of the
PA active module. The early days of the microwave era
were characterized by an extensive, widespread use of
vacuum-tube devices [1316] for both generation and am-
plication of microwave signals. If microwave electronics
may be dated back to the pioneering work of H. Hertz and
J. C. Bose [1720], a major push toward high-power mi-
crowave generation and use came from the World War II
military application in the radar eld, with the introduc-
tion and use of the cavity magnetron by British research-
ers and the klystron as a high-power source in 1939/40.
The klystron evolved in Stanford University toward a high
average power amplier, leading to modern applications
involving clustered cavity klystron and traveling-wave
tubes [2123].
In such a scenario, solid-state devices and related am-
pliers are relatively recent players, being the rst GaAs
MESFET, with good performance at X band, commercially
available since the early 1970s, despite the device intro-
duction in the early work of Stuetzer and Shockley [24,25].
The technology rapidly advanced and, in the 1970s, the
development of techniques for semiconductor crystal
growth such as molecular-beam epitaxy (MBE) permitted
optimized p-n-junction structures to be realized, allowing
a series of two- and three-terminal devices to be fabricated
with frequency operation ranging from a few gigahertz
well into the millimeter-wave region [2628]. To the two-
terminal device category can be assigned IMPATT transit-
time and Gunn transferred electron devices, which have
been and are still used for both frequency generation and
amplication, as negative-resistance ampliers, in the
millimeter-wave frequency region. The broader and
much more frequently adopted category of three-terminal
active devices includes the already mentioned metal semi-
conductor eld-effect device (MESFET), the high-electron-
mobility transistor (HEMT; demonstrated by Mimura in
MICROWAVE POWER AMPLIFIERS 2871
1980 [26]) with its pseudomorphic (PHEMT) and meta-
morphic (MHEMT) variants, the heterojunction bipolar
transistor (HBT; introduced by Kroemer in 1957 [29]), and
nally, due to the more recent major advances of high-fre-
quency silicon technology, MOS and bipolar silicon tran-
sistors, including laterally diffused MOS structures
(LDMOS, [30,31]) and silicongermanium (SiGe) HBTs
[32,33] as key representatives. Nevertheless, the work
horse technology for microwave power amplication is
indeed based on IIIV technologies, mainly of the GaAs
type. The latter technology is capable of providing, as the
output of a single device, output power levels close to
50 dBm [34] with operating frequencies approaching W
band. In the upper frequency range, InP solutions are
more appropriate, even if they provide very limited output
power levels.
Even if combined device concepts utilizing both solid-
state and vacuum-tube devices have been proposed, the
former are generally utilized for lowmoderate power out-
put while the latter is indeed necessary whenever high
power and high frequency of operation are addressed as
schematically indicated in Fig. 1.
Performance of the given device type may be readily
demonstrated to exhibit a well-known behavior if consid-
ered in their frequency limit region, i.e. following the law:
P
.
f
2
- const (1)
where P and f represent output power and the operating
(narrowband) frequency, respectively.
The single solid-state device output can be combined
utilizing a series of different techniques, and therefore
solid-state power ampliers with output powers compara-
ble to the ones of a vacuum-tube source (i.e., in the kilo-
watt region) may be obtained in the microwave frequency
region (up to X band). On the other hand, practical limi-
tations arise from the systematic application of combining
techniques as operating frequency increases, thus impos-
ing the use of vacuum-tube sources.
The increase in solid-state single-device performances
will therefore lead to new device concepts and new devel-
oping technologies. Among the latter ones, SiC [35] and
GaN wide-bandgap semiconductor technologies are actual-
ly being explored; whereas SiC MESFET and HEMT have
demonstrated 45W/mm output power density throughout
X band [36], nitride-based components are extremely prom-
ising, setting a new upper limit to device performance in
the range of 20W per millimeter of device periphery, as
compared to the 12W limit of GaAs FET-based technolo-
gies and the 24W of the HBT ones [3739]. Even if exper-
imentally demonstrated in the microwave range, such
performances are expected to be exported to higher oper-
ating frequencies, well into the millimeter-wave spectrum.
The more recent trends toward increasingly high
power densities, pushed mainly by radar and electronic
warfare applications, is the latest development of the
tremendous growth experienced by the high-frequency
semiconductor industry. In fact, since the early 1990s,
mobile and personal communications systems, ranging
from cellular telephony to wireless LAN, with the corre-
sponding demand for high-quality radiolinks, is posing
a major challenge to high-frequency technologies and
subsystem performances, especially in the microwave fre-
quency region.
The demand for portable apparatuses, whose main cha-
racteristic is battery duration and overall size, logically
Gridded tube
SIT
BJT
Klystron
Gyrotron
Vaccum
devices
Solenoid focused
CC - TWT
BWO
MESFET
IMPATT
FEL
FEL
PHEMT
Solid state
devices
PPM focused
helix TWT
CFA
Frequency (GHz)
A
v
e
r
a
g
e

p
o
w
e
r

(
W
)
10
2
.1 1 10 100 1,000 10,000 100,000
10
1
10
2
10
3
10
4
10
5
10
6
10
7
1
10
Figure 1. Single-device output power as a function of frequency for solid-state and vacuum de-
vices. [From V. L. Granatstein et al., Vacuum electronics at the dawn of the twenty-rst century,
Proc. IEEE 87(5):702716 (May 1999).]
2872 MICROWAVE POWER AMPLIFIERS
translates into a low-power electronic system. Since the
PA in the transmitter section clearly represents the main
source of supply power consumption, such feature is di-
rectly transferred to its specications. A difcult tradeoff
is therefore demanded of the PA designer, balancing
among the contrasting goals of high transmitted power,
low power consumption, and, for many telecommunication
systems, linear operation. Given the widespread diffusion
of many telecom applications, all of the abovementioned
specications have to be fullled keeping unit cost to a
minimum. Such goals and the resulting compromise may
vary depending on the type of radiolink to be established
and overall system specications, but their challenge
has heavily inuenced industrial, technical, and research
directions in the PA eld since early 1990s.
As a consequence, the high-frequency semiconductor
industry has nally moved toward high-volume produc-
tion. In particular, the search for an high output power is
pushing, from the technological point of view, toward the
design of active devices with high power densities, simul-
taneously exhibiting high reliability, and reproducibility,
and at a reasonable cost. Six-inch wafers are currently
adopted for high-volume production in GaAs, as compared
with the former prototyping 3- or 4-in. low-volume pro-
ductions, with a corresponding increase in reproducibility,
wafer uniformity, and yield, at a lower cost. As a conse-
quence, many benecial effects on PA performance in
terms of process maturity and stage performance have
been realized. From the device side, GaAs FET discrete
packaged power devices are actually produced and com-
mercialized, allowing high output power (r40 W in S
bands in partially matched conditions or 4100W direct-
ly in pushpull conguration for base-station applications
[34]) even in X or Ku band (420 W in internally matched
conguration in 1414.5GHz range [34]).
Where intrinsic difculties of material growth and re-
sulting quality are encountered, metamorphic solutions
have been proposed, allowing the growth of high-quality
materials on a solid and reliable bulk substrate; this is the
case for high-indium-content devices and structures uti-
lizing GaAs wafers. The inherent volume and cost advan-
tages are evident, with minor performance degradation;
the latter is due mainly to quality of the transition be-
tween materials, gradually accommodating the lattice
mismatch, which can be appropriately optimized.
2. BASIC DEFINITIONS AND PERFORMANCE PARAMETERS
Regardless of the specific application, a PA may be ulti-
mately considered, from the energy perspective, as a com-
ponent converting DC power from supplies (P
DC
) into
microwave power (P
out
), driven by an input power (P
in
).
This process is schematically depicted in Fig. 2.
The effectiveness of this conversion process is usually
described in terms of the ampliers efciency Z, or con-
version efciency, dened as the ratio between output RF
and supplied DC power:
Z
P
out
P
DC
.
100 (%) (2)
Efciency is often further specied as drain efciency or
collector efciency, in the case of a solid-state PA based
on eld-effect or bipolar transistor respectively.
Amplier efciency is in one of the key parameters in
specifying overall system performances; as noted above,
for a given amount of output power required by system
specication for the PA, the efciency actually xes the DC
power budget and hence the supply power. A reduced sup-
ply power resulting from high-efciency performance is a
key goal of mobile apparatuses, typically battery-operat-
ed, whose operating time depends strictly on the trans-
mitting section power requests.
Moreover, since practical and physical constraints
impose an actual efciency lower than the theoretical
maximum of 100%, high-efciency performances imply in
turn a low power dissipated on the power-amplifying de-
vice, therefore reducing actual size and weight of the
eventually required heatsinks. On the other hand, for
a given available DC power, high-efciency performance
allows higher transmitted power with a corresponding
increase in overall system capabilities.
As frequency increases, however, the PA gain decreas-
es, as a result of its active constituents gain rolloff behav-
ior. Considering the inputoutput transfer characteristics
of the PA, its power gain is dened as
G
P
out
P
in
(3)
and usually expressed in logarithmic units [decibels(dB)].
The contribution to the output power coming directly from
the input drive cannot be neglected since it constitutes, at
microwave frequencies and beyond, a significant portion
of the total amount. As a consequence, dening the added
power P
add
, that is, the net increase in the signal power
from the PA input to its output, we obtain
P
add
P
out
P
in
=P
out
.
1
1
G
_ _
(4)
Power-added efciency (PAE or Z
add
) is dened as the ratio
between the added power and the supplied DC power:
Z
add

P
add
P
DC
=
P
out
P
in
P
DC
=
P
out
.
1
1
G
_ _
P
DC
=Z
.
1
1
G
_ _
(5)
dc power
P
dc
r f output power
P
out
r f input power
P
in
PA
Figure 2. Energetic schematic representation of PA operation.
An alternative definition of Z
add
, less frequently used in
common practice, is [40]
Z
add

P
out
P
DC
P
in
=Z
.
1
1
Z
G
(6)
i.e. the ratio of output power to the total input power fed
into the amplier (RF plus DC; see Fig. 2). The two def-
initions (5) and (6) practically converge for high-gain am-
pliers, while giving substantially different results for
low-gain ampliers, especially when hardly driven into
compression (note that the conventional Z
add
definition
may lead to negative results). If nonconstant envelope sig-
nals have to be treated by the PA, an average efciency
can be introduced [40,41], dened as in (2), where the
quantities in the expression are replaced by input and
output powers averaged over an envelope period and
weighted by the envelope probability density function.
The conversion from DC to RF power implies that a
fraction of the supplied power is actually dissipated on the
active power device. The main contribution to the latter is
spent at the active device output, and is given by
P
diss;out

1
T
.
_
T
v t ( )
.
i t ( )
.
dt (7)
where v(t) and i(t) are the device output voltage and cur-
rent, integrated over a period (T) of the RF signal. To in-
crease the conversion efciency, as will be discussed later,
a possible strategy consists in the minimization of such
dissipated power, namely, in the proper shaping of device
output voltage and current waveforms.
For cascaded (matched) PAs, as depicted in Fig. 3, the
overall efciency Z
tot
is easily computed by
Z
tot
=
P
out
P
DC;1
P
DC;2
=
Z
2
1
P
DC;1
P
DC;2
=
Z
2
1
Z
2
Z
1
.
G
2
(8)
Since the DC supply power for the nal stages (P
DC,2
) is
usually much higher than the driver supply (P
DC,1
), over-
all efciency is dominated from the former amplier. On
the contrary, for a low-gain nal amplier, the drivers ef-
fect also becomes crucial for overall conversion efciency.
Regardless of the definition adopted for Z
add
, its maxi-
mization is to be achieved at the nominal drive level for
the PA, that is, while the latter is delivering the specied
output power. In such operating conditions, the amplier
is typically driven into compression, thus leaving its
almost linear region and exploring the nonlinear active-
device physical limitations, as depicted in Fig. 4, where a
typical P
in
P
out
power sweep is plotted.
From the power sweep in Fig. 4, a commonly used com-
pression level gure of merit, P
out,1 dB
, can be derived,
dened as the output power corresponding to a deviation
of 1 dB from the ideal linear behaviour. The corresponding
input power level, P
in,1 dB
, usually marks the border be-
tween highly nonlinear and almost linear operating
conditions.
The peak drain/collector or power-added efciency
usually occur at higher drive levels, corresponding to
24 dB PA gain compression (Fig. 5). In such a region
the active-device behavior is therefore highly nonlinear,
and correspondingly design methodologies for high ef-
ciency operation must cope with such intrinsic deviation
from linearity.
PAs are normally classied into operating classes
where the device output current conduction angle a (i.e.,
the fraction of the RF signal period where a nonzero cur-
rent is owing) is considered in detail, and where the clas-
sication in Table 1 holds.
The classication above is typically adopted regardless
of the PA drive level, in order to simply indicate the bias-
ing region of the active device, as determined by its qui-
escent supply conditions (Fig. 6).
Even for class A or B PAs, for which the conduction
angle is xed by definition, if the drive level is increased
P
in
P
dc,1
G
1
G
2
P
dc,2
P
out
Figure 3. Cascade connection of two PAs.
P
in
(dBm)
P
out
(dBm)
P
out,-1dB
P
in,-1dB
1dB
P
in
(dBm)
G
(dB)
G
lin
P
in,-1dB
1dB
(a)
(b)
Figure 4. Sample P
in
P
out
power sweep (a) and corresponding
amplier gain (b); P
1 dB
can be derived from both.
p
add
(%)
2 4
dB
P
in
(dBm)
P
in,-1dB
P
in, p max
Figure 5. Typical power-added efciency in a PA as a function of
input drive.
up to compression regions (i.e., if the ampliers are over-
driven or saturated), a variation of current conduction an-
gles occurs. It is clear that the two definitions may lead to
misleading interpretations. Nevertheless, above all in
communication systems characterized by non-constant-
envelope transmitted signals (as in the case of QAM or
in digital cellular communications with GSM and NADC
standards), the transmitter has to satisfy tight require-
ments, not only in terms of efciency but also linearity and
signal spectral purity.
Several indicators of PA linearity are used, depending
on the system specications and modulation schemes that
are adopted. In order to introduce and dene them, a sim-
ple third-order approximation of the PA transfer charac-
teristic is usually assumed
y t ( ) =A
.
x t ( ) k
2
.
x
2
t ( ) k
3
.
x
3
t ( )
_
(9)
where x(t) and y(t) are the input and output signals to the
amplier, respectively (they may be regarded as normal-
ized voltages or currents), A is the small-signal voltage (or
current) gain, and k
2
,k
3
are the rst two coefcients of a
McLaurin expansion of the PA transfer characteristic,
truncated to the third order. Please note that the approxi-
mation above, relating the output signal to the instan-
taneous input value, actually describes a memoryless
system, and therefore memory effects cannot be account-
ed for using this description.
If a single-tone excitation is assumed for the input
signal, with amplitude X and frequency f
x t ( ) =X
.
cos 2
.
p
.
f
.
t ( ) =X
.
cos o
.
t ( ) (10)
the corresponding input power P
x
(on a unitary normaliz-
ing resistor) is
P
x
=
X
2
2
(11)
Output power at frequency f(P
y
) and large-signal gain G
are easily obtained as
P
y;f
=A
2
.
1
3
2
.
k
3
.
P
x
_ _
2
.
P
x
=G
lin
.
1
3
2
.
k
3
.
P
x
_ _
2
.
P
x
(12)
G=
P
y;f
P
x
=G
lin
.
1
3
2
.
k
3
.
P
x
_ _
2
(13)
where G
lin
=A
2
. Since usually k
3
is negative, the previous
derivation accounts for large-signal gain compression,
namely, decrease from the ideal linear constant value
(G
lin
above).
In the same way, from the single-tone excitation, har-
monic generation at 2f and 3f arise, leading to a corre-
sponding output power at harmonic frequencies P
y,2f
and
P
y,3f
, given by
P
y;2f
=
G
lin
.
k
2
2
2
.
P
2
x
P
y;3f
=
G
lin
.
k
2
3
4
.
P
3
x
(14)
therefore justifying the increase in harmonic power by n
dB per dB of input power increase, where n is the order of
the harmonic under consideration, as in Fig. 7.
The harmonic distortion (HD) due to the nth output
harmonic component is therefore easily dened by
HD
nf

P
y;nf
P
y;f
(15)
resulting in the approximated expressions for second- and
third-harmonic distortion (HD
2f
, HD
3f
) in the simple cubic
memoryless model [9]:
HD
y;2f
=
G
lin
.
k
2
2
2
.
P
x
HD
y;3f
=
G
lin
.
k
2
3
4
.
P
2
x
(16)
Similarly, a total harmonic distortion (THD) is dened
summing up all harmonic distortion components in the
A
AB
B
C
I
d
C
B
AB
A
I
d
(A)
V
ds
(V)
0.6
0
0 14
Figure 6. Class of operation dened as out-
put current conduction angle (left) or simply
by the device quiescent bias point (right).
Table 1. Classication of PAs in Terms of Output Current
Conduction Angle a
Operating
Class
Current
Conduction
Angle (a)
Dependence on
Drive Level Bias
A a =2p No Midway between
Device Pinchoff
and Saturation
Regions
AB poao2p Yes Above pinchoff
B a =p No Device pinchoff
C aop Yes Below pinchoff
output signal:
THD
n_2
P
y;nf
P
y;f
(17)
The quantities above are typically measured in decibel
over the carrier power (dBc).
The model (9) is indeed an instantaneous model, that
is, a memoryless description of the PA inputoutput char-
acteristics. Real-world ampliers are dynamic systems
with memory, whose nonlinear behavior affects not only
the amplitude but also the phase of the output signal. In
fact, if the input signal to the PA is assumed to be
x t ( ) =X t ( )
.
cos 2
.
p
.
f
.
t j t ( ) [ ] (18)
the output signal may exhibit nonlinear phenomena in
both amplitude and phase
y t ( ) =G X t ( ) [ ]
.
cos 2
.
p
.
f
.
t j t ( ) F X t ( ) [ ]
_ _
(19)
giving rise to the AM/AM compression and AM/PM con-
version effects (Fig. 8), described by a nonlinear relation-
ship between input and output amplitudes (which actually
represents the ampliers output power compression) and
a phase that depends on the input signal drive level:
G X t ( ) [ ]OA
.
X t ( ) F X t ( ) [ ]Oconst (20)
In particular, the AM/PM conversion effect represents a
change in the phase of the output that depends on the in-
put drive level; this effect is potentially dangerous not only
in communication systems, giving rise for instance to dis-
torted QAM constellations, but also in phased-array ap-
plications where the phase of each signal exiting the
respective PA actually determines the active antenna
pointing characteristics. Nevertheless, the PA description
in terms of its AM/AM and AM/PM characteristics leads to
a narrowband model; it is valid until the amplitude and
phase modulating signal frequencies are much lower than
the carrier one. Regardless of its applicability to narrow-
band operation of PA, the AM/AM compressionAM/PM
conversion model has been widely adopted, also forming
the base for blackbox (or behavioral) modeling of PA op-
eration. Such a behavioral model may be derived from
Volterra analysis [42,43] or directly from experimental
data [44]. The Volterra series approach may also be gen-
eralized to extend the model validity to broadband oper-
ation of the PA [45].
Resorting to the memoryless PA model (9), a two-tone
test may be performed, attempting in this way to simulate
the simultaneous treatment of two different signals and
therefore their mutual interaction caused by the nonlin-
ear PA. Much in the same way, this test may give insight
on a narrowband signal, whose components (the tones in
the test) may interfere, leading to a distorted output. This
is clearly an approximation since the two signals are in
reality much more complex than a simple sinusoid; on the
other hand, the two-tone test is simple enough to be easily
carried out experimentally. The input signal in this case is
given by two closely spaced tones at frequencies f
1
and f
2
(f
1
of
2
) with amplitudes X
1
and X
2
, respectively:
x t ( ) =X
1
.
cos 2
.
p
.
f
1
.
t ( ) X
2
.
cos 2
.
p
.
f
2
.
t ( )
=X
1
.
cos o
1
.
t ( ) X
2
.
cos o
2
.
t ( )
(21)
In such a two-tone test, the frequency spacing (f
2
f
1
) is
much lower than the single frequencies, to replicate a
narrowband excitation. By inserting an input drive such
1
2
3
P
out
(dBm)
P
in
(dBm)
P
y,2f
P
y,f
P
y,3f
Figure 7. Output power in a single-tone test at fundamental
frequency (red), second- (green) and third- (blue) harmonic com-
ponents.
AM / AM
P
out
(dBm)
P
in
(dBm)
30
25
20
15
10
0 5 10 15 19
AM / PM
P
in
(dBm)
1(P
in
)+

(deg)
0
150
5 10 15 19
155
160
165
170
Figure 8. Typical AM/AM compression and AM/PM conversion
curves for a PA.
as that described above into the PA truncated expansion, a
series of output frequency components results, which are
grouped in Table 2, by the term in the expansion (linear,
quadratic, or cubic) from which they originated.
The terms in Table 2 are plotted as a function of fre-
quency in Fig. 9. From Table 2 and Fig. 9, a series of con-
clusions can be drawn. Firstly, the interaction between the
two input frequencies does create, in the nonlinear PA, a
series of frequency components that are not present if the
PA is separately excited by the single frequencies. In fact,
while DC, harmonic components, and compression terms
are already generated from the single-tone excitation, in-
termodulation frequencies and capture terms arise from
the interaction between the two tones. The latter contri-
butions, giving rise to out-of-band and in-band compo-
nents, actually have a power rising as 3 dB per dB increase
of the single tones power (Table 2). In particular, the cap-
ture term (often referred to as suppression term) tends
to decrease power output at a given fundamental frequen-
cy (say, f
1
) proportionally to the square of the power of the
other (f
2
) fundamental frequency; this phenomenon is par-
ticularly effective for high drive levels, and may eventu-
ally lead to the cancellation of one of the signal
components at the PA output, thus justifying the sup-
pression denomination.
Moreover, if signal purity is concerned, harmonic
contributions (at DC, second, and third harmonic of each
input excitation) together with second-order intermodula-
tion and the terms at 2f
2
f
1
and 2f
2
f
1
(all in blue in
Fig. 9) are far away from the useful part of the output
signal (at f
1
and f
2
, in green in Fig. 9), and are therefore
easily eliminated by simple ltering. Other contribu-
tions, much closer to the desired input replica, cannot be
ltered out; from the simple derivation above, they consist
in frequency terms located at 2f
2
f
1
and 2f
1
f
2
(com-
monly referred to as third-order intermodulation com-
ponents, giving rise to intermodulation distortion, (IMD)
and at the input signal frequencies f
1
and f
2
(in-band
distortion, given by the compression and suppression/
capture terms).
Considering one of the two third-order intermodulation
components and sweeping the input tones power, the
third-order intercept point (IP3) is dened as the output
(IP3
out
, or input, IP3
in
) power level at which the third-or-
der IMD component level equals the ideal linear output
power of the PA. Such definition is graphically depicted in
Fig. 10. Even if both useful output signal power and IMD
power tend to saturate for some input drive level, the IP3
definition consists in the ideal extrapolation of both output
signal components (black and blue lines, respectively, in
Fig. 10), ideally rising by 1 and 3dB per dB increase in
input drive respectively, and in the search for their inter-
cept (the IP3). On the basis of such construction, it is clear
that the resulting input drive level (IP3
in
in Fig. 10) is well
into the PA nonlinear operating region and is by far
beyond typical PA operating drives. Similar intercept
points can be dened (even if seldom used) by extension
for higher-order intermodulation products, such us IP5
(for fth-order distortion, located at 3f
2
2f
1
) or IP7 (for
seventh-order distortion, located at 4f
2
3f
1
).
Resorting to the power-series expansion (9), it can be
demonstrated that the IP3 output power level (IP3
out
) can
Table 2. Output Components in a Two-Tone Test Grouped by Originating Term in Truncated Series Expansion
Originating Term Output Frequencies Corresponding Amplitude
x t ( ) f
1
; f
2
X
1
; X
2
Linear term
x
2
t ( ) 2f
1
; 2f
2
X
2
1
; X
2
2
Second harmonic
DC X
2
1
; X
2
2
Rectied component
f
1
f
2
X
1
.
X
2
Second-order intermodulation
f
1
f
2
X
1
.
X
2
Second-order intermodulation
x
3
t ( ) f
1
; f
2
X
3
1
; X
3
2
Compression
f
1
; f
2
X
1
.
X
2
2
; X
2
1
.
X
2
Capture
3f
1
; 3f
2
X
3
1
; X
3
2
Third harmonic
2f
1
f
2
; 2f
2
f
1
X
2
1
.
X
2
; X
1
.
X
2
2
Third-order intermodulation
2f
1
f
2
; 2f
2
f
1
X
2
1
.
X
2
; X
1
.
X
2
2
Third-order intermodulation
dc f
2
f
1
2
.
f
1
f
2
2
.
f
2
f
1
2
.
f
1
2
.
f
2
2
.
f
2
2
.
f
1
3
.
f
1
3
.
f
2
f
2
+ f
1
f
2
+
+ f
1
f
f
1
, f
2
Figure 9. Frequency allocation of the output components originating in a two-tone test.
be related to the 1-dB compression point (P
1dB,out
) by
IP3
out
- P
1dB;out
10:6 dB (dBm) (22)
In real-world PAs, this expression typically overestimates
IP3
out
by 23 dB. The approximations inherent to the
truncated expansion adopted are in fact valid in a limit-
ed range for input drive level, which is typically violated
for the IP3 region. For such large drive levels, fth- and
higher-order contributions arise and significantly modify
the results of the simplied approach. In any case, once
the PA IP3
out
is known, actual IMD level relative to the
output signal level may be estimated from
P
IMD
=3P
out;dBm
2IP3
out
(23)
Another frequently adopted indicator of the PA nonlinear
behavior is the C/I or carrier-to-intermodulation ratio,
dened as the ratio between useful output power and IMD
output power, usually measured in dB below the carrier
(dBc):
C=I
P
out
P
IMD
(24)
where P
IMD
=P
2f
2
f
1
(or P
IMD
=P
2f
1
f
2
). Such C/I is clearly
dependent on the input power to the PA, decreasing by
2 dB per dB increase of the input drive. Combining the two
relationships (23) and (24):
C=I =2 IP3
out
P
out;dBm
_ _
(dBc) (25)
For moderate drive levels (say, up to 10 dB below
P
1 dB,out
), the third-order intermodulation component is
the dominant distortion. It is therefore possible to dene a
linearity range for the PA as the range of input drive levels
for which the P
IMD
stays below the noise oor of the am-
plier, usually specied as thermal noise at ambient tem-
perature; such linearity range, whose lower bound is
determined by the useful output power emerging from
the noise oor, is usually indicated as the spurious-free
dynamic range (SFDR; Fig. 11).
From knowledge of the PA noise gure (NF), its band-
width B and (available) gain G, the SFDR is given, at room
temperature, by
SFDR
dB
=
2
3
.
[IP3
out;dBm
NF
dB
G
dB
B
1 Hz
_ _
dB
174 dBm
_
(dB)
(26)
The quality factors dened above for linearity evaluation
are related to single- or two-tone tests, in an attempt
to mimic in this way the behavior of the PA in response
to a narrowband or multicarrier input. Real-world
input signals to a PA may deviate substantially from the
P
out
(dBm)
IP3
out
IP3
in
P
in
(dBm)
IP3
P
out
= P
in
+ G
lin,dB
P
IMD
= P
2f2f1
1 3
Figure 10. Third-order intercept point definition.
P
out
P
IMD
P
out
(dBm)
IP3
IP3
out
P
out
= P
in
+ G
lin,dB
P
IMD
= P
2f2-f1
IP3
in
P
in
(dBm)
SFDR
SFDR
1
3
Figure 11. Definition of the spurious-free dynamic range; shaded area represents thermal output
noise power at ambient temperature.
single-tone approximation, since modulation formats and
bandwidth occupation may differ to a great extent. In or-
der to account for signal distortion and the related spec-
tral regrowth in the case of bandlimited input signals, an
adjacent-channel power ratio (ACPR) is introduced. With
reference to Fig. 12, several definitions are adopted in this
respect; the most commonly used refers to the total ACPR
(ACPR
TOT
), the ratio between the total output power in
the signal bandwidth to the total output power in adjacent
channels:
ACPR
TOT

_
B
P
out
f ( )
.
df
_
LS
P
out
f ( )
.
df
_
US
P
out
f ( )
.
df
(27)
Clearly, if a single sideband is concerned, a lower-sideband
ACPR (ACPR
LS
) or upper-sideband ACPR (ACPR
US
) can
be dened, using the proper adjacent-channel power in
the definition
ACPR
LS

_
B
P
out
f ( )
.
df
_
LS
P
out
f ( )
.
df
ACPR
US

_
B
P
out
f ( )
.
df
_
US
P
out
f ( )
.
df
(28)
Moreover, a spot ACPR (ACPR
SPOT
) can be introduced,
utilizing the adjacent-channel power contained in a
predened bandwidth (B
x
) at a given offset (see Fig. 12),
dened by
ACPR
SPOT

_
B
P
out
f ( )
.
df
_
B
x
;offset
P
out
f ( )
.
df
(29)
The various ACPR gures clearly give a deeper insight
into the distortion properties of a PA than do their single-
or two-tone counterparts, as they are related to a specific
bandlimited input signal. Nevertheless, if the input signal
is approximated by a number of equally spaced equal-
amplitude tones, closed-form relationships may be found
between the gures [46,48].
Other indicators of PA linearity and appropriate tests
are the noise power ratio and cochannel power ratio
[47,48].
3. BASIC CONCEPTS IN PA DESIGN
Power limiting mechanisms in active devices reside in
their inherent physical constraints. For a FET device (but
also for a bipolar one, with different physical effects taking
place), such constraints are both on output current and
voltage: for the former, current saturation related to input
junction forward conduction and device channel pinchoff;
for the latter, device ohmic behavior and breakdown
(related to both channel and gatedrain junction). Such
physical limitations are graphically depicted in Fig. 13,
where sample output device characteristics are shown.
Power
density
[W/Hz]
Lower
sideband
LS
Upper
sideband
US
Output
signal
Input
signal
f Signal
bandwidth
B
Offset
B
x
Figure 12. Input and output power densities
for adjacent-channel power ratio definitions.
Maximum voltage swing
Maximum
current
swing
V
ds
[V] 0
0
I
d
[A]
Figure 13. Sample device output characteristics and physical
limitations on output current and voltage.
Collectively, such limitations pose an upper limit to the
maximum swings that output current and voltage may
experience, reecting in a limit corresponding to the de-
vice output power.
Small-signal amplier design is based on well-estab-
lished techniques. After the active device to be employed is
selected together with its operating (bias) point, from
knowledge of the device scattering parameters and de-
pending on the amplier specications, input and output
matching network characteristics (i.e., the impedances to
be presented at the device input and output ports) are
readily obtained via closed-form expressions [4951].
If a simplied device model is considered, as depicted in
Fig. 14, composed of a controlled current source (con-
trolled by the input voltage if a FET is considered) shunt-
ed by its output small-signal admittance (represented by
an output conductance g
ds
and capacitance C
ds
), the con-
dition that is imposed, compatibly with device stability, for
maximum power transfer from the device output to the
external load is the well-known conjugate matching
Z
L
f ( ) =Z
+
ds
f ( ) -
G
L
f ( ) =g
ds
B
L
f ( ) = j
.
2pf
.
C
ds
_
(30)
where
Z
L
f ( ) =G
L
f ( ) j
.
B
L
f ( ) (31)
Z
ds
f ( ) =G
ds
f ( ) j
.
B
ds
f ( ) =g
ds
j
.
2pf
.
C
ds
(32)
The conjugate matching condition therefore implies the
compensation of the active-device output reactive part and
a match in its small-signal output conductance, ensuring
at the same time the maximization of the amplier small-
signal gain. If the active-device output characteristics are
now considered, the load line dictated by the small-signal
matching condition (curve A) can be superimposed, as de-
picted in Fig. 15. If the same condition is adopted, driving
the amplier to operate into large-signal regime, a re-
duced current swing results, producing in turn an earlier
compression of the device output power (current-limited
operation). On the other hand, if the external load is se-
lected in order to fully exploit the maximum current
swing, (as in B, Fig. 15), a reduced voltage swing is pro-
duced (voltage-limited operation), again driving active de-
vice output into compression. The optimum situation is in
the simultaneous maximization of current and voltage
swings (as in C, Fig. 15), often referred to as load-line
matching condition. The typical output power behavior
corresponding to the three situations described in Fig. 15
is shown in Fig. 16 as a function of the input drive level.
Nevertheless, PAs that are load-line-matched exhibit
poor output VSWR in the system in which they are em-
ployed. If necessary, this problem can be solved through
the use of output isolators (clearly decreasing output pow-
er and efciency by their losses) or resorting to balanced
congurations if possible: also in this case the insertion
loss of the combining structure actually affects overall
power performance.
In order to have some quantitative indication of the
optimum loading in the simplied conditions described
above, the device output characteristics are often substi-
tuted by their piecewise linear approximation, leading to a
box model for the active-device current source. Such a
model is depicted in Fig. 17 for the case of a constant
trandsconductance (g
m
) device (equally spaced output
characteristics, left) and linear transconductance (linear-
ly increasing transconductance, right). The case of a linear
variation, introduced by Kushner [52,53], tries to mimic
the decrease in transconductance typically observed while
Active device output
I
d
g
ds
Z
L
(f)
Z
ds
(f)
S
D
C
ds
Figure 14. Schematic representation of the active-device output
connected to an external load Z
L.
V
ds
(V) V
ds,DC
V
gs,DC
0
0
I
d
(Amp)
Reduced
current
swing
Reduced voltage swing
(A)
(B)
(C)
Slope
g
ds
Slope
g
ds
Figure 15. Active-device output characteristics with superim-
posed the conjugate-matched load line (A); with voltage-limited
(B), and optimum loading (C). (This gure is available in full color
at http://www.mrw.interscience.wiley.com/erfme.)
P
out
(dBm)
P
in
(dBm)
(B) (A)
(C)
P
out,1dB,(C)
P
out,1dB,(A)
P
in,-1dB,A
P
in,-1dB,C
Figure 16. Output power for three loading conditions: current-
limited (A), voltage-limited (B), and optimum loading (C).
decreasing the device gate bias toward pinchoff. On the
other hand, such approximation actually generates arti-
cial distortion components even with low driving signals,
while requiring, for its actual implementation, a nonuni-
form doping prole.
In Fig. 17 V
k
is the device knee voltage, marking the
transition between the device ohmic and saturation re-
gions, V
br
the breakdown voltage (draingate junction
breakdown), I
max
the maximum drain current (occurring
for fully opened channel, that is, for V
gs
=V
bi
where V
bi
is
the input junction built-in voltage) and V
po
the device
pinchoff voltage. We nowassume a purely resistive loading
of the active device at fundamental and all harmonic fre-
quencies, and we drive the device up to its limit linear
behavior (i.e., without overdriving it into highly nonlinear
regimes). In both cases, if the device is biased at V
ds,DC
for
class A and class B operation as in Fig. 18, and the min-
imum drain voltage is assumed to be V
ds,min
=V
k
, dening
g =
V
k
V
ds;DC
; R
A
=2
.
V
ds;DC
V
k
I
max
(33)
the optimum resistive loads R
L,opt
for class A and B bias
conditions (i.e., for the device biased at I
d;DC
=I
max
=2 and
I
d;DC
=0 respectively) are reported in Table 3, together
with maximum drainsource voltage (V
ds,max
), dissipated
power (P
diss
), drain efciency (Z), and output power at fun-
damental frequency (P
out
). Several observations may be
performed on the results in Table 3.
Clearly, the output power directly depends on the max-
imum currents and voltages sustained by the active device,
namely, I
max
and V
br
(implicitly in Table 3 through the
limit it imposes on the maximum voltage V
ds,max
and hence
on V
ds,DC
). The limit imposed by the device maximum cur-
rent may be overcome by increasing device gate periphery
and/or through the use of device power combining tech-
niques. The former approach, often used in monolithic de-
sign, can be carried out by increasing device nger width
and/or increasing the number of device ngers (if a inter-
digitated device structure is considered, as in the case of
power transistors), as schematically depicted in Fig. 19.
Such device scaling actually enables one to increase
power output capabilities. Nevertheless, such an option
is related to the availability of an affordable monolithic
technology (it is clearly not applicable to hybrid design
approaches). Moreover, the scaling properties of a given
technology are strictly valid for moderate scaling only. For
the same total gate periphery, in fact, while output power
is almost constant (at a given frequency), large-signal
gain decreases with unit gate width, thus affecting
efciency performances as demonstrated experimentally
in Fig. 20.
Moreover, an increase in device periphery actually in-
creases device parasitic effects, thus leading to a further
reduction in gain and frequency of operation. From the
thermal management point of view, scaling up device
periphery induces an increase in dissipated power; the
resulting increase in device junction operating tempera-
ture has detrimental effects on device performance in
terms of both actual performance and reliability, especial-
ly if high-reliability applications such as satellite payload
transmitters are concerned. The commonly adopted solu-
tion to this effect consists in device substrate thinning,
thus decreasing thermal resistance and the temperature
difference between the device channel and its backside.
Clearly, from the design point of view, a viable solution is
in the exploitation of cold biasing regions (i.e., decrease
in bias toward class B), accurately accounting for the con-
sequences both on output power and linearity.
A second possibility for increasing output power perfor-
mances consists in an increase of the drain voltage swing.
This task can be accomplished mainly by increasing the
Constant-g
m
device
I
max
g
m
V
gs
= V
bi
V
gs
= V
po
V
br
V
bi
V
po
I
d
(A)
V
gs
(V)
V
ds
(V)
V
k
0 0
g
m
V
gs
(V) V
bi
V
po
0
Linear-g
m
device
g
m
V
gs
(V) V
bi
V
po
0
I
max
g
m
V
gs
= V
bi
V
gs
= V
po
V
br
V
bi
V
po
I
d
(A)
V
gs
(V)
V
ds
(V)
V
k
0 0
(a) (b)
Figure 17. Piecewise linear approximation of the device output characteristics in the case of
constant (a) and linear (b) transconductance.
device breakdown voltage. Unfortunately the latter is an
intensive quantity, depending mainly on material proper-
ties and device fabrication process. The more recent in-
creasing interest in wide-bandgap materials [viz., gallium
nitride (GaN) and silicon carbide (SiC)] is in fact motivated
from the intrinsic high breakdown eld that devices based
on such alloys exhibit, with typical values in the range of
100V, as contrasted to the fewtens of the traditional GaAs-
based FET devices. On the other hand, as shown in Table
3, PA drain efciency and output power are limited by the
presence of a nonnegligible knee voltage. The effect of
such low-voltage limit is especially important if a low-volt-
age PA has to be designed; this is the case in mobile hand-
sets and in general in every portable battery-operated
transmitter, in which the actual size and weight of the
overall apparatus are dictated by the choice of a light-
weight and small battery pack. In this case a few volts are
available for the PA bias, and any lower limit in the device
voltage swing is very important. As an example, assuming
a 3V supply and a 0.9V knee voltage for the active device,
a PA operating in class A decreases its efciency from the
ideal textbook 50% down to a modest 35% (as per Table 3,
class A, constant transconductance, with g =0.7). The
problem becomes even harder to deal with if high-perfor-
mance deep-submicrometer CMOS devices are employed,
with knee voltages that can be 23 times higher than the
ones of typical power transistors.
Up to this point, the two simple cases of class A and B
PAs under the maximum linear operation have been con-
sidered. The output power and efciency performances can,
however, be greatly improved by making use of harmonic
tuning strategies or switching-mode operating classes. To
this goal, a rst step resides in the use of the tuned-load
(TL [52,53]) operating mode of the power stage. With ref-
erence to Fig. 21, TL operation consists in loading the ac-
tive device output with short-circuit terminations (idlers) at
harmonic frequencies, maximizing at the same time the
fundamental frequency voltage and current swing.
The device output is assumed to operate as a voltage-
controlled current source (controlled by the input gate
source voltage) and therefore its time-domain waveform
can be expressed as a truncated sinusoid
i
d
t ( ) =
I
max
1 cos
a
2
_ _.
cos o
.
t ( ) cos
a
2
_ _ _ _
if o
.
t [ [ _
a
2
0 otherwise
_
_
(34)
I
max
I
d,rf max
V
ds,rf max
V
k
V
ds,DC
V
ds
(V)
I
d
(A)
slope = 1/R
L,opt
I
max
V
k
V
ds,DC
V
ds
(V)
I
d
(A)
slope = 1/R
L,opt
Class B Class A
Figure 18. Class A and B operating conditions for purely resistive loading.
Table 3. Single-Device PA Performance with Resistive Loading for Classes A and B Bias and Constant and Linear
Transconductance
Biasing Class g
m
R
L,opt
(O) V
ds,max
(V) P
diss
(W) Z (%) P
out
(W)
A Constant R
A
V
ds;DC
V
ds;DC
V
k
_ _ V
ds;DC
V
k
_ _
.
I
max
4
50
.
1 g ( )
V
ds;DC
V
k
_ _
.
I
max
4
A Linear
4
5
.
R
A
V
ds;DC
3
5
V
ds;DC
V
k
_ _ 0:65V
ds;DC
0:85V
k
_ _
.
I
max
4
53
.
1 g ( )
V
ds;DC
V
k
_ _
.
I
max
5
B Constant
p
p 1
.
R
A
2
V
ds;DC
V
ds;DC
V
k
p 1
0:4 V
ds;DC
0:87 V
k
_ _
.
I
max
4
58
.
1 g ( )
p V
ds;DC
V
k
_ _
.
I
max
8
.
p 1 ( )
B Linear
2
3
.
R
A
V
ds;DC
V
ds;DC
V
k
3
0:33V
ds;DC
0:66V
k
_ _
.
I
max
4
48
.
1 g ( )
32 V
ds;DC
V
k
_ _
.
I
max
27
.
p
2
+
Increase the number
of device gate fingers
Basic
device
Increase the device
unit gate width
(a) (b) (c)
Figure 19. Increasing device maximum current by scaling the number of gate ngers (a) or device
unit gate width (c) from a basic device (b).
P
out
PAE
Gain
Unit gate width [jm]
35.0
30.0
25.0
20.0
15.0
10.0
50 100 150 200 250
30.0
40.0
50.0
60.0
70.0
80.0
PAE
[%]
Gain
[dB]
P
out
[dBm]
Figure 20. Effect of device unit gate width
scaling for a xed total periphery (1.2 mm). Sol-
id and dashed lines indicate 18 and 14GHz,
respectively. (Data from Raytheon, 1998 IEEE
MTT-S Symposium, Workshop WFF.)
I
max
V
ds,DC
V
ds
(V) V
k
I
d
(A)
:/2 :/2
V
ds,DC
Z
L,f
Z
L,nf
= 0
@nf n2
@f
OUT
IN
(a) (b)
Figure 21. Tuned load loading conditions (a) together with typical voltage and current waveforms
superimposed on piecewise linear output characteristics (b).
where o=2p
.
f and a is the drain current conduction
angle (Fig. 21), while the output drain voltage, thanks to
the short-circuit terminations, is a purely sinusoidal
waveform:
v
ds
t ( ) =V
ds;DC
V
1
.
cos o
.
t ( ) (35)
Treatment of the actual device output admittance will be
omitted in the following in order to simplify the approach.
Nevertheless, the hypothesis of short-circuit output ter-
minations actually help in this assumption (if a real short
circuit could be imposed at the device intrinsic terminals,
not accounting for device parasitic effects), together with
the device output capacitive behavior, effectively short-cir-
cuiting very-high-frequency output components. Drain
current can be expressed in terms of its Fourier series ex-
pansion with coefcients I
n
graphically depicted in Fig. 22
as a function of the current conduction angle a.
The fundamental frequency current component is, un-
der the assumptions performed, a function of the current
conduction angle only. The drain voltage fundamental
component is related to the current one via the funda-
mental frequency output load Z
L,f
. The latter is to be se-
lected as a purely resistive termination in order to
maximize active power generation together with output
voltage swing. To this goal, if the device is biased at V
ds,DC
,
maximum output voltage fundamental frequency ampli-
tude is given by V
1;max
=V
ds;DC
V
k
, thus obtaining, for
the optimum load at fundamental R
TL
R
TL
a ( ) =
V
1;max
I
1
a ( )
=R
A
.
p
.
1 cos
a
2
_ _
a sin a ( )
(36)
where the optimum value is expressed evidencing the R
A
value (optimum resistance in the case of resistive loading,
or load-line match for the case of a class A PA; see Table 3).
Moreover, in the same way, DC supply power P
DC,TL
, out-
put power at fundamental P
out,TL
, and drain efciency Z
TL
can be easily obtained, again normalized to the class A
reference values in Table 3:
P
DC;TL
=I
0
.
V
ds;DC
=
P
DC;A
p
.
2
.
sin
a
2
_ _
a
.
cos
a
2
_ _
1 cos
a
2
_ _ (37)
P
out;TL
=
I
1
.
V
1
2
=
P
RF;A
p
.
a sin a ( )
1 cos
a
2
_ _ (38)
Z
TL
=
P
out;TL
P
DC;TL
=Z
A
.
a sin a ( )
2
.
sin
a
2
_ _
a
.
cos
a
2
_ _ (39)
The quantities described above are presented in graphical
form as functions of a in Fig. 23.
The results attainable via a TL approach are briefly
discussed:
1. An increase in output power over the reference class
A design is possible utilizing a class AB bias with the
TL output loading. Such increase in power perfor-
mance is maintained even moving toward class B
bias, where a simple resistive loading implies a de-
crease in output power with respect to a class A bias.
Clearly, such a result does not account for the gain
performance, since a maximum swing is assumed,
independently on the necessary drive level. In par-
ticular, for class A and B biases, the same output
power is obtained, but utilizing, for the class B drive,
twice the input drive signal (i.e., 6 dB higher input
power is necessary).
2. The optimum load exhibit a weak dependence on the
current conduction angle, thus allowing almost the
same resistive loading from class A to class B bias.
Efciency performances exhibit a constant increase
moving toward low-bias regions, with a limiting value
that doubles the class A performance (thus toward the
theoretical limit of 100% efciency). This is true, however,
only to a rst approximation, since the simplifying as-
sumptions performed on the device characteristics impose
a constant breakdown voltage, regardless of the device
gate bias; real-world devices have an almost constant
gatedrain breakdown, and therefore biasing the device
below pinchoff actually decreases the drainsource maxi-
mum voltage (V
br,ds
=V
br,gd
V
gs
). Finally, the power dis-
sipated on the active device, P
diss,TL
, shown in Fig. 24, is
P
diss;TL
=
P
DC;A
p
.
sin
a
2
_ _
.
2 cos
a
2
_ _ _ _
a
2
.
12
.
cos
a
2
_ _ _ _
1 cos
a
2
_ _
(40)
As expected, moving from class C bias toward class A
operation actually increases the power dissipated on the
active device, therefore decreasing the efciency corres-
pondingly.
A nal consideration can be drawn regarding device in-
put terminations. Input loading at fundamental frequency
Biasing class C B AB A
0.6
0.5
0.4
0.3
0.2
0.1
0.1
0
I
n

