07b

GROUND PENETRATING RADAR
DAVID J. DANIELS
ERA Technology
Surrey, United Kingdom
1. INTRODUCTION
Ground penetrating radar (GPR) is a nondestructive mea-
surement technique, which uses electromagnetic waves to
locate targets or interfaces buried within a visually opa-
que substance or Earth material.
GPR is also termed ground probing, surface penetrat-
ing (SPR), or subsurface radar. A GPR transmits a regular
sequence of low-power packets of electromagnetic energy
into the material or ground, and receives and detects the
weak reected signal from the buried target. The energy is
in the form of either a very short-duration impulse, a
sweep over a range of frequencies, radiation of noise over a
dened band, or a pseudorandom coded sequence of puls-
es. Most GPR systems, which all need to comply with the
relevant national and international regulations regarding
radio transmitters, operate within the range of frequen-
cies from 10 MHz to 10 GHz and can have a bandwidth of
several GHz. GPR systems are a special class of ultra-
wideband (UWB) radar systems. The typical average ra-
diated power is in the order of a thousandth of a watt. The
receiver is highly sensitive and can detect reected signals
of less than one millionth, of one millionth, of a watt. The
topic of radar system design is covered in many texts, and
useful information relating to GPR will be found in several
texts [17].
The buried target can be a conductor, a dielectric, or
combinations of both. The surrounding host material can
be soil, Earth materials, wood, rocks, ice, fresh water, or
manmade (synthetic) materials such as concrete or brick.
A typical GPR achieves a range of up to a few meters, but
some special systems can penetrate up to hundreds of me-
ters or even kilometers. A few GPR systems have been
operated from aircraft and from satellites to image geo-
logic features buried beneath the Sahara Desert as well as
measuring the depth of the (Earths) Moon and features on
Mars or comets.
The range of the GPR in the ground is limited because
of the absorption that the signal undergoes, while it trav-
els, on its two-way path, through the ground material.
GPR works well through materials such as granite, dry
sand, snow, ice, and freshwater, but will not penetrate
certain clays that are high in salt content or saltwater
because of the high absorption of electromagnetic energy
of such materials. In air, the GPR signal travels at the
speed of light, but is slowed down in ground materials by
their dielectric constant; hence true range needs calibrat-
ing for each material. GPR will not penetrate metal
because of the latters conductivity.
There are now a number of commercially available
equipments, and the technique is gradually developing
in scope and capability. GPR has also been used success-
fully to provide forensic information in the course of crim-
inal investigations, detect buried mines, survey roads,
detect utilities, measure geophysical strata, and in other
applications. Many GPR systems are handheld, but sys-
tems can be used on vehicles for rapid survey, by means of
an array of antennas. Other GPR systems are designed to
be inserted into boreholes to provide images of the inter-
vening rock. Typical GPR system attributes are given in
Table 1.
Most GPR systems use separate, human-portable,
transmit and receive antennas, which are placed on the
surface of the ground and moved in a known pattern over
the surface of the ground or material under investigation,
and an image can be generated, in real time, on a display
either in gray scale or in color. By systematically survey-
ing the area in a regular grid pattern, a radar image of the
ground can be built up. GPR images are displayed either
as two-dimensional representations, using horizontal (x or y)
and depth (z) axes or a horizontal plane representation (x,
y) at a given depth (z) or as a three-dimensional recon-
struction. GPR data may be classied as A scan, B scan, or
C scan depending on the plane of image. A scan is a mea-
surement at a single xed point in space and is displayed
in amplitude (y) and range (x). B scan is a representation
usually in grayscale or color-coded image intensity of a
plane (x, z or y, z) of scan, while C scan represents a hor-
izontal plane (x, y) at a given depth (z). Alternatively, the
GPR may be designed to provide an audible warning of
target presence.
The radar image is very different from an optical image
because the wavelengths of the illuminating radiation are
similar in dimension to those of the target. This results in
a much lower definition in the GPR image and one that is
highly dependent on the propagation characteristics of the
ground. The beam pattern of the antenna is widely spread
in the dielectric, and this degrades the spatial resolution
of the image, unless corrected. Refraction and anisotropic
characteristics of the ground may also distort the image.
For some longer-range systems, synthetic aperture tech-
niques processing techniques are used to optimise the res-
olution of the image.
Unprocessed GPR images often show bright spots
caused by multiple internal reections as well as a distor-
tion of the aspect ratio of the image of the target caused by
variations in the velocity of propagation. Symmetric tar-
gets, such as spheres or pipes, cause migration of the re-
ected energy to a hyperbolic pattern. GPR images can be
processed to compensate for these effects, and this is usu-
ally carried out offline. A GPR can be designed to detect
specific targets such as interfaces in roads, pipes, and
Table 1. Characteristics of GPR System
Pulse
Duration
(ns)
Center
Frequency
(mHz)
Target
Depth
(m)
Depth
Resolution
(m)
0.5 2000 o0.25 0.025
1.0 1000 o0.5 0.05
2.0 500 o1.0 0.1
4.0 250 o2.0 0.2
8.0 125 o4.0 0.4
16.0 63 o8.0 0.8
32.0 31 o16.0 1.6
GROUND PENETRATING RADAR 1833
Previous Page
cables and localized objects such as cubes, spheres, and
cylinders. GPR is capable of detecting features many hun-
dreds of years old; hence a prospective site should remain
unexcavated, prior to survey, so as to preserve its infor-
mation and aid interpretation of the GPR image.
2. PHYSICS OF PROPAGATION
2.1. Introduction
Maxwells equations are the foundation for the consider-
ation of the propagation of electromagnetic waves. In free
space the magnetic susceptibility and electric permittivity
are constants; that is, they are independent of frequency
and the medium is not dispersive. In a dielectric with zero
loss tangent, no losses due to attenuation are encountered
and hence there is no consideration of the attenuation,
which occurs in real dielectric media.
If an alternating electric eld is applied to a material,
the individual molecules will be induced to rotate in an
oscillatory manner about an axis through their centers,
the inertia of the molecules preventing them from re-
sponding instantaneously. Similar translational effects
can occur. The polarization produced by an applied eld
(such as a propagating radar wave) is closely related to the
thermal mobility of the molecules and is, therefore,
strongly temperature-dependent. In general, the relax-
ation time (which may be expressed as a relaxation fre-
quency) depends on activation energy, the natural
frequency of oscillation of the polarized particles, and tem-
perature. Relaxation frequencies vary widely between dif-
ferent materials.
For example, maximum absorption occurs at very low
frequencies in ice (10
3
Hz), whereas it takes place in the
microwave region in water (10
6
10
10
Hz); thus the effects
of this phenomenon can have a direct bearing on the di-
electric properties of materials at the frequencies em-
ployed by surface penetrating radars, especially if
moisture is present within a material. There are a num-
ber of other mechanisms that cause a separation of posi-
tively and negatively charged ions resulting in electric
polarization. These mechanisms can be associated with
ionic atmospheres surrounding colloidal particles (partic-
ularly clay minerals), absorbed water and pore effects, as
well as interfacial phenomenon between particles. The
general form of the model that describes the frequency
dependence of such systems is the Debye [8] relaxation
equation
e
0
ie
00
e
1
e
s
e
1
1 iot
where
e
0
real part of the dielectric permittivity
e
0 0
imaginary part of the dielectric permittivity
e
N
high-frequency limiting value of the permittivity
e
s
low-frequency limiting value of the permittivity
o radian frequency ( 2pf )
t relaxation time constant
The frequency of maximum movement and loss occurs
at o1/t.
In general, single relaxations are rarely observed in
natural systems. Instead, there are distributions of relax-
ations corresponding to distributions of size scales that
inuence movement of charge. There are several equa-
tions describing such distributed systems, with the most
common experimental observations in agreement with the
model from Cole and Cole [9]
e
0
ie
00
e
1
e
s
e
1
1iot
a
where a describes the breadth of the time constant distri-
bution, from a single relaxation, a 1, to an innitely
broad distribution, a 0, with a common process. Different
polarization processes may be described by a series of
ColeCole equations with different values of a and other
parameters.
The electromagnetic properties of a buried target must
be different from those of the surrounding soil or material,
and this means that to a rst order its relative dielectric
constant should be significantly lesser or greater than the
host soil. Typically, most soils exhibit a relative dielectric
constant, which ranges between 2 and 25. Freshwater has
a relative dielectric constant of approximately 80. It
should be noted that the ground and surface are quite
likely to be inhomogeneous and contain inclusions of other
rocks of various size as well as manmade debris. This sug-
gests that the signal to clutter performance of the sensor is
likely to be an important performance factor. Clutter may
be regarded as any radar return that is not associated
with the intended target and needs to be carefully dened.
2.2. Attenuation
Electromagnetic waves propagating through natural me-
dia experience losses, to both the electric (E) and magnetic
(H) elds. This causes attenuation of the original electro-
magnetic wave. Plane waves are good approximations to
real waves in many practical situations. More complicated
electromagnetic wavefronts can be considered as a super-
imposition of plane waves, and this method may be used to
gain an insight into more complex situations. For most
soils of interest in ground penetrating radar the magnetic
response is weak and need not be considered as a complex
quantity, unlike the permittivity and conductivity. Howev-
er, in certain soil types such as those derived from volcanic
rocks or otherwise high in iron content, full consideration
of the magnetic properties is necessary. In the case of lossy
dielectric materials, both conduction and dielectric effects
cause absorption of electromagnetic radiation.
The electromagnetic material properties that describe
such a coupled system are in the complex propagation
constant or circular wavenumber
g ika ib
where
g propagation constant
k circular wavenumber
a attenuation constant (Np/m) (nepers per meter)
k
2
o(m
0
im
00
)(o(e
0
ie
00
) is)
b phase constant (rads/m)
1834 GROUND PENETRATING RADAR
with
a o
_
m
0
e
0
m
00
e
00
s
o
_ __
2
_
m
00
e
0
m
0
e
00
s
o
_ __
2
_

_
m
0
e
0
m
00
e
00
s
o
_ __
2
_
b o
_
m
0
e
0
m
00
e
00
s
o
_ __
2
_
m
00
e
0
m
0
e
00
s
o
_ __
2
_

_
m
0
e
0
m
00
e
00
s
o
_ __
2
_
The eld at a distance z from the source is given by
Ez; t E
0
e
az
e
jotbz
The wavelength l in the medium is given in meters, and
the frequency f is expressed in hertz:
l
2p
b

n
f
The losses in such systems are described in terms of tan-
gents of loss angles d between elds and uxes:
tan d
e
e
00
e
0

s
oe
0
The electrical loss tangent represents the sum of the
charge transport and polarization relaxation losses, and
the phase angle between electric eld and current density.
The skin depth or attenuation length is 1/a(m); the dis-
tance electromagnetic energy travels while being attenu-
ated by 1/e in amplitude. This distance is known as the
skin depth d and provides an initial guide to the useful
penetration depth of a ground penetrating radar system,
although in some media the useful range may be greater.
In a lossy conducting dielectric the parameters associ-
ated with susceptibility can be rearranged to give
a o
me
0
2
1
e
0
e
00
_ _
2
1
_
_
_
_
_
_
_
_
_
b o
me
0
2
1
e
0
e
00
_ _
2
1
_
_
_
_
_
_
_
_
_
and where the losses are low, then
tan d
e
s
oe
0
This article has not considered the electromagnetic and
magnetic loss tangent, and these may need to be consid-
ered in special cases.
It can be seen from the expressions above that the at-
tenuation constant of a material is, to a rst order, linearly
related (in dB/m) to frequency. It is not sufcient to con-
sider only the low-frequency conductivity when attempting
to determine the loss tangent over the frequency range
10
7
10
10
Hz. In the case of a material that is dry and rel-
atively lossless, it may be reasonable to consider that tan d
is constant over that frequency range. However, for mate-
rials that are wet and lossy such an approximation is in-
valid. There are, nevertheless, a number of other factors,
that inuence the effective penetration depth, notably the
strength of reection from the target sought, and the de-
gree of clutter suppression of which the system is capable.
A rst-order estimate of the various contributions to
signal loss can be carried out using the standard radar
range equation, although this is only applicable for far-
eld conditions and thus has restrictions:
P
r
P
t
AGsk
4pR
2
2
e
a2R
where
P
t
transmitted power in watts
P
r
received power in watts
A antenna gain
G antenna effective aperture
R range in meters
s target radar cross section
k calibration coefcient
The cumulative losses include the transmission coef-
cients into the ground, the spreading losses describe the
R
4
losses for a target of 1 m
2
, and the attenuation losses
are for a soil with a e
r
of 9 and tand of 0.1. Fixed losses
include the transmission losses into the soil and the re-
ection loss from the target. In Fig. 1 the rst meter (1m)
has not been plotted, as the radar range equation is not an
accurate model in this range and the purpose of the ex-
planation is to provide a basic introduction to rst-order
signal estimation.
2.3. Velocity
The velocity of propagation of electromagnetic waves in free
space is approximately 310
8
m/s (2.99792510
8
m/s),
0
0 10 R
1 2 3 4 5 7 6 8 9 10
100
90
80
70
60
50
40
30
20
10
0
Reference
Fixed losses
Spreading loss
Attenuation
Noise floor
Clutter
Range in metres
S
i
g
n
a
l

l
e
v
e
l

i
n

d
B
Figure 1. Graph of the typical losses encountered at 100MHz by
a GPR probing into the ground. (This gure is available in full
color at http://www.mrw.interscience.wiley.com/erfme.)
but slows in a material depending on its relative permit-
tivity and relative magnetic permeability. The velocity of
propagation of electromagnetic waves in a soil with a val-
ue for e
r
of 9 would be slowed to 1.10
8
m/s. The time to a
target at a range of 1m is therefore 20 ns, and GPR sys-
tems operate at time ranges between a few nanoseconds
up to 200ns, although some systems for probing through
ice may use ranges up to several tens of milliseconds.
In general, it is not possible to make a reliable estimate
of propagation velocity or relative permittivity in a medi-
um from a single measurement without trial holing or
other supplementary information. Even in the case where
a measurement is carried out at one location, it is often
found that significant variations in velocity will occur
within comparatively short distances from the original lo-
cation. This can lead to significant errors in the estimation
of depths of reectors. One procedure that overcomes this
limitation is known as common-depth-point surveying,
which utilizes two antennas in bistatic operation at a
number of transmit and receive positions. The velocity of
propagation is also slowed by tand and e
r
and assumes
m
r
1 and is given by
n c
e
0
2e
0
1
e
0
e
00
_ _
2
1
_
_
_
_
_
It is also possible to derive velocity from multiple mea-
surements scanning over a target, but this works well only
in relatively uncluttered situations where the media has
no anisotropic characteristics.
2.4. Reection
In any estimation of received signal level it is necessary to
consider the coefcients of reection and transmission, as
the wave passes through the dielectric to the target and
Snells laws describe the associated angles of incidence,
reection, transmission, and refraction. Where lossy ma-
terials are involved, complex angles of refraction may oc-
cur, unlike the simple classical case, and polarization and
the Stokes matrix may also be required for oriented high-
aspect-ratio features such as pipes, wires, and fractures.
The intrinsic impedance Z of a medium is the relationship
between the electric eld E and the magnetic eld H and is
a complex quantity given by
Z
jom
s joe
At the boundary between two media, some energy will be