/

I
m
a
x
I
1
I
2
I
3
0

2
3
2
2
: [rad]
Figure 22. Fourier components I
1
, I
2
, I
3
normalized to the device
maximum current I
max
as functions of the drain current conduc-
tion angle a.
is to be performed in order to fulll maximum power trans-
fer to the active device; this is the well-known conjugate
matching condition. This condition ensures maximum gain
for a given output loading, while ensuring optimum input
VSWR. Clearly the device input is, for moderate drive lev-
els, mildly nonlinear and therefore the conjugate matching
must be attained under large-signal operating conditions
(i.e., the maximum power transfer condition depends on
the drive level). Harmonic input loading is seldom inves-
tigated, leaving this task to harmonic tuning strategies.
4. POWER BALANCE AND DIFFERENT
DESIGN APPROACHES
In order to generalise the approach outlined for the TL
operating condition and the approach outlined in Refs. 54
and 55 for switched-mode stages (see Section 6s), with the
aim to introduce high-efciency operating classes, let us
consider the power balance in a PA, distinguishing, in the
0.5
0.4
0.3
0.2
0.1
0
0

2
3
2
2
: (rad)
P
diss,TL
(:) / P
DC,A
Figure 24. Power dissipated on the active device P
diss,TL
for the
tuned load operating condition normalized to the class A (resistive
load) DC power as a function of the drain currentdrain conduc-
tion angle.
0.0
0.2
0.4
0.6
0.8
1.0
0

2
3
2
2
: (rad)
P
DC,TL
(:)/P
DC,A
0
0
0.25
0.5
0.75
1.25
1.0
2
3
2
2
: (rad)
p
TL
(:)
0
0.0
0.25
0.5
0.75
1.25
1
2
3
2
2
: (rad )
R
A
/ R
TL
(:)
0
0.0
0.30
0.60
0.90
1.5
1.2
2
3
2
2
: (rad)
P
out,TL
(:)/P
out,A
Figure 23. Optimum load R
TL
, DC power P
DC,TL
, output power P
out,TL
, and drain efciency Z
TL
for
the tuned load operating condition normalized to the corresponding class A (resistive load) quan-
tities.
output power, the two contributions to fundamental (P
out,f
)
and harmonic terminations (P
out,nf
, nZ2). In greater de-
tail, the power balance of a PA system can be stated as
P
DC
P
in
=P
diss;in
P
diss;out
P
out;f
P
out;nf
(41)
where P
diss,in
and P
diss,out
are the two contributions to the
dissipated power (i.e., converted to heat) in the input and
output sections of the PA stage, respectively. In this equa-
tion, the input power can be thought to be entirely dissi-
pated in the device input (i.e., P
diss,in
EP
in
), therefore
leading to
P
DC
=P
diss;out
P
out;f
P
out;nf
(42)
which represents a power balance equation related to the
output circuit only. The result in (42) is valid at the lower
microwave frequencies and, in general, if the active device
internal feedback can be neglected. In the output circuit,
device current and voltage may be expressed by their
Fourier series expansion
i
D
t ( ) =I
0
o
n=1
I
n
.
cos n
.
o
.
t x
n
( ) (43)
v
DS
t ( ) =V
ds;DC
o
n=1
V
n
.
cos n
.
o
.
t c
n
_ _
(44)
where x
n
and c
n
are the phases of the nth output current
(I
n
) and voltage (V
n
) harmonic components, respectively.
On the basis of such assumptions, harmonic termina-
tions to the output of the active device can be dened as
the ratio of the respective voltage and current harmonic
components:
Z
L;nf
=
V
n
.
e
jc
n
I
n
.
e
jx
n
=
V
n
I
n
.
e
j c
n
x
n ( )
=
V
n
I
n
.
e
jj
n
(45)
obtaining, for the bias DC power, the power dissipated on
the device and the power delivered to fundamental and
harmonic frequencies:
P
DC
=V
ds;DC
.
I
0
(46)
P
diss
=
1
T
.
_
T
i
d
t ( )
.
v
ds
t ( )
.
dt =P
DC
P
out;f
o
n=2
P
out;nf
(47)
P
out;nf
=
1
2
.
V
n
.
I
n
.
cos j
n
_ _
n=1; 2. . . (48)
Utilising these expressions in the definition of drain
efciency, it follows that
Z =
P
out;f
P
DC
=
P
out;f
P
diss
P
out;f

o
n=2
P
out;nf
(49)
From this expression, the necessary and sufcient condi-
tion needed to reach a complete conversion of DC power
into useful fundamental power (i.e., 100% efciency) is
that both the following conditions be fullled simulta-
neously:
P
diss
=0 (50)
o
n=2
P
out;nf
=
1
2
.
o
n=2
V
n
.
I
n
.
cos j
n
_ _
=0 (51)
More explicitly, on one hand dissipated power on the ac-
tive device has to be nulled, and this can be accomplished
by correctly shaping voltage and current waveforms in or-
der to ensure a null overlapping; thus the current on the
device must vanish for a nonzero voltage and vice versa.
Moreover, harmonic active power has to be nulled, thus
implying either the zeroing of current/voltage harmonic
components (V
n
=0 or I
n
=0 or both) or their proper phas-
ing (j
n
=p/2). Therefore, in low-frequency applications,
assuming an innite number of controllable harmonic ter-
minations, two possibilities are available to ensure the
fulllment of the latter condition:
*
Class F [5658] or inverse class F [59] strategies, in
which V
n
I
n
=0 for n41. Note that these are idealized
approaches since voltage and current harmonic com-
ponents, which in a real device are related by load
impedances as in Eq. (45), and are considered sepa-
rately. The class F strategy has been successfully
applied in class AB [61,62] amplier stages. Never-
theless, its application may lead to poor efciency
performances if class C or deep-AB (near-B) biasing
classes are adopted [60].
*
Class E or switched-mode strategies, in which j
n
=
p/2 for n41. The active device is operated as a
switch, and closed-form design expressions are avail-
able [63]. In such conditions, the stage acts more as a
DC/RF converter rather than an amplier. In this
case, the power gain of the stage is not controlled and
specied during the design phase; it is a specication
to be fullled by a separately designed driver circuit,
using information about the input port characteris-
tics of the power transistor to be driven. Moreover,
nothing is said about the input network, except that
the input voltage waveform must properly drive the
device to operate as a switch (i.e., be deeply pinched
off and saturated). In any case, an inspection of ex-
pression (49) suggests two possible directions for the
efciency optimization: (1) either the maximization of
device output power (for the same DC supply) or (2)
the minimization of the dissipated power (considered
as the sum of internally dissipated power P
diss,out
and
harmonic active power P
out,nf
).
The designers task is therefore in the appropriate
shaping of current and voltage waveforms, consisting in
the selection of appropriate harmonic terminations lead-
ing to the fulllment of the conditions described above.
5. HARMONIC TUNING APPROACHES
Harmonic tuning approaches for microwave PA design
represent indeed a hot discussion topic for both the aca-
demic and the industrial communities given the benecial
effect on stage efciency that can be obtained by a proper
selection of harmonic terminations. Contributions in this
eld range from experimental observation of performance
levels [6668] to introduction of novel topologies as in the
case of harmonic reaction ampliers [64,65], from system-
atic investigations of experimental performances [70] to
design methods for special harmonic tuning congura-
tions [69].
In fact, however, if the operating frequency enters the
microwave region, both approaches cited in the previous
section (viz., the standard class F and the switching-mode
class E) exhibit a degradation in performance. For in-
stance, actual class F ampliers are usually designed
making use of two or three idlers only to control second
and third output harmonic impedances. As frequency in-
creases (e.g., 420 GHz), the control of both the second and
third harmonic output impedances becomes troublesome,
since the active-device output capacitive behavior practi-
cally short-circuits higher components, not allowing the
desired waveshaping. Moreover, for low-voltage applica-
tions, a class F strategy is not the best solution, since dif-
ferent methodologies (based on second-harmonic output
impedance tuning) have demonstrated better performance
[71]. On the other hand, the switching mode operation of
the active device, necessary to implement class E strategy,
is not feasible in microwave communication systems, since
it requires that the PA operate in saturated conditions,
thus often increasing IMD levels and potentially affecting
active-device reliability.
Circuit complexity issues and effectiveness of the har-
monic control therefore suggest the control of the rst two
harmonic terminations for a PA stage, leaving higher ter-
minations dominated by the shunting effect of the output
device capacitive behavior. Assuming purely resistive ter-
minations for the sake of simplicity, the drain voltage
waveform can be expressed as [72]
v
DS
(t) =V
ds;DC
V
1
.
[cos(o
.
t) k
2
.
cos(2
.
o
.
t)
k
3
.
cos(3
.
o
.
t)]
(52)
where
k
2

V
2
V
1
k
3

V
3
V
1
(53)
and
V
1
=R
1
.
I
1
V
2
=R
2
.
I
2
V
3
=R
3
.
I
3
(54)
Hence, for a given device drive and bias, that is, for given
current source harmonic components (I
n
), the designers
task is in the selection of proper terminations at funda-
mental (R
1
) and harmonic frequencies (R
2
,R
3
) maximizing
the fundamental component (V
1
) of the swing experienced
by v
DS
(t), with the constraints dictated by the device
physical limitations (knee and breakdown voltages). This
is equivalent to the determination of the values of the pair
(k
2
,k
3
), which, for given physical limits and therefore for a
given voltage swing, allows maximization of the V
1
com-
ponent. An example of this effect is reported in Fig. 25,
utilising, for simplicity, a third-harmonic component only
summed to the fundamental one. In this scenario, for a
xed swing, dictated in this case by the bias voltage V
ds,DC
and V
k
physical limitation, the introduction of a third-
harmonic component actually increases the fundamental
frequency amplitude beyond the value of the TL case (viz.,
V
ds,DC
V
k
in this case).
Analytically, this is accomplished dening the voltage
gain function d by the ratio of the resulting fundamental
voltage component normalized to the unmanipulated case
(the tuned load value, numerically coincident with the
distance between the drain bias and the closer physical
limit):
d k
2
; k
3
( )
V
1
V
1;TL
(55)
The voltage gain d is plotted in Figs. 26 and 27 as a function
of (k
2
,k
3
) in a three-dimensional and contour plot format.
As it is possible to note, the d function exhibits a
clear maximum for (k
2
,k
3
) =( 0.55,0.17) whose value is
V
ds,DC
V
ds,3.f
(t)
V
ds,f
(t) v
DS
(t)
V
k
2

0 2
ct
Figure 25. Voltage waveform v
DS
(t) (red curve) obtained via the
introduction of a third-harmonic component (dashed line) summed
to a fundamental frequency component (continuous line) exceed-
ing the physical limitation (V
k
). (This gure is available in full
color at http://www.mrw.interscience.wiley.com/erfme.)
1.8
1.6
1.4
1.2
0.8
0.6
0.4
0.2
1
1
0.5
0
0.5
0.5
0
0.5
1
1 1
k
3 k
2
Figure 26. Voltage gain function d versus k

2
and k
3
.
approximately 1.62. The function is higher than unity
(i.e., use of the second and third harmonics effectively in-
creases fundamental drain voltage harmonic component)
within the marked region in Fig. 26.
The increase in fundamental frequency output voltage
component obtained via a proper selection of the (k
2
,k
3
)
pair, is directly reected in a corresponding increase in
power performances of the PA stage. In fact, under the
hypothesis that the controlled source is not affected by the
device harmonic terminations (i.e., its harmonics compo-
nents I
n
remain unchanged), the output power, large-sig-
nal gain, drain, and power-added efciency of the stage
become
P
out
k
2
; k
3
( ) =P
out;TL
.
d k
2
; k
3
( ) (56)
G k
2
; k
3
( ) =G
TL
.
d k
2
; k
3
( ) (57)
Z k
2
; k
3
( ) =Z
TL
.
d k
2
; k
3
( ) (58)
Z
add
k
2
; k
3
( ) =Z
add;TL
.
d k
2
; k
3
( )
.
G
TL
1=d k
2
; k
3
( )
G
TL
1
_ _
(59)
The harmonic terminations that actually shape output
voltage with a given selection of the (k
2
,k
3
) pair are there-
fore
R
1
=d k
2
; k
3
( )
.
R
TL
(60)
R
2
=d k
2
; k
3
( )
.
k
2
.
I
1
I
2
.
R
TL
(61)
R
3
=d k
2
; k
3
( )
.
k
3
.
I
1
I
3
.
R
TL
(62)
In the derivations above, clearly the role of the device out-
put conductance (g
ds
) is not evidenced; the latter clearly
poses an upper limit to the impedance values that can
effectively be presented to the active-device internal
current source. The effect of actual device output conduc-
tance leads to a modication of the approach presented
here [60].
Furthermore, the use of even-harmonic components
(the second one in this case) actually breaks the natural
symmetry of the voltage waveform; the desired attening
while approaching the physical limitation is accompanied
by a peaking in the remaining part of the RF cycle, which
can actually reach the opposite physical limitation. This
effect has to be carefully controlled and is quantitatively
evaluated via a voltage overshoot function [72].
However, several important cases can be extracted
from Fig. 26. The rst one consists in the previously dis-
cussed TL condition, corresponding to the origin of the
(k
2
,k
3
) plane, where d(0,0) =1. If a single voltage harmonic
is controlled at a time, terminating the remaining one on a
short-circuit termination, two possibilities arise. The rst
one is obtained for k
2
=0, consisting in the control of third
harmonic only: this situation corresponds to the high-fre-
quency class F design approach [60,73], which is repre-
sented in Fig. 26 by the k
3
axis. In this harmonic control
scheme the maximum value for the voltage gain function
is obtained for d(0, 0.167)E1.15. In this case, therefore,
the use of a high-frequency class F approach results in an
increase of 15% in output power, efciency, and large-sig-
nal gain over the tuned-load condition. Maximally at
waveforms are a special case of such design approach [73].
The second possibility consists in setting k
3
=0, there-
fore controlling the second-harmonic component only [71].
Moving along the k
2
axis in Fig. 26 results in a maximum
value for d( 0.35, 0)E1.41 and in the corresponding in-
crease in power performances. The results attainable via
harmonic tuning of the second and third output harmonic
components are summarized in Table 4.
It should be noted, however, that the potential increas-
es in the PA power performances are obtained only if the
proper loading is imposed on the active-device output.
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
1.6
1.5
1.4
1.4
1.3
1.3
1.3
1.2
1.2
1.2
1.1
1.1
1.1
1.1
1
1
1
1
0.9
0.9
0.9
0.9
0.9
0.8
0.8
0.8
0.8
0.8
0.8
0.7
0.7
0.7
0.7
0.7
0.6
0.6
0.6
0.5
0.5 0.4
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
k
3
k
2
Figure 27. Constant voltage gain function d
contours in the (k
2
,k
3
) plane.
This is not always possible. In fact, with reference to
Fig. 22, output current harmonic components, generated
via the clipping effect of the device pinchoff on a sinusoidal
input drive, show a different behavior as a function of the
current conduction angle. In particular, focusing on the
region aA[p,2p] (i.e., from class B to class A bias), the sec-
ond harmonic current component is always positive, while
the third one is negative (clearly taking the fundamental
one as a reference). Since actual terminations are passive
(i.e., R
i
Z0, i =1,...,3), this consideration implies that pos-
itive k
2
and negative k
3
are feasible:
k
2
=
V
2
V
1
=
R
2
.
I
2
R
1
.
I
1
_0 k
3
=
V
3
V
1
=
R
3
.
I
3
R
1
.
I
1
_0 (63)
Therefore, if current clipping effects only are exploited,
only third-harmonic (high-frequency class F) tuning is di-
rectly possible; the simultaneous use of second- and third-
harmonic components implies, in fact, from Fig. 26, that
k
2
o0 and k
3
40. An improper use of harmonic tuning may,
on the other hand, bring detrimental results on overall
power performances, as exemplied in Fig. 28, where the
case in Fig. 25 has been repeated with an incorrect phas-
ing of the third-harmonic component.
Clearly the possibility to generate the properly phased
current harmonic components is related to the driving
waveform, up to now assumed to be purely sinusoidal. In
fact, even if a perfectly sinusoidal drive is assumed, the
drain current waveform actually reects not only the
hardlimiting clipping phenomena due to the device pinch-
off but also the effect of local mild nonlinearities, slightly
modifying the truncated sinusoid assumption adopted
up to now [74]. On the other hand, the driving waveform
can be properly selected, eventually leading to quartic or
rectangular input signals [75], allowing in this way the
application of harmonic tuning strategies in limited rang-
es for the drain current conduction angle. The alternative
approach (i.e., the exploitation of the device input nonlin-
earities and therefore the input terminations to the active
device) has been the subject of a series of investigations,
leading to experimental and theoretical studies, with fre-
quently contrasting results [7680]; the input nonlinear-
ities of the device itself can be used to generate the
harmonic of interest [81], with effects that can be effec-
tively predicted using Volterra-like approaches [82], lead-
ing to an harmonic control scheme including input and
output terminations, schematically depicted in Fig. 29.
The possibility of effectively controlling input and out-
put harmonic control has been experimentally veried by
several authors. In Fig. 30 the measured results are re-
ported for the power performances of four single-device PA
stages, designed utilizing TL, second-harmonic tuning,
third-harmonic tuning, and second/third-tuning com-
bined. The PA stages, realized in this case using hybrid
technology, operate with a 5GHz fundamental frequency,
with a 5% operating bandwidth, and adopt a power 1-mm
MESFET device from Alenia Marconi Systems.
A nal consideration regarding the linearity perfor-
mances of harmonic-controlled power ampliers has to be
drawn. The high-efciency tuning of PA stages designed
according to a harmonic control scheme leaves in fact
some open questions on the resulting linearity. The proper
tradeoff among such requirements is not clearly evidenced
to date, even if experimental results demonstrate that a
high-efciency tuning does not necessarily imply detri-
mental linearity effects [83]. Moreover, harmonic injection
techniques have been proposed to optimize linearity
performance [84]. In any case however, proper subsystem
linearization techniques can be adopted in the case of
harmonic tuned PA stages, regardless of the adopted
modulation scheme, using predistortion [85], feedback
[86], or feedforward [87] approaches. Clearly a system-
level lineariation approach implies a decrease in efciency
performance and often a major increase in subsystem
complexity.
6. SWITCHING-MODE POWER AMPLIFIERS
In this category, a number of different approaches can be
grouped, all characterized by the assumption that the ac-
tive device (or devices) utilized in the power stage is (are)
operated as switches, commuted between the two com-
pletely ON (i.e., short-circuit-like) and completely OFF
V
DS
(t)
V
ds,3f
(t)
V
ds,f
(t)
V
ds,DC
V
k
2 2 0
ct
Figure 28. Voltage waveform v
DS
(t) (red curve) exceeding the
physical limitation (V
k
) obtained via the introduction of a third-
harmonic component (dashed line) summed to a fundamental fre-
quency component (continuous line). (This gure is available in
full color at http://www.mrw.interscience.wiley.com/erfme.)
Table 4. Maximum Theoretical Improvements in Drain Efciency over the Tuned Load Approach Attainable with Tuning
Second-, Third-, and Second/Third-Harmonic Output Terminations
Controlled
Frequencies k
2
k
3
d Z Improvement (%)
F Tuned load 0 0 1 0
f, 3f High-frequency class F 0 0.17 1.15 15
f, 2f 2nd-harmonic tuning 0.35 0 1.41 41
f, 2f, 3f 2nd/3
rd
-harmonic tuning 0.55 0.17 1.62 62
(i.e., open-circuit-like) states. As compared to the ap-
proaches described previously, in which the active device
is basically represented as a controlled current source
replicating with good delity the input driving stimulus,
switching-mode PAs, ideally driving the active device in
the ON and OFF states, indeed exhibit highly nonlinear
characteristics. Such inherent lack of linearity is not a
problem for constant-envelope modulations, for which a
switched-mode operation may infer the great benet of
high-efciency operation of the transmitter, while pre-
venting their use if amplitude-modulated signals are fed
into the amplier. On the other hand, techniques consist-
ing in the envelope elimination and restoration [88,89],
outphasing [90,91], or other high linearity modulating
schemes may alleviate the linearity drawback.
A further consideration applies to the large majority of
switched-mode PA strategies. As they are based on the
assumption of an active device operated as a switch, their
effective implementation and resulting performance levels
rely largely on the validity of such an assumption. It is
clear that the switching behavior of the device is strictly
possible only when the effect of device parasitics is negli-
gible, up to moderately high frequencies, including the low
microwave range. The class D and class E ampliers will
be briefly addressed in the following text. For a more com-
plete listing of switched-mode PAs and the relevant refer-
ences, see Chap. 14 in Ref. 92. The class D amplier,
introduced by Baxandall [93], is composed, in its basic
form, of a pair of active devices whose output is fed to an
output tuning circuit, as depicted in Fig. 31.
In such a basic conguration, referred to as comple-
mentary voltage switching, the two devices act as a two-
pole switch, whose output voltage is tends to be an ideal
square waveform, whose harmonic components are effec-
tively ltered by the series-tuned resonator formed by L
and C in Fig. 31, therefore delivering a sinusoidal signal
to the output load R. The two transistors are driven, in
fact, via the transformer, to deliver two output (drain)
1801 out-of-phase currents T
1
and T
2
, which pass alter-
nately between ON and OFF states. Assuming a 50% duty
19
20
21
22
23
24
25
26
12 14 16 18 20
P
in
(dBm)
P
out
(dBm)
Tuned load
Class F
2
nd
harmonic tuning
2
nd
& 3
rd
harmonic tuning
0
10
20
30
40
50
60
70
12 14 16 18 20
P
in
(dBm)
p
add
(%)
Tuned load
Class F
2
nd
harmonic tuning
2
nd
& 3
rd
harmonic tuning
(a) (b)
Figure 30. Output power (a) and power-added efciency (b) measured performances of four single-
device PA stages designed utilizing different harmonic tuning strategies.
IN
Z
S1
Z
S2
Z
S3
@f
@2f @3f @nf n4
Z
L1
V
gs,DC
V
ds,DC
Z
L2
Z
L3
@f @2f @3f @nf n4
OUT
Figure 29. Harmonic control scheme including input and output terminations.
cycle, the voltage V
x
in Fig. 31 may be represented by a
square waveform, whose Fourier expansion is given by
V
x
(t) =
V
DD
2
.
1
4
p
.
sin(2pf
.
t)
_

4
3p
.
sin(3
.
2pf
.
t)
4
5p
.
sin(5
.
2pf
.
t)
_
(64)
This voltage waveform is applied to the series output RL
C network. If the LC series-resonant circuit is tuned to
the fundamental switching frequency of operation f and
exhibits an acceptable quality factor, the fundamental fre-
quency component of V
x
is directly obtained at the output
load R, thus obtaining a purely sinusoidal output current:
i
out
t ( ) =
2
.
V
DD
p
.
R
.
sin 2pf
.
t ( ) (65)
Under the hypothesis described above, the drain currents
in the two transistors (i
T
1
andi
T
2
) are therefore two half-
sinusoids, as in Fig. 31, and, from inspection of the previ-
ous expression, the peak drain current value for both
transistors is therefore
i
T
1
;max
=i
T
2
;max
=
2
.
V
DD
p
.
R
(66)
while the output power delivered to the load is
P
out
=
1
2
.
2
.
V
DD
p
.
2
.
V
DD
p
.
R
=
2
.
V
2
DD
p
2
.
R
(67)
and the DC input power (i.e., from the power supply) is
given by the product of the applied voltage V
DD
times the
mean value of the current through T
1
P
DC
=V
DD
.
2
.
V
DD
p
2
.
R
=
2
.
V
2
DD
p
2
.
R
(68)
thus obtaining a theoretical 100% drain efciency:
Z
class D
=
P
out
P
DC
=1 (69)
The series-resonant LC circuit actually provides the ad-
equate harmonic ltering, which can be further improved
if the circuit quality factor is not adequate by a subsequent
lowpass lter. The bypass capacitor C
0
actually has a
DC ltering purpose, which can be further improved by
a series choke inductor directly connected to the bias
supply V
DD
.
Maintaining the same operating principle, transformer
coupling congurations may be applied, introducing cen-
ter-tapped output transformers. The dual conguration,
that is, the current-switching one, can be implemented as
well [92]. The idealized operation of the class D amplier,
however, has to be complemented by some practical con-
straints arising from actual device operation. In fact, the
device ON-state resistance is in series with the output load
(regardless of the specific complementary or transformer-
coupled voltage switching conguration), thus decreasing
the effective output voltage and efciency. The T
1
transis-
tor is, on the other hand, ungrounded, posing serious
problems for the actual implementation of the scheme at
high frequencies, for both driver requirements and con-
nection parasitics [94]. Moreover, since actual transistors
exhibit a parasitic output capacitance that sums up to the
circuit stray capacitances, the result is that, for each RF
cycle, the stored energy in such reactance is dissipated
during the ON state in the transistor itself. Such energy,
whose amount is approximately proportional to the sus-
ceptance associated with this parasitic [95], actually pro-
duces an increase in the input (DC) power required to the
proper operation of the stage for the same output power,
with the resulting decrease in stage efciency.
The abovementioned parasitic output capacitance is
one of the major factors responsible for nite transistor
switching times, which up to now have been assumed as
vanishingly small. As a result, nite transitions between
ON and OFF states produce a net decrease in stage drain
efciency [92].
t
t
t
t
V
DD V
x
V
out
V
out,max
I
T1
I
T1,max
I
T2
I
T2,max
T
1
C
0
V
DD
T
2
V
x
i
T2
i
out
i
T1
R
C L
OUT
Figure 31. Schematic representation of a ba-
sic class D amplier and basic waveforms.
A very popular and effective switching mode PA cong-
uration is the class E amplier. Introduced and patented
by Sokal in 1975 [96,97], the class E conguration has
been extensively studied [54,98], receiving, with expira-
tion of the issued patent, renewed interest [63,99102].
The basic circuit conguration of a class E PA is depicted
in Fig. 32, where the active device is represented by a
switch.
The capacitor C
p
includes both device output parasitic
reactance and circuit stray capacitances and is eventually
increased by an external capacitor. Its value is not such,
however, to short-circuit all voltage harmonics. Ideal
current and voltages on the active device, represented in
Fig. 33, are such that dissipated power is nulled. If a
Fourier analysis is further performed on such waveforms,
the phase difference between each current and voltage
harmonic component is equal to p/2, therefore fullling
both conditions indicated by (50) and (51) for efciency
maximization.
Analysis of the circuit in Fig. 32 can be carried out
considering a 50% duty cycle, together with an ideal
switch operation (i.e., assuming a null ON-state switch
resistance and an innite OFF-state one). Furthermore, the
RF choke is assumed ideal, and no losses are accounted
for the external elements in the circuit, the active power
is dissipated in the external load R only. The current
owing in the external load may be assumed to be a
purely sinusoidal one thanks to the series resonator
formed by the L
s
C
s
pair, tuned to operate at the circuit
fundamental frequency and with an innite quality
factor.
The two currents i
L
(t) and i
DC
(t) are therefore always a
sinusoidal one and a constant one, respectively, regardless
of the switch position. The current owing through the
parallel combination of the active device (switch) and the
capacitor C
p
is therefore always a DC-offset sinusoid.
Clearly, when the switch is ON, all such current ows
through it, and when the switch is OFF, the offset sinusoi-
dal current passes through the capacitor only:
i
DC
t ( ) =i
DC
i
L
t ( ) =i
L
.
cos o
.
t ( )
=i
d
t ( ) i
C
p
t ( ) =i
DC
i
L
.
cos o
.
t ( )
(70)
The onset of the switch ON state will be determined there-
fore from the zero crossing of such DC-offset current. An
example of the resulting switch current is shown in
Fig. 34. The corresponding voltage v
ds
(t) will be obtained
by the integration, over the OFF state of the switch period,
of the capacitor charge
v
DS
t ( ) =v
Cp
t ( ) =
1
C
p
.
_
i
Cp
t ( )
.
dt (71)
obtaining the corresponding voltage waveform depicted in
Fig. 34.
Closed-form design expressions may be derived for the
circuit components in Fig. 32. The latter may be found in
Ref. 63, and the step-by-step derivation of voltages and
currents may be found in Refs. 92 and 94 following a time-
domain approach.
With reference to Fig. 32, current and voltage in the ON
state of the switch are expressed as
i
D
(t) =i
DC
.
1 a
.
sin ot f ( ) [ ]
v
DS
t ( ) =0
(72)
on off
t
i
D
(t) i
D
(t) i
D
,v
DS
v
DS
(t)
Figure 33. Idealized current (i
D
) and voltage (v
DS
) waveforms.
t
on off
on off
t
i
D
(t)
v
DS
(t)
i
dc
al
DD
i
L
+i
dc
[/c
Figure 34. Switch current and voltage.
V
DD
L
s
C
s
C
p
v
DS
(t)
i
dc
(t)
i
L
(t)
i
D
(t)
i
Cp
(t)
L
RFC
+
s
active
device
R
OUT
Figure 32. Circuit schematic of a class E power amplier.
while in the OFF state as
i
D
t ( ) =0
v
DS
t ( ) =
1
C
p
.
_
t
0
i
D
t ( ) dt
=
i
DC
o
.
C
p
o
.
t a
.
cos o
.
t f ( ) cos f ( ) [ ]
(73)
Imposing a null switch (capacitor) voltage at the onset of
the ON state, together with its rst derivative (optimum
operation)
v
ds
t ( )
t =(T=2)
=0;
dv
ds
t ( )
dt
t =(T=2)
=0 (74)
it is possible to obtain
a=
1
p
2
4
_
; f= tan
1
2
p
_ _
; V
DD
=
i
DC
p
.
o
.
C
p
(75)
From the expression derived, noting that the dissipated
power is null on the active device (neglecting the switch
ON-state resistance), the output power P
out
equals the in-
put supply power P
DC
, given by
P
DC
=P
out
=V
DD
.
i
DC
=p
.
o
.
C
p
.
V
2
DD
(76)
bringing the efciency of the class E PA stage to unity. If a
nonzero ON resistance for the switch (R
on
) is considered,
the stage efciency becomes
Z =
1
p
2
o
.
C
p
.
R
on
1
p
2
4
_ _
.
1 p
.
o
.
C
p
.
R
on
_ _
2
(77)
An alternative derivation, based on a phasor-based fre-
quency-domain approach, can be found in Ref. 101. In this
case the result can be expressed in terms of harmonic load
impedances Z
L
, namely, the total impedance loading the
active device, given by
Z
L
=
0:28015
oC
p
e
j49:0524
at f
o at nf ; n > 1
_
_
(78)
where the parallel capacitance C
p
is related to the rf out-
put power P
out
by
C
p
=
P
out
p
.
o
.
V
2
DD
(79)
This expression implicitly poses an upper limit to the
maximum operating frequency of the class E amplier.
In fact, once the capacitor value is determined from
optimum circuit conditions, the maximum operating
frequency can be obtained as
f
max
=0:0177
.
I
max
V
DD
.
C
p
(80)
where
I
max
=a
.
I
DD
i
DC
(81)
is the maximum value of the current owing into the ac-
tive device. Furthermore, the maximum voltage across the
switching device obtained by expression (71) may rise well
beyond the DC supply, being given by
v
ds;max
- 3:562
.
V
DD
(82)
For safe and reliable operation, a value such as this must
be kept well below the device breakdown region. On the
other hand, the limitation imposed by (82) actually de-
creases the maximum output power that can be achieved
utilizing the class E approach if compared to conventional
class AB linear PA design, forcing a much lower bias. In-
deed, the class E scheme exhibits several advantages, in-
cluding the evident scheme simplicity, together with the
high efciency attainable.
Constraint imposed on the output circuit high-Q oper-
ation may be severe, however, if high-frequency operation
is concerned. The result of a low quality factor for the
output series resonant circuit will be a nonzero harmonic
current owing into the output, implying a nonoptimum
drain voltage waveform, with possible ringing phenome-
na. Nevertheless, since the class E amplier has to be over
driven by a driver stage well into device compression re-
gions, its large-signal gain may be well below the level
that can be attained by a comparable linear power ampli-
er (say, 35 dB [94]). Given the typical frequency rolloff of
power devices, this poses an intrinsic frequency limitation
to the applicability of the class E scheme. The frequency
limit is further decreased by the intrinsic reactive behav-
ior of the device output (dominated by the device output
capacitance) that actually restrict the possible values for
the capacitance C
p
in the class E scheme, as suggested
from (80).
7. LOAD-PULL TECHNIQUES
As widely noted previously, PA performance is determined
mainly by the terminations imposed to the active device,
mainly at its output but also considering its input nonlin-
earities. This consideration naturally leads us to consider
the possibility of experimentally characterizing the device
power performance by means of a setup that actually
modies device terminations and records the related per-
formance [106]. The loads are therefore pulled to test
device power performance levels at both the device output
(load-pull) and input (source-pull) ports. Such investiga-
tion is carried out at fundamental frequency (fundamental
load pull, i.e., controlling the loading conditions at funda-
mental frequency) or even at harmonic frequencies
(harmonic load pull, i.e. controlling the loading conditions
at one or more harmonic frequencies [107110]).
The result of such a measurement campaign, creating a
database of power performances corresponding to a given
combination of terminations, driving power, biases, and
frequency is typically organized in graphical form. Usual-
ly constant-parameter curves are drawn on the corre-
sponding load Smith chart, in order to immediately
detect optimality condition. For example, in Fig. 35 the
constant output power and drain efciency curves at 1dB
and 2 dB gain compression respectively are shown in the
load reection coefcient plane.
The load pull therefore gives directly all that is neces-
sary for optimum load selection, as the result of a tradeoff
between the performance behaviors of interest. The PA
designer therefore needs an automated load-pull measure-
ment bench and physically the selected device, that must
be available in the appropriate form, to avoid tedious and
risky deembedding procedures. It should be noted, how-
ever, that load-pull systems may be fruitfully employed
not only as a design tool but also for active-device nonlin-
ear model extraction/verication setup, where the model
can be compared and tested using data that have not been
used for its extraction.
Before describing the details of the load-pull character-
ization setup, we present a technique due to Cripps
[103,104] and further rened by Kondoh [105], allowing
determination of load-pull output power contours for a
class A biased active device. The technique, based on a
series of approximations performed on both the active de-
vice and the external circuit, is widely adopted and typi-
cally forms the basis for successive investigation (for both
load-pull characterization and actual PA design). The ac-
tive-device output is represented through its voltage-con-
trolled current source I
d
, assuming the remaining
components and parasitic elements grouped together in
the network between Sections 1 and 2 in Fig. 36, eventu-
ally constituting device mounting parasitics.
A piecewise linear description of the controlled source
is adopted, as outlined in Section 3, and a class A bias is
selected (corresponding to I
ds,DC
=I
max
/2, V
ds,DC
=V
DD
) as
in Fig. 37.
The optimum load Z
L,opt
imposed at the current source
terminals in Fig. 36, simultaneously maximizing current
and voltage swings, already determined in Section 3 (see
1 2
Z
L,ext
Z
L,ext
Z
L
I
d
Active device
and mounting
parasitics
Figure 36. Simplied scheme of the active-device output and
parasitics for application of the techniques in Ref. 103.
2 5
j5
j5
j2
j2
j1
j1
j0.5
j0.2
8 %
6 %
4 %
2 %
p
max
2
0
5
j5
j5
j2
j2
j1
j1
j0.5
j0.2
2.0 dB
1.5 dB
0.5 dB
P
out,max
1.0 dB
Figure 35. Typical load-pull contours for constant output power and drain efciency at 1 and
2dB gain compression, respectively.
0
I
max
I
d
[A]
V
k
V
DD
V
ds
[V]
V
gs
=V
po
V
gs
=V
bi
Sub-optimum complex load
Optimum load
R
L,opt
Figure 37. Piecewise linear representation of the device con-
trolled current source, with superimposed optimum and subopti-
mum load curves corresponding to a resistive and complex load.
Table 3) as R
A
, is a purely resistive one. Its value and the
corresponding power delivered by the source are given by
Z
L;opt
=R
L;opt
=2
.
V
ds;DC
V
k
I
max
=R
A
;
P
L;opt
=
V
ds;DC
V
k
_ _
.
I
max
4
(83)
If the source is loaded by a complex termination with the
same magnitude Z
L;opt
, a lower active output power re-

sults as a consequence of the phase difference between
voltage and current waveforms. The resulting load curve,
that is, the instantaneous plot of current and voltage at
the source terminals, is an ellipse, as indicated in Fig. 38.
A lower resistance load (R
L,I
oR
L,opt
) will cause the de-
vice current hard limitations (viz., pinchoff and fully
opened channel) to limit the current swing, occurring for
a reduced voltage swing and resulting in a reduced output
power. The same applies for a higher resistive loading
(R
L,V
4R
L,opt
) in which the voltage limiting mechanism
(viz., the device ohmic region) will occur for a reduced
current swing. The resulting output power for the current-
(P
L,I
) and voltage- (P
L,V
) limited operation are easily com-
puted as follows:
P
L;I
=
R
L;I
R
L;opt
.
P
L;opt
; P
L;I
=
R
L;opt
R
L;V
.
P
L;opt
(84)
If a reactive part is added in series to the lower-resistance
load R
L,I
, the active output power will remain at same
level, causing, on the other hand, the load curve to trans-
form into an ellipse, as depicted in Fig. 38. The same ap-
plies to the voltage-limited case if a susceptance is added
in parallel to the selected load R
L,V
. Clearly in both cases,
the output power will be constant until the resulting load
curve will incur the voltage (for the current-limited case)
or current (for the voltage-limited case) physical limit.
In order to draw a X-dB contour plot (i.e., the locus, on
the Smith chart, characterized by an output power given
by P
L,opt
XdB), the resistive values in the current- and
voltage-limited cases (R
L,I
, and R
L,V
respectively) are
found from (84):
R
L;I
=R
L;opt
.
10
(X=10)
; R
L;V
=R
L;opt
.
10
(X=10)
(85)
The X-dB contour can be drawn therefore following con-
stant-resistance (with a value R
L,I
in the current-limited
case) and constant-conductance (with a value R
L,V
in the
voltage-limited case) loci on the Smith chart. The proce-
dure can be repeated for a number of constant-output
power loci, resulting in the typical contour plot in Fig. 39,
where a 1 dB step has been adopted.
The load at the current source terminals must now be
transformed into an external load, thus including device
reactive elements and parasitics, together with mounting
ones. To this purpose, standard Smith chart transforma-
tions apply, moving the reference plane from Section 1 to
Section 2 in Fig. 36, and resulting in the nal plot on the
right side of Fig. 39.
Clearly a series of strong approximations have been
applied to get the nal result. Nevertheless, the results of
the procedure are often in very good agreement with mea-
sured load-pull contours, justifying the popularity and
widespread use of the Cripps technique. However, if a
complete characterization and design charts involving the
actual nonlinear nature of the active device are necessary,
a load-pull measurement setup has to be used. The mea-
surement systems adopted for load-pull setup may differ
greatly. They range from relatively simple scalar setups,
based on power meters and/or scalar network analysers,
up to vector receivers. In the former case scalar informa-
tion is available only on the device power performance, but
their cost and relative simplicity is indeed a strong point.
Alternatively, vector information can be gained from
the use of vector network analyzers (VNA, or from six-port
approaches [111]) or by means of time-domain measure-
ment systems [adopting, e.g., a microwave transition
analyzer (MTA)]; in this case more complete vector infor-
mation on the device can be extracted, together with high-
er accuracy on the results, due to the vector calibrations
that can be adopted in the case of VNA (e.g., [112117]) or
with time-domain waveform capabilities in the case of a
0
I
max
I
d
(A)
V
k
V
DD
V
ds
(V)
V
gs
= V
po
V
gs
= V
bi
Complex load
Resistive load
R
L,V
0
I
max
I
d
(A)
V
k
V
DD
V
ds
(V)
V
gs
= V
po
V
gs
= V
bi
Complex load
Resistive load
R
L,I
(a) (b)
Figure 38. Current- (a) or voltage-limited (b) load lines.
MTA [118,119]. In both cases, however, such improved ca-
pabilities are obtained at the expense of an higher bench
complexity and cost.
The load to be synthesized can be realized by means of
passive systems (passive load, resulting in passive load-pull
systems); the latter consist in mechanical (typically realized
via a slotted transmission line with slugs) or electrical (i.e.,
PIN-diode-based) tuners [120122]. A typical passive scalar
load-pull system is schematically depicted in Fig. 40.
The adopted single- or double-slug tuners have to be
accurately precharacterized by means of a VNA, or alter-
natively the scheme can be extended to include a series of
electromechanical switches connecting alternatively the
tuners to the DUT or to the VNA input for on-site imped-
ance measurements. Tuners are generally GP-IB-con-
trolled, and tuner position repeatability is guaranteed by
precision stepped motors. Moreover, above all for mechan-
ical tuners, the power-handling capability sufces for most
power applications.
On the other hand, the synthesized reection coef-
cient is limited in magnitude by the unavoidable tuner
and setup (cables, on-wafer probes, etc.) losses. Such lim-
itation implies that highly reective loads cannot be
realized and presented to the active device. Especially
for high-power devices this precludes the investigation
of maximum power regions for the output fundamental
load, since the optimum termination (approximately)
results from the ratio of the bias voltage to device maxi-
mum current [explicitly for the class A, PA (33)]. This
effect is clearly amplied if higher and higher frequencies
(and hence higher losses) are explored. Moreover, if an
harmonic load pull has to be performed, where optimum
harmonic terminations may be typically close to purely
reactive ones (see Section 5), a further drawback arises.
In order to overcome the problem, several solutions
have been proposed [123], based on the use of either
tuner prematching networks (at the device level) or pre-
matching tuners (at the measurement system level; see
Fig. 41).
In this way, higher reection coefcient magnitudes are
attainable. Nevertheless, especially in the case of pre-
matching tuners, precharacterization is indeed a problem,
Power
sensor
Power
sensor
Power meter
Tuner
I
S
Tuner
I
L
DUT
Figure 40. Schematic representation of a
passive scalar load-pull setup.
2 5
j5
j5
j2
j2
j1
j1
j0.5
j0.5
j0.2
j0.2
P
out,max
R
L,opt
1 dBm
2 dBm
3 dBm
4 dBm
R
L,ext
2 5
j5
j5
j2
j2
j1
j1
j0.5
j0.2
P
out,max
R
L,opt
1 dBm
2 dBm
3 dBm
4 dBm
R
L
(a) (b)
Figure 39. Derivation of constant output power contours, at the device intrinsic terminals (a) and
at the extrinsic ones (b). (This gure is available in full color at http://www.mrw.interscience.
wiley.com/erfme.)
since it implies the use of a ve-dimensional variables
space; moreover, harmonic terminations are not controlled
(even if they can be monitored). In the latter case, in order
to control harmonic terminations, more than a single tun-
er can be used or a combination of harmonic resonators,
one for each harmonic to be controlled (Fig. 42).
Clearly the rst solution increases the bench complex-
ity and losses, but at the same time exhibits a greater
exibility for frequency control and variation, as contrast-
ed with harmonic resonators, whose resonant frequency
actually xes the measurement frequency.
An alternative solution to the use of passive loads con-
sist in the synthesis of the load to be presented at the ac-
tive-device terminals with active techniques (active load,
leading to active load-pull systems). Basically the active
load is synthesized reecting back to the device port a
power wave whose amplitude and phase are in an exter-
nally controlled magnitudephase relationship with the
emitted wave emerging from the device.
Two basic schemes are usually adopted to this goal,
following the two-signal path and active loop tech-
niques, both schematically depicted in Fig. 43.
In further detail, in the two-signal path technique, the
device output is fed by a fundamental frequency signal
directly generated by the same signal source feeding the
device input, while in the active loop technique the re-
ected power to the device output is the device output
signal itself, appropriately amplied and shifted in phase.
In the former, technique however, the synthesized output
reection coefcient does vary in magnitude and phase
while the source power is swept. Hence, for each input
power a different load is presented to the active-device
output, and therefore attenuation and phase shift must be
controlled to keep the load constant. The same problem
holds when the device heats up. On the other hand, such a
technique is inherently stable, since no oscillation loop is
created in the measurement setup. The latter problem
does exist in the active loop scheme, which must be care-
fully dimensioned in order to avoid spurious oscillations
(even at harmonics of the fundamental frequency). The
active loop technique exhibits, on the other hand, a stable
load termination with input power to the active device.
Both techniques allow great exibility in load control and
selection, if compared to the passive approaches. In par-
ticular, setup losses are easily compensated by the loop
amplier, in most cases a TWT power amplier, given the
Pre-matching network
DUT
Pre-match
network
Losses
Tuner
I
L
DUT
Pre-match
tuner
Losses
Tuner
I
L
Pre-matching tuner
(a)
(b)
Figure 41. Overcoming setup losses with prematching net-
works (a) or prematching tuners (b).
DUT
Tuner
I
L,f
Tuner
I
L,2f
Fundamental
control slug
Harmonic
control slug
(a) (b)
Figure 42. Harmonic control in passive load-pull sys-
tems by means of harmonic tuners (a) or resonators (b).
DUT
Signal
source
Input
amplifier
Variable
attenuator
Power
splitter
Phase
shifter
Loop
amplifier
DUT
Signal
source
Input
amplifier
Variable
attenuator
Phase
shifter
Loop
amplifier
Loop coupler
(a) (b)
Figure 43. Two-signal path (a) and active loop (b) active load-pull techniques.
power and linearity requirements. Highly mismatched (or
even active) terminations are therefore easily synthesized
with an active load-pull approach, clearly with an in-
creased test bench complexity and overall cost.
If accuracy is a concern, in passive load-pull systems it
depends mainly on tuner position repeatability, residual
S-parameter measurement uncertainty, and power mea-
surement uncertainty. For active systems [124], the main
uncertainty source is in power calibration coefcients,
which are strongly dependent on the presence of an input
power amplier (TWTA) before the active device under
test. Such uncertainty is a strong function of the load re-
ection coefcient magnitude, rapidly increasing with it.
The result is that for low to moderate reection coefcient
magnitudes the output power uncertainty performances of
active systems are better than those for passive ones, but
rapidly degrade while approaching highly mismatched
load regions.
If an active harmonic load pull has to be performed, the
previous schemes can be extended for harmonic control,
leading to the setups schematically depicted in Figs. 44
and 45, respectively.
The load-pull techniques can be adopted to test a series
of device output performances, not restricted to classical
output power, large-signal gain, drain, or power-added ef-
ciency. In particular, depending on the measurement set-
up, time-domain waveform capabilities can be tested,
together with linearity indicators: ACPR or two-tone tests
can be performed as well [125,126], thus providing in situ
information on the device linearity and allowing the de-
signer to trade off between often contrasting goals directly
by means of load-pull charts. As an example, in Fig. 46 a
DUT
Signal
source
Input
amplifier
Variable
attenuator
Phase
shifter
Isolator
Variable
attenuator
Phase
shifter
Isolator
Variable
attenuator
Phase
shifter
Isolator
Refl.
bridge
Refl.
bridge
Combiner
x 2
x 3
Figure 44. Extension for harmonic load pull
of the two-signal path active load-pull tech-
nique.
DUT

Loop coupler
Signal
source
Input
amplifier
Loop
amplifier
Variable
attenuator
Phase
shifter
YIG @2f
Variable
attenuator
Phase
shifter
YIG @2f
Power
splitter
Power
combiner
Figure 45. Extension for harmonic load pull of the active loop active load-pull technique.
harmonic measurement setup [127] is shown, which com-
bines S-parameter capability, real-time load and source-
pull (single tone or harmonic) and intermodulation
measurements with time-domain waveformmeasurements.
Any linear vector network analyzer with at least two
samplers can be used as linear receiver, while a microwave
transition analyzer is used as a nonlinear receiver, mea-
suring the phase relationships between harmonics of the
signals at the device under test (DUT) ports. The setup is
completed with two active loops (but more loops could be
added), controlled independently. The loops can be set at
the input or output of the device, and tuned for both sin-
gle-tone and harmonic measurements.
A nal and general consideration regarding load-pull-
based PA design has to be stressed. A complete power
characterization of a given device is indeed an heavy task,
especially if harmonic terminations have to be investigat-
ed. If frequencies up to the third harmonic are considered
at both device input and output, this easily brings in the
control of a six-variable state space, to which a further
variable (drive level) is necessarily added. This formidable
task is further complicated if fundamental frequency vari-
ation and device bias are varied. It is clear, therefore, that
load pull has to be matched with some physical insight
gained on the power generating mechanisms in the device
by means of the harmonic tuning strategies and design
procedures described before. Furthermore, an approach
consisting in the control and consequent variation of a
termination at a time is risky and may lead to subopti-
mum conditions [128].
8. CAD-BASED TECHNIQUES
Computer-aided design (CAD) tools have experienced a
dramatic improvement both in performances and simula-
tion capabilities. Such improvement is due to both in-
crease in computing speed and hardware progress
(workstations and PCs), and in the assessment of appro-
priate analysis techniques.
Modern CAD tools include linear and nonlinear simu-
lation capabilities, coupled with electromagnetic eld solv-
ers, layout generation, and system simulation capabilities
[129135]. For the analysis and design of PA circuits
obviously the nonlinear simulation capabilities are indeed
necessary; the latter may be based on time-domain tech-
niques (derived from widely diffused SPICE-like ap-
proaches [136], adapted to deal with microwave
problems), adopted mainly for transient analysis, mixed
timefrequency-domain techniques [of the harmonic-bal-
ance (HB) type, [138]] for steady-state analysis, fully fre-
quency-domain algorithms [137], and nally Volterra
series approaches [82]. While Volterra-based approaches
nd their intended application to circuits exhibiting mild
nonlinearities and specifically to the analysis of PA lin-
earity [48], the work horse of all commercial nonlinear
CAD software is indeed the HB technique [140,141]. This
approach has been extended from the piecewise formula-
tion to include nasty nonlinear problems, such as non-
linear noise analysis, forced autonomous circuits, and
input signals with complex modulation schemes. In the
latter case, especially important for the analysis of PA
Microwave
source
Sampler 1
Sampler 2
Power splitter 1
Power splitter 3
Power splitter 2
Ref
Test
Microwave
transition
analyzer
Port 1 -Port 2
source switch
Microwave
auxiliary
source
Meas switch
Port 1 Port 2
DUT
Power combiner
Input loop
amplifier
Bias T Bias T
Variable
attenuator
Phase
shifter
YIG
tunable
filter
Port 3 Circulator
Output loop
amplifier
Spectrum
analyzer
a
1
a
m1
a
m2
b
m2
b
m1
a
2
b
1
b
2
1 2
3
Figure 46. Schematic active load-pull bench with time-domain waveform and spectrum cap-
abilities [127].
performance in actual operating conditions in communi-
cation systems, envelope-based approaches and multitone
ones are adopted ([139,142,143]), demonstrating excellent
simulation capabilities and numerical efciency.
The availability, in commercial CAD software pack-
ages, of powerful HB simulation engines, together with
2D or 2.5D EM simulators efciently modeling passive
structures and interconnects, is a valuable support for
microwave integrated circuit design. Nevertheless, the
delity of simulated results to measured data is not com-
pletely achieved yet; whereas the accuracy of the analysis
tool is in fact far beyond the process parameter variations,
this is not true if active-device models are considered.
Indeed, the effectiveness of a simulated result is to a great
extent based on the availability of an accurate device
model, especially if nonlinear operation is concerned.
Without entering the wide eld of nonlinear device
modeling, well beyond the scope of this contribution, a
few key points have to be addressed. The active-device
nonlinear models currently adopted in CAD tools may be
basically classied into two main categories: empirical
blackbox models [144146] and equivalent-circuit based
models [147,148]. The former are generally built on a large
database of measured data and appropriate expressions to
model the active device in the region of interest, while the
latter consist in tting predened expression for the ele-
ments of a given model topology to model parameter data
extracted from selected bias and operating conditions.
As a general consideration, the blackbox modeling ap-
proach does exhibit a better agreement with measured
data, while the equivalent-circuit approach is more com-
pact, is easier to hardwire into commercial CAD tools, and
gives a deeper insight into device physical behavior, since
its equivalent-circuit elements are directly related to
physical mechanisms inside the device.
Nevertheless, equivalent-circuit approaches may suffer
from physical inconsistencies [149,150] that may arise
from either an improperly selected topology on tting
functions. Moreover, frequency dispersion, temperature,
and trapping effects are not easily incorporated into a
nonlinear equivalent-circuit device model [151,152].
Pulsed (DC and RF) measurements are utilized to evi-
dence the abovementioned effects, but the resulting model
may account for them in a limited range of parameter
values (e.g., in a range of biases). Intermodulation predic-
tion is a typical test where most nonlinear general-pur-
pose active-device models may deviate substantially from
their corresponding measured behavior. The commonly
adopted solution is therefore in the development of a spe-
cial-purpose model, valid in a limited parameter range
that ts well within the desired application. Clearly such
special-purpose models are not always available, nor they
can be easily extracted. Scaling dimensions of the active
device and the properties of the respective model further
complicate the picture.
In any case, the adoption of a given device model in the
design process is a key point, and the available one (if any)
must be investigated in depth in both its eld of validity
and extraction methods. As previously noted, the out-
comes of a design by a nonlinear CAD tool are strongly
linked to the accuracy of the device model adopted, and
the adequacy of the latter to the specific design to be per-
formed has to be carefully evaluated beforehand. For ex-
ample, suppose that a large-signal model for a given
device is available, extracted from DC and multibias
small-signal S-parameter measured data. If such a mod-
el has to be used for optimization of the device output
power performance (even in class A operation) and there-
fore for optimum load impedance determination, the de-
signer cannot expect a high degree of accuracy versus
measured power data; DC output curves do not represent
actual large-signal RF operating conditions, due to tem-
perature and trapping effects. Only an estimate can there-
fore be obtained, even with a powerful CADtool. Similarly,
a model optimized for the active-device (i.e., class A)
operating region does not represent the same device
biased in class C, and the model adopted for tradeoff pur-
poses between class A and B performances should average
between the two situations.
Nevertheless, with the precautions stated above, non-
linear CAD tools and active-device models are frequently
and effectively utilized for the design of PA stages. After
the operating mode of the PA stage is appropriately
selected (i.e., a harmonic tuning strategy or a switching
mode operating class) on the basis of a preliminary eval-
uation of attainable performances as compared to the de-
sired specication, together with a preliminary device
bias, commonly adopted nonlinear CAD strategies are es-
sentially based on a series of repeated analyses. In par-
ticular, a possibility resides in the implementation of a
simulated load pull, in which the measurement bench is
replaced by the nonlinear simulator and its analysis
capabilities and the active device is replaced by its mod-
el. With reference to Fig. 47, a simulated load pull may be
performed in subsequent steps. First, a series of power
sweep analyses are performed, each obtained by varying
the fundamental frequency load (Z
L1
in Fig. 47) and ad-
justing the input termination (Z
S1
) in order to conjugately
match the subcircuit input at fundamental frequency. In
such power sweeps, harmonic terminations are kept xed
at a reference value (typically short-circuit termination, so
as to simulate a TL scheme or 50 O). The data gathered are
then organized in order to obtain, for each load terminat-
ing value, the relevant performances at a given compres-
sion level (e.g., 1dB gain compression or maximum
efciency). The resulting data are organized and plotted
as shown in Fig. 35. If the tradeoff attainable with the
generated charts fullls the imposed design goals, the
next step is in the actual synthesis of the resulting
fundamental and harmonic terminations, that is, in a
linear synthesis problem, which can be solved in classical
fashion. Otherwise, keeping the fundamental frequency
terminations at their optimum values, harmonic termina-
tions may be varied and contour plots of power perfor-
mances as functions of harmonic terminations are drawn,
as in Fig. 48, where constant-power-added-efciency loci
are plotted as a function of the output reection coefcient
at second harmonic.
Applying the information presented above, a new de-
sign choice is performed regarding second-harmonic ter-
mination. The procedure is then repeated. Clearly the
limit is in the number of harmonic terminations that can
be effectively controlled and implemented, without sacri-
cing circuit complexity, size, and clarity. The simulated
load pull can be applied as well to the determination of
optimum bias levels; the technique is extremely versatile
and much faster than the corresponding real load pull. In
fact, if a series of operating frequencies have to be inves-
tigated , an experimental load-pull characterization may
be impractical and time-consuming.
An alternative technique may be implemented, how-
ever. Since any iterative analysis technique for a nonlinear
microwave circuit (such as the harmonic-balance one) is
based on the actual linearization at each analysis step of
the nonlinear part of the circuit, the latter can be used,
during the analysis, for external linear circuit optimization
[154156]. In other words, simultaneously and consistently
with the nonlinear analysis, a circuit optimization may
be carried out, consisting in the fulllment of optimum
design conditions. The latter, for a microwave PA stage,
may consist in the imposition of a purely resistive termi-
nation at the intrinsic current source terminals of the
active-device model (thereby maximizing the stage ef-
ciency) or in the attainment of maximum large-signal
P
in Z
S1
Z
S2
Z
S3
Z
Sn
Z
L2
Z
L3
Z
Ln
Z
L1
@f
@f @2f @3f @nf n4 @2f @3f @nf n4
Active
device or
sub
circuit
Bias lines
OUT
Figure 47. Simulated load-pull scheme.
0
10
30
40
50
60
70
80
90
100
110
120
130
140
160
0.00 0.20 0.50 1.00 2.00 5.00
150
170
180
20
10
30
40
50
60
70
80
90
100
110
120
130
140
160
150
170
20
0.0 1.0 1.0
43.18
44.18
46.18
45.18
47.18
43.68
46.68
45.68
44.68
0.20
0.50
1.00
2.00
5.00
5.00
2.00
1.00
0.50
0.20
0.00
Figure 48. Power-added efciency on the sec-
ond-harmonic load plane (from Rohde and
Newkirk [153]).
power transfer (i.e., conjugate matching) at the device
input terminals.
However, both direct synthesis and simulated load-pull
techniques are inevitably based on a proper device model
and the difculties reside in its extraction apply for both
validity range and accuracy of the latter.
BIBLIOGRAPHY
1. H. Sobol and K. Tomiyasu, Milestones of microwaves, IEEE
Trans. Microwave Theory Tech. 50(3):594611 (March 2002).
2. M. Skolnik, Role of radar in microwaves, IEEE Trans.
Microwave Theory Tech. 50(3):625632 (March 2002).
3. W. Keydel, Perspectives and visions for future SAR systems,
IEE Proc. Radar Sonar Nav. 150(3):97103 (June 2003).
4. B. A. Kopp, M. Borkowski, and G. Jerinic, Transmit/receive
modules, IEEE Trans. Microwave Theory Tech. 50(3):
827834 (March 2002).
5. J. M. Osepchuk, Microwave power applications, IEEE Trans.
6. J. M. Osepchuk, Ahistory of microwave heating applications,
IEEE Trans. Microwave Theory Tech. 32(9):12001224
(Sept. 1984).
7. A. Rosen, M. A. Stuchly, and A. Vander Vorst, Applications of
RF/microwaves in medicine, IEEE Trans. Microwave Theory
Tech. 50(3):963974 (March 2002).
8. G. C. Giakos, M. Pastorino, F. Russo, S. Chowdhury, N. Shah,
and W. Davros, Noninvasive imaging for the new century,
IEEE Instrum. Meas. Mag. 2(2):3235, 49 (June 1999).
9. E. C. Fear, X. Li, S. C. Hagness, and M. A. Stuchly, Confocal
microwave imaging for breast cancer direction: Localization
of tumors in three dimensions, IEEE Trans. Biomed. Eng.
49(8): 812822 (Aug. 2002).
10. I. T. Rekanos and A. Raisanen, Microwave imaging in the
time domain of buried multiple scatterers by using an
FDTD-based optimization technique, IEEE Trans. Magn.
39(3):13811384 (May 2003).
11. L. Q. Huo, W. Zhong, T. T. Qing Zhang, J. A. Bryan, G. A.
Ybarra, L. W. Nolte, and W. T. Joines, Active microwave
imaging. I. 2-D forward and inverse scattering methods,
IEEE Trans. Microwave Theory Tech. 50(1):123133 (Jan.
2002).
12. E. C. Fear, P. M. Meaney, and M. A. Stuchly, Microwaves for
breast cancer detection? IEEE Potentials 22(1):1218 (Feb.
March 2003).
13. R. S. Symons, Modern microwave power sources, IEEE Aero-
space Electron. Syst. Mag. 17(1):1926 (Jan. 2002).
14. F. Murgadella, F. Payen, and P. Coulon, SSPAs & TWTAs: An
evolutive situation for electronic warfare applications, 23rd
Annual Gallium Arsenide Integrated Circuit (GaAs IC)
Symp. Tech. Digest, Oct. 2124, 2001, pp. 149152.
15. F. Murgadella, P. Coulon, and C. Moreau, Comparisons of
technologies and MMICS results for military needs, 23rd
Annual Gallium Arsenide Integrated circuit (GaAs IC)
Symp. Tech. Digest, Oct. 2124, 2001, pp. 223227.
16. R. K. Parker, R. H. Abrams, B. G. Danly, and B. Levush,
Vacuum electronics, IEEE Trans. Microwave Theory Tech.
50(3):835845 (March 2002).
17. J. M. Anderson, Irving Langmuir and the origins of elec-
tronics, IEEE Power Engg. Rev. 22(3):3839 (March 2002).
18. J. F. Ramsay, Microwave antenna and waveguide techniques
before 1900, Proc. IRE 46:405414 (Feb. 1958).
19. H. Hertz, Electric Waves, Macmillan, New York, 1893, Chap.
XI.
20. T. K. Sarkar and D. L. Sengupta, An appreciation of J. C.
Boses pioneering work in millimeter waves, IEEE Anten.
Propag. Mag. 39(5):5562 (Oct. 1997).
21. R. H. Varian and S. F. Varian, A high frequency oscillator
and amplier, J. Appl. Phys. 10:321327 (May 1939).
22. S. Nakajima, The history of Japanese radar development to
1945, in R. Burns, ed., Radar Development to 1945, Peter
Peregrinus, Stevenage, UK, 1988, Chap. 18, pp. 243258.
23. M. Chodorow, E. L. Ginzton, I. R. Nielson, and S. Sonkin,
Design and performance of a high-power pulsed klystron,
Proc. IRE 41:15841602 (Nov. 1953).
24. O. M. Stuetzer, A crystal amplier with high input imped-
ance, Proc. IRE 38:868 (Aug. 1950).
25. W. Shockley, A unipolar eld-effect transistor, Proc. IRE
40:365 (Nov. 1952).
26. T. Mimura, S. Hiyamizuk, T. Fujii, and K. Nanbu, A new
eld effect transistor with selectively doped GaAs/nAlGaAs
heterostructures, Jpn. J. Appl. Phys. 19:L225L227 (1980).
27. W. P. Dumke, J. M. Woodall, and V. L. Rideout, GaAsGa-
AlAs heterojunction transistor for high frequency operation,
Solid State Electron. 15:13391334 (Dec. 1972).
28. P. M. Asbeck, D. L. Miller, W. C. Petersen, and C. G. Kirk-
patrick, GaAs/GaAlAs heterojunction bipolar transistors
with cutoff frequencies above 10GHz, IEEE Electron.
Device Lett. EDL-3:366368 (Dec. 1982).
29. H. Kroemer, Theory of a wide-gap emitter for transistors,
Proc. IRE 45:15351537 (Nov. 1957).
30. E. Gebara, N. Rorsman, J. Olsson, H. Zirath, K. Eklund, and
J. Laskar, Power characteristics of high-voltage LDMOS
transistors, Proc. 30th European Microwave Conf., 2000,
Vol. 3, pp. 811.
31. S. Nuttinck, E. Gebara, J. Laskar, N. Rorsman, J. Olsson,
H. Zirath, K. Eklund, and M. Harris, Comparison between
Si-LDMOS and GaN-based microwave power transistors,
Proc. IEEE Lester Eastman Conf. High Performance Devices,
Aug. 2002, pp. 149154.
32. R. Strong, A. Agarwal, T. Smith, R. Messham, S. Mani,
V. Hegde, M. Hanes, H. Nathanson, P. Potyraj, K. Petrosky,
T. Knight, and P. Brabant, X-band SiGe power HBTs, 1998
Topical Meeting on Silicon Monolithic Integrated Circuits in
RF Systems, Digest of Papers, Sept. 1998, pp. 5760.
33. J.-S. Rieh, L.-H. Lu, Z. Ma, X. Liu, P. B. Katehi, P. Bhatta-
charya, and E. T. Croke, Small- and large-signal operation of
X-band CE and CB SiGe/Si power HBTs, IEEE MTT-S Int.
Microwave Symp. Digest, June 1999, Vol. 3, pp. 11911194.
34. http://www.fcsi.fujitsu.com.
35. R. J. Trew, J.-B. Yan, and P. M. Mock, The potential of dia-
mond and SiC electronic devices for microwave and milli-
meter-wave power applications, Proc. IEEE 79(5):598620
(May 1991).
36. C. M. Johnson, N. G. Wright, M. J. Uren, K. P. Hilton,
M. Rahimo, D. A. Hinchley, A. P. Knights, D. J. Morrison,
A. B. Horsfall, S. Ortolland, and A. G. ONeill, Recent pro-
gress and current issues in SiC semiconductor devices
for power applications, IEE Proc. Circ. Devices Syst.
148(2):101108 (April 2001).
37. C. Lee, P. Saunier, and Y. Jinwei, and M. A. Khan, AlGaN-
GaN HEMTs on SiC with CW power performance of 44 W/
mm and 23% PAE at 35GHz, IEEE Electron. Device Lett.
24(10):616618 (Oct. 2003).
38. C. Lee, Y. Jinwei, M. A. Khan, and P. Saunier, Ka-band CW
power performance by AlGaN/GaN HEMTs on SiC, Proc.
2003 Device Research Conf., June 2325, 2003, pp. 1718.
39. U. K.Mishra, P. Parikh,Y.-F. Wu, AlGaN/GaN HEMTs-an
overview of device operation and applications, Proc. IEEE
90(6):10221031 (June 2002).
40. F. H. Raab, P. Asbeck, S. Cripps, P. B. Kenington, Z. B. Popo-
vic, N. Pothecary, J. F. Sevic, and N. O. Sokal, Power ampli-
ers and transmitters for RF and microwave, IEEE Trans.
41. F. H. Raab, Average efciency of power ampliers, Proc. RF
Technology Expo, Anaheim, CA, Jan.Feb. 1986, pp. 474
486.
42. A. Leke and J. S. Kenney, Behavioral modeling of narrow-
band microwave power ampliers with applications in sim-
ulating spectral regrowth, IEEE MTT-S Int. Microwave
Symp. Digest, June 1996, Vol. 3, pp. 13851388.
43. H. Ku and J. S. Kenney, Behavioral modeling of nonlinear
RF power ampliers considering memory effects, IEEE
Trans. Microwave Theory Tech. 51(12):24952504 (Dec.
2003).
44. H. Ku, M. D. Mckinley, and J. S. Kenney, Extraction of ac-
curate behavioral models for power ampliers with memory
effects using two-tone measurements, IEEE MTT-S Int.
Microwave Symp. Digest, June 2002, Vol. 1, pp. 139142.
45. A. Zhu, M. Wren, and T. J. Brazil, An efcient Volterra-based
behavioral model for wideband RF power ampliers, IEEE
MTT-S Int. Microwave Symp. Digest, June 2003, Vol. 2, pp.
787790.
46. J. C. Pedro and N. B. Carvalho, On the use of multi-tone
techniques for assessing RF components intermodulation
properties, IEEE Trans. Microwave Theory Tech. MTT-
47(12):23932402 (Dec. 1999).
47. P. Kenington, High Linearity RF Amplier Design, Artech
House, Norwood, MA, 2000.
48. J. C. Pedro and N. B. Carvalho, Intermodulation Distortion
in Microwave and Wireless Circuits, Artech House,
Norwood, MA, 2003.
49. T. T. Ha, Solid-State Microwave Amplier Design, Wiley,
New York, 1981.
50. I. Bahl and P. Bhartia, Microwave Solid State Circuit
Design, Wiley, New York, 1988.
51. R. J. Weber, Introduction to Microwave Circuits, IEEE Press,
Piscataway, NJ, 2001.
52. L. J. Kushner, Output performances of idealised microwave
power ampliers, Microwave J. 103110 (Oct. 1989).
53. L. J. Kushner, Estimating power amplier large signal gain,
Microwave J. 87102 (June 1990).
54. B. Molnar, Basic limitations on waveforms achievable in
single-ended switching mode tuned (class-E) power ampli-
ers, IEEE J. Solid-State Circ. SC-19(2):144146 (Feb. 1984).
55. M. K. Kazimierczuk, Generalization of conditions for 100-
percent efciency and nonzero output power in power am-
pliers and frequency multipliers, IEEE Trans. Circ. Syst.
CAS-33(8):805807 (Aug. 1986).
56. V. J. Tyler, A new high efciency high power amplier, Mar-
coni Rev. 21(130):96109 (fall 1958).
57. D. M. Snider, A theoretical analysis and experimental con-
rmation of the optimally loaded and overdriven RF power
ampliers, IEEE Trans. Electron. Devices ED-14(6):851857
(June 1967).
58. F. H. Raab, Introduction to class-F power ampliers, RF
Design 19(5):7984 (May 1996).
59. A. Inoue, T. Heima, A. Ohta, R. Hattori, and Y. Mitsui, Ana-
lysis of class-F and inverse class-F ampliers, IEEE MTT-S
Symp. Digest, 2000, pp. 775778.
60. P. Colantonio, F. Giannini, G. Leuzzi, and E. Limiti, On the
class-F power amplier design, Int. J. RF Microwave Com-
put. Aided Eng. 9(2):129149 (March 1999).
61. C. Duvanaud, S. Dietsche, G. Pataut, and J. Obregon, High-
efcient class F GaAs FET ampliers operating with very
low bias voltages for use in mobile telephones at 1.75GHz,
IEEE Microwave Guided Wave Lett. 3(8):268270 (Aug.
1993).
62. P. Colantonio, F. Giannini, E. Limiti, and G. Saggio, Exper-
imental performances of 5 GHz harmonic-manipulated high
efciency microwave power ampliers, Electron. Lett.
36(9):800801 (April 2000).
63. N. O. Sokal, Class-E switching-mode high-efciency tuned
RF/microwave power amplier: Improved design equations,
IEEE MTT-S Symp. Digest, 2000, pp. 779782.
64. S. Nishiki and T. Nojima, High efciency microwave har-
monic reaction amplier, IEEE MTT-S Symp. Digest, 1988,
pp. 10071010.
65. T. Nojima and S. Nishiki, Harmonic reaction ampliera
novel high-efciency and high-power microwave amplier,
66. B. Kopp and D. D. Heston, High-efciency 5-Watt power am-
plier with harmonic tuning, IEEE MTT-S Int. Microwave
Symp. Digest, 1988, pp. 839842.
67. M. A. Khatibzadeh and H. Q. Tserng, Harmonic tuning of
power FETs at X-band, IEEE MTT-S Int. Microwave Symp.
Digest, 1990, pp. 989992.
68. S. Toyoda, High efciency ampliers, IEEE MTT-S Symp.
Digest, San Diego, CA, May 1994, pp. 253256.
69. A. Mallet, T. Peyretaillade, R. Sommet, D. Floriot, S. Delage,
J. M. Nebus, and J. Obregon, A design method for high ef-
ciency class-F HBT ampliers, IEEE MTT-S Symp. Digest,
San Francisco, CA, May 1996, pp. 855858.
70. J. Staudinger, Multiharmonic load termination effects on
GaAs power ampliers, Microwave J. 6077 (April 1996).
71. P. Colantonio, F. Giannini, G. Leuzzi, and E. Limiti, Class G
approach for high efciency PA design, Int. J. RF Microwave
Comput. Aided Eng. 10(6):366378 (Nov. 2000).
72. P.Colantonio, F. Giannini, G. Leuzzi,E. Limiti, Multi har-
monic manipulation for highly efcient microwave power
ampliers, Int. J. RF Microwave Comput. Aided Eng.
11(6):366384 (Nov. 2001).
73. F. H. Raab, Class-F power ampliers with maximally at
waveforms, IEEE Trans. Microwave Theory Tech. MTT-
45(11):20072012 (Nov. 1997).
74. P. Colantonio, F. Giannini, G. Leuzzi, and E. Limiti, Input/
output optimum 2nd harmonic terminations in low-voltage
high-efciency power ampliers, Proc. 10th MICROCOLL,
Budapest, Hungary, March 1999, pp. 401406.
75. T. M. Scott, Tuned power ampliers, IEEE Trans. Circ.
Theory 385389 (Sept. 1964).
76. M. Maeda et al., Source second-harmonic control for high
efciency power ampliers, IEEE Trans. Microwave Theory
Tech. MTT-43(12):29522958 (Dec. 1995).
77. P. M. White, Effect of input harmonic terminations on high
efciency class-B and class-F operation of PHEMT devices,
78. K. Jeon, Y. Kwon, and S. Hong, Input harmonics control us-
ing nonlinear capacitor in GaAs FET power amplier, IEEE-
MTT-S Symp. Digest, 1997, pp. 817820.
79. S. Watanabe, S. Takatuka, K. Takagi, H. Works, and Y. Oda,
Simulation and experimental results of source harmonic
tuning on linearity of power GaAs FET under class AB
operation, IEEE MTT-S Symp. Digest, 1996, pp. 17711774.
80. S. R. Mazumder, A. Azizi, F. E. Gardiol, Improvement of
a class-C transistor power amplier by second-harmonic
tuning, IEEE Trans. Microwave Theory Tech. MTT-
27(5):430433 (May 1979).
81. P. Colantonio, F. Giannini, G. Leuzzi, and E. Limiti, High
efciency low-voltage power amplier design by second har-
monic manipulation, Int. J. RF Microwave Comput. Aided
Eng. 10(1):1932 (Jan. 2000).
82. S. L. Bussgang, L. Ehrman, and J. W. Graham, Analysis of
non-linear systems with multiple inputs, Proc. IEEE
62:10881119 (Aug. 1974).
83. P. Colantonio, F. Giannini, G. Leuzzi, and E. Limiti, High-
efciency low-IM microwave PA design, IEEE MTT-S Symp.
Digest, Phoenix, AZ, May 2001, Vol. 1, pp. 511514.
84. M. R. Moazzam and C. S. Aitchison, A low third order
intermodulation amplier with harmonic feedback cir-
cuitry, IEEE MTT-S Int. Microwave Symp. Digest, 1996,
pp. 827830.
85. C. G. Rey, Predistorter linearizes CDMA power ampliers,
Microwave RF J. 114123 (Oct. 1998).
86. E. Ballestreros, Analysis and design of microwave linearised
ampliers using active feedback, IEEE Trans. Microwave
Theory Tech. MTT-36(3) (March 1988).
87. N. Potechary, Feedforward Linear Ampliers, Artech House,
1999.
88. L. R. Kahn, Single sideband transmission by envelope elim-
ination and restoration, Proc. IRE 40:803806 (July 1952).
89. F. H. Raab and D. J. Rupp, High efciency SSB transmitter
based upon envelope elimination and restoration, Proc. 6th
Int. Conf. HF Radio Systems and Techniques, York, UK, July
1994, pp. 2125.
90. H. Chireix, High power outphasing modulation, Proc. IRE
23(11):13701392 (Nov. 1935).
91. P. H. Raab, Efciency of outphasing power amplier sys-
tems, IEEE Trans. Commun. 33:10941099 (Oct. 1983).
92. H. L. Krauss, C. W. Bostian, and F. H. Raab, Solid State
Radio Engineering, Wiley, New York, 1980.
93. P. J. Baxandall, Transistor sinewave oscillators, IEE Proc. B
106(Suppl. 16):748758 (May 1959).
94. S. C. Cripps, RF Power Ampliers for Wireless Communica-
tions, Artech House, Norwood, MA, 1999.
95. W. J. Chudobiak and D. F. Page, Frequency and power lim-
itations of class D transistor ampliers, IEEE J. Solid-State
Circ. 4:2537 (Feb. 1969).
96. N. O. Sokal and A. D. Sokal, Class Ea new class of high-
efciency tuned single-ended switching power ampliers,
IEEE J. Solid State Circ. SC-10(3):168176 (June 1975).
97. N. O. Sokal, A. D. Sokal, High Efciency tuned Switching
Power Amplier, U.S. Patent 3,919,656 (Nov. 11, 1975)
(expired).
98. F. H. Raab, Idealised operation of the class E tuned power
amplier, IEEE Trans. Circ. Syst. CAS-24(12):725735
(Dec. 1977).
99. T. Mader and Z. Popovic, The transmission-line high ef-
ciency class-E ampliers, IEEE Trans. Microwave Theory
Tech. MTT-9(5):290292 (Sept. 1995).
100. T. Sowlati, C. A. T. Salama, J. Sitch, G. Rabjohn, and D.
Smith, Low voltage, high efciency GaAs class E power
ampliers for wireless transmitters, IEEE J. Solid State
Circ. 30(10):10741079 (Oct. 1995).
101. T. Mader, E. Bryerton, M. Markovic, M. Forman, and
Z. Popovic, Switched-mode high-efciency microwave
power ampliers in a free-space power-combiner array,
IEEE Trans. Microwave Theory Tech. MTT-46(10):
13911398 (Oct. 1998).
102. N. O. Sokal, Class-E RF power ampliers, QEX (published
by American Radio Relay League, 225 Main St., Newington,
CT 06111-1494, USA (204):920 (Jan./Feb. 2001).
103. S. C. Cripps, A theory for the prediction of GaAs FET Load-
Pull power contours, IEEE MTT-S Symp. Digest, Boston,
MA, May 1983, pp. 221223.
104. S. C. Cripps, Old-fashioned remedies for GaAs FET power
amplier designers, IEEE MTT-S Newsl. 1317 (Summer
1991).
105. H. Kondoh, FET power performance prediction using a
linearized device model, IEEE MTT-S Symp. Digest, 1989,
pp. 569572.
106. J. M. Cusak et al., Automatic load-pull contour mapping for
microwave power transistors, IEEE Trans. Microwave The-
ory Tech. 11461152 (Dec. 1974).
107. F. Larose, F. Ghannouchi, and R. Bosisio, A new multi-har-
monic load-pull method for non-linear device characteriza-
tion and modeling. IEEE Int. Microwave Symp. Digest, June
1990, pp. 443446.
108. R. Hajji, F. Beauregard, F. Ghannouchi, and R. G. Bosisio,
Multitone power and intermodulation load-pull characteri-
sation of microwave transistors suitable for linear SSPA
design, IEEE Trans. Microwave Theory Tech. MTT-45:1093
1099 (1997).
109. P. Berini, M. Desgagn, F. Ghannouchi, and R. G. Bosisio, An
experimental study of the effects of harmonic loading on mi-
crowave MESFET oscillators and ampliers, IEEE Trans.
Microwave Theory Tech. MTT-42:943950 (1994).
110. F. Blache, J. Nebus, P. Bouysse, and J. Villotte, A novel com-
puterized multi-harmonic load-pull system for the optimi-
zation of high-efciency operating classes in power
transistors, IEEE Int. Microwave Symp. Digest, Orlando,
FL, June 1995, pp. 10371040.
111. G. Berghoff, E. Bergeault, B. Huyart, and L. Jallet, Auto-
mated characterization of HF power transistor by source-
pull and multiharmonic load-pull measurements based on
six-port techniques, IEEE Trans. Microwave Theory Tech.
MTT-46:20682073 (Dec. 1998).
112. D. Rytting, ARFTG Short-Course: Network Analyzer Cali-
bration Theory, 1997.
113. R. Marks, Formulation of the basic vector network analyzer
error model including switch terms, Proc. 50th ARFTG
Conf., Dec. 1997, pp. 115126.
114. A. Ferrero and U. Pisani, An improved calibration tech-
nique for on-wafer large-signal transistor characteriza-
tion, IEEE Trans. Instrum. Meas. IM-47:360364 (April
1993).
115. G. Engen and C. Hoer, Thru-reect-line: An improved tech-
nique for calibrating the dual six-port automatic network
analyzer, IEEE Trans. Microwave Theory Tech. 987993
(Dec. 1979).
116. R. Marks, A multi-line method of network analyzer calibra-
tion, IEEE Trans. Microwave Theory Tech. 12051215 (July
1990).
117. MultiCalt Users Manual, v. 1.0, National Institute of Stan-
dards and Technology, 1997.
118. J. Benedikt, R. Gaddi, P.J. Tasker, and M. Goss, High-power
time-domain measurement system with active harmonic
load-pull for high-efciency base-station amplier design,
IEEE Trans. Microwave Theory Tech. MTT-48(12):2617
2624 (Dec. 2000).
119. G. Kompa and F. van Raay, Error-corrected large-signal
waveform measurement system combining network ana-
lyzer and sampling oscilloscope capabilities, IEEE Trans.
Microwave Theory Tech. MTT-38(4):358365 (April 1990).
120. Automated Tuner System Users Manual, v. 1.9, Maury
Microwave Corp., 1998.
121. Computer Controlled Tuner System Users Manual, v.6.0,
Focus Microwave Corp., 1998.
122. LP2 Automated Load Pull System Users Manual, ATN
Microwave Corp., 1997.
123. J. Sevic, A sub 1 O load-pull quarter-wave pre-matching
network based on a two-tier TRL calibration, Proc. 52nd
ARFTG Conf., Dec. 1998, pp. 7381.
124. A.Ferrero, V. Teppati, A. Carullo, Accuracy evaluation of on-
wafer load-pull measurements, IEEE Trans. Microwave
Theory Tech. MTT-49(1):3943 (Jan. 2001).
125. J. Sevic, R. Baeten, G. Simpson, and M. Steer, Automated
large-signal load-pull characterization of adjacent-channel
power ratio for digital wireless communication system, Proc.
45th ARFTG Conf. Nov. 1995, pp. 6470.
126. J. Sevic, K. Burger, and M. Steer, A novel envelope-termi-
nation load-pull method for the ACPR optimization of RF/
microwave power ampliers, IEEE Int. Microwave Symp.
Digest, June 1998, pp. 723726.
127. A. Ferrero and V. Teppati, A complete measurement test-set
for non-linear device characterization, 58th ARFTG Conf.
Digest, San Diego, CA, Nov. 2001.
128. P. Colantonio, F. Giannini, E. Limiti, and V. Teppati, An ap-
proach to harmonic load- and source-pull measurements for
high-efciency PA design, IEEE Trans. Microwave Theory
Tech. 52(1):191198 (Jan. 2004).
129. Advanced Design System, ADS Users Manual, Agilent
EEsof EDA, Palo Alto, CA.
130. Microwave Ofce Users Manual, Applied Wave Research
Inc., El Segundo, CA.
131. Ansoft Designer Users Manual, Ansoft Corp., Pittsburgh,
PA.
132. APLAC 7.91 Users Manual, APLAC Solutions Corp., Espoo,
Finland.
133. LINMIC 6.2 Users Manual, AC Microwave GmbH, Aachen,
Germany.
134. MIMICAD Users Manual, Optotek Ltd., Ottawa, Ontario,
Canada.
135. WaveMaker, Barnard Microsystems Ltd., London, United
Kingdom.
136. L. W. Nagel, SPICE, 2, a Computer Program to Simulate
Semiconductor Circuits, Ph.D. thesis, Univ. California,
Berkeley, 1975.
137. M. B. Steer, C.-R. Chang, and G. W. Rhyne, Computer-aided
analysis of non-linear microwave circuits using frequency-
domain non-linear analysis techniques: The state of the art,
Int. J. Microwave Millimeter-Wave Comput. Aided Eng.
1(2):181200 (Feb. 1991).
138. V. Rizzoli, A. Lipparini, and E. Marazzi, A general-
purpose program for nonlinear microwave circuit design,
IEEE Trans. Microwave Theory Tech. 83(9):762770 (Sept.
1983).
139. E. Ngoya and R. Larcheveque, Envelop transient analysis: a
new method for the transient and steady state analysis of
microwave communication circuits and systems, IEEE
MTT-S Int. Microwave Symp. Digest, June 1996, pp. 1365
1368.
140. V. Rizzoli and A. Neri, State of the art and present trends in
nonlinear microwave CAD techniques, IEEE Trans. Micro-
wave Theory Tech. MTT-36:343365 (1988).
141. V. Rizzoli et al., State-of-the-art harmonic-balance simula-
tion of forced non-linear microwave circuits by the piecewise
technique, IEEE Trans. Microwave Theory Tech. MTT-
40(1):1228 (Jan. 1992).
142. V. Rizzoli, A. Neri, and F. Mastri, A modulation-oriented
piecewise harmonic balance technique suitable for transient
analysis and digitally modulated signals, Proc. 26th Euro-
pean Microwave Conf. Prague, Sept. 1996, pp. 546550.
143. D. Sharrit, New circuit simulation analysis methods
for communication systems, IEEE MTT-S Workshop
WMFA: Nonlinear CAD, Digest, San Francisco, June 1996,
pp. 2941.
144. F. Filicori and G. Vannini, Mathematical approach to large-
signal modelling of electron devices, Electron. Lett.
27(4):357359 (Feb. 14, 1991).
145. F.Filicori, G. Vannini,V.A. Monaco, A nonlinear integral
model of electron devices for HB circuit analysis, IEEE
Trans. Microwave Theory Tech. 40(7):14561465 (July 1992).
146. D. E. Root, S. Fan, and J. Meyer, Technology independent
non quasi-static FET models by direct construction from
automatically characterized device data, Proc. 21th Europe-
an Microwave Conf. Stuttgard, Germany, Sept. 1991, pp.
927932.
147. I. Angelov, H. Zirath, and N. Rosman, A new empirical
nonlinear model for HEMT and MESFET devices, IEEE
Trans. Microwave Theory Tech. 40(12):22582266 (Dec.
1992).
148. T. Kacprzak and A. Materka, Compact DC model of GaAs
FETs for large-signal computer calculation, IEEE J. Solid-
State Circ. 18(2):211213 (April 1983).
149. V. Rizzoli, A. Costanzo, C. Cecchetti, and A. Chiarini, A non-
linear model of the power MESFET including temperature
and breakdown effects, IEEE MTT-S Int. Microwave Symp.
Digest, June 712, 1998, Vol. 3, pp. 16031606.
150. V. Rizzoli and A. Costanzo, A fully conservative non-linear
empirical model of the microwave FET, Proc. 24th European
Microwave Conf. Cannes, Sept. 1994, pp. 13071312.
151. Non-linear measurement and modelling, Paper presented at
IEEE MTT-S 97 Int. Symp. Workshop, Denver, CO, June
1997.
152. A. K. Jastrzebski, Characterisation and modelling of tem-
perature and dispersion effects in power MESFETs, Proc.
24th European Microwave Conf., Cannes, Sept. 1994, pp.
13191324.
153. U. L. Rohde and D. P. Newkirk, RF/Microwave Circuit
Design for Wireless Applications, Wiley, New York, 2000,
App. B.
154. F. Giannini, G. Leuzzi, E. Limiti, and L. Scucchia, Harmon-
ic-loaded microwave power ampliers: Non-linear design
procedure, Int. J. Microwave Millimeter-Wave Comput.
Aided Eng. 5(1):2025 (Jan. 1995).
155. P. Colantonio, F. Giannini, G. Leuzzi, and E. Limiti, Direct-
synthesis design technique for non-linear microwave
circuits, IEEE Trans. Microwave Theory Tech. MTT-
43(12):28512855 (Dec. 1995).
156. P. Colantonio, F. Giannini, G. Leuzzi, and E. Limiti, A fast
tool for high-efciency microwave power amplier design,
Microwave Engineering Europe, May 1997, pp. 3341.
MICROWAVE POWER TRANSMISSION
BERNDIE H. STRASSNER
Sandia National Laboratories
Albuquerque, NM
KAI CHANG
Texas A & M University
College Station, TX
1. INTRODUCTION
Microwave power transmission (MPT) is the wireless
transfer of large amounts of power at microwave frequen-
cies from one location to another. MPT is often referred to
in the literature as wireless power transmission (WPT) at
microwave frequencies. MPT research has been driven
primarily by the desire to remotely power unmanned ae-
rial vehicles (UAVs) and by the concept of space solar
power rst conceived by Dr. Peter Glaser of the Arthur D.
Little Company in 1968 [1]. Figure 1 shows a block dia-
gram for the specific application of MPT pertaining to
space solar power (SSP). SSP is an MPT system with the
addition of solar cells and magnetrons for microwave pow-
er generation.
The SSP idea calls for a constellation of solar power
satellites (SPSs) to be placed in geosynchronous orbit
(geo; 36,800km above Earth) in order to capture the
suns energy using arrays of solar cells. The satellites each
measuring several miles across, would be located in geo to
keep them in view of the sun 99% of the time, marking a
double improvement over terrestrial solar cells. In addi-
tion, the closer the satellites are placed to the sun, the
larger their effective collection area since light intensity
decreases by the inverse-square of distance. The solar cell
panels output large DC voltages to awaiting cavity mag-
netrons positioned on various subarrays within each of the
SPSs phased-array apertures. These magnetrons convert
the high-voltage DC outputs of the solar panel arrays to
microwave power. The microwave energy is then beamed
to Earth to farms of rectifying antenna (rectenna) arrays
that convert the incoming microwave energy back to DC
power [3].
Since its inception, SSP has gained considerable atten-
tion since it has the potential of providing clean, renew-
able, and continuous power for generations to come. With
the widespread belief that fossil fuel supplies cannot sup-
port the projected energy demand based on population
growth and increased development, SSP is seen as a pos-
sible remedy. In addition, SSP is seen as circumventing
pollution problems associated with currently used energy
enablers such as nuclear energy and coal. SSP is also seen
as meeting future energy demands where other limited
clean resources such as hydroelectric power and wind fall
short. As an added incentive, SSP can be delivered to the
most remote locations without connective infrastructure
such as pipes or powerlines. By making SSP an available
technology, developed nations will no longer be at the
mercy of OPEC, which includes many unstable and/or un-
friendly nations [4].
Both the fruition of SSP and the present-day desire to
remotely power unmanned aerial vehicles (UAVs) serve as
the main driving forces behind current advancements be-
ing made in MPT. The use of UAVs for communication and
surveillance is seen as an essential capability, especially
for the U.S. military. MPT has been shown experimentally
as a way for people on Earth to remotely power unmanned
high-altitude platforms such as UAVs. Additional uses for
MPT include powering space probes from future space
stations into deep space and powering robots to enter dan-
gerous environments such as nuclear contaminated areas.
2. MPT SYSTEM ARCHITECTURE
2.1. Traditional Transmitting Aperture
In order to achieve maximum transfer of the microwave
energy from the source to the receiver, the transmitting
antenna must be designed such that its sidelobes are
Solar Cell Array
Covert incident sunlight to dc power
Magnitrons
Convert high voltage dc power
to microwave energy
Phased Array
Beam microwave energy to Earth
Rectenna Array
Collect inbound microwave energy
and convert it back to dc power
Aperture
Transfer
Figure 1. Block diagram of an SSP MPT system, including a BP solar cell array, magnetron, a
circularly polarized (CP) phased array [2], and CP rectenna array [2].
2906 MICROWAVE POWER TRANSMISSION
reduced to the lowest possible levels and that its proper
beamwidth keeps spillover losses to a minimum. In the
past, high-gain reectors and horns antennas have been
used to transmit large amounts of power; however, modern
systems call for electronically steered phased arrays for
greater exibility in keeping the microwave beam on tar-
get. Figure 2 illustrates the geometry of a typical array.
The mn array of Fig. 1 contains m=2M columns and
n=2Nrows. One way to achieve low sidelobes is to apply a
tapered weighting to the arrays aperture. Such tapered
weightings include Taylor and Hamming distributions.
For its simplicity, Hamming weighting is analyzed in the
following. Each u,v element in the array receives power
based on the amplitude weighting given by
A
u;v
=
27
50