reected and the remainder transmitted. The reected
eld strength is described by the reection coefcient r
r
Z
2
Z
1
Z
2
Z
1
where Z
1
and Z
2
are the impedances of media 1 and 2,
respectively.
The reection coefcient has a positive value when
Z
2
4Z
1
, such as where an air-lled void exists in a dielec-
tric material. The effect on a pulse waveform is to change
the phase of the reected wavelet so that targets with rel-
ative dielectric constants different from those of the host
material show different phase patterns of the reected
signal. However, the propagation dielectric of the host
material, and the geometric characteristics of the target
and its dielectric parameters affect the amplitude of the
reected signal.
2.5. Polarization
A complete description of the radar scattering cross sec-
tion of a target includes a description of its polarization
scattering characteristics. The polarizing properties of tar-
gets are described by the Stokes parameters, and the po-
larization coordinates can be represented on the Poincare
sphere. All of these are well described in standard texts. In
summary, these descriptions allow the state of an electro-
magnetic wave to be described in terms of linear, elliptical,
and circular polarization (left-handed or right-handed). It
is well known that linear targets as depolarizing features
and a linearly polarized crossed dipole antenna rotated
about an axis normal to a linear target produces a sinu-
soidal variation in received signal. However, the null
points are a distinct disadvantage because the operator
is required to make two separate, axially rotated mea-
surements at every point to be sure of detecting pipes at
unknown orientations. An attractive technique is to radi-
ate a circularly polarized wave, which automatically
rotates the polarized vector in space and hence removes
the direction of signal nulls.
2.6. Dispersion
The frequency-dependent nature of the dielectric proper-
ties of the material causes the phase velocity of the com-
ponent frequencies of a wideband signal to suffer
differential propagation values. Hence, there will be vari-
ation in the velocity of propagation with frequency. Di-
electrics exhibiting this phenomenon are termed
dispersive. In this situation, the different frequency com-
ponents within a broadband radar pulse would travel at
slightly different speeds, causing the pulseshape to change
with time. However, the propagation characteristics of oc-
tave band radar signals remain largely unaffected by dis-
persion In many instances, the potential variation in the
velocity of wave propagation over the frequency range of
interest is small and will be ignored.
2.7. Clutter
A major difculty for operation of GPR systems is the
presence of clutter within the material. Clutter is dened
as sources of unwanted reections that occur within the
effective bandwidth and search window of the radar and
present as spatially coherent reectors. The definition of
clutter depends heavily on the intended target. The oper-
ator of a GPR system searching for pipes may classify the
interfaces between road layers as clutter, whereas the op-
erator of a system measuring road layer thickness might
consider pipes and cables as sources of clutter. Careful
definition and understanding are critically important in
selecting and operating the best system and processing
algorithms. Clutter can completely obscure the buried tar-
get and a proper understanding of its source and impact
on the radar is essential.
2.8. Depth Resolution
Normally the range, or in the case of GPR, the depth res-
olution, of a radar is dened by considering the case of
targets with identical radar scattering cross sections. In
the case of Gaussian pulses, resolution can be achieved at
pulse separations between 1.25 and 1.5 of the pulsewidth
and can be dened by the Rayleigh criteria for resolution.
Essentially, range resolution is dened by the bandwidth
of the received signal. A receiver bandwidth in excess of
500MHz and typically 1 GHz is required to provide a typ-
ical resolution of 520cm, depending on the relative per-
mittivity of the material.
When a number of features may be present, a signal
having a larger bandwidth is required to be able to dis-
tinguish between the various targets and to show the de-
tailed structure of a target. In this context it is the
bandwidth of the received signal that is important, rath-
er than that of the transmitted wavelet. The material acts
a lowpass lter that modies the transmitted spectrum in
accordance with the electrical properties of the propagat-
ing medium. There are some applications of subsurface
radar, such as road layer thickness measurement, where
the feature of interest is a single interface. Under such
circumstances, it is possible to determine the depth suf-
ciently accurately by measuring the elapsed time between
the leading edge of the received wavelet provided the
propagation velocity is accurately known.
Although a greater depth resolution is achieved in wet-
ter materials for a given transmitted bandwidth, Earth
materials with significant water content tend to have
higher attenuation properties. This characteristic reduc-
es the effective bandwidth, tending to balance out the
change so that within certain bounds the resolution is ap-
proximately independent of loss within the propagating
material.
Where interfaces are spaced more closely than a half-
wavelength, the reected signal from one interface will
become convolved with that from the other. In such cir-
cumstances, some form of deconvolution processing would
be required in order to recognize the responses from the
individual interfaces and to enable them to be character-
ized and traced. Such processing is seldom carried out
during standard commercial radar surveys.
It should be noted that the Rayleigh criteria for range
resolution are less appropriate for the case of a weak tar-
get adjacent to strong target. For example, if the ampli-
tude of the weak target is 20 dB relative to that of the
strong target, different criteria are needed. Thus the res-
olution at the 20 dB amplitude level may be a more rel-
evant criterion on which to assess the merits of competing
system designs. This is because of the need to detect tar-
gets of low radar cross section in close proximity to targets
of high radar cross section.
It can be shown [10] that for equivalent main 3 dB
bandwidths, a radar design using a time-domain, direct
receiver has a considerably better resolution at the
20 dB level than does a frequency-domain radar. This
is primarily because the receiver of the frequency-domain
radar cross-correlates the received signal. In essence, res-
olution has been achieved at the expense of signal-to-noise
ratio, but this may not be so relevant to attenuation-
limited operation.
2.9. Plan Resolution
The plan resolution of a subsurface radar system is im-
portant when localized targets are sought and when there
is a need to distinguish between more than one at the
same depth. Where the requirement is for location accu-
racy, which is primarily a topographic surveying function,
the system requirement is less demanding.
The plan resolution is dened by the characteristics of
the antenna and the signal processing employed. In gen-
eral, to achieve an acceptable plan resolution, one must
employ a high-gain antenna. This necessitates transmis-
sion of a sufciently large aperture at the lowest frequen-
cy. To achieve small antenna dimensions and high gain
therefore requires the use of a high carrier frequency,
which may not penetrate the material to sufcient depth.
When selecting equipment for a particular application, it
is necessary to compromise between plan resolution, size
of antenna, the scope for signal processing, and the ability
to penetrate the material. Plan resolution improves as at-
tenuation increases, provided there is sufcient signal to
discriminate under the prevailing clutter conditions. In
low-attenuation media the resolution obtained by the hor-
izontal scanning technique is degraded, but under these
conditions the use of advanced signal processing tech-
niques becomes feasible and synthetic aperture tech-
niques can dramatically increase the plan resolution.
These techniques typically require measurements to be
made using transmitter/receiver pairs at a number of an-
tenna positions to generate a synthetic aperture or focus
the image. Unlike conventional radars, which generally
use a single antenna, most ground penetrating radar sys-
tems use separate transmit and receive antennas in what
has been termed a bistatic mode.
3. PROPERTIES OF MATERIALS
Determination of the dielectric properties of Earth mate-
rials remains largely experimental. Rocks, soils, and con-
crete are complex materials composed of many different
minerals in widely varying proportions, and their dielec-
tric parameters may differ greatly even within materials,
that are nominally similar. Most Earth materials contain
moisture, usually with some measure of salinity. Since the
relative permittivity of water is in the order of 80, even
small amounts of moisture cause a significant increase of
the relative permittivity of the material. A large number
of workers have investigated the relationships between
the physical, chemical, and mechanical properties of ma-
terials and their electrical and in particular microwave
properties. In general, they have sought to develop
suitable models to link the properties of the material to its
electromagnetic parameters. Such models provide a basis
for understanding the behavior of electromagnetic waves
within these media. The real and imaginary dielectric
losses as a function of frequency can be plotted over a wide
frequency range, and a typical result is shown in Fig. 2.
Information on the geologic properties of Earth soils
can be found in the Digital Soil Map of the World and De-
rived Soil Properties CD r published by the Food and
Agriculture Organisation of the United Nations. This en-
ables the 10 map sheets of the world to be classied in
terms of parameters such as pH, organic carbon content,
carbon/nitrogen ratio, clay mineralogy, soil depth, soil
moisture capacity, and soil drainage class. Such informa-
tion is useful in assessing the potential of RF techniques
and particularly GPR for particular geographic regions.
There are two benets to understanding soil properties
in relation to GPR. (1) to understand the applicability of
GPR to particular soils and hence the possibility of using
GPR to detect buried targets such as pipe, cables, and
landmines and (2) to use GPR to characterise soils and soil
properties.
GPR can provide a detailed map of the subsurface,
which, combined with traditional soil survey methods, can
provide information on the type, lateral extent, and depth
of the soil; the water table; and the layering and features
of the soil and hence its local geology and history.
Since the mid-1970s GPR has been used in the united
states by the Department of AgricultureNatural Resourc-
es Conservation Service (USDA-NRCS) as a quality con-
trol tool for soil mapping and investigations. The use of
GPR in soil survey activities has provided information
about the soil resource that would have been unobtainable
by other means or would have been uneconomic to obtain.
An example of the results of this work is shown in Fig. 3.
4. GPR SYSTEMS
The choice of system design is to a large extent governed
by the type of target, the resolution required, and the an-
ticipated ground attenuation and clutter. The depth range
of the radar system is likely to be primarily dened by the
soil attenuation, once a particular frequency range has
110
8
110
8
110
9
110
10
110
11
110
11
0 0
4
8
12
16
20
20
Frequency
V
a
l
u
e
f