23
50
cos
2u 2M1 ( )
2
_ _
1
2
M1
p
_
_
_
_
_
_
_
_
_
_
_
27
50

23
50
cos
2v 2N1 ( )
2
_ _
1
2
N1
p
_
_
_
_
_
_
_
_
_
_
_
_
u=1; 2; . . . ; m
v =1; 2; . . . ; n
(1)
where the subscript u pertains to the uth element in ^ xx and
v identies the vth element in ^ yy. This applied aperture
taper will lower the sidelobes to more than 30 dB below
the mainbeam peak so that power is not radiated in un-
desirable directions. This becomes a serious concern in
SSP because of the large amount of power being trans-
mitted. If no taper is applied to the transmitting aperture,
that is, if the array has uniform weighting, the worst-case
sidelobes will be around 12.5 dB, or 12.5 dB below the
peak gain of the radiation pattern.
The array factor (AF) of the transmitting antenna ar-
ray is dened as
AF=
m
u=1
n
v =1
A
u;v
expj
2u (2M1)
2
_ _
1
_ _
(kDx siny
t
cos f
t
b
x
)
exp j
2v (2N 1)
2
_ _
1
_ _
(kDy siny
t
sinf
t
b
y
)
(2)
where k =2p/l
0
, and Dx and Dy are the element spacings
in the ^ xx and ^ yy directions. The elevation angle y
t
is the an-
gle measured from the ^ zz axis toward the ^ xx^ yy plane, and the
azimuth angle f
t
is measured from the ^ xx axis toward the ^ yy
axis. b
x
and b
y
represent the progressive phase shift in the
^ xx and ^ yy directions, respectively, for steering the beam. The
array elements should be spaced approximately l
0
/2 from
each other in both ^ xx and ^ yy directions to avoid harmful
gradient lopes. It is important to be aware that aperture
tapering for sidelobe reduction lowers antenna radiation
efciency and broadens the mainbeam of the arrays radi-
ation pattern. The efciency reduction is caused simply by
the fact that the outer elements of the array are contrib-
uting little to the radiated power. Some of the antenna
arrays elements positioned farthest from the arrays cen-
ter radiate very little, especially in large arrays, but these
remotely positioned elements are still vital for sidelobe
reduction.
In the proposed SSP system, a magnetron will be
located on a subarray with each antenna element in the
subarray having equal power. A graphical representation
of multiple subarrays within an array aperture is shown
in Fig. 3a. Figure 3a shows a particular array aperture
composed of 99 subarrays, each having 88 elements of
equal weight. Thus, the transmitting aperture consists of
7272 antenna elements. The Hamming taper is applied
at the subarray level. If the 7272 antenna elements are
spaced l
0
/2 in both ^ xx and ^ yy directions with no progressive
.
.
.
. . . . . .
.
.
.
3
2
1
1
1 1
(0,0)
2
2
2
Row
x
y
x/2
y/2
[
t
y
Figure 2. Two-dimensional even array geom-

etry containing m columns and n rows. In this
analysis the number of elements is considered
even, and the number of rows is even.
MICROWAVE POWER TRANSMISSION 2907
phase shift, the resultant array factor shown in Fig. 3b is
generated. This array factor has worst-case sidelobes of
25dB at 713.51. The sidelobe suppression can be im-
proved by reducing the subarray size, by increasing the
overall size of the array or by applying a monotonic Taylor
" nn-bar distribution over the arrays aperture. In addition,
the size of a subarray is chosen according to the amount of
microwave power released by the magnetron into the sub-
array input port. The overall size of the transmitting ap-
erture is based on the desired transmitting gain and the
beamwidth necessary to avoid unnecessary spillover losses.
2.2. Split Gaussian Transmitting Aperture
The temperature of the transmitting aperture becomes an
issue in SSP and other MPT applications where the gen-
erated power fed to the transmitting array is significantly
large. Any array inefciencies will result in heat being
propagated through the transmitting aperture. This heat
can destroy sensitive electronics such as phase shifters
and could potentially melt the array elements. The array
radiation pattern discussed in the previous section is
formed using an aperture taper with maximum current
weighting delivered to the centermost antenna elements.
Consequently, the centermost elements are more suscep-
tible to melting than those moving away from the center.
Two solutions to reducing the temperature at the cen-
ter of the transmit array are the split Gaussian and the
Gaussian with an attenuated center region [5]. These two
distributions are illustrated in Fig. 4. These novel weight-
ing schemes distribute power throughout the aperture
more uniformly than in the aforementioned traditional
transmitting aperture case. As a result, the heat caused by
losses in the transmitter is distributed more evenly in the
aperture lowering the chance of failure. The split Gauss-
ian has a center weighting equal to the outermost edges.
The distribution in Fig. 4b has a center level that can vary.
Both tapers shown in Fig. 4 are radially symmetric for
all f
t
. If we consider changing the coefcients c
a
and c
b
of
the more versatile Gaussian with the center attenuation
region depicted in Fig. 4b, the aperture tapers shown in
Fig. 5 result.
These tapers were analyzed by P. Zepeda [5] to a trans-
mit array 250m in diameter with an operating frequency of
5.8GHz. The power density of the arrays aperture versus
center power reduction based on the varying coefcients of
Fig. 5 is shown in Fig. 6a. The coefcients also affect the
sidelobe performance of the array as illustrated in Fig. 6b.
Surprisingly, sidelobe levels below 20dB are possible with
this unconventional aperture distribution. This split Gaus-
sian taper produces sidelobes in the range of the example
weighting given in the aforementioned conventional
(a) (b)
Figure 3. SSP array: (a) amplitude taper for an array containing 99 Hamming weighted sub-
arrays with each subarray consisting of 8 8 equal-power antenna elements denoting an aperture
of 72 72 antennas; (b) corresponding array factor for elements spaced l
0
/2 in both ^ xx and ^ yy direc-
tions and no progressive phase shift between the elements.
(a)
y(x)=
max(y(x))
e
c
a
[c
b
(xc
c
)]
2
(b)
y(x)=
max(y(x))
e
c
a
x
2

c
b
e
c
a
(c
c
x)
2
Figure 4. (a) Normalized split Gaussian taper;
(b) normalized Gaussian taper with an attenu-
ated center region.
Hamming case with great heat reduction, which is vital to
the reliability of the SPS transmitting array.
2.3. Aperture-to-Aperture Transfer
For MPT systems, the transmitting antenna is normally
either a high-gain reecting antenna, a horn antenna, or a
large array consisting of many individual elements. His-
torically, the receiving unit in MPT systems has been a
rectenna array made up of cascaded rectennas. Each of
these rectennas is a combination receiving antenna and
rectifying circuit consisting of a rectifying semiconductor
diode. A typical MPT setup is shown in Fig. 7.
The power in watts received at the receiving antenna
based on Friis free-space transmission equation is
P
r
=P
t
e
cdt
e
cdr
(1 [G
t
[
2
)(1 [G
r
[
2
)
l
0
4pR
_ _
2
D
t
(y
t
; f
t
)D
r
(y
r
; f
r
)[ ^ rr
t
.
^ rr
+
r
[
2
(10
(L
a
(z)=10)
)(10
(L
ra
(t)=10)
)
(3)
where P
t
is the source power transmitted, e
cdt
and e
cdr
are
the cumulative conductor and dielectric losses of the
transmitter and receiver antennas, l
0
is the free-space
wavelength of the energy at the frequency of operation f,
and R is the distance separating the midpoints of the
transmitting and receiving antenna apertures. The quan-
tity (l=4pR)
2
is declarative of the path loss L
p
between the
two antennas.
The quantities (1 [G
t
[
2
) and (1 [G
r
[
2
) represent the
mismatch losses at the inputs of the transmitting and re-
ceiving antennas. For any elded MPT system, these mis-
match losses should be mitigated to less than 1% reected
power relating to a return loss better than 20 dB. Any re-
ected power at the receiver port could cause large
amounts of reradiated or reected energy back into free
space. This could affect electronic devices in the vicinity of
the receiver, especially for an SSP system where large
amounts of power could be reradiated. Since large power
levels are supplied to the transmitting antenna, any re-
ected power could damage the transmitter through col-
lective heating. The efciency of the transmitter falls
rapidly as temperature increases.
The variables D
t
and D
r
are the directive gains (direct-
ivities) of the antennas in the oriented direction (y
t
,f
t
) and
(y
r
,f
r
), respectively. y
t
and y
r
represent the elevation an-
gles and f
t
and f
r
are the azimuth angles of the trans-
mitting and receiving antennas, respectively. The
directive gains of the antenna are related to the antenna
N
o
r
m
a
l
i
z
e
d

a
p
e
r
t
u
r
e

f
i
e
l
d

m
a
g
n
i
t
u
d
e
N
o
r
m
a
l
i
z
e
d

a
p
e
r
t
u
r
e

f
i
e
l
d

m
a
g
n
i
t
u
d
e
Normalized antenna radial position
Normalized antenna radial position
0.2
0.0
1.0
0.8
0.6
0.4
0 1 0.5 1 0.5
c
b
=0.4
c
b
=0.5
c
b
= 0.7
c
b
= 0.8
c
b
= 0.9
c
a
= 4.0
c
a
= 3.0
c
a
=1.9
c
a
=10.0
c
a
=15.0
c
a
= 21.7
c
b
= 0.2
0.2
0.0
1.0
0.8
0.6
0.4
0 1 0.5 1 0.5
(a)
(b)
Figure 5. Aperture tapers: (a) Varying C
b
for the normalized
Gaussian taper with an attenuated center region; (b) varying C
a
for the normalized Gaussian taper with an attenuated center region.
1
0 20 40 60
(a)
(b)
80 100
1.5
28
26
24
22
20
18
16
14
2
2.5
3
3.5
P
d
(
k
W
/
m
2
)
S
L
L

(
d
B
)
P
red
(%)
P
red
(%)
0 20 40 60 80 100
Variation on c
c
Variation on c
b
Variation on c
c
Variation on c
b
Figure 6. (a) Power density of the transmit array as a function of
the center power reduction in percent; (b) sidelobe levels in dB as
a function of the center power reduction in percent.
measurable gains by
G
t
(y
t
; f
t
) =e
cdt
(1 [G
t
[
2
)D
t
(y
t
; f
t
) (4)
G
r
(y
r
; f
r
) =e
cdr
(1 [G
r
[
2
)D
r
(y
r
; f
r
) (5)
If the antennas both have peak directivity broadside of the
arrays aperture (normal to the aperture plane), maximum
power transfer will occur if y
t
=y
r
=01.
The polarization mismatch is calculated by
[ ^ rr
t
.
^ rr
+
r
[ =[ cos c
p
[
2
, where ^ rr
t
and ^ rr
r
are the polarizations
of the transmitting and receiving antennas, respectively,
and c
p
is the angle between their corresponding f-unit
polarization vectors. To avoid polarization mismatch using
linearly polarized antennas, f
t
and f
r
must be such that
the time-harmonic electric eld vectors of each antenna lie
in the same f- plane. This polarization mismatch problem
can be avoided by making both antennas circularly polar-
ized. Circular polarization also avoids depolarization
caused by polarization rotation of the electric eld vector
of the microwave energy as it propagates through the at-
mosphere. Depolarization is primarily caused by water
present in the propagation path and becomes a serious
problem during rainfall.
2.4. Atmospheric Absorption
Another important consideration in MPT design is atmo-
spheric attenuation. This attenuation is caused by the
presence of oxygen and water in the atmosphere during
normal calm conditions and increases as the vertical dis-
tance z from the Earth increases. Attenuation is greatest
near sea level since oxygen and water levels decrease
moving away from the Earth. The atmospheric attenua-
tion is denoted L
a
(z) to reect the height dependence and
can be deduced from Fig. 8. Other gases such as carbon
dioxide can contribute attenuation, but oxygen and water
dominate because of their prevalence.
The choice of operating frequency for an MPTsystem is
governed by atmospheric absorption. The industrial, sci-
entific, and medical (ISM) bands at 2.45 and 5.8 GHz have
been chosen for MPT in the past because there attenua-
tion is low relative to higher frequencies and the sizes of
the transmitting and receiving antennas are of reasonable
size. More importantly, the ISM bands are permissible for
individual use by the Federal Communications Commis-
sion (FCC). For this reason, consumer microwave ovens
have been designed at 2.45 GHz, and the technology be-
hind their microwave source magnetrons has matured to
the point of providing efciencies over 80% DC-to-RF con-
version efciency at 2.45 GHz. Similar source performance
at 5.8GHz still needs some work, but 5.8GHz appears to
be the frequency of choice for future MPT SSP since it al-
lows smaller antenna apertures. An MPTsystem designed
at 22 GHz would see large amounts of attenuation due to
water vapor, especially in humid climates near sea level.
Similarly, oxygen would hamper MPT at 60 GHz.
Inclement weather further complicates the problem by
adding variable amounts of attenuation. The attenuation
loss in dB due to rainfall is [8]
L
ra
(t) =
_
R(t)
0
a[A(z; t)]
b
dz (6)
where
a=
4:21 10
5
f
2:42
; 2:9 _ f _ 54 GHz
4:09 10
2
f
0:699
; 54 _ f _ 180GHz
(7)
b =
1:41 f
0:0779
; 8:5 _ f _ 25 GHz
2:63 f
0:272
; 25 _ f _ 164GHz
(8)
and R(t) is the time-dependent portion of the path between
the transmitting and receiving antennas that contains the
rain. A(z,t) is the amount of rainfall in mm/h at time t at a
distance z km measured from the ground along the path.
These atmospheric attenuation problems are well known
to satellite communication designers and have the poten-
tial of greatly undermining an MPT system. In a dry lab-
oratory environment where R is just within the far eld of
the transmitter, both L
a
(z) and L
ra
(t) from Eq. (3) can be
neglected.
2.5. Rectenna Array
The rectenna array serves as both the absorber of the mi-
crowave energy from the transmitter and the rectier of
the microwave energy to DC power [9]. A diagram of a
typical rectenna array is shown in Fig. 9.
R
Transmitting antenna
(P
t
,G
0t
,D
gt
,e
cdt
,
t
,j
t
)
^
Receiving antenna
(P
r
,G
0r
,D
gr
,e
cdr
,
r
,j
r
)
^
(0
r
,[
r
)
(0
t
,[
t
)
Figure 7. Geometric depiction of a typical MPT setup [6].
Each rectenna is isolated RF-wise from the next adja-
cent rectenna by the capacitors that appear as short
circuits to the incident microwave energy that strikes
them. The antennas in this case are circularly polarized
folded dipoles that send captured microwave energy at
frequency f through the harmonic rejection lters (Fs) to
the Schottky rectifying semiconductor diodes. The diodes
mixing processes create power at DC (f, 2f, 3f, etc.). In
Minimum
values for
starting
heights of:
U.S. standard atmosphere
July, 45 N, latitude
moderate humidity
Standard deviation
(over all seasons,
many locations)
H
2
O
Starting heights
above sea level:
0 km
4
8
12
16
2
4
5
0
5
0

M
H
z
I
S
M

B
a
n
d
s
5
.
8
0
.
0
7
5

G
H
z
Frequency (GHz)
1 2 3 5 10 20 30 50 60 70 80 90 100
1000
.0001
.001
.01
1
100
10
1
T
o
t
a
l

(
o
n
e
-
w
a
y
)

z
e
n
i
t
h

a
t
t
e
n
u
a
t
i
o
n

L
a

(
d
B
)
Standard deviation with
surface temperature and
humidity corrections
Millimeter wave region
Range of
values
0 km
4
8
12
16
O
2
O
2
Figure 8. Atmospheric absorption under normal calm conditions [7].
each rectenna, these energies are then led to both the
lter Fand the capacitor. The DC power is passed through
the capacitors along the current collection bus to the
peripheral voltage collection bus. Each capacitor appears
as a short circuit to the microwave energy returning
the microwaves to the diode for further mixing. The
capacitors also tune out the imaginary part of the diode
impedance to avoid mismatch losses at the diode termi-
nals. Similarly, all microwave circuit components should
be matched to each other. Much of the matching of
the various components is done experimentally. The l-
ter F rejects harmonic energy at 2f and higher back to the
diodes for further mixing. F keeps these higher order
harmonic frequencies from reradiating, which could in-
terfere with electronic devices in the vicinity of the re-
ctenna array. This additional mixing produces more DC
power, increasing the RF-to-DC conversion efciency Z
A
of the rectenna array. Some fundamental f energy is lost
to the antenna for reradiation, but has been shown to
be minimal since the Schottky diodes used currently
have very high RF-to-DC conversion efciencies on the
order of 80%.
If the power density on the rectenna array is uniform,
the maximum Z
A
will occur when the rectenna array is
loaded with a real resistance equal to [10]
R
A
=R
L
N
x
N
y
(9)
where N
x
is the number of rows in the array and N
y
is the
number of rows connected by the voltage collection bus. R
L
is the optimal load resistance for each individual rectenna,
which needs to be calculated or experimentally deter-
mined. The diodes are connected in parallel in each row,
and the rows are connected in series. The setup for deter-
mining the Z
A
is depicted in Fig. 10.
The RF-to-DC conversion efciency Z
A
is dened in
terms of the rectenna arrays aperture area A
eff
A
as
Z
A
=
P
DC
P
r
=
4pR
2
V
2
A
R
A
_ _
P
t
G
t
y
t
; f
t
_ _
A
eff
A
[ ^ rr
t
.
^ rr
+
r
[
2
10
La(z)=10
_ _
10
Lra(t)=10
_ _
(10)
where A
eff
A
=4ab. It is important to make sure the re-
ctennas aperture is positioned such that each of its an-
tennas point toward the transmitter: y
r
=01. The rectenna
array is composed of numerous elements, each corre-
sponding to a particular transmit gain distribution
G
xy
(x,y,R). The average transmit gain seen across the
rectenna arrays aperture is [10]
G
avg
(a; b; R) =
1
4ab
_
b
b
_
a
a
G
xy
(x; y; R) dx dy (11)
This takes into account the fact that the power striking
the rectenna array aperture is normally not a plane wave.
In other words, the transmit power density is greatest
at (0,0) and decreases toward the rectenna arrays edges.
The RF-to-DC conversion efciency can now be expressed
as [10]
Z
A
=
pR
2
V
2
A
R
A
_ _
abP
t
G
avg
(a; b; R)[ ^ rr
t
.
^ rr
+
r
[
2
(10
La(z)=10
)(10
Lra(t)=10
)
(12)
Some previous designs compensated for the tapered power
density present at the rectenna arrays surface to increase
Z
A
. One way this compensation is accomplished is by tun-
ing the load resistance or by changing the lengths between
F F F F
F F F F
F F F F

+
To Load
+
_
+
+
Dc Current Collection Bus
Dc
Voltage
Collection
Bus
Microwave
Short-Circuit
Capacitors
Circularly
Polarized
Folded
Dipoles
Harmonic
Rejection
Filters
Schottky
Rectifying
Diodes
Figure 9. Rectenna array consisting of 12 in-
dividual rectennas. Currents add along each
row on the current collection bus, and the
three row voltage outputs are summed on the
voltage collection bus to result in a collective
power delivered to the load. The rectenna ar-
ray is position approximately l
0
/4 above a re-
ecting metal plane to enhance the gain of
each circularly polarized antenna by 3dB in
the desired direction of radiation coming out of
the page.
various components within each rectenna. Typical curves
for rectenna performance are shown in Fig. 11.
2.6. Retrodirectivity
The nal thing to consider in an MPT design is retrodi-
rectivity. For the SSP example, the rectenna array such as
the one illustrated in Fig. 9 is xed, although some
thought has been given to mechanically steering the re-
ctenna panels to vary y
r
. The SPS is proposed to be in
geostationary orbit, but if phase errors occur in the phased
array or the SPS, which is extremely large in size, under-
goes some inertial structural bending even in the slightest
amount, the microwave beam can veer off of the rectenna
array. To avoid this problem, a feedback loop is established
by a pilot beam sent from the rectenna array back to the
transmitting phased array in order to determine the prop-
er angle y
t
and set the proper aperture phase taper to keep
the beam on the rectenna array [11]. In the case of SSP, a
microwave beam rampaging through the countryside
could cause some public alarm even if the power density
at Earths surface is at acceptable safe levels. The pilot
beam can also send information about air trafc in the
vicinity overhead of the rectenna array so that the micro-
wave beam coming from the SPS can be turned off.
2.7. Other Applications of MPT
Apart from SSP, MPT can be applied to numerous other
applications. One such application is in radiofrequency
identication or (RFID). In RFID systems, power is trans-
mitted from a reader to a tag device that identies that
which it is mounted to. In some applications the tags are
passive, meaning that they contain no battery to drive the
tags onboard electronics. Some environments such as
ones with extreme heat can render batteries useless.
Passive tags rectify a majority of the incident RF power
to DC with the use of a rectifying diode in order to drive
the electronics. In essence, each of these tags is a rectenna
combined with identifying digital electronics. The use of
MPT has even been associated with such ideas as remotely
powering airplanes, tanks, and naval ships. The main
problem is that the required power to move such objects
is immense and the rectenna arrays would need to be of
ridiculous size such that they become impractical. How-
ever, MPT applied to lightweight mobile craft is feasible
and has been done in the past. Such craft include high-
altitude platforms such as UAVs and blimp airships in
which helium is used in conjunction with MPT to move the
airships. The main thing for MPT to be applicable is how
much power on the receiving end is needed to perform
a particular task. For space-to-space applications, MPT
is seen as a realizable technology since gravitational ef-
fects are minimized. MPT could be a real asset in the fu-
ture with regard to powering space probes from space
stations or even other planets. Some plans have called for
making the moon both an MPT transmission site and/or a
rectenna array site.
3. HISTORICAL MILESTONES FOR MPT
3.1. Early Years
The earliest example of power transmission by radiowaves
was carried out by Heinrich Hertz [12]. Hertz used a spark-
gap to generate high-frequency power and to detect the
same power on the receiving end. Reecting antennas were
used for transmitting and receiving the energy. In essence,
Hertz created a complete system for energy transfer.
Most of the early advances in WPTwere achieved at the
turn of the twentieth century by Nikola Tesla [13]. Using
R
b
a
x
y
Synthesized
sweeper
Power
meter
Incident energy
Rectifying
antenna or
array
Voltmeter V
A
R
A
0
t
(0,0)
G
xy
(x,y,R)
Transmitting antenna
Directional
coupler
Amplifier
+
Figure 10. Typical laboratory setup for measuring the RF-to-DC conversion efciency of a re-
ctenna array. The rectenna array output power is dened by the square of the voltage V
A
divided by
the load resistance R
A
[9].
the concepts of resonance rst displayed by Hertz, Nikola
Tesla demonstrated the transmission of low-frequency
electrical power using wires over long distances. Tesla
built several alternating current (AC) power grids and
proved their numerous advantages over the then com-
monly used direct current (DC) systems backed by Thom-
as Edison. Much to the dismay of Edisson, Tesla proved
that AC had much lower conductor losses than did DC
and therefore could be transferred over much greater
distances. Telsa continued to carry out experiments in
his New York City laboratory and turned his attention to
MPTor the transfer of high power wirelessly. While at his
laboratory, Tesla drew up plans to apply alternating surg-
es of current running up and down a metallic mast in or-
der to set up oscillations of electrical energy that would
propagate over large areas on Earth. These oscillations
would create a standing wave around Earth into which
receiving antennas could be positioned at the standing-
wave maximum amplitude locations. In other words, Telsa
wanted to connect the world without wires. This would
become Teslas obsession.
Teslas rst attempted to transmit power without wires
at Colorado Springs in 1899. Under a $30000 grant from
Colonel John Jacob Astor, owner of the Waldorf-Astoria
Hotel in New York City, Tesla built the huge Tesla coil
shown in Fig. 12a in a square building over which rose a
200-ft metallic mast with a 3-ft-diameter ball positioned at
the top. The Tesla coil resonated 300kW of low-frequency
electric energy at 150 kHz. According to Tesla, when the
RF output of the Tesla coil was unleashed into the mast,
100 MV of RF potential was produced on the sphere. Very
large discharges of electrical energy were seen by people
living in and around Colorado Springs. Unfortunately, no
data were collected on whether any significant amount of
power would be collected at any distant point.
With the self-heralded success of the Colorado Springs
experiment, Tesla obtained nancial backing from J. P.
Morgan to construct a setup similar to the one in Colorado
Springs on 2000 acres of land 60 mi east of New York City
at Shoreham, in Suffolk County, Long Island. The building
plans called for a wooden tower, namely, the Wardenclyffe
Tower shown in Fig. 12b, 154 ft high that would support a
giant copper electrode 100ft in diameter shaped like a do-
nut at its top. The structure was nearly completed when
the nancial resources ran dry, and Tesla was forced to
halt construction. The installation was eventually torn
down, during World War I, by the U.S. government be-
cause of its belief that the structure could constitute a
possible target. Tesla continued to pursue his dream of
connected the world without wires, but his efforts went
unnoticed, and, with little outside interest and no nan-
cial supporters, Tesla took a step backwards into seclu-
sion. The rst radio transmission was achieved not by
Tesla but instead by Guglielmo Marconi in 1901. After
Teslas death, the U.S. government seized Teslas docu-
mented works on WPT. The U.S. government saw the
technology as the scientific bases for their proposed death
ray weapon in which WPT would be used to destroy en-
emy weapon systems. Many of these concealed records
were later released to the general public.
From a historical point of view, Tesla was decades
ahead of his time. Not until the 1930s was another at-
tempt on WPTcarried out. This experiment, performed by
H. V. Noble at the Westinghouse Laboratory, consisted of
identical transmitting and receiving 100-MHz dipoles sep-
arated by 25 ft. No attempts to focus the energy were
made, but several hundred watts of power were trans-
ferred between the two dipoles. This experiment was dem-
onstrated again to the general public at the Chicagos
World Fair of 1933/34.
(a)
0
2
4
6
8
10
12
14
16
R
L
= 50
R
L
= 100
R
L
= 200
R
L
= 150
10 12 14 0 2 4 6 8
Circularly polarized power density (mW/cm
2
)
O
u
t
p
u
t

v
o
l
t
a
g
e

(
V
)
(b)
0
0.2
0.4
0.6
0.8
1
1.2
1.4
R
L
= 50
R
L
= 100
R
L
= 200
R
L
= 150
10 12 14 0 2 4 6 8
2
)
O
u
t
p
u
t

p
o
w
e
r

(
W
)
(c)
0
10
20
30
40
50
60
70
80
R
L
= 50
R
L
= 100
R
L
= 200
R
L
= 150
10 12 14 0 2 4 6 8
2
)
R
F
-
t
o
-
D
C

C
o
n
v
e
r
s
i
o
n

e
f
f
i
c
i
e
n
c
y

(
%
)
Figure 11. Rectenna array circularly polarized measured per-
formance: (a) rectied voltage; (b) output power; (c) RF-to-DC
conversion efciency for various resistive loading values.
The primary reason why WPTreceived little interest in
the rst part of the twentieth century was that knowl-
edgeable engineers and scientists knew that, in order to
achieve efcient point-to-point transmission of power, the
electromagnetic energy had to be concentrated into a nar-
row beam, reducing what is commonly referred to today as
spillover loss. It was theorized at this time that the only
way to obtain such conned energy would be to utilize en-
ergy at high frequencies and use radiating elements of
reasonable size. The other problem was that the existing
sources that created high-frequency energy outputted
only a few milliwatts of energynot enough for a feasible
WPT system.
In the late 1930s, two inventions were made that solved
the high-frequency source problem. The rst was the ve-
locity-modulated beam tube rst described by O. Heil
which after a few modications became the well-known
klystron tube. The second invention was the microwave
cavity magnetron developed by Randall and Boot in Great
Britain in 1940 and passed to the United States under
great secrecy during World War II [14]. These develop-
ments allowed the transition of WPT from lower frequen-
cies to microwave frequencies or MPT.
During World War II, with the advent of radar made
possible by the introduction of both the klystron and mag-
netron, antenna development and microwave generation
technologies so basic to MPT improved greatly. The U.S.
government took notice of the emerging technologies and
started proposing applications for the new capabilities.
3.2. Modern U.S. Contributions
In the late 1950s, a number of developments occurred that
revealed that WPT approaching 100% was possible. Cal-
culations and experimental results gathered by Goubau
and Schwering demonstrated that microwave power could
be transmitted with close to 100% efciency by a beam
waveguide consisting of lenses and/or reecting mirrors
[15]. These ndings dispelled the previously held assump-
tion that power density always decays by the square of the
distance. Another vital development was the high-pow-
ered microwave tube amplier or Amplitron. The last de-
velopment propelling MPT was the realization of the
growing need to communicate by line of sight over long
distances, which a platform placed at high altitudes in
Earths atmosphere could afford. Later satellites would be
used for this purpose.
The combination of the aforementioned developments
motivated the Raytheon Company to propose the Ray-
theon airborne microwave platform (RAMP) concept in
1959 to the U.S. Department of Defense as a solution to
surveillance and communication problems. The proposed
platform was a large helicopter positioned at 50,000 ft in a
region above the jetstream, where the atmospheric winds
are almost nonexistent. To y at this altitude, the heli-
copter needed to be powered from Earth by an Amplitron
having an output of 400 kW of energy at 3 GHz with an
efciency over 80%. This high-powered Amplitron was de-
veloped at Raytheons Spencer Laboratory in 1960 by Wil-
liam Brown, who is largely regarded as the principal
pioneer of practical MPT [16]. The only capability miss-
ing was the ability to convert microwave energy to DC
power in order to drive motors attached to the rotor
blades. The U.S. Air Force awarded several contracts to
study this rectication problem. One of the studies, car-
ried out by R. George and E. Sabbagh at Purdue Univer-
sity, showed that a semiconductor diode could be used as
an effective rectier [17]. At the same time, W. Brown at
Raytheon carried out research on the use of a thermionic
diode rectier [18]. Now that both high-powered sources
on the transmitting side and efcient rectiers on the re-
ceiving side were obtainable, MPT for the rst time be-
came both a feasible and possibly useful technology.
The Air Force continued to partner with W. Brown and
Raytheon during the early 1960s in order to pursue the
emerging MPT technological possibilities. One of the best-
known examples of MPT occurred on July 1, 1964 inside
Raytheons Spencer Laboratory. There, a microwave-pow-
ered helicopter much smaller than the one proposed in
RAMP was own a few inches off the ground. It was the
rst heavier-than-air vehicle to be own and was sus-
tained solely by a 2.45-GHz microwave beam. This heli-
copter experiment was demonstrated again to the mass
media on October 28, 1964. The helicopter shown in Fig.
13 was own for 10 h at an altitude of 50 ft [19]. The pre-
sentation was covered by Walter Cronkites CBS news
program and displayed to the world the real possibilities of
MPT. Dipole antennas were used to collect the incoming
microwave energy, and the DC energy that powered the
propeller was obtained using 4480 semiconductor diodes.
This rectifying circuit was built and tested by R. George at
Purdue University. It was the rst demonstration of a
rectifying antenna array in which each antenna element
and its corresponding semiconductor rectifying diode
(a) (b)
Figure 12. Teslas MPT experiments: (a) Colo-
rado Springs; (b) Wardenclyffe tower.
circuit are integrated together. Today such integrated cir-
cuits are known as rectennas.
After the helicopter ight, the U.S. Air Force elected to
discontinue their MPT endeavors. In 1967 W. Brown be-
gan to court Dr. Werner von Braun and his staff at NASAs
Marshall Space Flight Center (MSFC) on MPT possibili-
ties in space. In 1970 MSFC awarded Raytheon a contract
to improve the overall DC-to-DC efciency of the MPT
system. This DC-to-DC efciency includes the conversion
from DC to RF in the magnetron, the aperture transfer
efciency, and the RF-to-DC conversion of the rectenna
array. By multiplying these three efciencies, an overall
system efciency can be determined.
Raytheon continued to improve various rectenna de-
signs throughout the 1970s under the MSFC contract.
Another vital improvement to the MPT system was the
design of a dual-mode horn by P. D. Potter of the Jet Pro-
pulsion Laboratory (JPL) [20]. The modied horn
launched a Gaussian beam with negligible sidelobes to
improve the aperture transfer efciency. Advances to sol-
id-state rectifying diodes in the 1970s improved the RF-to-
DC conversion significantly. The MSFC program resulted
in drastic improvements in MPT system efciency.
In 1971, Brown of Raytheon and Glaser, the SPS mas-
termind, along with members of Northrop Grumman and
the solar photovoltaic company Textron, carried out a
6-month study on the SPS concept and concluded that
the idea was sound. A letter was then sent to the director
of NASA requesting funding [21]. As a result, NASAs
Lewis Research Center (LeRC) awarded a small contract
to Brown and his Raytheon colleagues to improve the
overall efciency of existing MPT systems in order to meet
the stringent requirements necessary for a elded SPS.
During the early 1970s NASA began to shift more and
more focus to SSP with JPL, under the guidance of Rich-
ard Dickinson, playing a major role in the process. The
culmination of efforts occurred in 1974 with the MPT set-
up shown in Fig. 14a having an overall DC-to-DC conver-
sion efciency of 5471%. The operating frequency was
2.446 GHz, and the rectennas output power level was
600 W. This efciency was certied by JPLs Quality As-
surance organization and to this day stands as the highest
MPT end-to-end efciency. The breakdown of the 54.18%
overall efciency is 68.9% for the DC-to-microwave power
conversion, 95% for the aperture-to-aperture transfer, and
82.4% for the beam collection and rectication [22,23].
In 1975, another important milestone was shown at the
Venus Site of JPLs Goldstone Facility. In this demonstra-
tion shown in Fig. 14b, microwave energy at 2.388 GHz
was sent over a 1mi distance to an awaiting 288 ft
2
rectenna array. The rectenna array was designed by
W. Brown at Raytheon and outputted 30 kW [24]. Both
the JPL certied and Goldstone experiments gave NASA
the condence it needed into the viability of MPT and its
possible use in Glasers SPS concept.
Even with the success of Goldstone, LeRC continued to
push for improvements to the transmitting antenna array
as well as the rectenna. In 1977, Brown improved the de-
sign of rectenna arrays by introducing thin-lm etched
rectennas in which the DC bussing is achieved in the
plane of the antennas [25]. Before etched rectennas, the
DC networks were attached behind the antennas, making
previous rectenna arrays more complex, much heavier,
and more costly. Almost all currently designed rectennas
are etched.
Between the years of 1977 to 1980, NASAworked joint-
ly with the U.S. Department of Energy (DoE) to further
evaluate SSP possibilities in providing affordable energy
to consumers on Earth. The study concluded in a 670-page
document that determined that SSP was a feasible tech-
nology and should be pursued in the future [26]. One idea
coming out of the study was the idea of retrodirectivity or
the ability to keep the microwave beam on target. Unfor-
tunately, the NASA sponsored program ended in 1980,
and the U.S. lead in SSP came to an end.
Figure 13. This U.S. Air ForceRaytheon-sponsored demonstra-
tion of a microwave-powered helicopter was made to public media
in October 1964.
(a) (b)
Figure 14. Raytheon/JPL MPTexperiments: (a) an
overall system efciency of 54.18% 70.94% was
obtained at the Raytheon Company with the
demonstration setup shownthe DC power output
was 495W, and the frequency was 2.446GHz; (b)
demonstration of beamed power over one mile dis-
tance at the JPL Goldstone facility in the Mojave
Desert, CA.
3.3. International Involvement
During the 1980s and early 1990s, the center of MPT re-
search and development shifted to Japan and to a lesser
extent Europe and Canada. Since Japan was and remains
a large energy consumer with little natural energy re-
sources, the promise of SPS warranted investigation. As a
result of the investigation into SPS, Japan carried out to
in-space experiments. The rst of Japans in-space exper-
iments was the Microwave Ionosphere Nonlinear Interac-
tion eXperiment (MINIX) conducted by Matsumoto and
colleagues in 1983. MINIX focused on how the plasma
wave dynamic spectrum changes when high-powered mi-
crowave energy is transmitted into ionospheric plasma
[27,28]. The second in-space experiment was the Interna-
tional Space YearMicrowave Energy Transmission in
Space (ISY-METS) in 1993. In ISY-METS microwave en-
ergy was transferred from one rocket to a second rocket
that carried two different rectenna arrays. ISY-METS rep-
resented the rst example of MPT in space [29].
In 1980, a program to develop a long-endurance high-
altitude platform called the Stationary High Altitude Re-
lay Program (SHARP) was proposed in Canada [30]. The
platform was to be the rst unmanned, fuel-less, light-
weight airplane powered remotely by microwaves, which
enabled it to stay aoat for long periods of time. On Sep-
tember 17, 1987, the
1
8
-scale prototype SHARP with a
wingspan of 4.5m seen in Fig. 15a ew on beamed micro-
wave power for 20 min at an altitude of 150m. A 2.45-GHz
microwave beam was transmitted by a parabolic dish an-
tenna, providing a power density at the airplane of 400W/
m
2
. The dual-polarized rectenna array received enough
microwave energy to generate 150 W of DC power to the
electric motor in order to lift and y the 4.1-kg airplane.
Another example of driving a model airplane using
microwave power was the MIcrowave Lifted Airplane
eXperiment (MILAX) conducted by Japan in 1992. The
experiment was the rst to use an electronic scanned
phased array to keep the 2.411-GHz microwave beam on
the moving target or, in this case, the airplane shown in
Fig. 15b. Two charge-coupled device (CCD) cameras rec-
ognized the airplanes pattern feeding the location to a
computer that scanned the array to the appropriate loca-
tion. The transmitting array was located on a sports util-
ity vehicle that was also in motion during the tests.
MILAX received nationwide media coverage in Japan
and impressed SSP favorably on the Japanese public [31].
3.4. Recent MPT Focus
NASA took notice of the Japanese successes and in 1995
undertook the Fresh Look Study to reconsider the chal-
lenges of large-scale SSP systems. The study emphasized
the most recent technological advancements, which ren-
dered SSP more viable than it was in the late 1980s [32].
In 1998, NASA conducted the SSP Concept Definition
Study, in which experts within NASA and outside the
agency were engaged. The second study backed up nd-
ings from the Fresh Look Study, but it also narrowed the
SSP concepts by invalidating some of the earlier ideas. In
2000, NASA MSFC conducted the SSP Scientific Explor-
atory Research and Technology (SERT) program. The pro-
gram broadened the scientific communitys involvement
and resulted in successful demonstrations on a variety of
system-level components.
The SERT program addressed numerous concerns pre-
viously outlined by Glaser. Some of these issues relate to
economic and societal assessment, environmental effects,
resource requirements, and legal issues. The economic as-
sessment studies provided a cost-effectiveness analysis of
the SSP system. The societal assessment included the un-
derstanding that SSP is for everyone even in the most re-
mote locations where SSP has an obvious advantage. This
avoids the situation where a country monopolizes the tech-
nology. The environment issues focused on human exposure
to the microwave energy, especially with regard to the peo-
ple working at or near the rectenna array. Studies have
shown that the rectenna arrays can be designed to accept
power densities within human exposure limits. Questions
still remain on how birds would be affected when ying
through the microwave beam, and how birds could be con-
vinced not to roost on the warmrectenna arrays. Some focus
was also given to possible climate change, although studies
have shown negligible effect even in heavy rainstorms when
the absorption of microwave power in the troposphere is
expected to increase. A concern resonated during SERT
meetings was the impact of reradiated energy from the re-
ctenna arrays. This energy could interfere with other elec-
tronic devices operating in the same frequency bands,
especially since the SPS is radiating large amounts of pow-
er with some spillover loss. Research in this area is ongoing,
but it has been shown that rectennas can be designed to
minimize harmonic energy reradiation with the use of har-
monic ltering. The land that a rectenna may need is also a
concern. The rectenna would most likely be placed in an
(a) (b)
Figure 15. MPT applied to unmanned remotely
powered model aircraft: (a) SHARP; (b) MILAX.
arid environment since rain can reduce rectenna efciency.
This would hopefully mitigate the number of birds and peo-
ple since both reside for the most part along the coasts and
waterways. Another concern is that the astronomers will be
upset that a large object that reects sunlight is appearing
in the night sky. This is a viable concern, especially since an
SSP system will call for a myriad of satellites. Finally, the
transport of materials to space to construct the SPSs, is a
daunting task. How to use the Space Shuttles payload bay
more efciently and the elevator to space concept are being
studied for transporting the necessary supplies. With the
recent nanotube technological breakthroughs, the elevator
idea is not as far-fetched as it might seem.
With the renewed U.S. involvement and continuing ef-
forts by Japanese researchers and others, SSP is progress-
ing steadily. Much of the research since 1990 has focused
on producing extremely efcient rectenna arrays. One such
etched rectenna design funded by the SSP SERT program
and demonstrated at the 2002 World Space Congress in
Houston, Texas, accepts circularly polarized energy at
5.8GHz and outputs DC at 82% efciency [2]. This ef-
ciency was made possible by the recent advances made in
reducing the parasitic losses of ip-chip Schottky diodes.
The frequency 5.8GHz was used for its ability to propagate
through the atmosphere with relatively low loss and be-
cause the receiving and transmitting antennas become
reasonably small in terms of SPS construction feasibility.
Circular polarization was chosen because of depolariza-
tion. Consequently, linearly polarized systems used for
SSP would most likely see degradations in efciency.
Currently, the part of the SSP system that warrants
the greatest focus is the photovoltaic solar cells. The state
of the art for solar cell sunlight to DC efciency has re-
mained around 30% since the mid-1990s. The other 70% is
predominately heat loss, which can heat the transmitting
aperture, lowering the radiated power. Improvements in
solar cell efciency are essential before elding a func-
tioning SSP system.
4. CONCLUSIONS
Future applications of MPT may apply to SSP, but in the
near term MPT will probably be applied to unmanned ae-
rial vehicles (UAVs). UAVs are capable of delivering ser-
vices such as communications and remote sensing and are
nding increasing use for synergistic military applica-
tions. The idea of powering space probes from transmit-
ters positioned in outer space also is seen as a realizable
and sensible technology. However the holy grail for MPT
is SSP. SSP is seen by some experts as a way of meeting
energy demands for all future generations. The reality is
that MPT SSP-associated systems will have relevance
until the sun stops emitting light, and by that time,
presumably billions of years from now, all life will have
ceased to exist, anyway.
BIBLIOGRAPHY
1. P. Glaser, Power from the sun: Its future, Sci. Mag.
162(3856):857861 (Nov. 1968).
2. B. Strassner, S. Kokel, and K. Chang, 5.8GHz circularly po-
larized low incident power density rectenna design and array
implementation, Proc. IEEE AP-S Symp., June 22, 2003, pp.
950953.
3. P. E. Glaser, F. P. Davidson, and K. I. Csigi, Solar Power Sat-
ellites, Wiley, New York, 1994.
4. R. Nansen, Sun Power, Ocean Press, Ocean Shores, WA, 1995.
5. P. Zepeda, Modeling of a sandwich transmitting array satel-
lite, World Space Congress, Houston, TX, 2002.
6. C. Balanis, Antenna Theory, Wiley, New York, 1982.
7. L. J. Ippolito, R. D. Kaul, and R. G. Wallace, Propagation Ef-
fects Handbook for Satellite Systems Design [NASA Reference
Publication 1082(03)], NASA, Washington, DC, June 1983.
8. R. L. Olsen, D. V. Rogers, and D. B. Hodge, The aR
b
Relation
in the culmination of rain attenuation, IEEE Trans. Anten.
Propag. AP-26:318329 (March 1978).
9. W. C. Brown and E. E. Eves, Beamed microwave power trans-
mission and its application to space, IEEE Trans. Microwave
Theory Tech., 40(6) (June 1992).
10. B. Strassner and K. Chang, Highly efcient C-band circularly
polarized rectifying antenna array for wireless microwave
power transmission, IEEE Trans. Anten. Propag. 51(5):1548
1553 (May 2003).
11. L. H. Hsieh, B. H. Strassner, S. L. Kokel, C. T. Rodenbeck,
M. Y. Li, and K. Chang, Development of a retrodirective
wireless microwave power transmission system, Proc. IEEE
AP-S Symp., June 22, 2003, pp. 393396.
12. H. Hertz, Dictionary of Scientific Biography, Vol. VI, Scriber,
New York; 1998, pp. 340349.
13. M. Cheney, Tesla, Man Out of Time, Prentice-Hall, Englewood
Cliffs, NJ, 1981.
14. H. Boot and J. Randall, Historical notes on the cavity mag-
netron, IEEE Trans. Electron. Devices ED-23(7), (July 1976).
15. G. Goubau and F. Schwerng, On the guided propagation of
electromagnetic wave beams, IRE Trans. Anten. Propag. AP-
9:248256 (May 1961).
16. J. F. Showron, G. H. MacMaster, and W. C. Brown, The super
power CW amplitron, Microwave J. (Oct. 1964).
17. R. H. George and E. M. Sabbagh, An efcient means of con-
verting microwave energy to dc using semiconductor diodes,
IEEE Microwave Theory Tech. 11(part 3):132141 (March
1963).
18. W. C. Brown, Thermionic diode rectier, in E. C. Okress, ed.,
Microwave Power Engineering, Academic Press, New York,
1968, Vol. I, pp. 295298.
19. W. C. Brown, Experimental Airborne Microwave Supported
Platform, Tech. Report RADC-TR-65-188, Contract AF30
(602) 3481, Dec. 1965.
20. W. C. Brown, Free-Space Microwave Power Transmission
Study, Combined Phase III and Final Report, Raytheon
Report PT-4601, NASA Contract NAS-8-25374, Sept. 1975.
21. W. C. Brown, Satellite power stationsa new source of ener-
gy? IEEE Spectrum, 10(3):3847 (March 1973).
22. R. M. Dickinson and W. C. Brown, Radiated Microwave Power
Transmission System Efciency Measurements, Tech. Memo
33-727, Jet Propulsion Lab., California Inst. Technology,
Pasadena (JPL/cal Tech) March 15, 1975.
23. R. M. Dickinson, Evaluation of a Microwave High-Power Re-
ception-Conversion Array for Wireless Power Transmission,
Tech. Memo 33-741, JPL/Cal Tech, Pasadena, CA, Sept. 1, 1975.
24. R. M. Dickinson, Reception-Conversion Subsystem (RXCV)
Transmission System, Raytheon Final Report Microwave
Power, ER75-4386, JPL Contract 953968, NASA Contract
NAS 7-100, Sept. 1975.
25. W. C. Brown, Electronic and Mechanical Improvement of the
Receiving Terminal of a Free-Space Microwave Power Trans-
mission System, Raytheon Contract Report PT-4964, NASA
CR-135194, Aug. 1977.
26. Final Proc. Solar Power Satellite Program Rev. 0DOE/NASA
Satellite Power System Concept Development Evaluation Pro-
gram, Conf.-800491, July 1980.
27. H. Matsumoto, N. Kaya, I. Kimura, S. Miyatake, M. Nagat-
omo, and T. Obayashi, MINIX project toward the solar power
satellite-rocket experiment of microwave energy transmission
and associated nonlinear plasma physics in the ionosphere,
Proc. ISAS Space Energy Symp., 1982, pp. 6976.
28. M. Nagatomo, N. Kaya, and H. Matsumoto, Engineering
aspect of the microwave ionosphere nonlinear interaction
experiment (MINIX) with a sounding rocket, Acta Astronaut.
13(1):2329 (1986).
29. R. Akiba, K. Miura, M. Hinada, H. Matsumoto, and N. Kaya,
ISY-METS rocket experiment, Inst. Space Astronaut.
Sci.652:113 (1993).
30. J. Schlesak, A. Alden, and T. Ohno, SHARP rectenna and low
altitude ight trials, Proc. IEEE Global Telecommunications
Conf., New Orleans, Dec. 25, 1985.
31. Y. Fujino, T. Ito, N. Kaya, H. Matsumoto, K. Kawabata,
H. Sawada, and T. Onodera, A rectenna for MILAX, Proc. Wire-
less Power Transmission Conf., Texas, Feb. 1993, pp. 273277.
32. J. M. Mcspadden and J. C. Mankins, Summary of recent
results from NASAs space solar power (SSP) programs and
the current capabilities of microwave WPT technology, IEEE
Microwave Mag. 3:4657 (Dec. 2002).
MICROWAVE RECEIVERS
JAMES BAO-YEN TSUI
Wright-Patterson Air Force Base
Ohio
A microwave receiver is used to receive information trans-
mitted at microwave frequencies (from 1 to 220GHz).
Most microwave receivers are integral parts of a system,
that is, communication or radar, and they are designed
together with the transmitter. In other words, the receiver
is designed to receive a specific signal with maximum ef-
ciency. For example, the video bandwidth of a pulsed ra-
dar receiver matches the bandwidth of the transmitted
pulse so as to receive maximum energy in the signal while
at the same time limit its noise bandwidth to a minimum
value. The receiver of a frequency modulation (FM) radar
has a dispersive delay line matching the transmitted sig-
nal and compresses it into a short pulse to increase the
processing gain. In the global positioning system (GPS)
the coarse acquisition (C/A) signals are code-division mul-
tiple access (CDMA). In a GPS receiver locally generated
C/A codes are used to correlate with an input signal to
perform acquisition and signal tracking. In an FM receiv-
er, the frequency variation information is converted into
amplitude information through a frequency discriminator.
From these examples one can see that each receiver is
uniquely designed.
Another type of microwave receiver detects uncooper-
ative signals. This type of receiver is usually referred to as
an intercept receiver. This type of receiver has some de-
signs in common, because each subtype is designed to de-
tect signals with very limited information. For example,
there are police radar receivers and police radar detectors.
The police radar receiver, designed to match the trans-
mitted signal to achieve the highest sensitivity, is an in-
tegral part of a radar system. The police radar detector
used in automobiles is an intercept receiver. The purpose
of the radar detector is to detect whether police trafc ra-
dar is being used. An intercept receiver does not have the
detailed information of the radar signal; instead, it uses
coarse information, that is, the frequency range over
which the radar operates. Although the intercept receiv-
er has less sensitivity than the radar receiver, the inter-
cepted signal is much stronger than the signal returned to
the radar. The signal strength received by a radar receiver
is proportional to 1/R
4
, where R is the distance between
the radar and the target (or intercept receiver). The in-
tercepted signal strength is proportional to 1/R
2
, which is
stronger than the returned signal.
In this article, intercept microwave receivers will be the
main subject because of their similarities in design goals.
Intercept receivers are very useful in military applica-
tions. Used in electronic warfare (EW), they are often re-
ferred to as EW receivers. These receivers are used to
intercept hostile communication as well as radar signals.
It is a more challenging task to design military intercept
receivers, because it is a common military practice to
design signals that cannot be detected by an intercept
receiver and cannot be jammed.
Receivers can sometimes be used to detect unintention-
al radiation. For example, one can use a microwave receiv-
er to check the microwave power leakage level of a
microwave oven. This type of receiver is also useful in mil-
itary applications. For example, the detection of sparkplug
radiation from automobiles can help locate enemy vehicles.
In the past, most microwave receivers were built using
analog techniques. Because of recent advances in digital
circuits, that is, with todays high-speed analog-to-digital
converters (ADC) with their large number of bits and re-
cent high-speed digital signal processing (DSP) tech-
niques, it appears that the trend is to build digital
receivers. Some narrowband receivers have already been
built using digital techniques. Wideband digital micro-
wave receivers are in the research stage. Digital receivers
should be more reliable because they require less mainte-
nance and adjustment.
This article includes descriptions of a generic receiver,
followed by a discussion of the important terminology and
definitions used in receiver design. A classication of re-
ceivers is then presented, with a discussion of analog and
digital receivers to follow in subsequent sections. A com-
parison of different types of receivers concludes the article.
1. GENERIC RECEIVERS
In general, the signals received by an antenna are very
weak. It is difcult to process the signals or even to detect
MICROWAVE RECEIVERS 2919
their presence directly. A common approach is to amplify
the signals to a higher power level before further process-
ing or detection. This amplication is accomplished
through a radiofrequency (RF) chain.
The RF chain usually contains the following compo-
nents: RF ampliers, lters, mixers, local oscillators, in-
termediate-frequency (IF) ampliers, IF lters, and
attenuators. Ampliers are used to raise the signal pow-
er level. Filters are used to limit out-of-band noise as well
as spurious responses (undesired frequencies) generated
from some components. The mixer and local oscillator are
used in combination to shift the input frequency to anoth-
er frequency, often referred to as the IF. At IF, additional
lters and amplication can be provided. In many receiv-
ers the input signals are not converted to a different fre-
quency and, therefore, the mixer and local oscillator are
not needed. Attenuators are used to adjust the overall
gain of the RF chain. The overall gain in the RF chain
must be of a specific value. Often commercial ampliers
with specific gain values are used in receiver design and it
is difcult to implement the desired value. A common
practice is to use ampliers to provide more gain than
the desired value and attenuators to lower the gain to the
correct value.
In communication receivers, automatic gain control
(AGC) is sometimes used. The AGC changes the gain of
the RF chain according to the input signal strength: lower
gain for strong signal and higher gain for weak signal. The
AGC is seldom used in receivers intercepting pulsed sig-
nals, because it is difcult to build an AGC with very fast
response time.
After the RF chain in an analog receiver, the signal is
detected by a crystal video detector. The detector lters
out the RF but retains the information of the signal, often
referred to as the video signal. Further processing is need-
ed to obtain the necessary information, which includes
digitizing the video signal. A basic analog receiver is
shown in Fig. 1a. In some analog receivers, there are no
RFampliers and a crystal video detector is used to detect
signals directly. These receivers usually have low sensi-
tivity and can detect only very strong signals.
After the RF chain in a digital microwave receiver, an
ADC is used to convert the input into digital data as
shown in Fig. 1b. Because the output of the ADC is digital,
digital signal processing (DSP) can be used to obtain the
necessary information.
2. DEFINITIONS USED IN RECEIVERS
The two most important specications to describe a
receiver are sensitivity and dynamic range. These two
parameters can be used to specify all kinds of receivers
whether they are intercept receivers or receivers designed
for a specific signal. Sensitivity can be briefly dened as
the capability to receive the weakest possible signal. Dy-
namic range is the maximum signal amplitude range that
a receiver can process without distortion. A strong signal
above the upper limit of the dynamic range can produce
distortion or generate spurious responses. It is desirable
to have high sensitivity to receive weak signals and high
dynamic range to receive a broad range of signals. How-
ever, high sensitivity often causes lower dynamic range
and vice versa. Therefore, in designing a receiver, the
tradeoff between sensitivity and dynamic range becomes
an important issue.
Sensitivity is closely related to noise oor, the noise
gure of the receiver, and the gain in the RF chain. In an
analog receiver, the sensitivity depends on the video band-
width after the detector. In a digital receiver, the DSP al-
gorithm used after the ADC affects the overall bandwidth
of the receiver and determines receiver sensitivity. When a
receiver can process only one signal, there is only one def-
inition of dynamic range. If a receiver can process simul-
taneous signals, there are usually three definitions of
dynamic range. They are denoted as the single signal,
two-signal spur-free, and two-signal instantaneous dy-
namic range. The lower limit of the dynamic range is al-
ways the sensitivity level. The upper limit depends on the
definition of the dynamic range. These definitions, termi-
nology and calculations to obtain some of the values, will
be discussed in the following paragraphs. Most equations
can be found in Ref. 1.
2.1. Receiver Input Bandwidth and Instantaneous Bandwidth
The input bandwidth of a receiver refers to the frequency
range in which the receiver can detect an input signal.
This is also referred to as the operational bandwidth or
the overall bandwidth of the receiver. The instantaneous
bandwidth means that any signals with sufcient ampli-
tude in the bandwidth will be detected immediately. Usu-
ally, the input bandwidth is wider than the instantaneous
bandwidth, but in some receivers they are the same. The
RF
chain
Crystal
detector
Analog
(a)
Analog or
digital
processor
RF
chain
ADC
Digital
(b)
DSP
Figure 1. (a) A basic analog receiver with RF amplier chain,
crystal detector, and analog/digital processor. The RF amplier
chain is used to amplify the input signal. The crystal detector
changes the RF signal into a video signal. The processor takes the
video signal and generates the desired digital information. (b) A
basic digital receiver with RFamplier chain, ADC, and DSP. The
ADC digitizes the analog input signals. The DSP processes the
digitized data and generates the desired digital information.
2920 MICROWAVE RECEIVERS
instantaneous bandwidth can be assigned to different por-
tions of the input bandwidth. For example, a receiver may
have a 16 GHz (218 GHz) input bandwidth but a 1 GHz
instantaneous bandwidth. This 1 GHz bandwidth can be
placed in any one of the bands in the 218 GHz range to
receive signals in that band. The input bandwidth and the
instantaneous bandwidth are not used to determine the
sensitivity. They are mentioned here only to distinguish
them from the RF bandwidth.
2.2. RF Bandwidth
The RF bandwidth is used to determine the sensitivity of a
receiver. In an analog receiver, the lter with the narrow-
est bandwidth in the RF chain is the RF bandwidth. In
some receivers, the input and the instantaneous and RF
bandwidths may be the same. In a digital receiver, usually
the inverse of the DSP length will be used as the RF
bandwidth. For example, if the input is digitized at
1000 MHz, each sample is separated by 1 ns. If 256 sam-
ples will be processed through the fast Fourier transform
(FFT), the RF bandwidth will be approximately 3.9MHz
(1/25610
9
s).
2.3. Video Bandwidth
The video bandwidth of a receiver is determined by the
video circuit following the crystal detector. The desired
bandwidth is determined by the input signals. For an elec-
tronic warfare (EW) receiver, the video bandwidth is de-
termined by the shortest pulse anticipated. In a digital
receiver, it is determined by the processing scheme and is
sometimes assumed to have the same as the RF band-
width.
2.4. Noise
Noise generated by a resistor R can be represented by a
noise generator in series with the resistor. Maximum pow-
er transfer from a generator to a load occurs when the load
impedance is matched to the generator impedance. Avail-
able power refers to the power that would be delivered to a
matched load. The available thermal noise power N
i
in
watts at the input of a receiver can be expressed as
N
i
=kTB (1)
where k is Boltzmanns constant ( =1.3810
23
J/K), T is
the temperature of resistor R, and B is the bandwidth of
the receiver in hertz. The power level in a typical receiver
system is very low and is usually expressed in milliwatts
or in dBm, which is dened as
P(dBm) =10 log(P) (2)
where the P on the right-hand side is power in milliwatts
and the base of the log is 10. The thermal noise at room
temperature where T=290 K can be expressed in dBm as
P(dBm) = 174dBm=Hz (3)
or
P(dBm) = 114dBm=MHz
These two values are commonly used in receiver designs.
It should be noted that with an antenna aimed skyward,
the noise temperature can be very low.
2.5. Gain
The gain of an amplier is dened as
G=
S
o
S
i
(4)
where S
o
and S
i
are the available output and input signal
powers, respectively. The gain is often dened in decibels as
G(dB) =10log(G) (5)
When N ampliers are connected in cascade, the overall
gain can be expressed as
G=G
1
G
2
G
N
(6)
or
G(dB) =G
1
(dB) G
2
(dB) G
N
(dB)
where G
1
, G
2
,y, G
N
are the gain of each individual ampli-
er.
2.6. Noise Figure
The noise gure is dened as
F =
N
o
GN
i
=
noise output of practical receiver
noise output of anideal receiver at temperature T
(7)
where N
o
is the noise at the output of the receiver, G is the
gain of the RF chain in the ratio form, and N
i
is the input
thermal noise ( =kTB). Substituting Eq. (4) into Eq. (7),
the result is
F =
S
i
=N
i
S
o
=S
o
=
signal-to-noise ratio at input of receiver
signal-to-noise ratio at output of receiver
(8)
The noise gure is often dened in decibels as
F(dB) =10 log(F) (9)
If there are N ampliers connected in cascade, the noise
gure ratio form can be expressed as
F =F
1
F
2
1
G
1
F
3
1
G
1
G
2