f
k.f
Figure 2. Real and imaginary dielectric losses plotted as a func-
tion of frequency for a typical Earth material with specied frac-
tions of water, sand, silt, and clay. (This gure is available in full
color at http://www.mrw.interscience.wiley.com/erfme.)
Figure 3. Soil suitability map for GPR applications in the United States. (Courtesy of USDA-
NRCS.) (This gure is available in full color at http://www.mrw.interscience.wiley.com/erfme.)
been chosen. However, it can be shown that considerable
variations (1030dB) in the sensitivity of competing sys-
tem designs actually translate to relatively small changes
in depth performance in lossy soils.
The selection of a suitable waveform for transmission,
at least in terms of resolution, can be considered a func-
tion of the duration of the complex envelope of the mea-
sured signal. The output from most ultrawideband radar
systems can be compared in terms of a time-domain rep-
resentation of the waveform. Almost all types of radar can
be assessed not just by their signal-to-noise and signal-to-
clutter ratios but also by comparing their inherent range
sensitivity. Such a procedure reveals the characteristics
that control radar performance. The design of a GPR sys-
tem is dened by the modulation technique and time-do-
main, frequency-domain, and noise-domain radar designs
are most likely to be encountered.
The frequency-domain, matched-lter, radar receiver
provides an optimum linear processing of radar in the
presence of noise. The radar signal is processed by a re-
ceiver that cross-correlates the received waveform with a
suitably time-delayed version of the transmitted wave-
form. The output results in a compressed pulse in which
the amplitude of the latter and its position in delay time is
related to the target radar characteristic. This type of re-
ceiver is widely used to process chirp, stepped-frequency,
coded, and noise waveforms, and the design of such wave-
forms is extensively described in the literature. Time-do-
main radar systems are simpler in design and use a
sampling receiver to downconvert the radar signals from
the nanosecond timeframe to a millisecond timeframe that
is easier to postprocess. However, a real disadvantage of
the sampling receiver is its limited dynamic range and
high noise oor.
A key parameter for most radar systems is the mean
power. For portable, close-in radars the capacity of the
power sources limits the radiated power. The frequency-
domain radar transmits, on a repetitive basis, a nominally
constant amplitude signal whose frequency increases in a
linear progression from the lowest to the highest values.
However, shaping of the envelope of the radiated signal
reduces the mean power compared with the instantaneous
peak value. The time-domain radar transmits, on a repet-
itive basis, a short-duration impulse. Consequently the
peak power is significantly greater than the mean power.
This is not the case with noise radar, whose radiated pow-
er per unit bandwidth is optimally low compared with the
frequency-domain and time-domain designs.
5. MODULATION TECHNIQUES
There are three basic modulation techniques: time-do-
main, frequency-domain, and noise-coded radar.
5.1. Time-Domain Radar
Most commercially available ground probing radar sys-
tems use short pulses or impulses such as the Ricker
wavelet shown in Fig. 4. The high-speed sequential sam-
pling approach used to acquire RF waveforms produces a
low SNR because the spectrum of the sampling pulse is a
poor match for that of the received pulse. In general, the
dynamic range of the sampling receiver is typically 60 dB,
without time varying gain, and with time varying gain is
equivalent to 90 dB or more. Signal averaging can increase
the effective sensitivity by the amount of averaging, and
this can be typically 1030dB. The overall dynamic range
is set largely by the ratio of the mean power to the effective
receiver sensitivity and can extend up to 150dB. The an-
tennas that can be used are limited to linear phase designs
such as resistively loaded dipoles, TEM horns, or impulse
radiating antennas (IRAs) unless the dispersive proper-
ties of the antenna used are compensated by suitable post-
processing ltering.
5.2. Frequency-Domain Radar
The main potential advantages of the frequency-domain
radar are the wider dynamic range, lower noise gure,
and higher mean powers that can be radiated. There are
two main types of frequency-domain radar: frequency-
modulated carrier wave (FMCW) and stepped-frequency
carrier wave (SFCW). FMCW radar transmits a continu-
ously changing carrier frequency by means of a voltage-
controlled oscillator (VCO) over a chosen frequency range
on a repetitive basis as shown in Fig. 5. The received sig-
nal is mixed with a sample of the transmitted waveform
and results in a difference frequency that is related to the
phase of the received signal; hence its time delay and
hence range of the target. The difference frequency or in-
termediate frequency (IF) must be derived from an I/Q
mixer pair if the information equivalent to a time-domain
representation is required, as a single-ended mixer pro-
vides only the modulus of the time-domain waveform. The
FMCW radar system is particularly sensitive to certain
parameters. In particular, it requires a high degree of lin-
earity of frequency sweep with time to avoid spectral wid-
ening of the IF and hence degradation of system
resolution. An assessment of the sensitivity of sidelobe
level to linearity was made by Dennis and Gibbs [11], who
showed the ratio of sidelobe to peak level was dependent
on the sweep linearity. Practically the effect of a nonlin-
earity of a few percent is to cause significant sidelobes.
0 16 32 48 64 80 96 112 128
1
0
1
2
t
(df)
t
Figure 4. Ricker wavelet (amplitude vs. time).
The SFCW radar transmits a series of incremental fre-
quencies and stores the received IF signal to then carry
out a fast fourier transform (FFT) reconstruction of the
time-domain equivalent waveform. The SFCW has found
many applications, and the impact of mobile communica-
tions technology has had a significant impact on reducing
the cost of radar components for this design. Two forms of
the synthesized radar can be considered. The rst and
simplest system is stepped-frequency continuous-wave ra-
dar. The second form is more complex in that each indi-
vidual frequency is appropriately weighted in amplitude
and phase prior to transmission. Normally the radar is
calibrated to establish a reference plane for measurement
as well as to reduce the effect of variations in the frequency
characteristics of components and antennas.
A much wider class of antenna is available for use by
the designer of frequency-domain radars. The sensitivity
of the receiver is much better than the time-domain equiv-
alent, simply by virtue of its lower bandwidth and hence
lower thermal noise. Typically, a sensitivity of 120dBm is
found, and a system dynamic range of 180 dB is feasible.
However, the FMCW ground probing radar system is par-
ticularly sensitive to certain parameters. In particular, it
requires a high degree of linearity of frequency sweep with
time to avoid spectral widening of the IF and hence
degradation of the system resolution.
The main potential advantages of a stepped-frequency
or FMCW ground probing radar are its ability to adjust
the range of frequencies of operation to suit the material
and targets and electromagnetic environment under in-
vestigation if the antenna has an adequate passband of
frequencies. It can radiate a higher mean power level per
spectral line, and its ability to integrate the received sig-
nal level improves the system sensitivity. The calibration
of the radar does, of course, depend on stable system char-
acteristics and antenna parameters that are invariant
with the spacing of the front surface and the antenna. Al-
though on rst consideration frequency-domain radars
should offer a sensitivity superior to that of time-domain
radars, because of their lower receiver bandwidth and,
hence thermal noise, both the type of receiver and the
range sidelobes of the radiated spectrum may result in an
equivalent or worse sensitivity in terms of range resolu-
tion as discussed above.
5.3. Noise or PseudoRandom Coded Radar
More recently, work has been carried out on noise modu-
lation and pseudorandom coded modulation techniques for
GPR. The main advantage of these methods is that the
energy transmitted is spread more evenly over the spec-
trum than with any other modulation method, and hence
the likelihood of interference to other users of the spec-
trum is minimized. In addition, the chances of other users
of, say, mobile phones interfering with the GPR operator
are also reduced. The mean power is the lowest of any of
the modulation schemes, and this is helpful in meeting
regulatory requirements.
The transmitted signal has noiselike characteristics,
and the received signal is cross-correlated with a sample
of the transmitted signal. The range of the target is given
by the time position of the cross-correlated signal and the
amplitude by the peak of the cross-correlated signal. Con-
trol of the cross-correlation sidelobes is vital to achieve
good range resolution, and the sidelobes are affected by
the antenna and system characteristics as well as the du-
ration and randomness of the transmitted waveform. Fur-
ther information is given by Narayan et al. [12] and Sachs
et al. [13].
6. ANTENNAS
In the ultrawideband case the radar antennas are consid-
ered in terms of their transfer function rather than their
gains or effective apertures. In many cases separate trans-
mit and receive antennas are used; hence their transfer
functions may not be identical. The type of antenna that is
used with ultrawideband radar has an important role in
dening the performance of the radar.
The types of antenna that are useful to the designer of
ultrawideband radar fall into two groups: dispersive an-
tennas and nondispersive antennas. However, in princi-
ple, all antennas are dispersive to some extent. Examples
of dispersive antennas that have been used in ultrawide-
band radar are the exponential spiral, the Archimedean
spiral, the logarithmic planar antenna, the Vivaldi anten-
na, slot antennas, and the exponential horn. The impulse
response of this class of antenna is extended, and gener-
ally results in a chirp waveform if the input is an im-
pulse. Examples of nondispersive antennas are the TEM
horn; the bicone; the bowtie; the resistive, lumped-ele-
ment loaded antenna; and the continuously, resistively
loaded antenna. The input voltage driving function to the
terminals of the antenna in an impulse radar is typically a
Gaussian pulse, and this requires the impulse response of
the antenna to be extremely short, mainly to prevent the
antenna from distorting the input function and generating
0 32 64 96 128 160 192 224 256
1
0.5
0
0.5
1
a
t
t
Figure 5. Typical FMCW transmitted
swept waveform (amplitude vs. time).
time sidelobes. These time sidelobes would obscure targets
that are close in range to the target of interest; in other
words, the resolution of the radar can become degraded if
the impulse response of the antenna is significantly
extended.
Element antennas are characterized by linear polar-
ization, low directivity, and relatively limited bandwidth
unless either end loading or distributed loading tech-
niques are employed, in which case bandwidth is in-
creased at the expense of radiation efciency. Horn
antennas have found most use with FMCW ultrawide-
band radars where the generally higher frequency of op-
eration and relaxation of the requirement for linear phase
response permit the consideration of this class of antenna.
FMCW ultrawideband radars have used an offset para-
boloid fed by a ridged horn. This arrangement was de-
signed to focus the radiation into the ground at a slant
angle to reduce the level of the reection from the ground.
Care needs to be taken in such arrangements to minimize
the effect of back- and sidelobes from the feed antenna,
which can easily generate reection from the ground
surface.
One method of radiating circular polarization is to use
an equiangular spiral antenna. The dispersive nature of
this type of antenna causes an increase in the duration of
the transmitted waveforms, and the radiated pulse takes
the form of a chirp in which high frequencies are radi-
ated rst, followed by the low frequencies. This effect,
however, may be compensated by a spiking lter, which
may take the form of a conventional matched lter or a
more sophisticated lter such as Wiener lter.
7. SIGNAL AND IMAGE PROCESSING
The most basic GPR data record is an A scan, and an ex-
ample is shown in Fig. 6. An A scan provides an ampli-
tudetime record of a single measurement over a target.
Only amplitude range information is plotted. Ground pen-
etrating radar is generally used in such a way as to gen-
erate a sequence of A scans related to the survey position
on the ground surface. This sequence can be termed a B
scan, and an example is shown in Fig. 7. This effectively
represents on one axis (z) depth and the orthogonal axis
(x or y) linear position. The amplitude of the signal may be
shown as a series of overlapping signals or alternatively a
wiggle plot (borrowed from the seismic terminology) or a
grayscale coded intensity plot or a pseudocolor image. In
the modeled example shown, the hyperbolic spreading of
the target spatial response can be seen. As shown in Fig. 8,
a C scan consists of a plan view (x,y plane over a dened
range of depth z).
As the antennas generally used for surface penetrating
radars have a poor directivity, the pattern of the reected
waveform in the B scan represented the spatial convolu-
tion of the antenna pattern with the point source of the
target. This pattern may be deconvolved using any of the
following processes: synthetic aperture processing, conju-
gate gradient methods, and reverse time migration. Many
of these techniques work well on isolated targets such as
pipes, which have well dened geometric boundaries. The
situation is more difcult with stratied layers and, of
course, anisotropic materials. When a focused image of the
buried object is created, whether as a B or C scan (area at
a particular range of depths), it is necessary to interpret
0 20 40 60 80 100 120 140 160 180 200
1
0.5
0
0.5
1
composite(toffset) output (t)
t
N
5
Figure 6. Typical example of GPR A
scan (amplitude/time-modeled data).
Figure 7. Typical example of B scan (y, z axis, gray scale image of
modeled data).
the radar image as being generated by a physical struc-
ture. This is not always easy in the case of a cluttered im-
age, and a great deal still depends on the eld experience
of the operator.
Automated methods based on standard image process-
ing methods (mask ltering), image matching, and similar
methods, as well as methods based on neural networks,
have been investigated. The variability of ground condi-
tions, as well as the physics of EM wave propagation and
reection, must be carefully taken into account in gener-
ating pattern libraries. For example, the depth image of a
void is always apparently smaller than the voids physical
size; corner reectors of any reasonable size generate
large, apparently discontinuous reection images; and con-
ductive targets, which reverberate by means of stored en-
ergy, create extended-depth images. The image of a buried
target generated by a subsurface radar will not, of course,
correspond to the targets geometric representation. The
fundamental reasons for this are related to the ratio of the
wavelength of the radiation and the physical dimensions of
the target. In most cases for surface penetrating radar, the
ratio is close to unity. This compares very differently with
an optical image, which is obtained with wavelengths such
that the ratio is considerably greater than unity. In ground
penetrating radar, applications, the effect of combinations
of scattering planes, for example, the corner reector, can
cause bright spots, in the image and variations in the
velocity of propagation can cause dilation of the aspect
ratio of the image. While many images can be focused to
reduce the effect of antenna beamspreading, regeneration
of a geometric model is a much more complex procedure
and is rarely attempted.
The general objective of signal processing as applied to
ground penetrating radar is to present either an image
that can readily be interpreted by the operator or to
classify the target return with respect to a known test
procedure or template. While much work has been carried
out on 3D SAR processing, there is a tendency for this to
be restricted to situations where the customer is prepared
to pay for the additional information.
The general processing problem encountered in dealing
with ground penetrating radar data is in the widest sense
the extraction of a localized wavelet function from a time
series that displays time-domain characteristics very sim-
ilar to those of the wavelet. This time series is generated
by signals from the ground and other reecting surfaces,
as well as internally from the radar system. Unlike con-
ventional radar systems in which the target can generally
be regarded as being in motion compared with the clutter,
in the ground penetrating radar case, the target and the
clutter are spatially xed and the radar antenna is moved
with respect to the environment.
It is assumed that data are recorded to an adequate
resolution and bandwidth. Most antennas used in surface
penetrating applications have a limited low-frequency re-
sponse and tend to act as highpass lters effectively dif-
ferentiating the applied impulse, hence creating a
wavelet. In the case of antennas operated in close prox-
imity to the ground, the antenna characteristics may vary
as a result of changes in the ground surface electrical pa-
rameters. Any processing scheme that relies on invariant
antenna parameters should take into account the mode of
operation of the antennas and the degree of stability that
is practically realizable. This is a particular issue for GPR
and needs careful attention to reduce the effect of anten-
naground surface interaction.
Alternative methods of processing are associated with
identifying target resonances using singularity expansion
methods (SEMs) to evaluate the target response. Early
research using Prony methods encountered difculties
due to the ill-conditioned nature of natural frequency
extraction from noisy data.
Some of the ancillary requirements of an operational
subsurface radar system need to be considered. There is a
need for an accurate, small-scale, low-cost position refer-
encing system for use with radar for subsurface survey
techniques. It is important that data can be related to a
true geographic reference particularly when led on dig-
ital mapping systems and used to dene areas of safe
working. It is necessary to provide some means of
scanning the antenna. Obviously a basic approach is the
Figure 8. Typical example of C scan (xy plane at dened z-range
color image of modeled data of buried landmines). (This gure is
available in full color at http://www.mrw.interscience.wiley.com/
erfme.)
Figure 9. GPR data from reinforced-concrete roadway taken with
1-GHz SPR radar system at 50-ns time window and 3m horizontal
distance. (Courtesy of ERATechnology.) (This gure is available in
full color at http://www.mrw.interscience.wiley.com/erfme.)
handheld device, but this places some limitations on the
signal processing strategies.
Another consideration is the plane of polarization of the
electromagnetic energy. For targets with one large area
dimension such as a pipe, the radar cross-scattering sec-
tion will be larger when the polarization vector is in line
with the pipe. This means that any area that is surveyed
with, say, parallel dipoles must be surveyed in orthogonal
directions to ensure that no targets are missed. The same
principle also relates to crossed dipole antennas.
8. APPLICATIONS
8.1. Introduction
It is possible to provide only a brief summary of the wide
variety of the applications for GPR, which has in some
cases become an established and routine method of sub-
surface investigation. In other applications, such as bore-
hole detection, the technique is under development,
whereas in others, such as mine detection, basic research
is being carried out, although some equipment is now be-
ing used in eld trials. GPR, in the hands of an expert,
provides a safe and noninvasive method of conducting
speculative searches without the need for unnecessary
disruption and excavation.
GPR has significantly improved the efciency of the
exploratory work that is fundamental to the construction
and civil engineering industries, the police and forensic
sectors, security/intelligence forces, and archaeological
surveys.
GPR has been very successfully used in forensic inves-
tigations. The most notorious cases are in the United
Kingdom in 1994, when the gravesites, under concrete
and in the house of Fred West, of the victims of the serial
murderer, were pinpointed. In Belgium, the gravesites of
the victims of the pedophile Dutroux were detected in
1996.
Archaeological applications of GPR have been varied,
ranging from attempts to detect the Ark to the exploration
of Egyptian (see Fig. 10) and North American Indian sites
as well as castles and monasteries in Europe. The quality
of the radar image can be exceptionally good, although
correct understanding normally requires joint interpreta-
tion by the archaeologists and radar specialists see
(Fig. 11).
Abandoned antipersonnel landmines and unexploded
ordnance are a major hindrance to the recovery of many
countries from war. Their effect on the civilian population
is disastrous, and major efforts are being made by the in-
ternational community to eliminate the problem. Most de-
tection is done with metal detectors, which respond to the
large amount of metallic debris in abandoned battleeld
areas and hence have difculty in detecting the minimum
metal or plastic, mine. GPR technology is being applied to
this problem as a means of reducing the false-alarm rate
and providing improved detection of low-metal-content
mines. An example of data gathered from a 4-m wide
GPR 32-channel array is shown in Fig. 12.
GPR has been used for surveying many different types
of geologic strata ranging from exploration of the Arctic
and Antarctic icecaps and the permafrost regions of North
America, to mapping of granite, limestone, marble, and
other hard rocks as well as geophysical strata.
The thickness of the various layers of a road can be
measured using radar techniques. The great advantage is
that this method is nondestructive and high-speed
(440 km/h) and can be applied dynamically to achieve a
continuous prole or rolling map. The accuracy of calibra-
tion tends to reduce as a function of depth because of the
attenuation characteristics of the ground. The accuracy
may be quite high (i.e., a few millimeters) for the surface
wearing course but will degrade to centimeters at depths
of 1m. An example is shown in Fig. 9.
While most GPR systems are used in close proximity to
the ground, airborne systems have been able to map ice
formations and glaciers, and penetrate through forest can-
opy. Airborne GPR, processed using synthetic aperture
techniques, has been used to detect buried metallic mines
from a height of several hundred meters. In addition, the
SIR-C satellite SAR radar has imaged buried artefacts in
desert conditions, and the JPL Website http://southport.
Figure 10. GPR system in use for archaeological surveying in
Saqqara, Eygpt. (Courtesy of ERA Technology.)
Grayel ridge (man-made)
Natural desert
floor
Sand filled
ditch North face of
South wall?
Section across the south wall (Line 62)
Figure 11. Radar cross section through buried wall (horizontal
15m; vertical 3.5m). (Courtesy of ERA Technology.)
jpl.nasa.gov/sir-c/ is an important source of radar
imagery.
A list of the applications is given below, and the reader
can nd further information from the literature and pub-
lications referenced:
Archaeological investigations
Bridge deck analysis
Borehole inspection
Building condition assessment
Contaminated-land investigation
Detection of buried mines (antipersonnel and antitank)
Evaluation of reinforced concrete
Forensic investigations
Geophysical investigations
Medical imaging
Pipes and cable detection
Planetary exploration
Rail track and bed inspection
Remote sensing from aircraft and satellites
Road condition survey
Security applications
Snow, ice, and glaciers
Timber condition
Tunnel linings
Wall condition
Readers are encouraged to review the applications de-
scribed in the following sources.
8.2. GPR Conferences
GPR 2000Gold Coast, Australia 8th Int. Conf.
Ground Penetrating Radar, David Noon, University
of Queensland; email noon@cssip.uq.edu.au.
GPR 98Lawrence, Kansas, USA 7th Int. Conf.
Ground Penetrating Radar Dr. Richard Plumb,
Univ. Kansas; email rplumb@binghamton.edu.
GPR 96Sendai, Japan 6th Int. Conf. Ground Pene-
trating Radar, Prof. Motoyuki Sato, Tohoku Univ.
email sato@cneas.tohoku.ac.jp.
GPR 94Kitchener, Ontario Canada 5th Int. Conf.
Ground Penetrating Radar, David Redman, Sensors
& Software; email dr@sensoft.ca.
GPR 92Rovaniemi, Finland 4th Int. Conf. Ground
Penetrating Radar, Pauli Hanninen, Geological Sur-
vey of Finland.
GPR 90Lakewood, Colorado (USA) 3rd Int. Conf.
Ground Penetrating Radar, Prof. Gary Olhoeft, Col-
orado School of Mines; email golhoeft@mines.edu.
1988Gainesville, Florida (USA) 2nd Int. Symp. Geo-
technical Applications of Ground Penetrating Radar,
Mary Collins, Univ. Florida; email mec@gnv.ifas.
u.edu.
1986Tifton, Georgia (USA) 1st Int. Conf. Geotechni-
cal Applications of Ground Penetrating Radar.
8.3. International Workshops on Advanced GPR
IWA GPR Delft 01, 1st Int. Workshop on Advanced
Ground Penetrating Radar (International Work-
shop), proceedings published in Subsurface Sensing
Technologies and Applications (Kluwer Academic
Publishers).
IWA GPR Delft 03, 2nd Int. Workshop on Advanced
Ground Penetrating Radar (International Work-
shop), Weblink http://irctr.et.tudelft.nl/IWAGPR/.
8.4. Institution of Electrical Engineers (UK)
See also Radar 2002; http://www.iee.org/Publish/
Digests/conf2002.cfm.
See the IEE Proceedings of Radar, Sonar, and Naviga-
tion; Weblink http://ioj.iee.org.uk/journals/ip-rsn.
See Edinburgh MD96, Detection of Abandoned Land-
mines (Main Past Conf.); Weblink http://
www.iee.org/Publish/Digests/conf1996.cfm.
See Edinburgh MD98, 2nd Int. Conf. Detection of
Abandoned Land mines (Main Past Conf.); Weblink
http://www.iee.org/Publish/Digests/con-
f1998.cfm.
8.5. Institution of Electrical and Electronic Engineers (USA)
See proceedings particularly of the societies for
Antennas and Propagation; http://www.ieeeaps.org/.
Aerospace and Electronic Systems; http://
ewh.ieee.org/soc/aes/.
For Radar conferences, see also http://www.
ewh.ieee.org/soc/aes/Conferences.html.
Geoscience and Remote Sensing; http://www.
ewh.ieee.org/soc/grss/.
Microwave Theory and Transactions; http://
www.mtt.org/.
Figure 12. Radar plan image of buried AT
mines (4m swathe width; 10m survey
length). (Courtesy of ERA Technology.) (This
gure is available in full color at http://
www.mrw.interscience.wiley.com/erfme.)
8.6. SPIE Conferences
SPIE OrlandoSPIE Detection and Remediation Tech-
nologies for Mines and Minelike Targets 1995 to 2003;
http://spie.org/app/conferences/index.cfm?fuseaction
archive & year 2002.
8.7. Geophysics
http://www.geo-online.org/.
8.8. Subsurface Sensing Technologies and Applications
http://www.kluweronline.com/issn/1566-0184.
8.9. Literature on GPR Applications
D. J. Daniels, Ground Penetrating Radar, IEE Radar
Sonar Navigation and Avionics Series, Vol. 6, 1996.
M. Skolnik, Introduction to Radar Systems, 2nd ed.,
McGraw-Hill, 1981.
S. Cloude, Introduction to Electromagnetic Wave Prop-
agation and Antennas, UCL Press, 1995.
J. Hunter, C. Roberts, and A. Martin, Studies in Crime:
An Introduction to Forensic Archaeology, B. T. Batsford
Ltd. 1996.
9. REGULATION
All countries require that GPR systems be properly regu-
lated and operated in accordance with national and inter-
national requirements. Users should consult with their
national authorities to determine the regulatory environ-
ment.
Within the European Union (EU) there are two main
considerations governing the use of ground penetrating
radar (GPR): (1) the use of the equipment as a deliberate
radiofrequency radiator and (2) usage as equipment that
must satisfy the EMC requirements of the EU. The ETSI
regulatory body is in the process of drafting specications,
and information can be found at http://www.etsi.org that
will cover the use of such equipment as a deliberate ra-
diofrequency radiator. Planned legislation and an ETSI
product specication at some undened time in the future
means that this equipment will eventually need to con-
form to the R&TTE directive. In the short term until a
new product specication is introduced and formally pub-
lished in the Ofcial Journal of the European Communi-
ties, the EMC directive should be applied. All equipment,
including ultrawideband radar or ground probing radar,
must be CE marked to demonstrate that it satises the
relevant directives of the European Union. The CE mark
may be applied only when the requirements of all other
relevant EU directives, such as safety, have also been
demonstrated.
In the United States the FCC Website http://
www.fcc.gov/aboutus.html provided the following infor-
mation in February 2002:
Feb 2002 Washington, D.C.The Federal Communication
Commission (FCC) adopted today a First Report and Order
that permits the marketing and operation of certain types
of new products incorporating ultra-wideband (UWB)
technology. UWB technology holds great promise for a vast
array of new applications that have the potential to provide
significant benets for public safety, businesses and consum-
ers in a variety of applications such as radar imaging of objects
buried under the ground or behind walls and short-range,
high-speed data transmissions.
UWB devices operate by employing very narrow or short du-
ration pulses that result in very large or wideband transmis-
sion bandwidths. With appropriate technical standards, UWB
devices can operate using spectrum occupied by existing radio
services without causing interference, thereby permitting
scarce spectrum resources to be used more efciently. This
First Report and Order (Order) includes standards designed
to ensure that existing and planned radio services, particular-
ly safety services, are adequately protected. The FCC will act
vigorously to enforce the rules and act quickly on any reports
of interference.
The standards adopted today represent a cautious rst step
with UWB technology. These standards are based in large
measure on standards that the National Telecommunications
and Information Administration (NTIA) believes are neces-
sary to protect against interference to vital federal govern-
ment operations. Since there is no production UWB equipment
available and there is little operational experience with the
impact of UWB on other radio services, the Commission chose
in this First Report and Order to err on the side of conserva-
tism in setting emission limits when there were unresolved
interference issues. The Commission intends within the next
six to twelve months to review the standards for UWB devices
and issue a further notice of proposed rule making to explore
more exible standards and address the operation of addition-
al types of UWB operations and technology.
Ground Penetrating Radar Systems: GPRs must be operated
below 960MHz or in the frequency band 3.110.6GHz. GPRs
operate only when in contact with or within close proximity of,
the ground for the purpose of detecting or obtaining the im-
ages of buried objects. The energy from the GPR is intention-
ally directed down into the ground for this purpose. Operation
is restricted to law enforcement, re and rescue organizations,
to scientific research institutions, to commercial mining com-
panies, and to construction companies.
It is understood that waivers have been granted to cer-
tain operators and the FCC are in the process of reviewing
the policy announced in 2002. In February 2003 the fol-
lowing announcement was made:
In response to the petitions, the Commission amended the
rules to facilitate the operation of through-wall imaging sys-
tems by law enforcement, emergency rescue and reghter
personnel in emergency situations; eliminated the require-
ment that GPRs and wall imaging systems operate with their
10dB bandwidths below 960MHz or above 3.1 GHz; speci-
ed the limitations on who may operate ground penetrating
radar (GPR) systems and wall imaging systems and for what
purposes; eliminated the requirement for non-hand held
GPRs to employ a dead man switch; claried the coordination
requirements for imaging devices; and claried the rules re-
garding emissions produced by digital circuitry used by UWB
transmitters.
For more information, see http://www.g-p-r.com/.
10. SUMMARY
Advances in the performance and design of lightweight
and compact GPR radar sensors have been significant
since the mid-1990s because of the impact of the telecom-
munications industry and availability of affordable high-
power computing; the volume of published papers and
conferences, some devoted entirely to GPR, has risen con-
siderably since then. Inevitably there have been some
claims for GPR capability that are simply outside the
realms of known physics and unfortunately have been ac-
cepted by some sections of the media. Tuley et al. [14] give
an excellent review of one such claim. There has also been
some duplication of basic research. However, develop-
ments in the eld of noise and pseudorandom coded [15]
modulation techniques together with better signal and
image processing algorithms as well as spinoff from im-
proved microwave technology originating from the tele-
communications industry will have a significant impact on
GPR in the future.
It is reasonable to conclude that GPR is now an estab-
lished branch of science. It has links to electronic engi-
neering, mathematics, signal processing, geophysics, civil
engineering, structural engineering, archaeology, forensic
science planetary exploration, as well as unexploded or-
dance (UXO) clearance and mining. GPR has achieved ac-
ademic recognition as well as commercial impact and
seems set to continue its impact into all branches of
human activity where buried objects are sought.
BIBLIOGRAPHY
1. D. J. Daniels, Ground Penetrating Radar, 2
nd
ed., IEE Radar
Sonar Navigation and Avionics Series, London, 2004.
2. C. E. Cook and M. Bernfeld, Radar Signals, An Introduction
to Theory and Application, Artech House, Norwood, MA,
1993, p. 9.
3. M. Skolnik, Radar Handbook, 2
nd
ed., McGraw-Hill, New
York, 1985, Chap. 10.
4. F. Nathanson, Radar Design Principles, SciTech Publishing,
1969, Chap. 8, pp. 276318.
5. D. R. Wehner, High Resolution Radar, Artech House, Nor-
wood, MA, Chap. 4.
6. G. Galati, Advanced Radar Techniques and Systems, IEE
Radar Sonar Navigation and Avionics Series, London, 1994,
Vol. 4, p. 104.
7. L. Y. Astanin and A. A. Kostylev, Ultra-Wideband Radar
Measurements Systems, IEE Radar Sonar Navigation and
Avionics Series, London, 1997, Vol. 7, Chap. 1.
8. P. Debye, Polar Molecules, Chemical Catalog Co., New York,
1929.
9. K. S. Cole and R. S. Cole, Dispersion and absorption in di-
electrics, I, alternating current characteristics, J. Phys. Chem.
9: 341351 (1941).
10. D. J. Daniels, Resolution of UWB signals, IEE Proc. Radar
Sonar and Navig. 146:189194 (Aug. 1999).
11. P. Dennis and S. E. Gibbs, Solid-state linear FMCW
systemstheir promise and their problems, Proc. IEEE Int.
Microwave Conf., Atlanta, 1974, pp. 340345.
12. R. M. Narayanan, Y. Xu, P. D. Hoffmeyer, and J. O. Curtis,
Design, performance, and applications of a coherent random
noise radar, Opt. Eng. 37(6):18551869 (June 1998).
13. J. Sachs, P. Peyerl, F. Tkac, and M. Kmec, Digital ultra-wide-
band-sensor electronics integrated in SiGe-technology, Proc.
EuMC, Milan (Italy), Sept. 2002, Vol. II, pp. 539542.
14. M. T. Tuley, J. M. Ralston, F. S. Rotondo, A. M. Andrews, and
E. M. Rosen, Evaluation of EarthRadar unexploded ordnance
testing at Fort A.P. Hill, Virginia, IEEE Aerospace Electron.
Syst. Mag. 17(5):1012 (May 2002).
15. J. Sachs and P. Peyerl, Chip integrated UWB radar electron-
ics, Proc. 3rd DTIF Workshop on Ground Penetrating Radar
in Support of Humanitarian Demining, JRC, Ispra (Italy),
Sept. 2002.
GUIDED ELECTROMAGNETIC WAVES
WILLIAM T. JOINES
Duke University
Closed-form solutions for TEM, TM, and TE waves in
guiding structures are obtained from Maxwells equations.
Propagation parameters, frequency ranges, broadbanding
techniques, and design examples are presented. Dielectric
waveguides for microwave and optical frequencies are also
presented.
1. MAXWELLS EQUATIONS
Maxwells equations in differential form, as determined
from Faradays, Amperes, and Gauss laws, respectively,
are
rE
@B
@t
1
rHJ
@D
@t
2
r
.
Dr 3
r
.
B0 4
In these equations, E is the electric eld intensity in V/m,
H is the magnetic eld intensity in A/m, DeE is the
electric ux density in C/m
2
, e is the electric permittivity
in F/m, JsE is the conduction current density in A/m
2
,
s is the electrical conductivity in S/m, r is electric charge
density in C/m
3
, BmH is the magnetic ux density
in Wb/m
2
(or tesla) and m is the magnetic permeability in
H/m.
2. PROPAGATING SINE WAVES AND THE
WAVE EQUATION
For signals oscillating sinusoidally in time it is convenient
to represent the time variation as e
jot
. Thus, the electro-
1846 GUIDED ELECTROMAGNETIC WAVES
magnetic elds are expressed as Ex; y; z; t Ex; y; ze
jot
,
and Hx; y; z; t Hx; y; ze
jot
. Substituting these expres-
sions into the differential form of Maxwells equations giv-
en in (1)(4) yields the sinusoidal steady-state form as
rE jomH 5
rHs joeE 6
r
.
Dr 7
r
.
B0 8
where the last two equations are unchanged since no time
derivative is involved.
A wave equation for E and H is derived by taking the
curl of both sides of (5) and (6), and then using the vector
identity
rrErr
.
E r
2
E 9
Thus, from (5), (6), and (9), we obtain
rrE jomrH 10
and
rr
.
E r
2
E joms joeE 11
If in the region of interest, r 0, and eOex; y; z, then
from (7), r
.
E0, and (11) takes a convenient form of the
wave equation as
r
2
Ejoms joeEg
2
E 12
A similar procedure, starting with the curl of both sides of
(6), yields an identical equation involving H as
r
2
Hjoms joeHg
2
H 13
In (12) and (13), we obtain
g
joms joe
_
a jb 14
where g is the propagation constant, a is the attenuation
constant, and b is the phase constant, all expressed per
unit length.
Solutions to the wave equations for E and H may be
obtained most conveniently by separation of variables, as
shown in the next section.
3. SOLUTIONS FOR TEM, TM, AND TE WAVES [1,2]
Allowing all polarizations of E in rectangular coordinates,
(12) may be expressed as
r
2
E
x
^ xx E
y
^ yy E
z
^ zz g
2
E
x
^ xx E
y
^ yy E
z
^ zz 15
where ^ xx; ^ yy, and ^ zz are unit vectors. Thus, a typical term is
r
2
E
z
g
2
E
z
16
Substituting the separated-variable solution E
z
(x, y, z)
X(x)Y(y)Z(z) into (16), and dividing by XYZ, produces the
result
1
X
d
2
X
dx
2

1
Y
d
2
Y
dy
2

1
Z
d
2
Z
dz
2
g
2
k
2
x
k
2
y
k
2
z
17
From (17), it will be convenient to let
k
2
z
g
2
k
2
x
k
2
y
G
2
18
or
G
g
2
k
2
x
k
2
y
_
19
where the propagation constants k
x
, k
y
, and G are to be
determined.
Equating terms in (17) yields three ordinary differen-
tial equations that have the implied solutions indicated in
the following:
d
2
X
dx
2
k
2
x
X 0 !Xx A
1
cos k
x
x B
1
sin k
x
x 20
d
2
Y
dy
2
k
2
y
Y 0 !Yy A
2
cos k
y
y B
2
sin k
y
y 21
d
2
Z
dz
2
G
2
Z0 !Zz A
3
e
Gz
B
3
e
Gz
22
We express a sinusoidally time-varying, z-polarized wave
propagating in the z direction as
E
z
x; y; z; t E
z
x; y; ze
jot
XxYye
jotGz
23
and all the other components of E and H are expressed in a
similar form.
In (23), in order to produce the term e
jotjGjz
necessary
for wave propagation to occur, G must be imaginary.
Taking the dielectric material through which the wave
propagates to be perfect (s 0), then g
2
joms joe
o
2
me. The propagation condition is
G
g
2
k
2
x
k
2
y
_
j
o
2
me k
2
x
k
2
y
_
jb
g
24
and wave propagation occurs if b
g
(the phase constant of
the wave-guiding structure) is real. At b
g
0, no propa-
gation occurs, which denes a cutoff frequency
f
c
o
c
2p