F
N
1
G
1
G
2
G
N1
(10)
where G
1
, G
2
,y and F
1
, F
2
,y are the gain and noise
gure of the rst, second,y ampliers, respectively, and
are expressed in power ratio rather than in decibels. From
Eq. (10) it can be shown that if the rst component in the
RF chain is a high-gain amplier, the overall noise gure
can be approximately equal to the noise gure of the rst
amplier.
2.7. Sensitivity
The sensitivity of a receiver depends on the noise power at
the input of the receiver, which is related to the bandwidth
of the RF chain, gain, and video bandwidth for an analog
receiver and the DSP algorithm used for a digital receiver.
It is often specied along with false-alarm rate and prob-
ability of detection. The false-alarm rate is dened as the
number of false measurements when there is no input sig-
nal. In some cases more restrictions can be added to the
definition of sensitivity; that is, the parameters measured
by the receiver must be within certain limits. In designing
a receiver, the sensitivity can be determined from curves
generated based on the RF bandwidth, the video band-
width, the probability of detection, and the probability of
false alarm from this authors work [1]. Once a receiver is
built, one can apply specific requirements to evaluate its
sensitivity. The sensitivity is usually frequency-depen-
dent, which means that its value varies across the fre-
quency range of the receiver.
2.8. Tangential Sensitivity
As the sensitivity of a receiver with given false alarm and
probability of detection is tedious to calculate and mea-
sure, an easily calculable and measurable sensitivity is
dened. For an analog receiver, the tangential sensitivity
is measured through visual display on an oscilloscope
that monitors the output of a video amplier following
the detector. The input must be a pulse signal. On the
scope display, when the bottom of the noise trace in
the pulse region is roughly tangential to the top of
the noise trace between pulses, as shown in Fig. 2a, the
receiver is at tangential sensitivity. At tangential sensi-
tivity, the signal-to-noise ratio is 8 dB at the output of
the detector with a standard deviation of 0.4 dB. Based
on these values, the tangential sensitivity of a digital
receiver can be illustrated from the signal-to-noise ratio
of 8 dB as shown in Fig. 2b. When there is sufcient
gain in the RF chain, the tangential sensitivity is inde-
pendent of the characteristics of the video detector. This
case is referred to as the noise-limited case. If there is
insufcient gain, the tangential sensitivity depends on
the characteristics of the detector and is referred to as
the gain-limited case. The minimum gain required for the
noise-limited case is to raise the noise oor to the tan-
gential sensitivity of the crystal detector (approximately
35 to 45 dBm). Above this gain value, the sensitivity
is gain-independent. Most modern microwave receivers
fulll this condition. The tangential sensitivity (TSS) for
the noise-limited case is
TSS= 114 10 log(F) 10 log
(3:15B
v
2:5
2B
R
B
v
B
2
o
_
) dBmfor
B
v
_ B
R
o2B
v
(11)
TSS= 114 10log(F) 10 log
(6:31B
v
2:5
2B
R
B
v
B
2
v
_
) dBmfor
B
R
_ B
v
where 114 is the noise oor of 1 MHz bandwidth, F is
the overall noise gure of the receiver, and B
v
and B
R
are
the video and RF bandwidths, respectively.
The tangential sensitivity is usually too low to be used
as the operating sensitivity level, because at this level the
8
6
4
2
0
2
4
2000 1500 1000
(b)
500 0
(a)
Figure 2. (a) Tangential sensitivity of an analog signal from an
oscilloscope display. The output signal-to-noise ratio =8 dB.
(b) Tangential sensitivity of a digital signal where the signal-to-
noise ratio =8 dB.
false-alarm rate is very high. The rule of thumb is that the
operating sensitivity is approximately 6 dB higher than
the tangential sensitivity level.
2.9. Single-Signal Dynamic Range
This dynamic range is applicable to all receivers. The low-
er limit is the sensitivity level. When an input signal is
very strong, it may cause some components in the RF
chain or the ADC in a digital receiver to become saturated.
Under this condition, the output signal will be distorted or
spurious signals will occasionally be generated. Depend-
ing on the specications of a certain receiver, the upper
limit is determined accordingly.
2.10. Two-Tone Third-Order Intermodulation Products and
Third-Order Intercept Point
If the passband of a receiver is less than an octave, the
third-order intermodulation products are the lowest-order
intermodulation products that can fall within the pass-
band. That is why they are used as the upper limit of the
two-tone spur-free dynamic range. The third-order inter-
modulation is measured with two input signals of equal
amplitude. When the input signals f
1
, and f
2
, are strong
enough to drive the RF chain into saturation, spurs will be
produced at frequencies 2f
1
f
2
and 2f
2
f
1
as shown in
Fig. 3. The outputs at these two frequencies are referred
to as third-order intermodulation. The amplitude of
the third-order intermodulation can be experimentally
measured.
Another quantity related to third-order intermodula-
tion is referred to as the third-order intercept point. The
third-order intercept point can be obtained from the third-
order intermodulation in Fig. 4. The straight line with a
1 : 1 slope represents the gain of an amplier. Another
straight line is drawn that passes the measured third-
order intermodulation point with a 3 : 1 slope that repre-
sents the anticipated amplitude change of the third-order
intermodulation as a function of input signal. The inter-
cept point between these two lines is the third-order
intercept point and the value is often read from the
output axis. It should be kept in mind that these two
straight lines are projected results. They cannot actually
be measured because, when a component is approaching
saturation, the actual measured lines will bend down-
ward. This quantity is used to determine the spur-free
dynamic range. The third-order intercept point of many
microwave components is provided by the manufacturers.
2.11. Third-Order Intercept Point of Cascade Components
The overall third-order intercept Q
i
of N components con-
nected in cascade is given by
Q=
G
1
G
2
G
N
G
1
Q
1
G
1
G
2
Q
2

G
1
G
2
G
N
Q
N
(12)
where Q
1
, Q
2
,yare the third-order intercept point of the
rst, second,ycomponents. If a component is passive, for
example, as with an attenuator, a high third-order inter-
cept value can be assigned to the device and its effect will
be negligible from that seen in Eq. (12).
2.12. Two-Signal Spur-Free Dynamic Range
This dynamic range (DR) is usually applied to receivers
that can process simultaneous signals. The lower limit is
the sensitivity level. The upper limit is reached when two
strong signals of equal amplitude begin to generate a de-
tectable third-order intermodulation. For convenience, the
detectable level is often chosen to be the noise oor. With
this assumption, the two-signal spur-free dynamic range
can be written as
DR=
2
3
(QGN
o
) dB (13)
where Q is the third-order intercept point, G is the gain of
the RF chain, and N
o
is the noise power at the output of
A
m
p
l
i
t
u
d
e
Third-order
intermodulation
Input
Frequency
2f
1
- f
2
2f
2
- f
1
f
1
f
2
Figure 3. Third-order intermodulation products. The input
signals are at f
1
and f
2
with equal amplitude and the third-order
intermodulations are at 2f
1
f
2
and 2f
2
f
1
.
Third-order
intermodulation
Third-order
intercept point
Actual
response
Actual
response
s
l
o
p
e

1
:
1
s
l
o
p
e

3
:
1
O
u
t
p
u
t

i
n

d
B
m
Input in dBm
Figure 4. Third-order intercept point. The fundamental output
has a slope of 1 : 1, and the third-order output has a slope of 3: 1.
The third-order intercept point is where the extrapolated funda-
mental and third-order outputs intercept. The third-order inter-
modulation is also shown.
the RF chain and can be expressed as
N
o
= 114F 10 log(B
R
) dBm (14)
where 114 is the noise oor of 1MHz bandwidth, F is
the overall noise gure of the receiver, and B
R
is the RF
bandwidth. To achieve a certain sensitivity and dynamic
range in a receiver, the gain of the RF chain must be of a
specific value. Although higher gain may improve sensi-
tivity, it degrades the dynamic range. However, a third-
order intercept point higher than the designed value will
not produce any adverse effect.
2.13. Two-Signal Instantaneous Dynamic Range
This dynamic range is applicable to receivers that can
process simultaneous signals. It represents the capability
of the receiver to receive a strong as well as a weak signal
simultaneously. The instantaneous dynamic range is de-
ned as the maximum amplitude separation between two
simultaneous signals such that the receiver can measure
both correctly. Generally speaking, this dynamic range
depends on the frequency separation of the two signals.
When two signals are close in frequency, the instanta-
neous dynamic range is low. When the frequencies of two
signals are widely separated, the instantaneous dynamic
range is high.
If a microwave receiver is used to intercept radar sig-
nals, two additional parameters are important: the
throughput time (throughput rate) and delay time.
2.14. Throughput Time (Throughput Rate)
If a receiver can process only one signal at a time, the
throughput time is the shortest time between two pulses
the receiver can process. If the receiver can process N si-
multaneous signals, the throughput time is the shortest
time between two groups of N simultaneous pulses the
receiver can process. The information of a pulsed signal,
that is, RF and pulse amplitude, can be obtained from the
front of the pulse, but the pulsewidth must be measured at
the end of the pulse. Therefore, the throughput time is
pulsewidth-dependent. To keep the throughput time an
intrinsic characteristic of the receiver, it should be mea-
sured with signals of minimum pulsewidth. The inverse of
the throughput time is called the throughput rate.
2.15. Delay Time
The delay time is measured from the time a pulse reaches
the input of a receiver to the time it is completely encoded.
If delay lines are used in the receiver to store information
temporarily, the delay time can be only a few microsec-
onds. Delay time is important for intercept receivers used
to generate information to respond on the same signal. If
delay time is too long, the information cannot be used by a
jammer to respond on the same signal.
3. CLASSIFICATION OF RECEIVERS
There are many ways to classify microwave receivers. One
popular classication uses operating frequency range,
that is, Ku band, X band, or extremely high-frequency
(EHF) receivers. Another way of classifying them is by
application, for example, satellite receiver, or global posi-
tioning system (GPS) receiver. Even the intercept receiv-
ers can be subdivided by application, such as warning or
electronic intelligent (ELINT) receivers. Although these
classications can reveal some specific information about
the receiver, they do not reveal the technology on which
the receiver is based.
In this article receivers are classied by their structure
and only intercept receivers will be discussed. These re-
ceivers are designed to receive different kinds of signals,
and all receivers have similar input. Different techniques
can be used to build intercept receivers. These techniques,
fundamental to receiver designs, are adopted in many
other types of receivers. For example, the superheterodyne
technique is used in most receiver designs. Almost all
communication receivers irrespective of their operating
frequency and applications often use superheterodyne
techniques.
A very important factor in a receiver is whether the re-
ceiver can process multiple simultaneous signals. Since it
is usually easy to isolate one desired signal, it is relatively
easy to build a receiver that can process only one signal at
a time. Most commercial communication receivers belong
to this category. A scanning receiver can listen to many
stations in a sequential manner, but it can process only one
signal at a time. It is relatively difcult to build a receiver
that processes more than one signal, especially when high
instantaneous dynamic range is required. High instanta-
neous dynamic range requires the receiver to distinguish a
weak signal from a spurious response, which is difcult to
achieve. For low instantaneous dynamic range require-
ment, the problem is not as severe. For example, a GPS
receiver must receive simultaneous signals from many sat-
ellites. As the signals from different satellites have about
the same amplitude, the required instantaneous dynamic
range is low and it is relatively easy to build such a re-
ceiver. If a large number of simultaneous signals need to be
processed, considerable hardware will be required and the
receiver can become rather complicated.
The receivers are divided by structure into two major
groups: analog and digital. Comparatively speaking, ana-
log receivers are technologically more mature. In fact,
most commercial and military receivers are analog. There-
fore, several techniques are used in designing analog in-
tercept receivers that will be discussed here. Although the
digital receiver is in its infancy, it is anticipated that this
kind of receiver will become popular because advances in
digital hardware and software can be applied to receiver
design and processing.
4. ANALOG INTERCEPT RECEIVERS
This discussion is limited to radar intercept receivers. The
receivers will be divided into six types depending on their
structures. The rst three types, which process only one
signal at a time, are the crystal video, superheterodyne,
and instantaneous frequency measurement (IFM) receiv-
ers. The next three types, which can process simultaneous
signals, are the channelized, Bragg cell, and compressive
receivers.
In a radar intercept receiver ve parameters will be
measured. These parameters are RF, angle of arrival
(AoA), pulse amplitude, pulsewidth, and time of arrival
(ToA). Pulse amplitude is measured from the amplitude of
the detected video pulse. Pulsewidth is measured from the
width of the video pulse. The ToA is measured from the
leading edge of the video pulse. These three parameters
are measured similarly for different receivers. The AoA is
measured from several antennas and receivers combined
together and will not be included in this article. A digital
EW receiver must also be able to generate these ve pa-
rameters as output.
4.1. Crystal Video Receiver
This is the simplest analog receiver and has existed for
many years. The receiver consists of an RF chain and a
crystal video detector. In the RF chain there is a wide
bandpass lter and RFampliers. Along with the detector
there is a video lter, a video amplier, and a comparator
as shown in Fig. 5. Sometimes, the RF ampliers are not
available to cover the desired bandwidth or they do not
have sufcient gain. Under this condition, the sensitivity
of the receiver depends on the sensitivity of the detector
and the tangential sensitivity cannot be calculated from
Eq. (11). As a result, sensitivity will be poor. To improve
sensitivity, the detector is occasionally biased in the for-
ward direction as shown in Fig. 5. However, improving the
sensitivity of the detector may decrease bandwidth. At the
output of the detector a video lter will be used to limit
the output bandwidth. The video lter bandwidth should
match the shortest pulsed signal anticipated. For exam-
ple, if the shortest pulse anticipated is 100ns, the video
lter will be approximately 10 MHz (1/100 10
9
). The
video amplier is used to amplify the video level to a level
that can be properly processed. Sometimes a logarithmic
(log) video amplier is used instead of a video amplier. A
log video amplier takes a video signal as input and gen-
erates a video output that is proportional to the logarithm
of the input signal. Finally, a comparator is used to deter-
mine whether a signal is crossing a certain threshold.
A crystal video receiver usually has a very wide input
bandwidth and often covers a bandwidth of an octave or
more. A crystal video receiver does not provide frequency
information. The only frequency information is the detect-
ed signal within the input band of the receiver. The re-
ceiver can measure pulse amplitude, pulsewidth, and ToA.
When simultaneous signals arrive at the input of the re-
ceiver, the receiver can receive and process all of them.
However, the receiver does not have the capability to in-
dicate the existence of simultaneous signals. The pulse
amplitude measured will be the sum of the amplitudes of
the simultaneous signals if the frequency separation of the
signals is much greater than the video bandwidth. How-
ever, when this condition is not met, the pulse amplitude
measured will be the vector sum of the signals. The pulse-
width measured will be from the rst leading edge to the
last trailing edge of the pulses in the group. The ToA mea-
sured will be the rst leading edge of the group.
Because of its relatively low sensitivity and poor fre-
quency accuracy, this type of receiver is no longer widely
used. However, due to simplicity of the receiver, some-
times it is used to provide AoA information. The AoA in-
formation is often obtained through amplitude comparison
from four directional antennas and four crystal receivers.
In this application, log video ampliers are used after the
detectors. The difference between two adjacent channels is
equal to the amplitude ratio due to the logarithmic rela-
tionship. The amplitude ratio can be converted into AoA
information.
4.2. Superheterodyne Receiver
In the superheterodyne concept, the input signal is chan-
ged from one frequency to a different frequency while also
maintaining all the signal information. This is very im-
portant to the technology, with most of todays communi-
cation receivers using superheterodyne techniques.
Almost all receiver systems use this methodology at one
stage or another. For example, an intercept receiver may
have an instantaneous bandwidth of 1GHz and an input
bandwidth of 18 GHz. One common approach is to divide
the input bandwidth into eighteen 1-GHz bands and
convert each band to a common frequency range through
the superheterodyne technique. The intercept receiver
is timeshared among these bands to measure the input
signals.
A generic superheterodyne receiver is shown in Fig. 6.
In this design, the rst element is a tunable lter and an
RF amplier. If higher sensitivity is desirable, an RF am-
plier can be placed in front of the lter. A combination
mixer/oscillator is used to change the input frequency to a
different frequency, usually referred to as the IF. The IF
can be either higher or lower than the input signal. If the
IF is higher than the input frequency, it is called upcon-
version. If the IF is lower than the input, it is referred to as
downconversion. The IF signal is further amplied and
ltered before reaching the crystal detector. Sometimes an
Wideband
filter
Video
amp
Video
filter
+
Detector
Comparator RF
amp
Figure 5. A basic crystal video receiver. This receiver consists of
a wideband lter, an RFamplier chain, a biased detector, a video
amplier, and a comparator. The wideband lter is used to limit
out-of-band signals. The RF amplier is used to amplify the input
signal. The detector is used to change the RF signal to a video
signal. The bias applied to the detector is to increase the detector
sensitivity. The video lter is designed to match the anticipated
minimum pulsewidth to maximize the detection efciency. The
video amplier is used to amplify the video signal. The compar-
ator is used to detect signals crossing a certain threshold.
IF logarithmic (log) video amplier is used. An IF log video
amplier takes IF as input and generates a video signal
proportional to the logarithm of the input. In an IF log
video amplier, many IF ampliers and detectors are
used. Each detector covers approximately 1015dB of dy-
namic range. Therefore, the amplier can cover a wide
dynamic range and provide high accuracy. Along with the
detector there are video ampliers and video lters to
shape the video signal.
A mixer is a nonlinear device that converts a signal
from one frequency to another. In order to change fre-
quency the device must be nonlinear. However, in receiver
design, a mixer is considered to be a linear component
with a negative gain (loss) and a third-order intercept
point. The desired output IF frequency f
IF
of a mixer is
f
IF
=f
i
f
o
or f
IF
=f
o
f
i
(15)
where f
i
is the input frequency and f
o
is the oscillatory
frequency. The plus sign is used for upconversion; the mi-
nus sign, for downconversion. If f
i
4f
o
, the rst part of the
equation is used; when f
o
4f
i
, the second part is used. The
input frequency is usually downconverted to a lower
IF, because ampliers and narrowband lters are more
available at lower frequencies. The IF is xed at a certain
value, and the frequency of the oscillator is tuned across
the input bandwidth of the receiver to nd signals. Once
the input signal is converted to the IF, the signal will be
detected and the video signal will be processed. The fre-
quency of the input signal can be measured from the pre-
ceding equation, because f
IF
and f
o
are known.
When the input frequency is higher or lower by f
IF
than
the local oscillator frequency, they will be detected by the
receiver; these are images of each other. If the input band-
width is not properly limited, both the signal and its image
can be received by the receiver. An image rejection mixer
can be used to separate signals above the local oscillator
frequency from signals below it. However, it is often de-
sirable to limit the input band of the receiver either above
or below the local oscillator frequency. Limiting the band-
width to one side of the local oscillator frequency also re-
duces the noise by 3 dB.
As a mixer is a nonlinear device, many other frequen-
cies will be generated in addition to the desired frequency.
These extraneous frequencies are referred to as spurs.
The frequencies f
I
of the spurs, including the desired IF,
can be determined by
f
I
=Mf
1
Nf
2
(16)
where f
1
and f
2
are used to represent the input and oscil-
lator frequencies; M and N are integers (either positive or
negative) and the output frequency f
I
must be a positive
value. In order to limit the spurious outputs, two lters
are often used in a superheterodyne receiver. An IF lter
with a xed center frequency f
IF
is used to reject spurs at
the output of the mixer. Because the input bandwidth is
much wider than the IF lter, several signals present at
the input of a mixer will cause it to generate spurs. A
tunable lter with bandwidth comparable to that of the IF
lter can be placed at the input of the receiver to limit the
instantaneous bandwidth and reduce spur generation.
The tunable lter and the local oscillator must be syn-
chronized, and the difference frequency between them
must equal f
IF
. Thus, one control unit is often used to
tune the tunable lter as well as the oscillator. The fre-
quency range of the input lter and the oscillator must be
wide enough to cover the input bandwidth of the receiver.
Because the lter bandwidth of a superheterodyne re-
ceiver is very narrow, the sensitivity is high. The spurious
responses generated by the mixer are carefully ltered, so
the dynamic range is usually high. It is relatively easy to
build superheterodyne receivers with matched perfor-
mance, that is, amplitude and phase, because of the
narrow bandwidth. The probability of intercept of a super-
heterodyne receiver is low and it cannot process simulta-
neous signals. Therefore, a superheterodyne receiver
alone is seldom used as an intercept receiver. It is often
used as part of an intercept system such as one that
measures AoA information through multiple antenna
and receiver combinations.
4.3. IFM Receiver
An IFM receiver uses the autocorrelations function to
measure the input frequency. A signal is correlated with
its delayed version. The outputs are lowpass ltered to
generate the desired video signals, which in turn can be
used to produce the frequency information. There are
many different ways to build a correlator by using differ-
ent RF components such as 901 hybrids and in-phase pow-
er dividers. One common approach is shown in Fig. 7.
The input signal is divided into two paths, and in one of
the paths a known delay time t is added. Four hybrids are
used to obtain the desired phase relations. The detectors
are used to perform multiplication. If the input frequency
is f, the outputs of the detectors consist basically of two
Tunable
filter
Mixer
Osc
Control
RF
amp
IF
filter
IF
amp
Detector
Video
filter
IF
amp
Figure 6. A basic superheterodyne receiver. This receiver con-
sists of a tunable lter, an RF amplier chain, a mixer and local
oscillator, an IF lter, IF ampliers, a detector, and a video am-
plier. The tunable lter and the local oscillator are controlled by
the same control circuit, and their frequencies are tracked to lter
out unwanted input signals. The RF amplier is used to amplify
the input signal. The mixer and local oscillator combined changes
the input frequency to the IF. The IF lter is used to limit spu-
rious signals generated by the mixer. The IF ampliers are used
to amplify the IF signal. The detector changes the RF signal into a
video signal, and the video amplier is used to amplify the video
signal.
termsa double frequency term and a DC term. The out-
puts of the detectors are lowpass-ltered to stop the high
frequency and pass the DC components. They represent
the autocorrelation functions of the input signal. The dif-
ferential ampliers are used to remove a constant term in
the DC components, and the outputs are sin(2pft) and
cos(2pft). As the delay time is known, the frequency of the
input signal can be found as
f =
y
2pt
(17)
where
y = tan
1
sin(2pf t)
cos(2pf t)
_ _
=2pf t
By measuring the sin(2pft) and cos(2pft), the angle y can
be found and the frequency can be calculated.
As the angle y calculated from the sine and cosine is
less than 2p, this relation limits fto1. If the desired input
bandwidth is 2 GHz, the delay time must be less than
0.5 ns. Using this t value to measure frequency, the accu-
racy is rather poor because of the poor angle resolution.
In order to cover a wide instantaneous bandwidth and at
the same time produce ne frequency accuracy, several
correlators are needed. Correlators with short delay times
are used to resolve ambiguity; correlators with the longest
delay line provide frequency accuracy. The longer the
delay line, the better the frequency accuracy a correlator
can provide. The longest delay line must be shorter than
the shortest pulse anticipated and must allow the pulse
and its delayed version to have sufcient overlap. The
delay-line lengths are commonly selected to be multiples
of each other. The two common delay-line ratios are 1: 2
and 1 : 4. Two examples will be used to illustrate these two
design ideas.
To cover a 2 GHz bandwidth, it is common practice to
select the unambiguous bandwidth wider than the desired
value, that is, 2.56GHz, and the corresponding shortest
delay time is 0.390625 ns (1/2.56 GHz). In the 1: 2 ratio
case, the delay-line lengths are 1, 2, 4, 8, 16, 32, and 64,
and the shortest delay line is considered as unit length.
There are seven correlators in this design. In the 1 : 4 ratio
case, the delay-line lengths are 1, 4, 16, and 64, and there
are only four correlators. The correlators with the longest
delay line have the same design in both cases and they
provide the frequency accuracy (usually about 1MHz).
Decoding schemes for the other correlators are slightly
different. The decoding scheme of the correlators with a
1 : 2 ratio is simpler than the ones with a 1: 4 ratio because
the former one is required to generate only 1 bit and the
latter one must generate 2 bits of information. Therefore,
both approaches are popularly adopted in IFM receiver
designs.
An IFM receiver uses a very unique RF front-end de-
sign. The RF chain of most receivers uses linear compo-
nents to avoid generating spurs and retain the amplitude
information of the input signals. In an IFM receiver it is
common practice to use a limiting amplier in the RF
chain. This kind of amplier raises all signals above a
certain threshold to a constant level. Therefore, amplitude
information is lost with the limiting amplier. A separate
circuit must be used to measure pulse amplitude. As the
limiting amplier is a nonlinear device, it generates
strong spurious responses. When two signals are present
in the amplier, the strong one will suppress the weak
one. This is referred to as the capture effect. Because an
IFM receiver can process only one signal at a time, the
capture effect will enhance the receiver performance
to measure the strong signal under simultaneous signal
conditions.
An IFM receiver has many advantages over other types
of intercept receivers. The receiver can cover a very wide
instantaneous bandwidth [possibly 16 GHz (218 GHz)]
and provide ne frequency accuracy on short pulses, for
example, 1 MHz accuracy on a 100ns pulse. As the struc-
ture of the receiver is very simple, the receiver can be very
compact, low cost, and reliable. The only deciency is that
the receiver cannot process simultaneous signals. Not only
is it unable to process simultaneous signals, but simulta-
neous signals with amplitude within about 5 dB may
cause the receiver to produce an erroneous frequency re-
port without reporting the mistake. This is considered a
major problem with the IFM receiver and limits its usage.
However, the concept of the IFM receiver is very impor-
tant and can be adopted in many receiver designs, includ-
ing digital receivers.
4.4. Channelized Receiver
A channelized receiver can intercept simultaneous sig-
nals. The basic concept is very simple. It uses a bank of
lters with adjacent frequencies to separate signals. Sig-
nals with different frequencies will exit from different l-
ters. By measuring the outputs of the lters one can
determine the frequencies of the input signals. A basic
channelized receiver, shown in Fig. 8, consists of four
90
Hybrid
Detector
Diff
amp
sin 2 f
90
Hybrid
Delay
line
Power
divider
80
hybrid
cos 2 f
90
hybrid