1
2p

me
p
k
2
x
k
2
y
_
25
GUIDED ELECTROMAGNETIC WAVES 1847
Substituting (25) into (24) yields
Gj
o
2
me o
2
c
me
_
jo

me
p
1
f
c
f
_ _
2
jb
g
26
These parameters will be examined further for different
wave-guiding structures.
All of the eld solutions, expressed in the form XYe
Gz
,
satisfy the wave equation, and the solutions must also
satisfy the boundary conditions imposed by the guiding
structure. For guided waves, the basic wave types are
transverse electromagnetic (TEM) waves, transverse mag-
netic (TM) waves, and transverse electric (TE) waves. For
propagation guided in the z direction, a TEM wave has
components of E and H that are perpendicular (or trans-
verse) to the z axis, and thus by definition, E
z
H
z
0. A
TM wave has components of H perpendicular to the z axis,
but by definition, H
z
0. A TE wave has components of E
perpendicular to the z axis, but by definition, E
z
0. These
wave types and polarizations are illustrated in Fig. 1,
where the TE and TM waves are guided by reecting rays
at specific angles (y) between the parallel metal plates at
x 0 and x a.
The guiding structures may take on a variety of forms,
but two important ones are diagramed in Fig. 2: Fig. 2a is
a parallel-plate transmission line with plate separation a
and width w; Fig. 2b is a rectangular waveguide with in-
side width a and height b. Within the parallel-plate line
depicted in Figs. 1 and 2a, a TEM wave can propagate at
all frequencies, and TM and TE waves can propagate at all
frequencies above a certain cutoff frequency (f
c
) that will
be determined later. Within the closed structure of the
rectangular waveguide, TM and TE waves can propagate
at all frequencies above their cutoff frequencies (f
c
differs
for TM and TE waves in the closed waveguide), but TEM
waves cannot exist within the closed waveguide.
3.1. Transverse Electromagnetic (TEM) Waves
To show that TEM waves cannot exist, and therefore prop-
agate, within the rectangular waveguide, proceed to solve
for a propagating wave as if one may exist. Assume solu-
tions of the form XYe
Gz
for E
x
, E
y
, H
x
, and H
y
, take E
z
H
z
0 by definition, and substitute the assumed solutions
into Maxwells curl equations. First, rE jomH
yields
0GE
y
jomH
x
27
GE
x
0 jomH
y
28
@E
y
@x

@E
x
@y
0 29
and rHs joeE yields
0 GH
y
sjoe E
x
30
GH
x
0 sjoe E
y
31
@H
y
@x

@H
x
@y
0 32
where the common term e
Gz
has been canceled in each
equation.
From (27), (28), (30), and (31), we obtain
E
y
H
x
E
x
H
y
jom
G

G
sjoe
33
Therefore, G
2
joms joe g
2
, and k
x
k
y
0, which
allows no variation of E and H in the xy plane. Thus,
boundary conditions cannot be satised within the closed
structure, and TEM waves cannot exist.
Since G
2
g
2
for TEM waves, the wave equations for E
x
and H
y
within the parallel-plate transmission line are
d
2
E
x
dz
2
g
2
E
x
and
d
2
H
y
dz
2
g
2
H
y
34
If wba for the parallel-plate line, then E
x
and H
y
repre-
sent uniform elds between the plates. By applying Fara-
days law between the conducting plates, and Amperes
law around the plates, the elds are related to the voltage
(a)
x
a
a
a
z
y
E
P=EH
(c)
x
z
y
H
E
P
P

(b)
x
z
y
H
P
P
E

H
Figure 1. Waves guided between parallel metal plates spaced
a distance a apart: (a) TEM wavesEE
x
^ xx; HH
y
^ yy; E
z
0;
(b) TM wavesE E
x
^ xx E
z
^ zz; H H
y
^ yy; H
z
0; (c) TE
wavesEE
y
^ yy; H H
x
^ xx H
z
^ zz; E
z
0.
x
y
a
b
z
x
y
w
a
(a) (b)
z
Figure 2. Two important structures for guiding waves: (a) par-
allel-plate line of width w and spacing a; (b) closed rectangular
waveguide of inside width a and height b.
and current at any point z by
E
x
Vz
a
and H
y
Iz
w
35
Thus, V(z) and I(z) also satisfy the wave equation, and for
waves propagating in the z direction they have the so-
lutions
Vz Ae
gz
and Iz
A
Z
0
e
gz
36
where A is a constant and Z
0
is the characteristic imped-
ance of the transmission line. As in (33), since E
x
/H
y
Z,
the intrinsic impedance of the material between the
plates, it is easily determined from foregoing relations
that
Z
0
Vz
Iz

E
x
a
H
y
w

a
w
Z
a
w
jom
s joe
37
for the parallel-plate line considered here.
3.2. Transverse Magnetic (TM) Waves
Returning to the rectangular waveguide of width a and
height b, since H
z
0 by definition for TM waves, select
E
z
XYe
Gz
as a solution that cannot vanish if a solution
does exist. Assuming solutions of the form XYe
Gz
for E
x
,
E
y
, H
x
, and H
y
, substitute the assumed solutions into Max-
wells curl equations. First, rE jomH yields
@E
z
@y
GE
y
jomH
x
38
GE
x
@E
z
@x
jomH
y
39
@E
y
@x

@E
x
@y
0 40
and rHsjoeE yields
0GH
y
sjoe E
x
41
GH
x
0 s joe E
y
42
@H
y
@x

@H
x
@y
sjoe E
z
43
The common term e
Gz
has been canceled in each equa-
tion.
From (41) and (42)
E
y
H
x
E
x
H
y
G
s joe
Z
0
44
which denes Z
0
as the characteristic impedance of TM
waves in the rectangular waveguide. Using this definition
of Z
0
, and the curl equations, all the eld components may
be expressed in terms of the one-component solution E
z
.
Substituting (42) into (38) and using (44), we obtain
H
x
@E
z
=@y
GZ
0
jom
45
Substituting (41) into (39) and using (44), we obtain
H
y
@E
z
=@x
GZ
0
jom
46
The remaining components are
E
x
Z
0
H
y
; E
y
Z
0
H
x
; and H
z
0 47
Again, the assumed solution for E
z
is
E
z
XYe
Gz
A
1
cos k
x
x B
1
sin k
x
x
A
2
cos k
y
y B
2
sin k
y
y e
Gz
48
and the boundary conditions are that E
z
0 on each wall
of the waveguide. Letting E
z
(0, y) 0 and E
z
(x, 0) 0
requires that A
1
0 and A
2
0. The remaining sinu-
soidal terms will satisfy the conditions E
z
(a, y) 0 and
E
z
(x, b) 0 if
k
x
mp
a
and k
y
np
b
49
where m and n are integers, including zero. Thus, the
solution for E
z
is
E
z
x; y; z; t E
mn
sin
mpx
a
sin
npy
b
e
Gz
e
jot
50
Depending on the choice of integers m and n, different
cross-sectional eld patterns may occur in the xy plane.
These different modes are designated by TM
mn
. Observe
from (50) that the TM
23
mode (for example) has two half-
sine-wave variations in the x direction and three half-sine-
wave variations in the y direction. The lowest-order mode
that can propagate is the TM
11
, since the TM
10
and the
TM
01
modes render E
z
0 in (50).
In (50), as stated earlier, wave propagation occurs if G is
imaginary. Taking the dielectric material (usually air) ll-
ing the waveguide to be perfect (s 0), then g
2
joms joe o
2
me, and
G
g
2
k
2
x
k
2
y
_
j
o
2
me
mp
a
_ _
2
np
b
_ _
2
_
jb
g
51
Thus, wave propagation occurs at frequencies above the
cutoff frequency
f
c
o
c
2p

1
2p

me
p
mp
a
_ _
2
np
b
_ _
2
_
v
m
2a
_ _
2
n
2b
_ _
2
_
52
The cutoff frequency of the TM
11
(lowest-order) mode is
f
c
v
2a
1
a
b
_ _
2
_
53
Note that if the dimension b approaches innity, and the
parallel sides remain a distance a apart, the cutoff fre-
quency for all TM modes in (52) becomes
f
c
mv
2a
54
3.3. Transverse Electric (TE) Waves
For TE waves within the rectangular waveguide of width a
and height b, since E
z
0 by definition, select H
z
XYe
Gz
as a solution that cannot vanish if a solution does exist.
Assuming solutions of the form XYe
Gz
for E
x
, E
y
, H
x
, and
H
y
, substitute the assumed solutions into Maxwells curl
equations. First, rE jomH yields
0 GE
y
jomH
x
55
GE
x
0 jomH
y
56
@E
y
@x

@E
x
@y
jomH
z
57
and rHs joeE yields (assuming s0 for a perfect
dielectric medium)
@H
z
@y
GH
y
joeE
x
58
GH
x
@H
z
@x
joeE
y
59
@H
y
@x

@H
x
@y
0 60
The common term e
Gz
has been canceled in each equa-
tion.
The characteristic impedance of TE waves is dened
from (55) and (56) as
E
y
H
x
E
x
H
y
jom
G
Z
0
61
Using this definition of Z
0
in (58) and (59) yields
@H
z
@y
joeE
x
GE
x
Z
0
joe
G
Z
0
_ _
E
x
62
@H
z
@x
joeE
y
GE
y
Z
0
joe
G
Z
0
_ _
E
y
63
For the rectangular waveguide under consideration, E
x
0
at y 0 and y b, and E
y
0 at x 0 and x a. Thus, from
(62) and (63), we obtain
@H
z
@y
0 at y 0 and y b 64
and
@H
z
@x
0 at x 0 and x a 65
are appropriate boundary conditions since H
z
cannot van-
ish if a solution exists.
Applying the boundary conditions to the assumed
solution
H
z
XYe
Gz
A
1
cos k
x
x B
1
sin k
x
x
A
2
cos k
y
y B
2
sin k
y
ye
Gz
66
yields
H
z
H
mn
cos
mpx
a
cos
npy
b
e
Gz
67
where k
x
mp=a, k
y
np=b, and m and n are integers cor-
responding to the mode variations in the xy plane.
The remaining components of the TE waves are ob-
tained as follows: From (62), we obtain
E
x
@H
z
@y
joe
G
Z
0
k
x
H
mn
cos k
x
x sin k
y
ye
Gz
joe
G
Z
0
68
From (63)
E
y
@H
z
@x
joe
G
Z
0
k
x
H
mn
sin k
x
x cos k
y
ye
Gz
joe
G
Z
0
69
From (61)
H
x

E
y
Z
0
70
and
H
y
E
x
Z
0
71
If a4b for the rectangular waveguide, the mode of lowest
order is the TE
10
, which has the cutoff frequency f
c
v/2a
and the eld components
H
z
H
10
cos
px
a
e
Gz
72
E
y
jZ
2a
l
H
10
sin
px
a
e
Gz
E
10
sin
px
a
e
Gz
73
H
x

E
y
Z
0
74
E
x
H
y
E
z
0 75
The cutoff frequency of the TE
10
mode is
f
c
v
2a
m
2
n
2
a
b
_ _
2
_

v
2a
76
Again, TE and TM waves share the same general ex-
pressions for propagation constant (G) and cutoff frequen-
cy (f
c
), as expressed in Eqs. (24), (25), (51), and (52). For
the two wave types, G and f
c
differ in the choice of the in-
teger mode numbers m and n. However, if the dimension b
goes to innity, creating a parallel-plate waveguide, then
G and f
c
are the same for both TE and TM waves. The
characteristic impedance differs for the two wave types, as
Z
0
G=joe Z
1 f
c
=f
2
_
for TM waves, and Z
0
jom=GZ=
1 f
c
=f
2
_
for TE waves, where Z
m=e
_
is
the intrinsic impedance of the medium lling the guide.
Thus, as frequency is increased from f
c
to 1, Z
0
increases
from 0 up to Z for TM waves and decreases from1down to
Z for TE waves.
For f of
c
, G is real, and the propagation term e
jotGz
from the wave equation becomes e
jota
g
z
, or
Go

me
p

f
c
=f
2
1
_
a
g
77
At operating frequencies below cutoff for the waveguide,
the waves are attenuated by the factor a
g
as they evanesce
or diffuse along the guide in the z direction with no real
wave propagation occurring. A waveguide below cutoff
may be used as an attenuator. For example, if f 10 GHz
and f
c
13.12 GHz (corresponding to a0.45 in. or
11.43 mm in an air-lled, TE
10
waveguide), then
a
g
1:78 Np/cm, or 15.46 dB/cm. Thus, the length is ad-
justed to yield a desired attenuation.
For f of
c
, the characteristic impedances are imaginary,
as Z
0
G=joe jZ
f
c
=f
2
1
_
for TM waves, and
Z
0
jom=GjZ=
f
c
=f
2
1
_
for TE waves. This puts the
transverse components of E and H 901 out of time phase,
and tells us once again that time-average power cannot be
transferred as a propagating wave. Thus, energy transfer
along the guide is by evanescence or diffusion.
4. INTERPRETING Z
0
FOR TEM, TM, AND TE WAVES
Taking the propagation medium to be a lossless dielectric
(s 0), the intrinsic impedance of the medium is Z
m=e
_
.
For TEM waves propagating in the z direction, the char-
acteristic impedance is
Z
0
Vz
Iz
ZFg
E
x
H
y
Fg 78
where, as in Fig. 2a, the geometry function F(g) a/w for
two parallel conducting strips of width w, separation a,
and wba; and Fg 1=2p lnb=a for a coaxial line, if a
is the inner conductor radius and b is the inner radius of
the outer conductor.
For TM waves guided in the z direction, the character-
istic impedance is
Z
0
G
joe

jo

me
p
joe
1
f
c
f
_ _
2
1
f
c
f
_ _
2
79
From a raypath diagram of the guided TM wave, as in
Fig. 1b, Z
0
may also be determined as
Z
0
E
x
H
y
Esin y
H
Z sin y Z
1 cos
2
y
p
80
But, b
z
b sin y and b
x
bcos y, and Z
0
becomes
Z
0
Z
l
l
z
Z
1
l
l
x
_ _
2
1
f
c
f
_ _
2
81
where l
z
l
g
is the wavelength along the direction of
propagation or the wavelength along the guide, and
l
x
l
c
is the cutoff wavelength.
For TE waves guided in the z direction, the character-
istic impedance is
Z
0
jom
G

jom
jo

me
p
1
f
c
f
_ _
2

Z
1
f
c
f
_ _
2
82
From a raypath diagram of the guided TE wave, as in Fig.
1c, Z
0
may also be determined as
Z
0