Figure 7. A correlator for the IFM receiver. The correlator con-
sists of a power divider, a delay line of known delay time, three 901
and one 1801 hybrids, four detectors, and two differential ampli-
ers. The power divider divides the input signal into two parallel
paths, and the signal in one path is delayed by a known time. The
four hybrids are used to provide the necessary phase shifts. The
detectors are used to convert RF signals into video signals as well
as to perform multiplication of two signals. The two differential
ampliers are used to cancel a DC bias term. The outputs from
the correlator are sin(2pft) and cos(2pft).
major components. The rst one is the RF chain. The sec-
ond component is a lterbank with consecutive center fre-
quencies. In order to retain the amplitude information, IF
log video ampliers are used after the lters. The last
component is a parameter encoder, which takes the out-
puts from the lters and converts them into the desired
information. These components are discussed here.
The RF chain usually consists of ampliers, lters, and
mixers to shift the input frequency. The RF chain must
operate in the linear region to avoid generation of spurs.
As the bandwidth of a channelized receiver is wide, some
spurs generated from mixers will not be ltered out but be
present in the output.
The output of the RF chain is fed to the input of the
lterbank. A common way to feed the lterbank is through
a power divider. Using a power divider can improve im-
pedance matching to the lters, but they cause insertion
loss. Every time a signal is divided into two paths there is
a 3dB loss. These losses must be recovered by placing ad-
ditional ampliers in the receiver. Although a frequency
multiplexer is a better approach to feed the input of the
lter, it is usually difcult to achieve the desired lter re-
quirements. In each lter usually only one signal can be
processed. If more than one signal is in one lter, they will
be processed as one signal and may produce erroneous re-
sults. Depending on the center frequency of the lters and
their bandwidth, different techniques can be used to build
the lters, for example, surface acoustic wave (SAW) tech-
nology and lumped LC elements. The two general require-
ments of these lters are low insertion loss and small size.
The lter must also have low sidelobes in the frequency
and time domains. The required shape of the lter is usu-
ally based on the encoding circuit design.
The outputs of the lters are further amplied by IF
log video ampliers. If the amplitude information after a
lter is not of interest, a limiting amplier followed by a
crystal detector can be used to convert the IF into a video
signal. With only one signal processed within a lter,
intermodulation and spurs are not of concern. Therefore,
in designing a receiver, the third-order intercept point in
the IF channel is not of concern. If amplitude is not re-
tained after the IF ampliers, pulse amplitude informa-
tion must be obtained from another part of the receiver.
A parameter encoder takes the video signals from all of
the lters as input and produces the desired information.
Most of the effort in encoder design is directed toward ob-
taining frequency information. There are usually two
ways to obtain frequency information. One is to compare
amplitudes from adjacent lter outputs. In this approach,
IF log video ampliers must be used to generate amplitude
information at the lter outputs. One problem with this
approach is that it is difcult to balance the gain in all the
channels. If the gain from one channel changes slightly,
that is, due to temperature drift, the encoder must be ad-
justed accordingly. Sometimes it can be a major problem in
a receiver with many channels. Another approach is to
detect the transient response of a lter output. If a pulsed
signal passes the center of a lter, the transient effect is
not significant, which means that the pulseshape is slight-
ly distorted. On the other hand, if a pulsed signal passes
the skirt of a lter, the transient effect is very significant,
which means that the pulseshape is drastically distorted.
By measuring the transient on the output pulse, one can
determine whether a signal is in the middle or on the skirt
of a lter. In this design, both limiting ampliers and IF
log ampliers can be used after the lters. The problem
with this approach is that the variation on the leading
edge of the pulse can change the transient response and
the performance of the encoding circuits. In some receiver
designs both the amplitude comparison and transient phe-
nomenon are used to obtain better results. The lter
shape, which controls the amplitude and the transient of
the video signal, is often determined from the encoding
circuit design. The encoder design is the most critical
element in building a channelized receiver.
In a receiver that can process simultaneous signals,
there are two quantities related to frequency measure-
ment. One is frequency resolution, which tells the receiver
to separate two simultaneous signals that are close in fre-
quency. The other one is frequency accuracy, which is re-
lated to the error in the frequency measurement. The
minimum pulsewidth determines the minimum lter
bandwidth, which determines the frequency resolution of
the receiver. Frequency measurement must be carried out
after the transient dies off. The transient is approximately
equal to the inverse of the lter bandwidth. During the
transient period both the shape and the RF of the pulse
change. If the desired minimum pulsewidth is 100ns, the
minimum bandwidth of the lter is approximately 10MHz
and the frequency resolution is close to 10 MHz. However,
the receiver can be designed with a lter bandwidth that
is much wider than the minimum lter bandwidth. A
wideband lter has a shorter transient time. If addition-
al processing, that is, the concept of an IFM receiver is
used, better frequency accuracy can be obtained. It is dif-
cult to design a receiver with both high-frequency reso-
lution and frequency accuracy on short pulses because it
takes a longer time to generate a ne frequency reading.
In general, the anticipated minimum pulsewidth deters
Input
RF
chain
Consecutive
filter bank
Parameter
encoder
IF Log video
amplifiers
Output
Figure 8. A basic channelized receiver. The receiver consists of
an RF amplier chain, a lterbank with contiguous lters, IF log
video ampliers, and a parameter encoder. The RF amplier is
used to amplify the input signal. The lterbank is used to sepa-
rate signals with different frequencies into different output ports.
The IF log video ampliers amplify the IF signals and convert
them into video signals. The parameter encoder generates the
desired digital information from the input video signals.
the minimum lter bandwidth and thus the minimum
frequency resolution. This design rule applies to all re-
ceivers with simultaneous signal capability.
Because the concept of channelized receivers is very
simple, a general misunderstanding is that it is very easy
to build. As a result, the design goals might be set too high
to achieve. In general, a receiver has fewer problems when
the lter bandwidth is wide because the receiver has fewer
parallel channels. However, wider bandwidth means poor
frequency resolution and lower sensitivity. When a high
instantaneous dynamic range is desired, the receiver must
detect a weak signal in the presence of strong signals.
Sometimes it is difcult to distinguish a spurious response
from a weak signal. Therefore, a high-dynamic-range
receiver may produce a spurious signal report. When the
instantaneous dynamic range requirement is low, the
receiver can be designed to produce fewer spurious res-
ponses.
The input bandwidth of a channelized receiver is pro-
portional to the number of channels and their bandwidth.
When a large number of channels are built, a channelized
receiver can be bulky and expensive but have better fre-
quency resolution. If the channel bandwidth is wide, a
small number of channels can cover a wide frequency
range. Because the encoding circuit also has fewer inputs,
the receiver can be relatively small.
4.5. Bragg Cell Receiver
A Bragg cell receiver can be considered another type of
channelized receiver where channelization is accom-
plished by optical means. The term Bragg cell receiver
derives from the concept of the Bragg angle of optical dif-
fraction. A basic Bragg cell is shown in Fig. 9. The optical
arrangement will be discussed rst. A laser is used as a
coherent light source. A diode laser is often preferred to a
gas laser because of its small size. The beam expander and
the collimator are used to form the lightbeam into the de-
sired shape to shine on the Bragg cell. A Bragg cell is used
to diffract input light. A Bragg cell is made of a crystal
transparent to the laser light. At one end of the crystal is a
transducer to change the input electric signals into acous-
tic signals. At the other end of the Bragg cell is absorption
material used to eliminate acoustic wave reection to
avoid the generation of standing waves. The acoustic
wave produces variations in the refractive index of the
crystal. This refractive index modulation causes the laser
beam to deect. The timebandwidth product (TBP) of a
Bragg cell is dened as the time of the acoustic wave trav-
eling through the window of the Bragg cell multiplied by
the bandwidth. The laser beam is incident on the Bragg
cell at the Bragg angle where the desired diffracted light is
at maximum intensity. The light output from the Bragg
cell passes through a Fourier transform lens and focuses
on a photodetector array. The photodetector changes the
optical signal back into a video signal.
The input signal to the receiver is applied to the trans-
ducer of the Bragg cell through an RF chain containing
ampliers and lters. There is an impedance matching
network to match the output of the RF chain to the input
impedance of the Bragg cell. The output power from the
RF chain should keep the Bragg cell operating in its linear
region. High input power can drive the Bragg cell into
the nonlinear region and generate spurious responses.
Transducer
Photodetector
array
Parameter
encoder
Fourier
transform
lens
Bragg
cell Collimeter
Amplifier
Input
Filter
Beam
expander
Laser
Outputs
Figure 9. A basic Bragg cell receiver. The receiver consists of an RF chain, a laser, a beam ex-
pander, a collimator, a Bragg cell, a Fourier transform lens, a photodetector array, and a parameter
encoder. The RF amplier is used to amplify the input signal. The beam expander and the colli-
mator shape the laser light into the desired shape. The input signal is converted into an acoustic
wave in the Bragg cell to form a grating and diffract the light from the laser. The Fourier transform
lens focuses the light on the photodetector array, which converts the light into video signals. The
light position on the photodetector array determines the frequency of the input signal. The pa-
rameter encoder generates the desired digital information from the input video signals.
The input signal is converted into a traveling acoustic
wave in the Bragg cell. The position of the diffracted light-
beam on the detector array is related to the frequency of
the input signal. If more than one signal is present in the
Bragg cell, there will be multiple outputs on the detector
array. The maximum number of outputs is equal to the
timebandwidth product of the Bragg cell. The amplitude
of the output is related to the power of the input signal.
The video outputs from the photodetector array can be
considered as outputs from a lterbank with adjacent fre-
quencies. The parameter encoder takes these outputs and
generates the desired information. In a channelized re-
ceiver, IF log video ampliers or limiting ampliers are
used after the lters to provide more RF gain. To provide
the same gain in a Bragg cell receiver, a light amplier
must be used in each output channel, however, such a de-
vice is not available with todays technology. Thus, the
sensitivity of the receiver depends on the laser power and
the characteristics of the photodetectors. A laser with
higher power can be used to improve sensitivity. As the
outputs from the photodetectors are video signals, some of
the encoding schemes applicable to a channelized receiver
may not be adopted for Bragg cell receivers. Another dif-
ference between the photodetector and the crystal detector
is that the output from a photodetector is proportional to
the input light power and, therefore, proportional to the
input power. The output from a crystal detector is propor-
tional to the voltage of the input signal, which is related to
the square root of the input power. Therefore, a crystal
detector covers more dynamic range.
Compared to a channelized receiver, a Bragg cell re-
ceiver has less dynamic range. The major advantage of a
Bragg cell receiver is its simplicity and compactness. A
Bragg cell receiver including the optical bench (consisting
of beamforming lenses and Bragg cell), laser and detector
array, but excluding the RF chain, is only a few cubic
inches in size. Such a small receiver can provide over 100
parallel outputs. The size of the optical bench can be fur-
ther reduced using integrated optics. In the integrated
optical approach, the entire optical bench can be fabricat-
ed on a single chip. The light is transmitted through a
light waveguide and the Bragg cell uses SAW technology.
All optical components can be made on the chip, and the
laser and detector array can be attached from the ends of
the chip. Thus, the size can be a few cubic centimeters.
The Bragg cell receiver already discussed here is often
referred to as the power Bragg cell receiver, because the
detector output is proportional to the power of the input
signal. In order to improve the dynamic range of a Bragg
cell receiver, an interferometric approach can be used. The
main goal of this approach is to make the photodetector
proportional to the voltage of the input signal rather than
the power. In this arrangement, the laser beam is split
into two paths through a beamsplitter with a Bragg cell in
each path. The input to one Bragg cell is a locally gener-
ated spread-spectrum signal covering the entire band-
width of the Bragg cell. The input signal is applied to
the other Bragg cell. The outputs of the two Bragg cells are
focused on a photodetector array, and the location repre-
sents the input frequency. Each photodetector is used as
an optical mixer and the output is an IF signal rather than
a video signal. Thus, IF ampliers and lters can be used
to improve the performance of the receiver. After the IF
chain, crystal detectors are used to convert the IF signals
into video signals. Although this approach has the poten-
tial to improve the dynamic range of the receiver, it
sacrices the simplicity of the Bragg cell receiver. Indi-
vidual IF channels have to be built separately and the
optical bench is equivalent to the lterbank in a chan-
nelized receiver.
As in designing channelized receivers, the parameter
encoder of a Bragg cell receiver is a major portion of the
effort.
4.6. Compressive Receiver
A compressive receiver can also process simultaneous sig-
nals. As opposed to a channelized receiver where all out-
puts are in parallel, the outputs from a compressive
receiver are in series. A basic compressive receiver is
shown in Fig. 10. In this gure the RF chain is not in-
cluded. The two major components in a compressive re-
ceiver are the local oscillator and the dispersive delay line.
The output from the local oscillator is a repetitive FM sig-
nal. The frequency range of the FM signal can be very
wide and the period very short, such as 1GHz/200 ns. This
receiver is also called a microscan receiver because the
scan time is short. The input signal after mixing with the
local oscillator output is converted into an FM signal. This
FM signal passes through the dispersive delay line (often
referred to as compressive line) and is compressed into a
pulse. The position of the pulse relative to the beginning of
the scan represents the frequency of the input signal. The
amplitude of the compressed pulse represents the ampli-
tude of the input signal.
The instantaneous bandwidth of the compressive line is
generally equal to the IF bandwidth of the receiver. The
input bandwidth can be either wider or narrower than the
IF bandwidth. In one common design, the local oscillator
scans the sum of the IF bandwidth and the receiver input
bandwidth. In order to compress the FM signal from the
output of the mixer into a short pulse, the frequencytime
slope of the local oscillator and the slope of the compres-
sive line must be properly matched. In most designs, the
compressive line is made from SAW technology with band-
widths of up to 1GHz and dispersive delays of a few hun-
dred nanoseconds. The local oscillator is often
implemented from a dispersive delay line because a short
pulse applied to the input of a dispersive delay line gen-
erates an FM signal at the output. This signal can be am-
plied and used as the local oscillator output. Because the
frequency of the local oscillator is wider than the com-
pressive line bandwidth, the requirements on the disper-
sive delay line to generate the local oscillator signal are
more stringent. An oscillator built from a dispersive delay
line is easier to match the frequencytime slope of the
compressive line. The timebandwidth product of the com-
pressive line is dened as the dispersive delay time mul-
tiplied by the bandwidth. The maximum number of
compressed pulses generated per scan is approximately
equal to the timebandwidth product of the compressive
line. The maximum number of compressed pulses can be
considered equivalent to the total number of parallel chan-
nels in a channelized receiver.
The compressed pulse is very short and is inversely
proportional to the bandwidth of the compressive line. The
pulse output rate equals approximately the bandwidth of
the compressive line. If the bandwidth of the compressive
line is 1 GHz, the pulsewidth is close to 1 ns (1/1 GHz) and
the output rate is about 1 GHz. The center frequency of
the pulse equals the center frequency of the compressive
line. The compressed pulse has a main peak and many
sidelobes. The sidelobes should be reduced to simplify the
frequency encoding circuit design. The sidelobes can be
reduced by adding a weighting (window) function to the
compressive line. An IF log video amplier is used to
convert the compressed pulse into video pulses for
further processing. The video bandwidth of the IF log
video amplier must be wide enough to accommodate the
compressed pulsewidth.
A compressive receiver usually intercepts a pulsed
signal in many consecutive scans. It is desirable to report
the information on the intercepted signal on a pulse-
by-pulse basis rather than a scan-by-scan basis. The pa-
rameter encoder will sort and combine the scan-by-scan
information into pulse-by-pulse information. The outputs
from each scan are the number of simultaneous signals
intercepted during the scan time. From this operation, it
is easily seen that the time resolution in generating
pulsewidth and ToA equals the scan time. Thus, it is
usually coarser than the time resolution in other types
of receivers.
The main function of the parameter encoder is to nd
the frequency of the input signal. Because the compressed
pulse output rate is usually rather high, the encoder
must be able to process the signals at the same rate.
Although the parameter encoder is quite different from a
channelized receiver, they face the same basic challenges:
detecting the signal and avoiding the sidelobes and spu-
rious responses. The bandwidth of a compressive receiver
is limited by the technology used in the compressive line
as well as the speed of the logic circuit in the parameter
encoder.
There are several different ways to design a compres-
sive receiver. One of the most common designs is to make
the receiver bandwidth equal to the IF bandwidth. Under
this condition, the outputs of the compressive line are
present only 50% of the time. This affects the probability of
intercept. If a short pulse (less than half the scan time)
falls in the silent half, the receiver will miss the pulse. To
improve the probability of intercept, another mixer and
local oscillator can be added in parallel. The beginning of
the FM signal from this local oscillator is shifted to the
middle of the original one. For example, if the scan time is
200 ns, the original scan is from 0 to 200 ns and the addi-
tional scan is from 100 to 300 ns. Outputs from both chan-
nels are combined and fed into the compressive line. With
this arrangement, the compressive line can have output
100% of the time. This technique is called the interlace
scan. The outputs from the IF log video amplier are in
series, so less hardware is required in the parameter en-
coder as compared with a channelized receiver; however,
operational speed is very high and matches the bandwidth
of the compressive line.
The minimum pulsewidth that a compressive receiver
can process is approximately equal to the scan time or half
the scan time if the interlace scan is used. The frequency
resolution is close to the inverse of the minimum pulse-
width. Compared with a channelized receiver, it shows
that the same laws of physics govern the performance of
both receivers; that is, the frequency resolution is inversely
proportional to the pulsewidth.
Mixer
Dispersive
delay line
Time
F
r
e
q
u
e
n
c
y
RF Log video
amplifier
Parameter
encoder
Output Input
Time
F
r
e
q
u
e
n
c
y
LO
Figure 10. A basic compressive receiver. The receiver consists of a local oscillator, a mixer, a dis-
persive delay line, an RF log video amplier, and a parameter encoder. The local oscillator gen-
erates a linear FM signal that changes the input to a linear FM signal through the mixer. The
dispersive delay line compresses the FM signal into a short pulse. The RF log video amplier is
used to generate the logarithm from the compressed pulses. By measuring the output time of the
compressed pulse with respect to the beginning of the scan, the signal frequency can be deter-
mined. The parameter encoder converts the scan to scan information to a pulse-by-pulse infor-
mation.
5. DIGITAL RECEIVERS
A digital receiver consists of three building blocks: the RF
chain, the ADCs, and the DSP. Once a signal is digitized
the data are less affected by ambient conditions such as
temperature changes. In an analog receiver, the perfor-
mance of the components may change because of temper-
ature variation and aging, while digital circuits do not
have these problems. The digitized data can be processed
with many different DSP approaches. Although most of
the DSP are still hardwired for receiver applications, in-
creasing the processing speed may permit changing re-
ceiver functions through software switching. Thus, the
software receiver concept is becoming popular. With this
concept many different receivers can be implemented us-
ing the same hardware. The software receiver idea is par-
ticularly popular for military communication receivers
because of its potential versatility.
Strictly speaking, an ADC is a nonlinear device. It can
be considered as a linear device for a large number of
quantization levels. The discussion in the following sec-
tions is based on the linear model of an ADC. Nonlinear
operation still exists if strong signals drive the ADC into
saturation. Under this condition, the output of the receiv-
er may produce many spurious responses. Special proce-
dures must be considered to deal with this phenomenon.
To cover a wide bandwidth, the ADC must operate at
high speed. The Nyquist sampling theorem requires that
the minimum sampling speed be twice the information
bandwidth. To cover a wide dynamic range, the ADC
must have a large number of bits. The number of quan-
tization levels is related to the number of bits b as 2
b
. If
the dynamic range is dened from a signal at the highest
level to a signal at the lowest level, it can be readily
expressed as
DR=20 log(2
b
) - 6b (18)
In this equation, the ADC is assumed ideal, that is, the
quantization levels are uniform and there is no jitter in
the sampling window. In a typical ADC, there are noise,
nonuniform quantization level and sampling window jit-
ter. An effective bit is often used to characterize a nonideal
ADC. The number of effective bits is less than the actual
number of bits and is dened as [2]
b
eff
=b log
2
RMSerror (actual)
RMSerror (ideal)
_ _
(19)
The effective bits change with frequency and have fewer
bits at higher frequency. The number of effective bits
should be used in Eq. (18) to determine the dynamic
range. If the highest spurious response is used as the low-
er limit of a receiver, the dynamic range varies. However,
Eq. (18) provides a simple estimation.
Digital receivers can be divided into two generic
groups: narrowband and wideband receivers. Usually,
one can consider a narrowband receiver to receive only
one kind of signal and the receiver is not required to
process simultaneous signals. For each type of signal a
narrowband receiver will be designed individually. There
are exceptions to this definition. For example, a GPS C/A
(coarse acquisition) code receiver has a narrow bandwidth,
because the input signals have the same frequency. How-
ever, this receiver can receive multiple signals with dif-
ferent code. Awideband receiver can receive simultaneous
signals over a wide frequency range. The input signals can
be either known or unknown. For a wideband communi-
cation receiver the signals are known. For an intercept
receiver the signals are unknown.
Another way of differentiating narrowband and wide-
band digital receivers is by the frequency tuning schemes.
In a narrowband receiver, the frequency tuning is accom-
plished through analog means by changing the frequency
of the local oscillator and selecting the proper lters. For
this type of tuning, input to the ADC is usually an isolated
signal rather than the full input bandwidth of the receiver.
Thus, the ADC and the DSP following it can operate at a
lower frequency.
In a wideband digital receiver the frequency tuning
method is implemented digitally. The sampling frequency
of the ADC and the following DSP must be high enough to
accommodate the input bandwidth of the receiver. For ex-
ample, one can build an FM radio through either a nar-
rowband or a wideband approach. The input bandwidth of
an FM radio is 20 MHz (88108 MHz). If a single FM sta-
tion is selected by analog tuning, the signal bandwidth is
200 kHz. If the sampling rate is 2.5 the signal bandwidth,
an ADC with a sampling speed of 500 (2.5200) kHz is
required. If the frequency will be tuned digitally, the sam-
pling speed must be 50 (2.520) MHz. Although the signal
information bandwidth is only 200 kHz, the following DSP
must match this operational speed of 50 MHz. The poten-
tial advantages of digital tuning are superior lter shape
and exible tuning ability.
In the following sections, a narrowband receiver with a
band folding concept and a wideband channelized receiver
will be presented.
5.1. Narrowband Digital Receiver
The Nyquist sampling theory requires that the minimum
sampling speed be twice the signal bandwidth, not twice
the highest frequency of signals. For example, the C/A
code of the GPS signal is at L
1
band (1575.42MHz) with a
bandwidth of 2 MHz from null to null. The minimum sam-
pling frequency to acquire this signal is 4MHz, although
the signal is at 1575.42 MHz. If the input bandwidth of the
ADC can accommodate the 1575.42 MHz frequency, the
signal can be sampled directly at slightly higher than
4 MHz. Sampling will alias the input signal to a baseband
as shown in Fig. 11. In this gure the sampling frequency
is f
s
; thus, the maximum unambiguous bandwidth is f
s
/2.
All the input bandwidth from 0 to (n1)f
s
will fold into
the bandwidth f
s
/2. From this frequency folding property,
the RF chain of the receiver should be designed with at
least two bandpass lters. The rst one should be placed
near the front of the RF chain to reject out-of-band signals
as in an analog receiver. The second lter should be placed
in front of the ADC to limit the out-of-band noise gener-
ated from the ampliers in the RF chain.
One can design a receiver to receive several narrow-
band signals separated in frequency through sampling at
a proper frequency. An example will be used to illustrate
this approach. The Y code of the GPS signal at L
2
band has
a bandwidth of 20 MHz centered at 1227.6 MHz. If one
desires to receive the C/A code at L
1
and the Y code at L
2
of
the GPS signals, the total bandwidth is 22 MHz (220).
The minimum required sampling frequency should be
44 MHz (2 22). Some specific sampling frequencies can
be used to fold the two signals into the baseband. One
sampling frequency is 51.6MHz. Under this condition, the
Y code is aliased to 525MHz and the C/A code is aliased
to 0.622.62 MHz. These two frequency bands are not
overlapped in the baseband. In general, it is desirable to
fold the two signals into separate regions of the baseband
to avoid interference. If interference is not a problem, it is
possible to overlap the two signals to save bandwidth.
However, overlapping two signals into one frequency
range also folds the noise together. The noise oor will in-
crease at the overlap region. This concept can be extended
to more than two signal bands.
5.2. Wideband Digital Receiver
The most popular wideband digital receiver is a digital
channelized receiver. The basic idea is the same as it is for
an analog channelized receiverto separate the input into
many parallel consecutive frequency channels. In a digital
channelized receiver, the lterbank is implemented digi-
tally. To cover a wide bandwidth and high dynamic range,
the ADC must operate at very high speed and have many
output bits.
The simplest way to build a channelized digital receiver
is to use discrete Fourier transform (DFT) implemented
using the fast Fourier transform (FFT). The FFT should
operate on the input data on a continuous manner. The
FFT outputs from different time intervals can be consid-
ered as the outputs of each individual lter. This tech-
nique requires the FFT to operate at high speed. If a
receiver covers a 1 GHz bandwidth, the sampling speed
must be above 2 GHz. Suppose that the sampling speed is
2560 MHz and a 256-point FFT can produce 256 parallel
channels. However, only 128 channels provide useful in-
formation, and the other 128 outputs are the complex con-
jugate of the rst ones. It takes 100ns to accumulate 256
data points and the output channel bandwidth is 10 MHz
(1/100 10
9
s). As the bandwidth of the receiver is
1 GHz, only the center 100 outputs will be used and the
remaining 28 end channels will not be monitored. To pre-
vent missing data, the FFT must perform 256 point FFTat
least every 100ns. If any overlap of the FFT is desired, the
processor must operate at a faster speed.
In general, it is difcult to match the digital processing
speed to the speed of the ADCs. One common way to solve
this problem is to reduce the total number of output chan-
nels, but keep the input at 256 data points. This approach
reduces the frequency resolution of the receiver. For ex-
ample, one can perform a 32-point FFT to reduce the out-
put to sixteen 80-MHz channels. Under this condition, a
bandpass lter will be used to limit the input bandwidth to
1 GHz, which will partially block the input of the two end
channels. The input data are decimated 32 times, then
each of the 8 data points will convolve with an 8-point l-
ter as shown in Fig. 12. The coefcients of the 32 lters are
obtained from decimating a 256-point lter, Which is de-
signed to shape the frequency response of the 16 lters.
The 32 outputs from the lters become the input to the 32-
point FFT. In order not to miss data, a 32-point FFT must
be performed every 100 ns, which is a less stringent re-
quirement than the previous case. If the FFT can operate
at a higher speed, the input data can be processed in an
overlap mode, that is, performing FFT every 50, 25, or
12.5 ns. This overlap mode is a common practice in chan-
nelized receiver design. Finally, the outputs from the FFT
must be correctly decoded to generate the desired infor-
mation. If the receiver is used to intercept radar signals,
the encoder will provide the ve parameters: frequency,
AoA, ToA, pulse amplitude, and pulsewidth.
5.3. Hybrid Receivers
One common hybrid receiver design is to build a wideband
receiver such as channelized or compressive receiver with
a coarse frequency resolution. For example, one can divide
the input bandwidth of 1GHz into 1020 uniform bands
with a frequency resolution of 10050 MHz. Once a signal
is detected and its coarse frequency is measured, a
narrowband receiver or receivers can be used to obtain
ne-grain information on the signal. One approach is to
rapidly tune a narrowband IFM receiver to measure the
Y
0
Y
1
Y
2
Y
3
Y
k
8 pt filter
h
0
h
1
h
2
h
3
h
k
Y
31
h
31
x
0
x
1
x
2
x
3
x
k
x
31
32 pt
FFt
Figure 12. A 32-output lterbank using decimation scheme. The
main goal is to perform a 32-point FFTwith 256 input data points.
The input is decimated into 32 parallel channels. In each channel,
there is an 8-point lter. The output from each lter contains
8 input data points. These outputs are used as the input to the
32-point FFT.
O
u
t
p
u
t
f
r
e
q
u
e
n
c
y
f
s
/2
f
s
/2 2f
s
Input frequency
nf
s
(n + 1)f
s
f
s
0
f
f
Figure 11. A sampling scheme transferring the input to base-
band. The sampling frequency is f
s
; the maximum unambiguous
bandwidth is f
s
/2. A signal with bandwidth of Dfof
s
/2 in the fre-
quency range of nf
s
to (n1/2)f
s
can be transferred to the base-
band of 0 to f
s
/2 through sampling.
frequency of the input signal, because the IFM receiver
can provide improved frequency resolution. Another de-
sign is to rapidly tune several narrowband receivers that
are connected to different antennas to measure the AoA of
the input signal. In this design, it is easier to match the
amplitude and phase of narrowband receivers. It is a
common practice to combine these two approaches in one
hybrid receiver design. These approaches can be imple-
mented in both analog and digital receivers. In an analog
hybrid receiver design, RF delay lines must be used to
temporarily store the input signal while the narrow-
band receiver or receivers can be tuned to the desired
frequency. Wideband low-loss RF delay lines are not avail-
able and this is one of the major problems in analog hybrid
receiver design. In digital design, since the digitized data
can be stored easily, delay lines are no longer required.
Once the basic properties of a receiver are understood,
they can be combined in various ways to solve specific
problems.
6. COMPARISON OF DIFFERENT TYPES OF RECEIVERS
It is difcult to compare the performance of different types
of receivers. For example, one can consider a channelized
receiver to be complicated if there are a large number of
channels. If there are only a few channels, the receiver can
be rather simple. The same argument holds for Bragg cell
and compressive receivers. To make the assessment mean-
ingful, it is assumed that the four types of receivers
channelized, Bragg cell, compressive, and digital chan-
nelizedall have the same number of output channels.
The other three types of receiverscrystal video, super-
heterodyne, and IFMcannot process simultaneous sig-
nals. It is also difcult to put quantitative measures on the
performance. For one specific receiver this should be the
correct way to present its performance. However, for one
type of receiver it may be difcult to do so. For example,
one channelized receiver is designed for higher sensitivity
and lower dynamic range, and another one is designed for
higher dynamic range and lower sensitivity. If the best
performance of each receiver is reported, the results can
be misleading because they cannot be achieved in one
receiver design. Therefore, the performance is listed in a
qualitative manner as in Table 1.
BIBLIOGRAPHY
1. J. B. Y. Tsui, Microwave Receivers with Electronic Warfare
Applications, Wiley, New York, 1986.
2. J. Keffel, T. R. McComb, and R. Malewski, Comparative eval-
uation of computer methods for calculating the best t sinu-
soid to the high purity sine wave, IEEE Trans. Instrum. Meas.
IM-36:418422 (June 1987).
MICROWAVE RESONANCE PLASMA SOURCE
ALEXANDER L. TAUBE
GREGORY M. DEMYASHEV
Swinburne University of
Technology
Melbourne, Australia
1. INTRODUCTION
Low-temperature thermal plasmas operating at atmo-
spheric pressure are used for a variety of contemporary
plasma-enhanced technologies, such as plasma chemistry,
materials processing, coating deposition, surface treat-
ment, and waste treatment. Conventional plasma spray
sources running at atmospheric pressure are based on
electric arc [1] or radiofrequency (RF) [26]. The currently
employed industrial plasma spray sources have a number
of disadvantages [7], including low efciency of energy
consumption, plasma contamination, and low efciency of
precursor consumption.
Generally a microwave plasma electrodeless concept
is based on using microwave electromagnetic energy in
order to form and sustain a plasma discharge. There are
various ways to supply the microwave energy for plasma
generation [8].
Ionization of feed gases and precursors by electric eld
is dened by the ratio of electric eld strength to gas
Table 1. Receiver Performance
Crystal Video Superheterodyne IFM Channelized Bragg
a
Cell Compressive Digital
b
Channel
Instantaneous BW Very wide Narrow Very wide Wide Wide Wide Wide
Simultaneous signal
capability
None None None Good Good Good Good
Frequency accuracy Poor Excellent Excellent Good Good Good Good
Sensitivity Poorfair Excellent Good Good FairGood Good Good
Single-signal DR Fair Excellent Excellent Good FairGood Good Good
Two-signal instant DR N/A N/A N/A Good Fair Good Good
Two-signal spur-free DR N/A N/A N/A Good Fair Good Good
Structure Simple Moderate Simple Complex Moderately
complex
Complex Complex
Size Small Small/moderate Small Bulky Smallmoderate Moderate Bulky
a
This represents the power Bragg cell receiver. The interferometric Bragg cell receiver performance should be comparable to a channelized receiver, but the
structure is more complicated.
b
The digital receiver is in the development stage, and performance should improve in the near future.
2934 MICROWAVE RESONANCE PLASMA SOURCE
pressure [9]. Depending on a working pressure in a cham-
ber, conventional microwave plasma discharges can be
classied as follows:
*
High-vacuum electron cyclotron resonance plasma
*
Low-pressure microwave plasma
*
Atmospheric-pressure waveguide-based setup micro-
wave plasma
The electron cyclotron resonance (ECR) plasma is
based on the cyclotron phenomenon of electrons in a stat-
ic magnetic eld, which allows transfer of the micro-
wave energy to plasma. The ECR plasma operates in
fairly high vacuum (10
1
10
3
Pa) [1013]. The ECR
can be realized through the proper combination of a mi-
crowave frequency and strength of magnetic eld. The
frequency of the alternating electric eld is made to match
the natural frequency of the electrons orbiting magnetic
lines of the magnetic eld. For 2.45 GHz the required
strength of magnetic eld is B875G. The ECR plasma
has the electron density of 10
10
10
12
cm
3
and a high
degree of ionization (10%). The disadvantages of ECR
plasma sources include
*
Difcult process control
*
Costly equipment due to the added variable magnetic
eld and due to high vacuum
*
Low rate of deposition (B2nm/min)
For example, ECR has been used for deposition of AlN
[14], SiO
2
[15], SiN
x
O
y
[12], microwave plasma heating
[16], tungsten incorporated carbon lms (W-C:H) [17], and
other applications.
Low-pressure microwave plasma discharge (LPMPD)
runs at relatively low pressure (133 Pa13kPa) [13,1822].
Low pressure implies more collisions between electrons
and neutral molecules.
The LPMPD can be realized in either a waveguide set-
up, where a dielectric tube evacuated to low pressure is
put through a waveguide [23], or a dielectric jar (or dis-
charge chamber [24,25] evacuated to low pressure, where
the dielectric jar is located inside a vessel supplied with
microwave power. Quartz (or alumina) is commonly used
as materials for the dielectric tube/jar.
The waveguide setup utilizes the standing-wave mode
of TE
01
. A movable plunger is used to tune the electric
eld maximum of the standing wave so that the maximum
coincidences with quartz tube location.
The LPMPD has a few peculiarities. The electrons have
a temperature of 10
3
10
4
K and a plasma electron density
of 10
9
10
14
cm
3
. The temperature of heavy particles
(ions, radicals, species, etc.) in LPMPD is approximately
500K. The LPMPD is commonly used for deposition of
diamond like carbon [23,2628], low-pressure plasma
cleaning [29], and other procedures.
The waveguide setup of atmospheric pressure micro-
wave plasma source is similar to that for the LPMPD. A
waveguide plasma applicator contains quartz tube at the
end of a rectangular waveguide. A movable short-circuit-
ing plunger allows the adjustment of a maximum electric
eld at the plasma-containing quartz tube. A trigger
(tungsten rod) is an integral part of this type of plasma
source in order to sustain a plasma discharge. The wave-
guide setup of atmospheric-pressure microwave plasma
source commonly operates with 2.45GHz magnetrons [30
33], but other frequencies, such as 896 MHz [34] and
915 MHz [35], are employed as well. The waveguide-based
plasma applicator is an inherent part of uidized-bed
microwave plasma reactors operating at atmospheric
pressure [36]. The ion density of the waveguide setup of
atmospheric-pressure microwave plasma varies within
10
11
10
12
cm
3
[30], B10
16
cm
3
[32,34].
From time to time other microwave plasma sources op-
erating at atmospheric pressure are mentioned in the lit-
erature. All of them are difcult to classify in scientific
terms because the basic information provided in the in-
dustry is very limited. For example, Ecotec (Taiwan) Co.
Ltd. enounced a new construction of a microwave plasma
surface operating at atmospheric pressure [37]. This mi-
crowave plasma burner allows one to obtain a plasma
temperature near the nozzle in the range of 30004500 K.
Plasma is produced just near the end of the central elec-
trode (tube), then drifts outward, and microwave energy is
released the gas apart from the nozzle.
2. BACKGROUND
2.1. Single Resonance Mode for Microwave Resonance
Plasma Source (MRPS)
To create a required E eld and ignite plasma, a high-ef-
ciency resonance cavity with high Q factor should be de-
signed. A high-efciency microwave resonance cavity is
dened as a resonant circuit element in a microwave net-
work, where the resonance frequency of the resonant cir-
cuit element is a function of its characteristic complex
impedance [38,39].
The Q factor of a high-efciency microwave resonance
cavity at the resonance frequency is described by
Q=o(energy
max stored
=power
average dissipated
)[
over acycle
When a high-efciency microwave resonance cavity runs
and a load in the form of a plasma slab is present in the
cavity, the Q factor is dened by the total energy absorbed
by both the resonance cavity and the plasma slab. The
plasma slab absorbs all microwave power, and the micro-
wave energy stored by the resonance cavity is negligible
small. However, when the plasma slab disappears, the
microwave energy stored by the resonance cavity is
maximised.
The microwave resonance cavity in question generally
employs a TM
01n
mode since this mode creates a displace-
ment current associated with longitudinal electric eld
E
z
along the axis of the cylindrical guide. In particular, the
TM
013
mode is widely used to increase the Q factor. A
schematic representation of this is shown in Fig. 1, where
E represents the lines of electric eld and H, the magnetic
lines.
MICROWAVE RESONANCE PLASMA SOURCE 2935
3. MRPS DESIGN AND OPERATION
3.1. Design
The geometry of a resonant cavity is the key element in
the design of the MRPS [40]. Cavity geometry will identify
the resonant mode of the microwave electric eld and,
therefore, ensure that only the required mode oscillates
the cavitys length, and the diameter should be carefully
calculated. As the radius of the resonant cylindrical cavity
is difcult to modify in order to optimize and tune the mi-
crowave resonance, the length (L) of the resonant cavity
can be tuned with a threaded plug (6) moving it in or out
during operation of the resonant cavity (numbers in pa-
rentheses refer to points in diagrams).
A block diagram of the instrumentation employed in
this research is shown in Fig. 2. The actual design of the
cylindrical resonance cavity is displayed in Fig. 3. The
resonant cavity is constructed of a highly conductive
material, such as brass and, to reduce ohmic losses (R
s
),
is covered with silver. Genuine setup of the MRPS, which
consists of the TM
013
cylindrical resonance cavity, wave-
guide, and a magnetron (1.3 KW, 2.45 GHz), is displayed in
Fig. 4. The six-port analyzer is inserted in the scheme for
measurement purposes.
In order to maintain two separate chambers of the res-
onant cavity, a dielectric partition has been built in and
hermetically sealed in order to provide vacuum evacuation
of chamber 1 (Ch1). The dielectric partition allows for
plasma to form only in the Ch1, limited by length l (Fig. 2).
The dielectric plate is made of hexagonal boron nitride
(hBN), which is transparent for microwave radiation. The
hBN plate acts as a pressure-dividing bafe between the
Ch1 and the chamber 2 (Ch2), which is always under
atmospheric pressure. It allows a lets plasma well to form
in Ch1. After evacuation of Ch1 up to 20 kPa, plasma
E field
E

f
i
e
l
d
z
q
2
l
Half-wavelength
change across
the diameter
No pattern change
around the
circumference H field
Figure 1. The eld conguration of the TM
013
mode in the cylindrical resonant cavity: longi-
tudinal section (above) and cross section (on
the left).
l
L
9
8
7
D
Chamber 2 Chamber 1
6
2 3
A
4
5
11
10
d
1
12
13
4
z
Figure 2. Sketch of the cylindrical resonant
cavity: 1magnetron, 1.3 kW, 2.45GHz;
2waveguide (WR-340); 3inlet nozzles;
4,5channels for precursors; 6,13adjusted
pistons; 7,9watching windows; 8hBN
partition; 10face lead; 11convergent
divergent (Laval-type) nozzle; 12rectangu-
lar aperture.
begins to form under microwave radiation only in the low-
pressure area (i.e., in Ch1). Once started, plasma is sus-
tained with microwave energy only in Ch1.
The outlet nozzle was manufactured from stainless
steel. It was found that the inlet geometry of the outlet
nozzle is crucial because the plasma can concentrate at
the sharp edges of the outlet nozzle. Stabilization of the
plasma in the center was achieved by rounding the inlet
prole of the outlet nozzle.
To introduce microwave energy, a magnetron launcher
on the base of a standard section of WR-340 waveguide
was attached to the cylindrical wall of Ch2. The aperture
(12)a symmetric iris, which can be an elliptical, circular,
or rectangular slotwas used for coupling between a cy-
lindrical cavity and a rectangular WR-340 waveguide op-
erating in the dominant TE
10
mode. The aperture ensures
the maximum amount of microwave energy absorbed
within the resonant cavity, instead of being reected
back toward the magnetron. The optimal shape of the ap-
erture is determined experimentally.
As in TE
10
mode, a magnetic eld component, which is
parallel to the long dimensions of the slot, will be coupled
through aperture. For the slot orientation perpendicular
to the longitudinal z axis of cylinder, the TE
10
magnetic
eld couples into the cavity, creating a magnetic eld pat-
tern within the cavity, which allows any of the TM
01n
res-
onance modes to excite. When the waveguide and aperture
is centrally located along the z dimension, the n mode
must be even, but for odd values (n=3 for TM
013
) the
magnetic eld is zero at that point in the cavity. Thus,
the axis of the aperture should be approximately l/4 from
the ends of the cavity [38].
In order to maintain a stable plasma stream at the La-
val nozzle, the gas ow should rotate within Ch1 (Fig. 5).
Chamber 1 has three inlet nozzles (IN). The IN (3) are
symmetrically placed close to the dividing BN plate. The
diameter of the calibrated orices must be optimized with
a owrate of gases, with power of microwave supply, plas-
ma parameters (ion concentration, ionization ratio, etc.)
for different atomic (Ar, He, etc.) or molecular (O
2
, N
2
,
CH
4
, etc.) gases, and with the throat of the Laval nozzle.
The IN should be directed tangentially to the wall and at a
small angle of 10201 to the hBN plate. The plasma torch
is formed from the Laval nozzle (5), which can have chan-
nels (5) for supplying powders/precursors into the plasma
torch. Chamber 1 can have channels (4) for introduction of
precursors directly into plasma slab.
The inlet and outlet can be optimized. The gases used
are considered to be ideal, and the gas ow through the
nozzles is isotropic. For a standard convergingdiverging
nozzle, the plasma ow is optimized when the plasma ve-
locity at the throat is at the speed of sound. This condition
is known as shocked ow. For shocked plasma ow, the
velocity of the plasma after the throat in the diverging
section of the nozzle will continue to increase until the
pressure is equal to the ambient outlet pressure. For op-
timized or shocked ow, the point at which the diverging
section of the nozzle ceases to accelerate the ow is de-
pendent on the half-angle of the outlet nozzle and the
length of the nozzle. The actual values of the length of the
exhaust nozzle can be found only by experiment. The ow
may also uctuate due in proportion to the temperature
uctuations within the resonant cavity and the ambient
outlet conditions.
Figure 4. Experimental setup of the micro-
wave resonance plasma source.
Figure 3. General view of the microwave resonant cavity.
The MRPS makes available a required ionizing electric
eld density for plasma discharge and uses microwave
energy to produce expanding plasma, which is directed
through the convergentdivergent nozzle. Plasma acts as
a resistive load that absorbs the incident microwave en-
ergy and dissipates it as thermal energy, which is trans-
ferred to a owing gas. It is common to refer to this type of
plasma as a free-oating plasma since it is located at
regions of maximum electric eld density within the inte-
rior of the cavity and does not have to be attached to an
interior solid surface of the resonance cavity.
The MRPS design with the rectangular waveguide for
transmission of microwave energy in Ch1 has a number of
major advantages:
*
It is easy to adjust during MRPS operation.
*
Less manufacturing is required.
*
It has higher power-handling capability.
*
It is more efcient (lower loss per unit length of
microwave transmitter).
3.2. Operation
The microwave plasma torch is a amelike discharge and
has a central plume (Fig. 6a). The plasma torch shown in
Fig. 6b is formed from the Laval nozzle (Fig. 6a). The mi-
crowave energy is coupled into the resonant cavity via a
rectangular waveguide and a carefully designed aperture
for the best coupling. Tuning can be accomplished by ad-
justing the position of the ne-tuned piston (6) (Fig. 2).
With the piston set near its optimal position, the plasma
can be ignited easily with a short burst inside the cylin-
drical resonance cavity (Fig. 6c).
In maximum region of electric eld density, ionization
of a carrier gas occurs, producing a microwave resonance-
induced plasma discharge. Ionization of a carrier gas by
electric eld is dependent on the ratio of electric eld
strength to pressure (i.e., the E/p ratio). The MRPS dis-
charge will be maintained when the production of newly
ionized species scarcely is equal to the rate of losses,
namely, recombination and formation of torch.
For ignition of plasma, the exhaust nozzle is capped
(Fig. 3) and the air is evacuated from the chamber 1, and
then microwave power is introduced to the resonant
cavity. After the formation of plasma in Ch1, a fore-vacu-
um pump is disconnected and the carrier gas begins to
ow into Ch1. On the commencement of ow into Ch1, the
plasma coalesces and moves toward the outlet nozzle.
Once atmospheric pressure is achieved in Ch1, the nozzle
cap is removed. From this moment, the microwave
B
(a) (b)
A A
3
:
3
AA
3
B
33
BB
Figure 5. A cross-sectional view (AA) of the
plasma outlet chamber in place of connection
of three calibrated orices (see AA cross sec-
tion positioned in Fig. 2).
(a) (b) (c)
Figure 6. (a) Top view of the plasma torch at
atmospheric pressure; (b) side view of the plas-
ma torch at atmospheric pressure; (c) a plasma
slab of MRPS operating at atmospheric pres-
sure in plasma-forming chamber.
resonance plasma torch begins to run. Adjustment of the
length (L) of the resonant cavity would also occur to max-
imize the microwave power absorption.
A typical position of the MRIPS is stabilization of the
plasma slab at the center of the Ch1 touching the Laval
nozzle. The plasma would concentrate at the edge of the
outlet nozzle, and swirl in the same direction as the inlet
swirl.
4. COMPARISON OF MICROWAVE PLASMA SOURCES
4.1. Equivalent Circuits of Microwave Plasma Sources
4.1.1. Waveguide Setup Plasma Source. An equivalent
scheme of waveguide-type plasma reactor is shown in
Fig. 7. In this equivalent circuit the distributed-circuit
parameters are given in the form of equivalent lumped
impedances.
Typically, the impedance of the plasma column can be
represented by two parallel inductive and active resis-
tances as a central part of a T-shape equivalent scheme in
which the symmetric lumped complex impedances Z
1
and
Z
2
depend on the geometric dimensions of the waveguide
and discharge tubes. It is important that the impedance of
plasma torch depend on the temperature, pressure, gas
ow, and absorbed microwave power. A lossless shorting
plunge is represented by shortcut transmitted line with
impedance:
Z
s
=jZ
0
tan(2pd=l
g
)
where
Z
S
=impedance of the shortcut line
Z
0
=characteristic of the shortcut line
d =length o the line
l
g
=wavelength
To get the maximum value of electric eld in the required
plasma location (r20,000V/m in the standard WR-340
waveguide at 2.45GHz), l is usually equal to (2n1)l
g
/4,
where n is an integer, so Z
s
-N, which is approximately
analogous to the open-circuit line. Analysis of this equiv-
alent circuit shows that before plasma have been gener-
ated, the input impedance of this system Z
in
approaches to
innity, as it equals to the series connection of lumped
complex impedances Z
2
and Z
s
.
The starting plasma is rather difcult because there are
no free electrons that act in response to the E eld. Inci-
dent microwave power is not absorbed by cold gases and is
almost completely reected. The corresponding reection
coefcient rises to |G|=0.95. Consequently, at atmospher-
ic pressure the microwave discharge in this type of plasma
source can be generated and sustained only by a trigger
(such as tungsten rod), which is an inherent part of this
type of plasma source. Moving off the tungsten trigger
leads to sparking of the plasma discharge; therefore, this
plasma discharge is classied as a trigger-sustaining
plasma. After plasma was generated, the incident power
begins to be partially absorbed by the plasma column;
however for this type of plasma source, the reection
coefcient still remains high up to |G|=0.5, so almost
25%of the power is reected [5]. The Qfactor of this type of
plasma source is evaluated at B100500 for 2.45GHz.
4.1.2. Waveguide-to-Coaxial Setup Plasma Source. Gen-
erally, this form of plasma generator is used as a coaxial
line which has a characteristic impedance lower than that
of waveguides. As a result, it is more easy to transform the
coaxial plasma apparatus impedance and to match it with
the characteristic impedance of a waveguide without
much reection. In this type of plasma source, the wave-
guide-to-coaxial transformer is loaded by the plasma
chamber at the end of the coaxial line. The coaxial line
is connected to the broadwall of the waveguide with its
outer conductor terminating on the wall. Another end of
coaxial line is connected with various kinds of the plasma
discharge chamber. One end of waveguide is connected to
power source and another to the lossless shorting plunge.
One such conguration of the waveguide-to-coaxial type of
plasma source together with an equivalent lumped circuit
and an equivalent circuit with distributed circuit elements
is shown in Fig. 8.
Equivalent circuits for coaxial line loaded with plasma
chamber to waveguide junction connected to microwave
generator is composed of complex impedances Z
1
, Z
2
and
Z
3
and, which are characterize electromagnetic eld be-
haviour and power losses due to evanescent higher-order
modes that are exited at each of discontinuity, and power
losses due to active losses in waveguide-to-coaxial junc-
tion. The equivalent scheme also comprises the lumped
impedance of coaxial plasma structure Z
p
, and Z
s
the
impedance of the waveguide transmitted line with short-
ing plunge [41].
a
(a) (b)
P
l
a
s
m
a

c
o
l
u
m
n
2l
z
g
tan =
Zo
Z
1
Magnetron
Z
2
Z
p
Z
s
j Z
0
Figure 7. Equivalent scheme of waveguide array plasma source: (a) sketch of plasma slab;
(b) equivalent circuit.
Figure 8b shows that lumped plasma formation imped-
ance is included in series in the transmitted waveguide;
thus there is a very narrow band of impedance matching
in order to ignite the plasma. Furthermore, any random
destabilizing processes exert inuence on absorbtion of
microwave energy by plasma and cause reection of
microwave power. Thus, this kind of microwave plasma
design also requires constant sparking ignition.
4.1.3. Microwave Resonance Plasma Source. There is a
different scenario when plasma starts inside the high-Q-
factor resonance cavity [42]. The equivalent scheme com-
prises the lumped complex impedances Z
1
, Z
2
related to
impedances of iris and impedances of high-Q-factor reso-
nance cavity X
r
, L
r
, R
r
and the lumped complex impedance
of plasma column Z
p
, which depend on various factors
such as temperature of plasma, gas ow, pressure, and
absorbed power, (Fig. 9).
This MRPS uses microwave energy to generate self-
sustaining plasma within the resonance cavity that is di-
rected as a plasma torch through a convergentdivergent
nozzle. A cold gas passes through the cavity, is ionized and
heated by microwave power, and passes out of the cavity
through a nozzle to produce the plasma torch.
4.1.4. Equivalent-Circuit Parameters. There are two
major theoretical techniques for obtaining the equivalent
circuits constants: the energy methods and the admit-
tance methods. The rst method analyses energy stored
by the circuits elements followed by a description of equiv-
alent intrinsic inductance, capacitance, and resistance.
Admittance/impedance method considers voltage and
current across the microwave circuit elements and conse-
quently computes the characteristics of the equivalent-
circuit elements. The equivalent circuits of stable, passive,
and geometrically simple elements can be obtained rela-
tively easily by integrating the electromagnetic eld in-
side the microwave system, using the energy methods or
admittance method [43]. But obtaining the equivalent mi-
crowave plasma circuit parameters with non-regular-
shaped microwave elements and with randomly changing
reected and absorbed by plasma microwave power is
usually complicated.
Some researchers applied method of geometric simi-
larity, when the equivalent complex impedance of part of
the plasma column within the waveguide can be found by
analogy with the ideally conducting metal rod. In the
same approach the equivalent parameters of holes in the
wider walls of waveguide, capacitance between plasma
column and waveguide walls, and impedance of other el-
ements of scheme were calculated [5]. However, in real
discharge systems, width and temperature of the plasma
column differ from point to point and can be represented
as a lossy dielectric rather than ideal conducting metal
rod; therefore, it seems inappropriate to use the method of
geometric similarity in the GHz frequency range be-
cause the wavelength becomes comparable to the wave-
guide system dimensions, and what geometrically appears
to be inductance can possess strong resonance properties
in real system.
The most accurate methods used to obtain character-
istics of the equivalent scheme is waveguide/cavity per-
turbation technique, which is one of the most important
and widely used techniques to study the microwave im-
pedance of materials and devices and offers both qualita-
tive and fairly accurate quantitative information for
engineering purposes [44]. However, even this highly
Magnetron
Incident wave
Precursor inlet
Plasma slab
TM
013
(a) (b)
Zo
Z
1
Z
2
Z
p
Figure 9. Equivalent scheme of resonance plasma source.
Zo
Magnetron Z
1
Z
2
Z
3
Z
p