E
y
H
x
E
Hsin y

Z
sin y

Z
1 cos
2
y
p 83
Again, b
z
b sin y and b
x
b cos y, and Z
0
for TE waves
becomes
Z
0
Z
l
z
l

Z
1
l
l
x
_ _
2

Z
1
f
c
f
_ _
2
84
and again, l
z
l
g
is the wavelength along the guide, and
l
x
l
c
is the cutoff wavelength.
For convenient reference, selected equations, proper-
ties and parameters of waveguiding structures are collect-
ed in Table 1 and 2.
5. FREQUENCY BANDWIDTH OF WAVEGUIDES
As discussed in earlier sections, waveguides operate at
frequencies above a specific cutoff frequency, and they may
be subject to higher-order modes (higher than the funda-
Table 1. Comparison Equations and Properties of TM, TE, and TEM Waves
TM Waves TE Waves TEM Waves
H
z
0 E
z
0 E
z
H
z
0
G
o
2
me k
2
x
k
2
y
_
G
o
2
me k
2
x
k
2
y
_
g
o
2
me
_
jo

me
p
f
c
v
2p
k
2
x
k
2
y
_
f
c
v
2p
k
2
x
k
2
y
_
f
c
0
l
c
v
f
c
2p
k
2
x
k
2
y
_ l
c
v
f
c
2p
k
2
x
k
2
y
_
l
c
N
b
g
o
2
me k
2
x
k
2
y
_
b
g
o
2
me k
2
x
k
2
y
_
b o

me
p
b
g
2p
l
1
l
l
c
_ _
2
b
g
2p
l
1
l
l
c
_ _
2
b
2p
l
l
g
2p
b
g
1 l=l
c
2
_ l
g
2p
b
g
1 l=l
c
2
_
l
Z
0
G
joe
Z
1 l=l
c
2
_
Z
0
jom
G

Z
1 l=l
c
2
_
Z
v
g
o
b
g
v
g
o
b
g
v
o
b
Zz Z
0
Z
L
jZ
0
tan b
g
z
Z
0
jZ
L
tan b
g
z
Zz Z
0
Z
L
jZ
0
tan b
g
z
Z
0
jZ
L
tan b
g
z
Zz Z
0
Z
L
jZ
0
tan bz
Z
0
jZ
L
tan bz
Table 2. Features of Some Commercially Available Rectangular Waveguides
EIA Inside Dimensions Cutoff Range for CW Power
Designation ab (in.) f
c
(MHz) TE
10
(MHz) Rating (MW)
WR1150 11.55.75 513 640960 35.0053.80
WR975 9.754.875 605 7501120 27.0038.50
WR770 7.703.85 766 9601450 17.2024.10
WR650 6.503.25 908 11201700 11.9017.20
WR510 5.102.55 1157 14502200 7.5010.70
WR430 4.302.15 1372 17002600 5.207.50
WR340 3.401.70 1735 22003300 3.404.71
WR284 2.841.34 2077 26003950 2.183.10
WR229 2.291.145 2590 33004900 1.562.14
WR187 1.870.87 3155 39505850 0.941.32
WR159 1.590.76 3710 49007050 0.750.98
WR137 1.370.62 4307 58508200 0.550.70
WR112 1.120.50 5260 705010000 0.360.45
WR90 0.900.40 6560 820012400 0.210.29
WR75 0.750.37 7867 983314950 0.190.27
WR62 0.620.31 9516 1190018080 0.160.22
WR42 0.420.17 14047 1756026690
WR28 0.280.14 21071 2634040035
WR19 0.1880.094 31383 3923059630
WR15 0.1480.074 39865 4983075740
WR10 0.1000.050 59000 73750112100
WR8 0.0800.040 73750 92200140125
WR6 0.0650.0325 90769 113460172460
WR5 0.0510.0255 115686 144600220000
mental mode). Single-mode propagation is necessary for
efcient transmission and detection of signal information
on the carrier wave, and, of course, more information may
be sent if the bandwidth is as large as possible. These re-
strictions may be troublesome at times, but there are ways
to avoid most of the trouble.
5.1. Lowering the Cutoff Frequency of Waveguides
The bandwidth of a rectangular waveguide operating in
the TE
10
mode is generally taken to be 1:25f
c
f 1:90f
c
,
where f
c
v=2a is the TE
10
-mode cutoff frequency. This is
the choice because f
1
1:25f
c
is a convenient distance
above f f
c
(where Z
0
1) and f
2
1.90f
c
is convenient-
ly less than f 2f
c
(where the rst higher-order mode,
TE
02
, occurs if 2boa. Thus, the operating bandwidth is
f
2
f
1
0.65f
c
.
To increase the bandwidth, we rst lower the cutoff
frequency of the TE
10
mode waveguide by increasing the
wide dimension a in f
c
v/2a, as in Fig. 3a. To accommo-
date a large increase in the width a, the waveguide
is folded such that the width is now 2b and the height is
(a2s)/2, as in Fig. 3b.
In Fig. 3, the dotted path a
0
is longer than a because of
the fringing capacitance across the gap s. The fringing
capacitance is [3]
C
f
e
4
p
ln coth
ps
2b
_ _
85
which is equivalent to a length increase given by
C
f
e
2Da
b
86
Thus
a
0
a2Daa
4b
p
ln coth
ps
2b
_ _
87
In (87), if s 0.2b and b 0.5a, then a
0
1.715a, and the
resulting lower cutoff frequency for the TE
10
mode is
f
0
c
v=2a
0
f
c
=1:7150:583f
c
. The lower operating fre-
quency is f
0
1
1:25f
0
c
0:729f
c
. The upper operating fre-
quency remains the same at f
0
2
f
2
1:90f
c
, since the
electric eld of the TE
20
mode is zero across the gap s
and little or no fringing capacitance occurs to increase the
pathlength. Thus, the new operating bandwidth is
f
0
2
f
0
1
1:90 0:729f
c
1:171f
c
, rather than 0.65f
c
.
Example 1. Given a12.5cm, b 0.5a6.25cm, and s
0.2b 1.25 cm, determine the operating frequencies and
bandwidth for exclusive TE
10
mode operation in (a) the
rectangular waveguide and (b) the folded rectangular
waveguide in Fig. 3b.
Solution: (a) For the rectangular waveguide
f
c
v
2a

310
8
0:25
1200 MHz 88
f
1
1:25f
c
1500 MHz 89
f
2
1:90f
c
2280 MHz 90
BWf
2
f
1
780 MHz 91
(b) For the folded rectangular waveguide in Fig. 3b, we
obtain
f
0
c
v
2a
0
0:583f
c
700 MHz 92
f
0
1
1:25f
0
c
875 MHz 93
f
0
2
f
2
1:90f
c
2280 MHz 94
BWf
0
2
f
0
1
1405 MHz 95
The folded waveguide considered here is more commonly
known as a single-ridged waveguide with a ridge thick-
ness approaching zero (t 0). A double-ridged version
using the same conguration (same operating frequencies
and bandwidth as in Fig. 3b) is shown in Fig. 3c. More
commonly used relative dimensions are shown in Figs. 4a
and 4b for single-ridged and double-ridged waveguides,
respectively [4]. Another variation on the same theme is
the T-septum waveguide, single and double, as shown in
Figs. 4c and 4d [57].
6. DIELECTRIC WAVEGUIDES
An electromagnetic wave is guided by a dielectric material
[8] of permittivity e
1
surrounded by another dielectric
material of permittivity e
2
, where e
1
> e
2
, as illustrated in
a
(a)
b
s
b b
a
a+2s
2
2
2s
(b) (c)
a'
a'
a'
Figure 3. (a) The cutoff frequency of TE
10
-mode waveguide is
lowered by increasing the width a; (b) the waveguide in (a) is
folded for a more compact structure, and the new equivalent
width is a
0
; (c) a double-ridged version of (b), imaged along the
horizontal bisector, has the same operating frequencies and band-
width as in (b).
Fig. 5a and the top view along the propagation direction in
Fig. 5c. If y is the angle of the guided ray to the interface
normal, and f is the angle to the normal interface of any
ray that transmits into material 2, then by Snells law of
refraction
e
1
p
sin y
e
2
p
sin f 96
Requiring f90
for total internal reection yields

y sin
1
e
2
e
1
_

D
y
c
97
Thus, with e
1
> e
2
, the wave is guided along material 1 for
all angles y ! y
c
. While most of the guided wave is in ma-
terial 1, part of the wave protrudes into material 2 and
propagates along the interface.
For TE waves, E
x
0, E
z
0, and
E ^ yyE
y
xe
jbz
^ yyE
y
x; z 98
where bo

me
e
p
is the propagation constant of the guided
wave that propagates through an effective permittivity
that has the range: e
2
e
e
e
1
. Note that @E=@z jbE,
and @E=@y 0, but @E=@xO0, since the permittivity
changes along the x dimension. Substituting these speci-
cations into Maxwells curl equations yields
@E
y
@x
jomH
z
99
and
jbE
y
jomH
x
100
from rE jomH, and
jbH
x
@H
z
@x
joeE
y
101
from rHjoeE.
a'
a'
(a) (b)
(c)
(d)
Figure 4. More commonly used relative dimensions for (a) a sin-
gle-ridged waveguide and (b) a double-ridged waveguide. Other
variants are (c) the single T-septum waveguide and (d) the double
T-septum waveguide.
h
x
z
(a)
2

1
y
2d
(b)
(c)
E
E
E
E
X
X
X
X
2d
E
H
X
y
Z
1

1
=
1
Sin
Figure 5. (a) Dielectric waveguide of permittivity e

1
surrounded
by permittivity e
2
, where e
1
> e
2
and 2d5h; (b) electric eld dis-
tributions along x for the rst two even modes (cosine terms) and
the rst two odd modes (sine terms)as in Fig. 6, the phase angle
from center to edge b
1x
d in increasing mode order is 0ob
1x
dop=2,
p=2ob
1x
dop, pob
1x
do3p=2, and 3p=2ob
1x
do2p; (c) a TE wave
guided in the z direction between the sides 2d apart.
Substituting (99) and (100) into (101) yields
@
2
E
y
@x
2
b
2
o
2
me E
y
102
which is valid in both material 1 and material 2 and has
the general solution
E
y
A
i
e
b
2
o
2
me
p
x
103
Thus, in material 1 (e
1
> e
2
), we obtain
E
y
A
i
e
b
2
o
2
me
1
x
_
A
i
e
j
o
2
me
1
b
2
x
_
A
i
e
jb
1x
x
Acos b
1x
x Bsin b
1x
x
104
where
b
2
1x
o
2
me
1
b
2
b
2
1
b
2
105
The cos b
1x
x and sin b
1x
x solutions in (104) allow harmonic
modes of the propagating wave across the x dimension in
material 1. In material 2 (e
2
oe
1
)
E
y
A
i
e
b
2
o
2
me
2
x
_
Ce
a
2
x
106
where
a
2
2
b
2
o
2
me
2
b
2
b
2
2
107
which shows that E
y
decreases exponentially in the x di-
rection.
Tangential components of E and H must be continuous
across the interface between materials 1 and 2, and this
condition is used to determine additional characteristics of
the propagating modes. Equating E
y
terms at x d, and
using the E
y
Acos b
1x
x solution from (104), we obtain
Acos b
1x
dCe
a
2
d
108
or
CAe
a
2
d
cos b
1x
d 109
Thus, in material 2 at x ! d, we have
E
y
Ae
a
2
d
cos b
1x
de
a
2
x
Acos b
1x
de
a
2
xd
110
Equating H
1z
H
2z
at x d by using @E
y
=@x jomH
z
as
given by (99), we obtain
H
1z
1
jom
@
@x
Acos b
1x
x
b
1x
A
jom
sin b
1x
x 111
H
2z
1
jom
@
@x
Ce
a
2
x

a
2
C
jom
e
a
2
x
112
and at x d
b
1x
Asin b
1x
da
2
Ce
a
2
d
a
2
Ae
a
2
d
cos b
1x
d e
a
2
d
113
or
tan b
1x
d
a
2
b
1x
114
is obtained for the choice E
y
Acos b
1x
x in material 1. If
the equally valid E
y
Bsin b
1x
x had been chosen as the
solution in material 1, the result of matching tangential
elds at x d would have been
tan b
1x
d
b
1x
a
2
115
The electric eld distributions along the x dimension for
the rst four modes, two even (cosine terms), and two odd
(sine terms), are shown in Fig. 5b. Note the corresponding
phase angle b
1x
d from center to edge for each mode as
listed in the caption for Fig. 5b. These phase angles and
the propagating modes that are allowed may be deter-
mined by graphically solving the transcendental eigenval-
ue equations tan b
1x
da
2
=b
1x
and tan b
1x
d b
1x
=a
2
.
From the expressions dened earlier, a
2
2
b
2
o
2
me
2
and b
2
1x
o
2
me
1
b
2
, express
a
2
b
1x
o
2
me
1
e
2
b
2
1x
1
116
Next, dene and then substitute the term
V
D

o
2
me
1
e
2
d
2
_
od

me
0
p
e
1r
e
2r
p
od
c
e
1r
e
2r
p

2pd
l
0
e
1r
e
2r
p
117
to obtain
tan b
1x
d
a
2
b
1x
V
b
1x
d
_ _
2
1
118
for the even modes, and
tan b
1x
d
b
1x
a
2
V
b
1x
d
_ _
2
1
_ 119
for the odd modes. The operating frequency and the pa-
rameters of the dielectric waveguide set the value of V
(called the normalized frequency). The rst and last terms
of (118) and (119) are plotted against b
1x
d, and the cross-
ing points of the curves give the allowed modes and the
particular value of b
1x
d for that mode. An example is plot-
ted in Fig. 6 for the case when V6. For this value of V,
two even modes and two odd modes are allowed to prop-
agate. Observe from the plot that if Vo3p=2, only three
modes could propagate (two evens and one odd). Also, if
Vop=2, only one mode could propagate (the lowest-order
even mode).
Example 2. A dielectric sheet of thickness 2d0.002 m
and permittivity e
1
3:25e
0
is surrounded by air (e
2
e
0
).
(a) Over what range of operating frequencies will no more
than three modes propagate? (b) Over what range of op-
erating frequencies will only one mode propagate?
Solution:
V
2pfd
c
e
1r
e
2r
p
2pf 0:001
3 10
8
3:25 1
p
pf 10
11
(a) For pf 10
11
o3p=2, three modes propagate in the
range fo150 GHz.
(b) For pf 10
11
op=2, only one mode propagates in
the range fo50 GHz.
As indicated earlier, b, the propagation constant of
waves or modes that are guided in the z direction, lies
within the range b
2
b b
1
, where b
1
o

m
0
e
1
p
and
b
2
o

m
0
e
2
p
. Also, the raypaths of guided waves may
have y angles that lie within the range y
c
y 90
, where
y
c
sin
1

e
2
=e
1
_
is the angle of the critical ray and y 901
is the angle of the axial ray. So, how do the angles of the
raypaths relate to the propagation constant of modes or
waves? From the raypath diagram in Fig. 5c, the propa-
gation constant of modes guided in the z direction is
b b
1
sin y o

m
0
e
1
p
sin y 120
For axial rays (y 901)
b b
1
o

m
0
e
1
p
121
This is the largest value of b, and it occurs at the highest
operating frequency of a particular mode, since b is di-
rectly proportional to frequency.
For rays at y y
c
, we obtain
b o

m
0
e
1
p
e
2
e
1
_
o

m
0
e
2
p
b
2
122
This is the smallest value of b, and it occurs at the lowest
operating frequency of a particular mode. Also, at y y
c
where bb
2
, we have from (105)
b
2
1x
o
2
me
1
b
2
o
2
me
1
o
2
me
2
123
or
b
1x
o

m
p
e
1
e
2
p

V
d
124
Thus, rays at y y
c
within a particular mode have b
1x
d
V, which is at the point of extinction or cutoff for the mode.
The relationships discussed above are indicated on the
plot of b versus radian frequency o in Fig. 7. Note from the
plot that at a single operating frequency, many different
modes may be propagating, and the highest-order modes
have ray paths at y y
c
, while the lowest-order modes
have axial rays (y 901).
Example 3. (a) A dielectric waveguide has e
1r
2:34;
e
2r
2:28, d5 mm and f 318.31 THz, which yields a
normalized frequency of V od=c

e
1r
e
2r
p
8:22. The
6
0 at
1x
d = V
1x
d
at
1x
d=V
5 2 4
1
3
tan
1x
d
e2
tan
1x
d
0
1
e
1
5
2
2
3
2
2
1
V
1x
d
1
2
1
V
1x
d
2

Figure 6. First and last terms of (118) and (119) plotted against
b
1x
d for the case when V6. For this value of V, two even modes
and two odd modes are allowed to propagate.
One mode
Lowest-order
mode
=
c
1
=
0
2
=
= 90
Figure 7. Plot of b versus radian frequency o. Note from the plot

that at a single operating frequency, many different modes may be
propagating, and the highest-order modes have raypaths near the
critical angle, while the lowest-order modes are nearly axial.
eigenvalues of each propagating mode are b
1x
d
1:400; 2:491; 4:179; 5:543; 6:865; and8:045. Find the
propagation angle (y to the interface normal) for each
mode. (b) Is the raypath of the highest-order mode
near the critical angle (y y
c
) or near the axial angle (y
901)?
Solution: The propagation constant along a ray path with-
in the waveguide is
b
1
o

m
0
e
1
p

o

e
1r
p
c

1:02 10
7
m
125
The propagation constant along the x direction is
b
1x
b
1
cos y
1
, where from the data, we have
b
1x
0:28; 0:498; 0:8358; 1:1086; 1:373; 1:60910

6
m
126
Thus
y cos
1
b
1x
b
1
88:43
; 87:20
; 85:30
; 83:76
; 82:26
; 80:92
127
Observe that the highest-order mode is near the critical
angle, y
c
sin
1 1:51
1:53
80:72
, and the lowest-order mode

is nearly axial (901).
Example 4. (a) A dielectric waveguide has e
1r
2:19 and
e
2r
2:16. For single-mode propagation at the following
wavelengths (in air), l
0
1:3; 1:4; 1:5; 1:6 mm, the eigen-
values are b
1x
d0:9336; 0:9027; 0:8734; 0:8456, respec-
tively. Find the propagation angle (y to the interface
normal) for each wavelength if d1.89mm. (b) Is the
shortest wavelength closest to the critical ray (y y
c
) or
to the axial ray (y 901)?
Solution: The normalized frequency (V) in the order of in-
creasing wavelength (in mm) is
V
2pd
l
e
1r
e
2r
p

2:042
l
0
1:5708; 1:4586; 1:3613; 1:2763
128
Using this result, the eigenvalue at each wavelength is
found by solving
tan b
1x
d
V
b
1x
d
_ _
2
1
2:042
lb
1x
d
_ _
2
1
129
to yield b
1x
d0:9336; 0:9027; 0:8734; 0:8456, and
b
1x
0:4940; 0:4776; 0:4621; 0:4474 10
6
=m. At each
wavelength,
b
1
2p

e
1r
p
=l
0
7:1532; 6:6422; 6:1994; 5:811910
6
=m.
Thus
y cos
1
b
1x
b
1
86:04
; 85:88
; 85:72
; 85:58
130
in the order of increasing wavelength. Note that the short-
est wavelength is closest to the axial ray (y 901), and the
longest wavelength is closest to the critical ray
(y y
c
83:34
).
Acknowledgement
This work was supported in part by the National Cancer
Institute under PHS Grant 2 pol CA42745-17.
BIBLIOGRAPHY
1. S. Ramo, J. R. Whinnery, and T. Van Duzer, 3rd ed., Fields
and Waves in Communication Electronics, Wiley, New York,
1994.
2. P. Rizzi, Microwave Engineering Passive Circuits, Prentice-
Hall, Englewood Cliffs, NJ, 1988.
3. S. B. Cohn, Problems in strip transmission lines, IRE Trans.
Microwave Theory Tech. 3:119126 (March 1955).
4. S. Hopfer, The design of ridged waveguides, IEEE Trans. Mi-
crowave Theory Tech. 3:2029 (Oct. 1955).
5. Y. Zhang and W. T. Joines, Some properties of T-septum wave-
guides, IEEE Trans. Microwave Theory Tech. 35:769775
(Aug. 1987).
6. Y. Zhang and W. T. Joines, Attenuation and power-handling
capacity of T-septum waveguides, IEEE Trans. Microwave
Theory Tech. 35:858861 (Sept. 1987).
7. G. G. Mazumder and P. K. Saha, Rectangular waveguide with
T-shaped septa, IEEE Trans. Microwave Theory Tech. 35:201
204 (Feb. 1987).
8. W. B. Jones, Introduction to Optical Fiber Communication
Systems, Oxford Univ. Press, New York, 1988.
GUNN OR TRANSFERRED-ELECTRON DEVICES
AND CIRCUITS
HERIBERT EISELE
University of Leeds
Leeds, United Kingdom
1. GENERAL
Among all solid-state microwave devices, transferred-elec-
tron devices (TEDs), often called Gunn devices, are quite
unique in that they utilize specific bulk-material proper-
ties of certain semiconductors. They are unipolar devices
and, generally, do not exhibit the distinctive diode char-
acteristic of p-n (positivenegative) junctions as seen, for
example, in other two-terminal microwave devices such as
impact ionization avalanche transit-time (IMPATT) or
tunnel injection transit-time (TUNNETT) diodes. Origi-
nally, TED structures and circuits were developed for both
GUNN OR TRANSFERRED-ELECTRON DEVICES AND CIRCUITS 1857
amplier and oscillator applications. However, rapid pro-
gress in high-speed and high-frequency three-terminal
devices with excellent noise performance has practically
been eliminating TEDs from all low-noise preamplier
applications up to the high-millimeter-wave frequencies.
Additionally, oscillators or ampliers with three-terminal
devices continue to reach higher and higher frequencies
and offer similar or even higher radiofrequency (RF) out-
put power levels and direct-current (DC)-to-RF conversion
efciencies compared to the most powerful two-terminal
devices such as IMPATT diodes. TEDs in oscillator appli-
cations are characterized by low-noise and medium-RF
output power P
RF
. Therefore, they are well suited for local
oscillators in sensitive heterodyne receivers and for trans-
mitters at operating frequencies above 30 GHz.
2. PRINCIPLES OF OPERATION
The transferred-electron effect only depends on a specific
band structure of the semiconductor material and, there-
fore, is present in the bulk material. Several materials,
mainly in groups IIIV and IIVI compound semiconduc-
tors and listed in Table 1 [13] exhibit such a band struc-
ture. These semiconductor materials have more than one
energy minimum (i.e., valley in the conduction band) and
meet the following criteria, which were proposed indepen-
dently by Ridley and Watkins [4] as well as Hilsum [5]
(RWH):
1. At least two valleys must be present in the conduc-
tion band.
2. The minimum (minima) of the upper valley(s) must
be several times the thermal energy of electrons
above the minimum of the lowest ( main) valley in
the conduction band for electrons to reside initially
in the lowest valley.
3. The energy difference (DE) between the minimum
(minima) of the upper valley(s) and the minimum of
the main valley in the conduction band must be less
than the energy bandgap E
g
to avoid the onset of
significant impact ionization in such a device.
4. The transfer of electrons from one conduction band
valley to another must require much less time than
one period of the intended operating frequency.
5. The effective masses and densities of states in the
upper valley(s) must be considerably higher than in
the main valley. As a consequence of the higher ef-
fective masses, mobilities in the upper valley(s)
must be much lower than in the main valley.
For the principles of operation, a homogeneous bulk semi-
conductor material and a simplied band structure as
shown in Fig. 1 are assumed. Electrons at low energies
initially reside in the main valley of the conduction band,
where a low effective mass corresponds to a high mobility
m
1
. When electrons acquire more energy (e.g., under an
electric eld E), most of them still remain in the main
valley if EoE
th
, where E
th
is referred to as the threshold
electric eld. As electrons acquire even more energy (for
E > E
th
), many of them are scattered (transferred) into
the upper valley, where a higher effective mass corre-
sponds to a lower mobility m
2
[1]. For an electric eld E,
assumed to be constant in the bulk material, an average
electron velocity v
avg
and average mobility m
avg
can be
dened as
v
avg
n
1
m
1
n
2
m
2
n
1
n
2
Em
avg
E 1
where n
1
and n
2
denote the number of electrons in the
lower and upper valleys, respectively. Since n
1
decreases
and n
2
increases, v
avg
decreases as shown in Fig. 2. At
large energies [i.e., high electric elds (EcE
th
)], most of
the electrons are transferred to the upper valley and
n
1
5n
2
. Therefore, after reaching the minimum value,
the average drift velocity again increases for higher elec-
tric elds. The decrease in the average drift velocity for
E > E
th
generates a region of negative differential
Table 1. Semiconductor Materials Related to the Transferred-Electron Effect
Valley Separation
Semiconductor E
g
(eV) Between DE (eV) E
T
(kV/cm) v
p
(10
7
cm/s) T (K)
GaAs 1.42 G and L 0.31 (0.33) 3.23.5 2.22.3 300
InP 1.35 G and L 0.53 (0.45) 1012 2.52.8 300
Ge 0.74 L and G 0.18 2.3 1.4 77
CdTe 1.50 G and L 0.51 11 1.5 300
InAs 0.36 G and L 1.28/0.87 1.6/2.5 3.6 300
InSb 0.28 (0.18) G and L 0.41 0.6 5.0 77
ZnSe 2.60 G and L 38 1.5 300
Ga
0.5
In
0.5
Sb 0.36 G and L 0.36 0.6 2.5 300
Ga
0.3
In
0.7
Sb 0.24 G and L 0.6 2.9 300
In
0.53
Ga
0.47
As 0.76 G and L 0.55 34 2.9 300
InAs
0.2
P
0.8
1.10 G and L 0.95 5.7 2.7 300
Ga
0.13
In
0.87
As
0.37
P
0.63
1.05 G and L 5.58.6 1.2 300
GaN 3.36 G and X 1.5 80160 2.54.5 300
Ga
0.5
Al
0.5
N 4.77 G and X 0.44 470 2.5 300
Source: Data From Refs. 13.
1858 GUNN OR TRANSFERRED-ELECTRON DEVICES AND CIRCUITS
mobility m
diff
with
m
diff
dv
avg
dE
2
If electrons in the upper valley reach a region where the
electric eld E drops below E
th
, they lose energy and sig-
nificantly more of them are scattered back to the main
valley.
James B. Gunn was the rst to observe current oscil-
lations experimentally in bulk GaAs and InP [6,7], which
were subsequently explained by this transferred-electron
effect [8]. As a result, the term Gunn device quickly be-
came common for this type of device. Out of more than 10
semiconductor materials known for the transferred-elec-
tron effect, only GaAs and InP have so far found wide-
spread use in system applications. GaAs and InP have
three valleys in the conduction band, and at the doping
concentrations required for operation at millimeter-wave
frequencies, high-eld mobilities are considerably lower
than the low-eld mobilities. As a consequence of two up-
per valleys and additional scattering mechanisms, the
electron drift velocity v monotonically decreases for elec-
tric elds above E
th
. Figure 3 shows simplied band struc-
ture diagrams for GaAs and InP, and Fig. 4 their
respective velocityelectric eld proles. Table 2 [911]
summarizes the relevant material characteristics of GaAs
and InP. The nite time it takes for electrons to gain or
lose energy in an electric eld causes a fundamental phys-
ical frequency limit in these devices. This frequency limit
was originally thought to be approximately 100 and
200 GHz [911] for TEDs based on the GaAs and InP ma-
terial systems, respectively. However, more recent theo-
retical and experimental work indicates a significantly
higher frequency limit in InP-based TEDs.
Except for some rare conditions in devices at lower mi-
crowave frequencies, the bulk negative differential mobil-
ity alone does not cause a static negative differential
resistance (NDR) to be used for RF power generation. A
mechanism based on the negative differential mobility re-
sults in a dynamic negative resistance as shown next. In a
region of bulk semiconductor material under uniform con-
ditions (doping concentration N
D
, electric eld E, average
differential mobility m
diff
), any space charge inhomogeneity
Q
S
(x,t) traveling at the velocity v grows or decays follow-
ing an exponential law that can be derived from Maxwells
equations:
Q
S
x; t Q
S
x vt; 0 exp
t
t
_ _
3
where
t
e
s
s

e
s
qN
D
m
diff
4
v
v
1
=
1
v
2
=
2
diff
=

Peak
Valley
dE
< 0
dv
diff
=

dE
> 0
dv
Figure 2. Velocityelectric eld prole for the two-valley semi-
conductor of Fig. 1.
(a)
k
n
1
n
2
n
1
n
2
n
1
n
2
E
L
--
--
--
(b)
k
E
L
--
--
- -
(c)
k
E
L
-
- - -
- -
Figure 1. Simplied energy-band diagram for a di-
rect two-valley semiconductor showing electron
transfer: (a) electron distribution for EoE
th
]; (b)
electron distribution for E > E
th
]; (c) electron distri-
bution for EbE
th
].
3
2
1
0
1
2
X L
<111> <100>
X L
<111> <100>
GaAs
Wave vector
E
n
e
r
g
y