2l
z
g
tan = Z
s
j Z
0
Plasma
chamber
(a) (b)
P
l
a
s
m
a

c
o
l
u
m
n
Figure 8. Equivalent circuit with distributed circuit elements.
sensitive and relatively simple technique cannot be ap-
plied to nonstable, irregular, and uctuating plasma when
most of or all parameters of the equivalent scheme of
plasma sources are interdependent and are similar with
respect to absorbed power, temperature, type of gas, ow-
rate, and so on.
4.2. Incident/Reected Microwave Power
Because of high rate of electron collisions with neutrals
and ions, the plasma in the MRPS is resistive and inher-
ently absorptive. Since the plasma occupies the region
where most of the electric eld energy is stored in the
cavity, the resistive plasma discharge exerts a strong dam-
ping on the cavity by absorbing approximately 98% of the
incident microwave power (Fig. 10) and transforms the
microwave energy into thermal plasma, which is trans-
ferred to the owing gas. Thus, the plasma transforms the
cavity from a high-Q to a low-Q cavity, and any disturbing
factors that render the plasma unstable will again in-
crease the Q factor of the resonance cavity. Therefore, in-
creasing the electric eld sufciently reignites the plasma.
As evaluated [5,45], the coefcient of microwave power
absorption for the nonresonance plasma sources very sel-
dom surpasses 50%.
4.3. Smith Chart
The Smith Chart (Fig. 11) illustrates the differences of
microwave plasma discharges generated in the resonance
microwave plasma source (Fig. 9) and in the nonresonance
plasma souse (Figs. 7 and 8). In resonance design, the
impedance point shifts to the center of the Smith chart
(Fig. 11a) after the plasma is ignited and the operation
spot is located predominantly, around 1. In the wave-
guide, coaxial-to-waveguide, or coaxial design, the operat-
ing conditions are randomly distributed throughout Smith
chart (Fig. 11b).
4.4. Stability of Microwave Plasma Sources
The most significant feature of these microwave plasma
sources is their stability [42]. Therefore it is very impor-
tant to analyze an inuence of disturbing factors, such as
gas ow instability, pressure, and absorbed power, doping
mass and ow variations, on the main types of microwave
plasma sources.
MRPS with air
0
200
400
600
800
1000
1200
1400
1
2
7
5
3
7
9
1
0
5
1
3
1
1
5
7
1
8
3
2
0
9
2
3
5
2
6
1
2
8
7
3
1
3
3
3
9
3
6
5
3
9
1
Time (1 count = 50 msec)
M
W

p
o
w
e
r

(
W
a
t
t
s
)
Incid (W)
Refl (W)
K-absorption: MRPS with air
(a)
(b)
0
0.2
0.4
0.6
0.8
1
1
2
9
5
7
8
5
1
1
3
1
4
1
1
6
9
1
9
7
2
2
5
2
5
3
2
8
1
3
0
9
3
3
7
3
6
5
3
9
3
Time (1 count = 50 msec)
K
-
a
b
s
o
r
p
t
i
o
n
K-absorbtion
Figure 10. Incident/absorbed microwave
power (a) and coefcient of absorption (b) as
a function of time for the MRPS.
The problem with plasma discharge stability can be
solved by analysis of currentvoltage characteristics [5].
As a result, the stability criterion was formulated as a
rising currentvoltage characteristic of the load or by con-
sidering the gradient of functional dependence of the re-
quired power and the plasma column admittance. But
analysis of plasma source performance by considering the
plasma sources equivalent scheme is unreliable as there
are uncertainties in parameter calculations and in deni-
tion of voltage and current in microwave schemes.
To analyze the instability effect of any of the perturba-
tion factors (temperature of plasma, gas owrate, pres-
sure, etc.) on plasma sources performance the Q-factor
method was proposed and applied [42]. If we describe per-
turbation injurious effect of any of the inuencing quan-
tities as P, so any change in perturbation factors will
represent the absolute value |DP|. Now the criterion for
stability (S) of plasma design can be expressed as
S=dQ=(d[DP[) > 0
Thus, when energy stored in the system is increased with
a change of any of the inuencing quantities, the micro-
wave plasma source will be stable and self-sustained.
Therefore, if the stability factor S40, any injurious ef-
fects of the inuencing conditions, which cause the plasma
to become unstable, will again increases the Q factor of the
resonance cavity, thus increasing the electric eld suf-
ciently to reignite the plasma.
For illustration, let us consider the conventional low Q-
factor waveguide setup plasma source (Fig. 7). It is useful
to represent a plasma slab as a quasidielectric rod inte-
grated in the microwave system with absorbed microwave
power depending on type, pressure, owrate of gases, and
other parameters.
For this type of design, there is a very narrow interval
of parameters in which plasma can exist and be stable.
Any interchange in condition will result in changes in im-
pedance matches and power balance. For example, when
the plasma temperature increases as a result of any desta-
bilizing factors, the concentration of charged particles in
plasma increases so that the active component of the plas-
ma column impedance changes. This in turn increases the
reection from the plasma slab. As a result, the Q factor of
the system and, therefore, the absorbed power decrease,
stability factor So0, that renders plasma generation un-
stable. When the plasma temperature decreases under the
inuence of uctuating factors, the local electric eld as
well as absorbed microwave power decreases and the plas-
ma extinguishes, and the stability factor So0. This kind
of plasma source requires high stability of the external
parameters or a constant ignition such as the tungsten
trigger.
The microwave plasma stability in the MRPS and in
the waveguide setup plasma source can be compared with
equilibrium between a ball in a valley and on a hill. Like
the MRPS, any destabilizing factors (DD) return the ball
to equilibrium position (Fig. 12a). In case of the MWTPS,
the destabilizing factors (DD) move the ball away from the
equilibrium state (Fig. 12b).
Stability of MRPS, intensity of microwave power
absorption, and the size of a plasma slab have been
R=1
1
.2 .5 1 2 5
1
(a)
2
5
5
.5
2 .5
.2
.2
R=1
1
.2 .5 1 2 5
1
(b)
2
5
5
.5
2 .5
.2
.2
Figure 11. Operational points of resonance
(a) and nonresonance (b) plasma sources on
Smith chart.
D D
D D
High Q
Low Q
(a) (b)
Figure 12. Mechanical analogy of plasma sta-
bility in the MRPS (a) and the waveguide array
plasma source (b).
investigated, depending on type and owrate of the gases
(air, CO
2
, N
2
, He, etc.) [46]. It was revealed that the vol-
ume of microwave plasma slab for given applied micro-
wave power is inversely proportional to ionization
potential of applied gases and their owrate. Because of
peculiarities in MRPS design, a precursor mixture totally
passes through the plasma stage, significantly enhancing
the efciency of the MRPS in comparison with microwave
waveguide-type plasma sources.
5. SOME PLASMA CHARACTERISTICS OF MRPS
A chemical kinetic modeling of the MRPS action on CO
2
reduction was realized [47] on the basis of experimental
data published earlier [46]. The software CHEMKIN [48]
was employed for modeling the chemical kinetics. The
plasma model was built up. It was shown [47] that the
predicted electron (T
e
) and ion (T
i
) temperatures are very
close at B7000K for this case [46]: T
e
E7500K and
T
i
E6500K when the gas mixture ow is around 2 L/min.
Therefore, this plasma can be classied as thermal plasma
with a temperature of approximately 7000K. These cal-
culations predicted that the nonthermal effect (i.e.,
T
e
E7500 K and T
i
E2500K) becomes remarkable when
the gas owrate is 410 L/min. At a owrate of 100 L/min,
the predicted temperatures electrons and ions were T
e
E6500 K and T
i
E500K respectively, which corresponds to
the nonequilibrium plasma state.
6. ADVANTAGES OF MRPS
The MRPS has a number of advantages over a conven-
tional thermal plasma sources (arc plasma, RF plasma,
etc.) in that no electrodes are necessary. No electrodes are
needed for the generation of the microwave resonance
plasma that eliminates a potential source of the plasma
contamination. The absence of a cathodic sheath with high
potentials leads to a very-low-energy ion bombardment of
a given surface, minimizing surface damage. The MRPS
provides a downstream plasma torch unidirectional-act-
ing on surfaces that allows surface engineering of 3D-
shaped workpieces, which have to be rotated.
Incorporation of waveguide for introduction of the mi-
crowave energy into the resonant cavity has a number of
advantages over the design using the coupling probe con-
nected to a coaxial cable, namely:
*
Noncontaminating plasma source
*
Easy to adjust during MRPS operation;
*
Simpler and cheaper design (coaxial is more expen-
sive than waveguide)
*
Less manufacturing required
*
Higher power-handling capability;
*
Higher efciency (lower loss per unit length of micro-
wave transmitter).
The novel MRPS described is a synergy of microwave
technique and a technique intended for materials process-
ing, plasma chemistry, plasma surface treatment, plasma
sterilization, and other applications. Microwave technolo-
gy, together with materials science, promises research op-
portunities with great potential in areas such as
nanotechnology, composite materials, and biomaterials.
Potential benets include enhanced performance, extend-
ed life, cheap lifecycle costs, reduced environmental im-
pact, improved cost-effectiveness and value-added use of
materials through advanced manufacturing.
7. POTENTIAL APPLICATIONS OF MRPS
7.1. Surface Treatment
The microwave resonance plasma technique proposed in
this research of both deposition of coatings and surface
treatment at atmospheric pressure has great potential in
surface engineering. It is common to use a conventional
plasma spray technology for coating at atmospheric pres-
sure [7]. A plasma surface treatment (nitriding, oxinitrid-
ing, nitrocarburizing, etc.) is applied mostly by vacuum-
based methods such as direct-current glow discharge
(DCGD), and plasma immersion ion implantation (PI
3
).
In order to demonstrate the possibility of the novel
MRPS for surface treatment by nitriding and oxinitriding,
typical materials such as titanium with a microhardness
of 300350 VH
50
and its alloys, stainless steels 304 and
316 and mild steel 5140, were chosen for nitriding by
MRPS at atmospheric pressure [40,49]. Operational
MRPS parameters for nitriding/oxinitriding were incident
power 8001000 W and gas ow 15 L/min. The gas ow-
rates of nitrogen and air were controlled by owmeters
accurate to approximately 5% of the maximum owrate.
The MRPS creates a stable plasma torch of 36 cm long.
The steel 5140 of chemical composition [(0.380.45)C,
(0.150.35)Si, (0.70.9)Cr] and stainless steels, 305 and
316 were nitrided by the MRPS. X-ray and metallographic
investigations of the nitrided layer on the steel surface
showed the formation of the amorphous layer (Fig. 13a).
Nitriding of steel 5140 in pure nitrogen plasma starts at
B501C of the substrate. Temperature increase of the sub-
strate up to 1001C leads to transformation of nitriding to
oxinitriding. This transformation is accompanied by
changing of color from light orange to dark pink. Above
1001C of steel 5140 substrate, the surface nitriding trans-
formed to oxidization and the color of the oxidized sur-
face became violet. Stainless-steel 304/316 substrates
(a) (b)
<100 m> <--100 m-->
Figure 13. Scanning electron microscopic cross sections of 5140-
steel (a) and Ti (b) after microwave plasma nitriding at atmo-
spheric pressure.
commenced nitriding at B601C. Above 1501C, the stain-
less steels started to oxidise, and the surface acquired a
violet color.
The MRPS nitiriding demonstrated that the Ti surface
exhibits a somewhat distinct golden color that is char-
acteristic of titanium nitride. Diffusion of implanted ni-
trogen seldom occurs in the operative temperature range
of the DCGD and the PI
3
, because of the afnity of Ti for
nitrogen. In case of the MRPS nitriding, the proper layer
of TiN
x
with a hardness of B1500 VH
200
is formed on the
Ti surface (Fig. 13b). It was observed that a thin surface
layer (510 mm) of X-ray amorphous TiN was formed im-
mediately. The introduction of (21 vol% of O
2
and 79 vol%
of N
2
) into the MRPS had no significant effect on the per-
formance of the nitrided layers for the given air plasma.
Titanium begin to nitride at B1001C of substrate. This
Ti nitriding transformed to oxidization and the treated Ti
surface became violet above B2001C. Addition of oxygen
(up to 21 vol% of O
2
) to the pure nitrogen plasma shifted
the starting temperature of nitriding (B601C) for steel
5140 and at B1001C for stainless steels 304/316. In the
case of Ti, nitirding started at 41201C. Transitional tem-
peratures, from which oxinitriding passes to oxidizing, be-
came lower, such as B751C for steel 5140, B1151C for
stainless steel (SS) 304/316, and B1751C for Ti.
Reectance Fourier transform infrared (FTIR) spec-
troscopy was employed to study the changes in chemical
composition of the nitrided layers. In particular, the re-
ectance FTIR spectra of steel 5140 nitrided by the MRPS
are displayed in Fig. 14. The maximum absorption (B0.7)
of the nitrided layer is reached at B14001600cm
1
. The
band, where absorption is more than 0.5, is between 1700
and 1000cm
1
.
FTIR spectroscopy of conventionally plasma (ion)
nitride steel 5140 by direct-current glow discharge
(DCGD) detected the strong absorption (40.5) within
the range of 8002000cm
1
. Absorption (B0.85) peaked
at B1150cm
1
. Comparing these spectra, one can note
the significant differences between plasmamaterial in-
teraction for MRPS and that for the conventional DCGD,
creating an equilibrium nitrided layer consisting of
microcrystalline nitrides on surface.
Titanium surface nitrided by MRPS radically differs
from that for steel 5140. An intense absorption peak was
detected at approximately 850cm
1
. The MRPS nitriding
forms a sharp band of nitrogentitanium bonding.
7.2. Amorphization and Nanostructuring of Surface
The MRPS operating at atmospheric pressure [50] allows
the modication of a microcrystalline surface structure,
which becomes amorphous to X-rays [51]. In particular,
the X-ray amorphous nitride layers were formed on the
surfaces of Ti, SS 304, SS 316, and plain steel 5140 [50
52], using a nitrogen-based mixture for the MRPS. Gen-
erally, it is possible to alloy the amorphous layers by bo-
ron, carbon, chromium, and other elements by employing
appropriate precursors in both gaseous and solid state.
This ability of novel MRPS to amorphize a surface mi-
crocrystalline layer up to B10mm thick, combined with
alloying, was used in this study to transform an original
microcrystalline surface layer into a nanocrystalline state
via an amorphous state. The novelty of this research is to
amorphize a microcrystalline surface by MRPS. The sub-
sequent heat treatment in order to transform the meta-
stable amorphous phase to a nanocrystalline state is
described in Ref. 53.
7.3. Nanostructured Biomaterials
A feasibility study of plasma nitriding/oxinitriding has
demonstrated that a novel low-cost plasma torch devel-
oped on the basis of the microwave resonance phenomenon
can be practical for nanotechnology [45,54], in parti-
cular for bioengineering coatings such a nanostructured
1.0
0.9
0.8
0.7
0.6
0.5
A
b
s
o
r
b
a
n
c
e
0.4
0.3
0.2
0.1
0.0
4000 3000 2000
Wavenumbers (cm-1)
Novel MRPS-nitriding
Conventional plasma (ion) nitriding
1000
Figure 14. FTIR spectra of 5140 steel treated
by the novel MRPS nitriding and the con-
ventional DCGD nitriding. (This gure is
available in full color at http://www.mrw.
interscience.wiley.com/erfme.)
zirconium/titanium oxide, carbyne-containing nanocom-
posites [5558], diamondlike carbon, and carbon nitride.
Novel materials are required for cardiovascular thera-
pies, such as replacement of blood vessels and heart valves
and the provision of devices that intimately contact com-
ponents of the cardiovascular system, for example, dialy-
sis and oxygenation membranes, interventional catheters,
and cardiovascular stents. Improved biocompatibility of
these implants could result in patients requiring lower
doses of medication to prevent blood clotting.
MRPS allows the deposition of nanostructured bioma-
terials and coatings on metal/nonmetal surfaces using a
variety of carrier gases and the precursor materials, which
can be powders, liquids, or gases. This technology is also
capable of surface engineering of complex 3D shapes that
might be required for articial implants. Advantages of
the proposed method overall include an enhancement
of biocompatibility and extended service life of articial
implants.
One current trend in biomedical applications is to use
novel carbon-based biomaterials. Carbynes, which possess
a higher degree of biocompatibility and low litho- and
thrombogenic activity [59], have shown promise in modi-
fying coating on a surface of biocompatible material for
use in reconstructive surgery. For example [55,57,58],
nanocomposite coatings (nc-W
3
C/nc-carbynes) containing
carbynes as a binder demonstrated a high hardness
(3540GPa), chemical inertness, a low friction coefcient
(0.10.12), and excellent biocompatibility.
7.4. Plasma Chemistry
Environmental pollution by exhaust/ue gases is a topical
subject of the greenhouse effect. For example, a petrol en-
gine, which consumes of B14 L of liquid petrol per 100km,
throws B38 L/s of exhaust gases. Currently, an incredible
amount of effort has been undertaken to reduce such emis-
sions and has been responsible for the longstanding inter-
est in the theoretical and experimental research of plasma
exhaust/ue gas treatment. Currently it is suggested that
microwave plasma treatment of exhaust/ue gases is able
to solve the problem of exhaust/ue gas environmental
pollution. Exhaust gases represent a multicomponent gas-
eous system. After, the combustion of petrol, which con-
sists of B96% of octane [CH
3
.
(CH
2
)
6
.
CH
3
], exhaust gases
correspond predominantly to a mixture of vapor (H
2
O),
carbon dioxide (CO
2
), and nitrogen (N
2
):
C
8
H
18
62:5(0:2O
2
0:8N
2
)
=8CO
2
9H
2
O50N
2
Nitrogen is the ballast gas component because air, which
is used for combustion, is a mixture of oxygen (B20 vol%
O
2
vol%) and nitrogen (B80 vol% N
2
). Nitrogen amounts
to 75% of total volumetric portion of exhaust gases: CO
2
H
2
ON
2
E116. These three gaseous components dene
microwave plasma chemistry treatment of exhaust gases.
Small additions of oxygen (25 vol%) and carbon monoxide
(B0.15 vol%) play a minor role.
The microwave plasma chemistry of the CO
2
H
2
ON
2
gas system is important for technical application. In par-
ticular [46], an attempt was made to investigate the plas-
ma chemistry of the gas mixture of CO
2
N
2
and CO
2
H
2
O
as potential precursors for a coating deposition of super-
hard carbon nitride, diamondlike carbon, carbyne-based
nanocomposites, and other compounds.
The results of the microwave plasma chemistry exper-
iment are shown in Figs. 1517. As can be seen in Figs. 15
and 16, carbon dioxide decomposes on carbon monoxide
and oxygen. According to the formula CO
2
=CO0.5O
2
(DH=29.78 kJ/mol), two decayed molecules of CO
2
create
two molecules of CO and one molecule of O
2
. Comparison
of the experimental volumetric portions of the gases after
microwave plasma treatment (Fig. 15) reveals good agree-
ment between these experiments with the equation given
above. Visible traces of solid carbon isolated in any form
have not been detected.
This scenario changed radically when nitrogen was
added on a ratio of CO
2
N
2
=6. Yield of CO increased sig-
nificantly. Every alternate CO
2
molecule was dissociated.
Visible traces of solid carbon were not detected. The
most unexpected result was that the addition of nitrogen
V
o
l

%
12
10
8
6
4
2
0
0.1 0.2 0.3
CO
2
flow (l/min)
Microwave plasma chemistry: CO
2
CO
2
(BP)
CO
2
(AP)
CO

(AP)
O
2
(AP)
O
2
(BP)
0.4 0.5
Figure 15. Volumetric portion of gases before (BP) and after (AP)
plasma treatment of carbon dioxide as a function of CO
2
owrate.
V
o
l

%
18
16
14
12
10
8
6
4
2
0
0.35 0.85
N
2
+CO
2
flow (l/min)
Microwave plasma chemistry: N
2
+CO
2
(N
2
/CO
2
= 6)
CO
2
(BP)
CO
2
(AP)
CO

(AP)
O
2
(BP)
O
2
(AP)
1.35
Figure 16. Volumetric portion of gases before (BP) and after (AP)
plasma treatment of (N
2
CO
2
) mixture as a function of (N
2
CO
2
) owrate where N
2
/CO
2
E6.
initiated an extra yield of carbon monoxide. As stated ear-
lier [5], a small addition of nitrogen does not affect the
yield of carbon monoxide. In this case, the CO yield in-
creases in by a factor of 1.5.
Nitrogen oxides are inherent byproducts of microwave
resonance plasma treatment when a gaseous mixture con-
tains nitrogen and oxygen in any combination. Nitrogen
oxides are usually formed in plasma at temperatures ex-
ceeding 20001C. In the course of the experiments (Fig. 17),
the concentration of nitrogen oxides increased in propor-
tion to the volumetric portion of nitrogen and carbon
dioxide beyond the upper limit of 4000ppm (particles
per million) when NO
x
was measured a using gas ana-
lyzer [46].
The experiments showed a very rapid decomposition of
carbon dioxide into the main stable species CO and O
2
.
Comparison of the decomposition of carbon dioxide with-
out and with nitrogen addition revealed that the ratio be-
tween carbon monoxide and oxygen is altered by a factor
of 1.5 at steady state.
The characteristics of atmospheric pressure microwave
resonance plasma chemistry of carbon dioxide are
*
No visible dissociation of carbon dioxide up to solid
carbon in any form when using the gas mixture of
CO
2
He
*
Effect of increasing the yield of CO by a factor of 1.5
when using nitrogen addition to gas mixture of CO
2
He
8. EVALUATION
The design of the novel MRPS source based on the TM
013
mode has a number of advantages over the conventional
plasma spray sources, including
*
High (B98%) efciency of microwave energy con-
sumption
*
High efciency of precursor consumption
*
Simple/and reliable design
*
Noncontaminating thermal plasma source
*
Ease of operation
*
Variety of carrier gases for thermal plasma
*
Selective surface treatment/coating
*
Operating at atmospheric pressure
*
Stability of the plasma torch
*
Reliability of the MRIP source
*
No inherent problem of erosion (no electrodes)
*
Low capital and running costs
*
Retrotable design
*
Compactness
*
Low maintenance requirements
*
Simple surface preparation for modication.
It is believed that the MRPS could nd interesting ap-
plications in plasma thermal spray technology for deposi-
tion of coatings and in plasma chemistry for materials
processing. The novel MRPS is innovative for deposition
and coating of bionanomaterials.
The research and development have demonstrated
the commercial potential of the MRPS [40,60]. The
MRPS can be a commercial processing tool for the solid-
state amorphization of a surface layer, which is a transi-
tional stage for nanocrystallisation under an appropriate
thermal annealing.
Acknowledgments
The authors would like to thank Prof. Kai Chang (Texas
A&M University) for this opportunity to contribute to the
Wiley Encyclopedia of RF and Microwave Engineering by
preparing this article, and Kris Parrish and Cassandra
Craig (John Wiley & Sons, Inc., Hoboken, NJ) for their
assistance in preparation of this article, Mrs. Fiona
ODonnell (Hawthorn library of Swinburne University of
Technology, Melbourne, Australia) for inestimable infor-
mation support, as well as Mrs. Andrea Meyer (IRIS,
Swinburne University of Technology, Melbourne, Austra-
lia) for their assistance in polishing the manuscript.
BIBLIOGRAPHY
1. M. Rahmane, G. Soucy, M. I. Boulos, and R. Henne, Fluid dy-
namic study of direct current plasma jets for plasma spraying
applications, J. Thermal Spray Technol. 7:349356 (1998).
2. S. V. Dresvin, A. V. Donskoi, V. M. Goldfarb, and V. S. Klubni-
kin, Physics and Technology of Low-Temperature Plasmas,
Iowa State Univ. Press, Ames, 1997.
3. A. Garscadden, PECVD RF discharge models review, in Proc.
Materials Research Soc. Symp. on Characterization of
Plasma-Enhanced CVD Processes, Nov. 2728, 1989, Boston,
MA, Materials Research Society, Pittsburgh, 1990, Vol. 165,
pp. 315.
4. J. L. Cecchi, Introduction to plasma concepts and discharge
conguration, in S. M. Rossnagel, J. J. Cuomo, and W. D.
Westwood, eds., Handbook of Plasma Processing Technology.
Fundamentals, Etching, Deposition and Surface Interactions,
Noyes Publications, Park Ridge, IL, 1990, pp. 1469.
N
O
x

(
p
p
m
)
5000
4500
4000
3500
3000
2500
2000
1500
1000
500
0
1.5 1 0.5
N
2
+CO
2
flow (l/min)
Microwave plasma chemistry:
N
2
+CO
2
(N
2
/CO
2
= 6)
0
Figure 17. Concentration of nitrogen oxides after plasma treat-
ment of N
2
CO
2
mixture as a function of (N
2
CO
2
) owrate
where N
2
/CO
2
E6.
5. V. M. Batenin, I. I. Klimovskii, G. V. Lysov, and V. N. Troitskii,
Superhigh Frequency Generators of Plasma, CRC Press, Boca
Raton, FL, 1994, p. 175.
6. M. I. Boulos, RF induction plasma spraying: State-of-the-art
review, J. Thermal Spray Technol. 1:3340 (1992).
7. C. C. Berndt, W. Brindley, A. N. Goland et al., Current prob-
lems in plasma spray processing, J. Thermal Spray Technol.
1:341356 (1992).
8. C. M. Ferreira, M. Moisan, and Z. Zakrzewski, in M. Moisan
and J. Pelletier, eds., Microwave Excited Plasmas, Elsevier,
Amsterdam, 1992, pp. 1152.
9. J. Feinman, Plasma Technology in Metallurgical Processing,
Iron & Steel Society, Warrendale, PA, 1987, p. 27.
10. J. Pelletier, Distributed ECR: Concept, performance and per-
spectives, in C. M. Ferreira and M. Moisan, eds., Microwave
Discharges: Fundamentals and Applications, Plenum Press,
New York, 1992, pp. 181204.
11. R. Willhelm, ECR plasma sources, in C. M. Ferreira and
M. Moisan, eds., Microwave Discharges: Fundamentals and
Applications, Plenum Press, New York, 1992, pp. 161180.
12. A. A. Chtcherbakov and P. L. Swart, Automatic microwave
tuner for plasma deposition applications using a gradient
search method, J. Microwave Power Electromagn. Energy,
32:2833 (1997).
13. H. Conrads and M. Schmidt, Plasma generation and plasma
sources, Plasma Sources Sci. Technol. 9:441454 (2000).
14. G. Eecke, G. Eichhorn, J. Pezpldt, C. Reinhold, T. Stauden,
and F. Supplieth, Deposition of aluminium nitride lms by
electron cyclotron resonance plasma-enhanced chemical va-
pour deposition, Surf. Coat. Technol. 98:15031509 (1998).
15. P. Bulkin, N. Bertrand, and Drevillon, Deposition of SiO
2
in
integrated distributed electron cyclotron resonance micro-
wave reactor, Thin Solid Film 296:6668 (1997).
16. A. J. Chacon Velasco and V. D. Dougar-Jabon, Physical
principles of microwave plasma heating and its technological
applications, Proc. 7th Int. Conf. Microwave and High
Frequency Heating, Valencia, Spain, Sept. 1317 1999, pp.
99102.
17. M. Rusli, S. F. Yoon, H. Yang, J. Ahn, Q. F. Huang, Q. Zhang,
Y. P. Guo, C. Y. Yang, E. J. Teo, A. T. S. Wee, A. C. H. Huan,
and F. Watt, Tungsten-carbon thin lms deposited using
screen grid technique in an electron Cyclotron resonance
chemical vapour deposition system, Surf. Coat. Technol.
123:134139 (2000).
18. M. L. Passow, M. L. Brake, P. Lopez, W. B. McColl, and T. E.
Repetti, Microwave resonant-cavity-produced air discharge,
IEEE Trans. Plasma Sci. 19:219228 (1991).
19. Z. Zakrzewski, M. Moisan, and G. Sauve, Surface-wave plas-
ma sources, in C. M. Ferreira and M. Moisan, eds., Microwave
Discharges: Fundamentals and Applications, Plenum Press,
New York, 1992, pp. 117140.
20. G. Bohm, Fabry-Perot-type microwave resonator, in C. M.
Ferreira and M. Moisan, eds., Microwave Discharges: Funda-
mentals and Applications, Plenum Press, New York, 1992, pp.
215224.
21. M. Moisan, J. Hubert, J. Margot, G. Sauve, and Z. Zakrzew-
ski, The contribution of surface-wave-sustained plasmas to
HF plasma generation. Modelling and applications: Status
and perspectives, in C. M. Ferreira and M. Moisan, eds.,
Microwave Discharges: Fundamentals and Applications,
Plenum Press, New York, 1992, pp. 124.
22. M. Moisan, Z. Zakrzewski, R, Grenier, and G. Sauve, Large
diameter plasma generation using a waveguide-based eld
application at 2.45 GHz, J. Microwave Power Electromagn.
Energy 30:5865 (1995).
23. J. Laimer and S. Matsumoto, Pulsed microwave plasma-as-
sisted chemical vapour deposition of diamond, Int. J. Refrac-
tory Metals Hard Mater. 14:179184 (1996).
24. P. K. Bachmann and H. Lydtin, Diamond forming discharges,
in Proc. Materials Research Society Symp. on Characteriza-
tion of Plasma-Enhanced CVD Processes, Boston, Nov. 2728,
1989, Materials Research Society, Pittsburgh, 1990, Vol. 165,
pp. 181197.
25. M. Baeva, X. Luo, B. Pfelzer, T. Repsilber, and J. Uhlenbusch,
Experimental investigation and modelling of a low-pressure
pulsed microwave discharge in oxygen, Plasma Source Sci.
Technol. 9:128145 (2000).
26. Y. Mitsuda, Y. Kojima, T. Yoshida, and K. Akashi, The growth
of diamond in microwave plasma under low pressure, J. Mat-
er. Sci. 22:15571562 (1987).
27. M. A. Cappelli, T. G. Owano, A. Gicquel, and X. Duten, Methyl
concentration measurements during microwave plasma-as-
sisted diamond deposition, Plasma Chem. Plasma Process.
20:112 (2000).
28. K. Donnelly, D. P. Dowling, R. V. Flood, M. L. McConnell,
O. Berkefelt, and J. Svennebrink, Diamond deposition using a
novel microwave applicator, Surf. Coat. Technol. 124:248252
(2000).
29. W. Perasch, B. Kegel, H. Schmid, K. Lendenmann, and H. U.
Keller, Low-pressure plasma cleaning: A process for precision
cleaning applications, Surf. Coat. Technol. 97:176181 (1997).
30. A. Muller, M. Emme, D. Korzec, and J. Engemann, Direct
power coupling into a waveguide cavity plasma source, Surf.
Coat. Technol. 116119:674678 (1999).
31. A. I. Al-Shammaa, S. R. Wylie, J. Lucas, and C. F. Pau, Design
and Construction of a 2.45GHz waveguide-based microwave
plasma jet at atmospheric pressure for material processing, J.
Phys. D: Appl. Phys. 34:27342741 (2001).
32. A. I. Al-Shammaa, S. R. Wylie, J. Lucas, and R. A. Stuart,
Microwave plasma jet for material processing at 2.45GHz, J.
Mater. Process. Technol. 121:143147 (2002).
33. M. Moisan, Z. Zakrzewski, and J. C. Rostaing, Waveguide-
based single and multiple nozzle plasma torches: The TIAGO
concept, Plasma Sources Sci. Technol. 10:387394 (2001).
34. J. D. Yan, C. F. Pau, S. R. Wylie, and M. T. Fang, Experimental
characterization of an atmospheric argon plasma jet generat-
ed by an 896 MHZ microwave system, J. Phys. D: Appl. Phys.
35:25942604 (2002).
35. J. B. Salsman, Microwave plasma process for the accelerated
synthesis of nanostructured carbides, in K. Upadhya, eds.,
Plasma Synthesis and Processing of Materials, TMS, Denver,
1993, pp. 155162.
36. M. Karches and P. R. von Rohr, Microwave plasma charac-
teristics of a circulating uidized bed plasma reactor for coat-
ing of powders, Surf. Coat. Technol. 142144:2833 (2001).
37. http://www.abba.com.tw/html/microwave_burner.htm.
38. P. A. Rizzi, Microwave resonators and lters, in Microwave
Engineering, Prentice-Hall, Englewood, NJ, 1988, p. 412.
39. J. F. Gerling, Techniques to improve the performance of mi-
crowave process systems which utilize a high Q-cavities, in
D. E. Clark, F. D. Gac, and W. H. Sutton, eds., Ceramic Trans-
actions, Vol. 21, Microwaves: Theory and Application in
Materials Processing, American Ceramic Society, Waterville,
1991, pp. 667674.
40. A. L. Taube, G. M. Demyashev, and E. Siores, Microwave res-
onance-induced plasma technologies at atmospheric pressure:
a commercial reality in Australia, in Microwave Technologies
in National Economy: Introduction, Problems, Prospects, Uk-
raine Science Technology Centre, KievOdessa, 2002, Part 5,
pp. 6068.
41. R. Chatterjee, Advance Microwave engineering, Ellis Hor-
wood Ltd., UK, 1988, p. 125.
42. A. L. Taube and G. M. Demyashev, Comparative analysis of
atmospheric pressure microwave plasma sources: TM
013
res-
onance cavity and waveguide-type, Proc. 2002 Asia-Pacific
Microwave Conference (APMC-2002), Kyoto, Japan, Nov.
2002, Vol. 2, pp. 811814.
43. T. Koryulishii, Microwave Engineering, Harcourt Brace
Jovanovich, Orlando, FL, 1989, p. 136.
44. J. L. Altman, Microwave Circuits, Van Nostrand, Princeton,
NJ, 1964, p. 173.
45. G. M. Demyashev, A. L. Taube, and E. Siores, Superhard
nanocomposite coatings, in H. S. Nalwa, ed., Handbook of Or-
ganic-Inorganic Hybrid Materials and Nanocomposites, Vol.
2, Nanocomposites, American Scientific Publishers, Los An-
geles, 2003, Chap. 1, pp. 161.
46. A. L. Taube and G. M. Demyashev, Peculiarities of plasma
formation in the microwave resonance system for plasma
chemistry applications, Proc. IMPI 37th Annual Int. Micro-
wave Symp. Atlantic City, NJ, July 2002, pp. 1215.
47. R. T. Deam, Reduction of CO
2
by microwave induced plasma,
Proc. 9th Int. Conf. Microwave and High Frequency Heating,
Loughborough Univ., UK, Sept. 2003, 15, pp. 417420.
48. CHEMKIN Collection Release 3.6. September 2000, http://
www.reactiondesign.com/.
49. G. M. Demyashev, A. L. Taube, and E. Siores, Feasibility of
plasma nitriding by atmospheric pressure microwave reso-
nance-induced plasma source, in D. Tonchich, ed. Proles in
Industrial Research: Knowledge & Innovation-2001, IRIS
(Swinburne Univ. Technology), Melbourne, Australia, 2001,
pp. 274299.
50. E. Siores, A. L. Taube, and G. M. Demyashev, TM
013
resonance
cavity microwave plasma torch, Proc. IMPI 36th Annual Int.
Microwave Symp., San Francisco, April 2001, pp. 7578.
onance-induced plasma source for surface modication tech-
nologies at atmospheric pressure, Proc. 2001 Asia-Pacific
Microwave Conf. (APMC-2001), Taipei, Taiwan, Dec. 2001,
Vol. 2, pp. 733736.
onance-induced plasma source for surface treatment at atmo-
spheric pressure, Abstracts 8th Int. Conf. Microwave and
High Frequency Heating, Bayreuth, Germany, Sept. 2001,
pp. 4951.
53. K. Lu, Thermodynamics and kinetics of the amorphous-
to-nanocrystalline transformation, in C. Suryanarayana,
J. Singh, and F. H. Froes, eds., Processing and Properties
of Nanocrystalline Materials, TMS, Warrendale, 1996,
pp. 2333.
54. G. M. Demyashev, A. L. Taube, and E. Siores, Superhard
nanocomposites, in H. S. Nalwa, ed., Encyclopedia of Nano-
science and Nanotechnology, American Scientific Publishers,
Los Angeles, 2004, Vol. 10, pp. 191236.
55. G. M. Demyashev, A. L. Taube, and E. Siores, Nanostructured
tungsten-based composite coatings and their applications,
NANO Lett. 1:183187 (2001).
56. G. M. Demyashev, A. L. Taube, and E. Siores, Surface mod-
ication of titanium carbide with carbyne-containing nano-
coatings, J. Nanosci. Nanotechnol. 2:133137 (2002).
57. G. M. Demyashev and A. L. Taube, Carbyne-containing nano-
structured coatings for bioapplications: Synthesis and struc-
ture, in A. Dorner-Reisel and C. Schurer, eds., Surface
Engineering of BiomaterialsProperties, Processing and
Potentials of Carbon Coatings, Wiley Verlag, Freiberg, 2003.
58. G. M. Demyashev and A. L. Taube, Carbyne-containing bio-
materials, in M. Akay ed., Wiley Encyclopedia of Biomedical
Engineering, Wiley, New York, (in press).
59. V. I. Kirpatovsky, Medical applications of carbynoid materi-
als, in R. B. Heimann, S. E. Evsyukov, and L. Kavan, eds.,
Carbynes and Carbynoid Structures, Kluwer, Dordrecht,
1999, pp. 427435.
60. A. L. Taube and G. M. Demyashev, Microwave Resonance
Plasma Source Operating at Atmospheric Pressure: Method
and Apparatus, Australian Provisional Patent (2003).
MICROWAVE SCATTERING MODELS FOR
EARTH TERRAIN
ADRIAN K. FUNG
University of Texas at Arlington
Arlington, Texas
1. INTRODUCTION
The scattering properties of a terrain generated by an
impinging electromagnetic wave are generally repre-
sented by a quantity proportional to the average scattered
power called the scattering coefcient. This quantity is
dependent on the exploring frequency, view angle, polar-
ization, and the geometric and electric properties of the
terrain. Hence, it is a quantity that relates the geometric
and electric properties of the terrain and the sensing
system parameters to the scattering phenomenon. Its
precise denition will be given in the next section.
The scattering of waves that takes place at a surface
boundary between two homogeneous media is called sur-
face scattering. For natural ground surfaces where the
roughness can be described only statistically, the scattered
eld will vary from location to location. Such a variation in
the received signal is called fading, and the associated
eld amplitude and power distributions are its fading
statistics. A meaningful signature of the rough surface is
the statistically averaged, received power. It follows that
this average power must be a function of the statistical
parameters of the surface such as the standard deviation
of the surface height (RMS height) and its height correla-
tion function. In remote sensing it is the scattered eld
that is received by the observing antenna. Thus, scatter-
ing is the key mechanism. However, in the presence of an
inhomogeneous medium such as a vegetation layer, sea
ice, or a snow layer, the propagating elds within the
medium are equally important, since they are part of the
sources of the scattered eld. For an inhomogeneous layer
with irregular boundaries, an incident wave will generate
2948 MICROWAVE SCATTERING MODELS FOR EARTH TERRAIN
Next Page

12i

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

12i

Transféré par

Droits d'auteur :

Formats disponibles

MICROWAVE PHOTONICS: TECHNOLOGICAL

EVOLUTION AND ITS APPLICATIONS

Figure 26. Voltage gain function d versus k

, a lower active output power re-

Figure 2. Two-dimensional even array geom-

Vous aimerez peut-être aussi