(
e
V
)
InP
T = 300 K T = 300 K
Upper
valley
Upper
valley
Lower
valley
Lower
valley
E = 0.31
E = 0.53
E
g
= 1.42
E
g
= 1.35
Figure 3. Simplied band diagram for the three-valley semicon-
ductor materials GaAs and InP.
where, q denotes the elementary charge; s and e
s
denote
specific conductivity and dielectric constant of the mate-
rial, respectively. At low electric elds E, where m
diff
40,
the charge inhomogeneity decays with t t
D
, the dielec-
tric relaxation time, and at higher electric elds, where
m
diff
o0, a charge inhomogeneity can grow. This charge in-
homogeneity reaches a significant level only if the growth
factor for the maximum traveled distance l, the length of
the active region of the device, is very large. Therefore the
condition
l
vjtj

lqN
D
jm
diff
j
e
s
v
> 1 5
must be satised, which corresponds to
N
D
l > 1 10
12
cm
2
6
for both GaAs and InP.
Typical Gunn devices at millimeter-wave frequencies
have N
D
l products between 1 10
12
cm
2
and 3
10
12
cm
2
, and doping concentrations N
D
in the active re-
gion exceed 10
15
cm
3
. For N
D
410
15
cm
3
, space charge
inhomogeneities typically grow into so-called dipole do-
mains where accumulation and depletion layers are
lumped together. Figure 5 shows the carrier distribution
and electric-eld prole for such a dipole domain under
uniform conditions. Electrons in the low-eld region travel
at a constant v
T
for a constant electric eld E
1
. Electrons
in region a are accelerated by the higher electric eld until
they reach region b, where they are transferred to the up-
per valley and slow down to be trapped in this accumula-
tion region. Electrons in region c lose energy and are
transferred back to the lower valley. Their average veloc-
ity now is higher than the average velocity in region b,
thus region c is depleted of electrons. After a domain forms
at the cathode, grows, and propagates through the active
region, the voltage drop across the domain increases and,
under a constant bias voltage, lowers the voltage drop
outside the domain. This voltage drop is equivalent to a
reduction in the electric eld E
1
outside the domain and
generally prevents formation of new domains in the active
region. It also limits the growth of the existing domain
because fewer electrons are trapped in the accumulation
layer or escape the depletion layer. Domains reaching the
anode collapse and induce a current ow in the external
circuit. A phase difference occurs between this current
ow and the applied RF voltage and, at the right frequency,
this phase difference corresponds to a dynamic negative
resistance and causes RF power generation in an appro-
priate RF circuit.
Distinct modes of operation have been investigated
and described for TEDs at microwave frequencies [12].
v

(
1
0
7

c
m
.
s
1
)
3
2
1
0
0 10 20 30 40 50 60 70
InP
E (kV
.
cm
1
)
GaAs
E
th
E
th
Figure 4. Velocityeld prole for the three-valley semiconduc-
tor materials GaAs and InP.
Table 2. Semiconductor Material Characteristics
Relevant to GaAs and InP TEDs
a
Semiconductor
Properties GaAs InP
Energy gap (eV) 1.42 1.34
Low-eld mobility (at 500K)
(cm
2
V
1
s
1
)
5000 3000
Thermal conductivity (Wcm
1
K
1
) 0.370.54 0.680.80
Velocity peak-to-valley ratio 2.2 3.5
Threshold eld E
th
(kV/cm) 3.5 10.5
Breakdown eld (at N
D
10
16
cm
3
)
(kV/cm)
400 500
Effective transit velocity v
T
(cm/s) 0.710
7
1.2 10
7
Temperature dependence of v
T
(K
1
) 0.0015 0.001
Diffusion coefcient-mobility ratio
at 2 E
th
(cm
2
/s)
72 142
Energy relaxation time due to
collisions (ps)
0.40.6 0.20.3
Intervalley relaxation time (ps) 0.25
Accelerationdeceleration time (ps)
(inertial energy time constant)
1.5 0.75
Source: After Wandinger Ref. 9, Fank et al. [10], and Eddison [11].
a
At a temperature of 300K unless noted otherwise.
N
Accumulation layer
Depletion layer
X
l
a b c
X
Figure 5. Carrier concentrations and electric eld prole for a
dipole domain.
However, as is shown next, at millimeter-wave frequen-
cies nite intervalley transfer and domain formation
times reach a significant fraction of the RF cycle. In
such a case, domains form, grow, and suppress formation
of new domains, but may never reach the stable state be-
fore they reach the anode as described above. Therefore,
modes get blurred, and devices generally operate in a near
transit-time mode, where the operating frequency f
op
is
given by
1
T
f
op
v
T
l
7
The effective transit velocity v
T
v(E
1
) v
D
(E
h
) can be de-
termined from Butchers equal-area rule [13], which is
_
E
h
E
l
vE v
D
dE0 8
for a constant diffusion coefcient throughout the active
region and is illustrated in Fig. 6.
If the operating frequency f
op
differs somewhat from
Eq. (7), the domain reaches the anode prematurely or is
delayed. Similar to the operation of transit-time diodes,
the current pulse from the collapsing domain still causes a
negative resistance and generates RF power. Therefore,
operation over a broad bandwidth can be achieved. Addi-
tionally, higher bias voltages increase electric elds in the
device, and higher electric elds reduce the domain veloc-
ity v
D
, as seen in Fig. 6. At higher electric elds, electrons
also acquire the energy for intervalley transfer over a
shorter distance as is described in more detail in a
subsequent section. Consequently, the portion of the ac-
tive region where domains form and travel is increased.
Lower v
D
and longer domain travel distance correspond to
a lower optimum f
op
.
Figure 7 gives an overview of typical structures and
schematic doping proles for TEDs that yielded (Figs. 7a,
7b, 7d, 7e) or are expected to yield (Fig. 7c) excellent RF
performance in oscillators. The three-zone at-doping and
the two-zone at-doping structures were the rst to be
exploited because they are easy to grow in more classical
growth systems such as liquid-phase or vapor-phase epi-
taxy (LPE or VPE). The three-zone at-doping structure
[14] consists of the n
-doped active region sandwiched

between the highly doped n
regions for the ohmic con-

tacts. Since low-ohmic alloyed contacts can be formed on
n-type GaAs and InP, the highly doped region on the cath-
ode side can be omitted and just a two-zone, at-doping
structure needs to be grown by VPE or LPE.
The advent of advanced growth techniques such as mo-
lecular-beam epitaxy (MBE), metallorganic chemical va-
por deposition (MOCVD), metallorganic molecular-beam
E
h
E
l
V
v
D
v
D
(E
h
)
Figure 6. Equal-area rule for TEDs.

+
+
+
+
+
N
Metal
GaAs or InP
GaAs or InP
Metal
n
+
n
n
+
(a)
N
(b)
N
Metal
Metal
Metal
Metal
GaAs or InP
(c)
N
Metal Metal
GaAs or InP
(d)
N
GaAs
AIGaAs
n
+
GaAs GaAs
GaAs
Metal Metal
Graded composition
(e)
n
+
n
n
+
n
+
n
n
+
n
+
n
n
+
n
+
n
n
+
Figure 7. Different device structures for TEDs: (a) three-zone
at doping; (b) two-zone at doping; (c) three-zone at doping
with cathode notch; (d) three-zone graded doping; (e) heterojunc-
tion barrier cathode.
epitaxy (MOMBE), and chemical-beam epitaxy (CBE) has
allowed more complicated structures to be grown. Using
these growth techniques, graded-doping proles and he-
terojunction barriers as shown in Fig. 7 can be incorpo-
rated into the device structures and suitably tailored to
optimize device performance at a particular frequency or
to extend the frequency limit of TEDs. Computer simula-
tions [11,15,16] have revealed that in a three-zone at-
doping structure, cold electrons with low energies enter
the active region from the contact zone at the cathode and
require some time to acquire enough energy before they
transfer to the upper valley. The results of such Monte
Carlo simulations [16] at a frequency of 95 GHz are illus-
trated in Fig. 8 for a three-zone at-doping structure in
InP with a doping of 110
16
cm
3
in the active region.
The nite-energy relaxation times, which are shown in
Fig. 9 as a function of the electron energy in GaAs and InP,
create a huge so-called dead space at the beginning of
the 1.7-mm-long active region. As can be seen from Fig. 8c,
the average energy E of electrons within the dead space
does not reach the threshold energy E
th
for intervalley
transfer, and electrons reside mostly in the main valley.
Therefore, the differential mobility remains positive, and
space charge inhomogeneities are prevented from grow-
ing, which is illustrated in Fig. 8b with insignificant elec-
tron accumulation within the dead space. As a
consequence, the resistance of the device R(x) [i.e., the
real part of Zx] as a function of the position x
Zx Rx jXx
_
x
0
Ex
0
dx
0
A
l
_
x
0
Jx
0
dx
0
9
E
l
e
c
t
r
i
c

f
i
e
l
d

(
k
V
.
c
m
1
)
40
20
0
20
40
60
80
100
120
140
6
0
2
4
D
e
n
s
i
t
y

(
c
m
3
)
310
17
210
17
110
17
110
16
Doping concentration
0
2
4
6
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
(a) (b)
E

(
e
V
)
0.6
0.5
0.4
0.3
0.2
0.1
0.0
E
th
X (m) X (m)
x (m) x (m)
(c)
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
4
2
0
2
4
(d)
0
5
6
7
1
2
3
4
R

(
)
Figure 8. Evolution of (a) electric eld, (b) electron density, (c) average electron energy E, and
(d) diode resistance R against position x (active region 0.11.8mm): f 95GHz, V
rf
1:0 V,
V
bias
5:0 V, I
bias
474mA, T500K. The graphs in parts (a), and (b) show the electric eld
and the electron density, respectively, at ot np/4, n0,2,4,6, and part (c) shows the electron
energy prole at ot np/4, n0y7, during one RF cycle.
remains positive for a large fraction of the active region
and contributes to losses in this dead-space region, where-
as a negative resistance contributes to the RF power gen-
eration only for a small fraction of the active region. In
Eq. (9), A and J, denote the device area and total current
density, respectively. In addition to this dead-space region,
the peak electric eld occurs near the anode, and, at a high
DC bias, the electric eld may reach values for the onset of
avalanche breakdown. The energy-dependent energy re-
laxation times of Fig. 9 lead to effective transfer time con-
stants as shown in Table 2 for GaAs and InP.
Fundamental frequency limits of 100GHz for GaAs and
200GHz for InP TEDs were originally estimated from
these effective transfer time constants and considered
hard limits. However, in subsequent sections, some solu-
tions that help reduce the dead-space region or have been
demonstrated to extend the useful frequency range even
far beyond these fundamental frequency limits are pre-
sented and discussed.
3. FABRICATION TECHNOLOGIES
TEDs are characterized by low to medium DC-to-RF con-
version efciencies ranging from approximately more than
15% down to less than 1%. As a consequence, most of the
DC input power P
DC
, namely, P
DC
P
RF
, needs to be dis-
sipated as heat in the device. In most cases, one of the
metal contacts near the active region of the device also
acts as the heatsink; therefore, TEDs for millimeter-wave
frequencies generally are mesa-type devices. Additionally,
operation at these frequencies requires thin devices to re-
duce losses in the substrate resulting from the skin effect.
The integrated heatsink technology is the most wide-
spread for devices at millimeter-wave frequencies. To re-
duce losses in the substrate, most of it needs to be removed
during fabrication.
In early fabrication technologies, vapor-phase epitaxy
(VPE) provided the layer structures. As a rst step in pro-
cessing, a few small holes (or grooves) across the sample
were etched through the epitaxial layers down into the
substrate. An appropriate depth of the holes was chosen to
gauge the thickness during substrate removal. The advent
of more advanced growth techniques, such as MBE, MOM-
BE, MOCVD, and CBE, allows the incorporation of a lat-
tice-matched, stopetch layer between the substrate and
the epitaxial layers of the device. This way, the substrate
is completely removed, and precise control of the mesa
height and, consequently, the device diameter is achieved.
Fabrication technologies for substrateless devices on inte-
gral heatsinks or on diamond heatsinks for better heat
removal have been developed and described in the litera-
ture. Selective etching technologies in the GaAs and InP
material systems [1618] employ as etchstop layers lat-
tice-matched Ga
x
Al
1 x
As (xo0.4) and In
0.53
Ga
0.47
As lay-
ers, respectively. Improved yield, reproducibility, and
performance characterize these substrateless devices.
Figure 10 summarizes the basic steps of these fabrica-
tion technologies. The batch fabrication of InP TEDs on
integral heatsinks serves as an example [16]. In the rst
step, the metallization for the n-ohmic contact (Ni/Ge/Au/
Ti/Au) is evaporated or sputtered onto the surface. A thick
gold layer is then electroplated onto this metallization to
form the integral heatsink. The sample is mounted on a
carrier to provide additional mechanical support and pro-
tect the heatsink during the subsequent processing steps.
The substrate is removed in a selective etchant of diluted
hydrochloric acid [16], which does not attack the In
0.53-
Ga
0.47
As etchstop layer. Good ohmic contacts can be
formed on n-type InP or, with lower specific contact resis-
tance r
c
, on n-type In
0.53
Ga
0.47
As. Therefore, this In
0.53-
Ga
0.47
As layer need not be removed, but may be etched
away selectively in a standard solution of phosphoric or
sulfuric acid, hydrogen peroxide, and water as indicated in
Fig. 10. Such a solution does not attack InP. A photoli-
thography step denes the openings on this InP surface
(or In
0.53
Ga
0.47
As surface if left in place), where the met-
allization (Ni/Ge/AuTi/Au)) for the other n-ohmic contacts
on the second heavily n
-doped layer is deposited. Excess

metal outside the contacts is lifted off with the photoresist
and, using another photolithography step, the contacts are
selectively electroplated with several micrometers of gold
to form a good bonding pad. The contact pad acts as a
mask when the mesa of the diode is etched in a nonselec-
tive etch. After the sample has been removed from the
carrier, the contacts are annealed, and the sample is diced
into individual diodes. Diodes are then mounted in pack-
ages for appropriate RF circuits.
4. DEVICE PACKAGES, OSCILLATOR CIRCUITS, AND
RF PERFORMANCE
Figure 11a shows a typical TED package. It consists of a
gold-plated threaded copper puck (which can be screwed
into the RF circuit), an alumina ring, and a top lid for a
hermetic seal. The device is soldered or thermocompres-
sion-bonded onto a pedestal inside the ring, and gold
straps are then thermocompression-bonded onto the de-
vice and the top metallization of the alumina ring. The
height and diameter of the ring depend on the operating
e

(
p
s
)
4
3
2
1
0
0 0.2 0.4 0.6 0.8 1.0
GaAs
InP
E (eV)
Figure 9. Energy relaxation times t
e
in GaAs and InP against
electron energy E. (After Rolland et al. [15].)
frequency as well as the device, and typical values are
given in Fig. 11a. This type of package is used up to fre-
quencies of 94 GHz, and its parasitic elements can be ap-
proximated by lumped elements as illustrated in Fig. 11b.
Different ribbon congurations are chosen to minimize the
inuence of the parasitic inductance L
p
, which is the high-
est for just one gold strap across and the lowest for the
star conguration. The useful frequency range of the
package can be extended to 140GHz and higher if the
alumina ring is replaced by a quartz ring for a lower par-
asitic capacitance C
p
. However, new devices for frequen-
cies above 100 GHz are still being developed, and, for
research purposes, a low-parasitic open package with
two or four standoffs at the highest millimeter- and
up to submillimeter-wave frequencies is also often
employed [18,19].
The heat-ow resistance R
th
from the active layer of the
device to the package causes an average temperature
increase DT
op
in the active layer
DT
op
R
th
P
DC
P
RF
10
Too high an active layer temperature degrades the RF
performance as well as the device reliability and lifetime
[11]. A larger valley separation of 0.53 eV in InP than in
GaAs (see Fig. 3) reduces the temperature dependence of
the transfer mechanism as well as the temperature-de-
pendence of the effective transit velocity v
T
(see Table 2).
As a result, DC-to-RF conversion efciencies and oscilla-
tion frequencies are generally less temperature-depen-
dent in InP Gunn devices. However, the higher
threshold electric eld of 10.5 kV/cm in InP (see Table 2)
requires higher bias voltages than those applied at GaAs
devices of the same length. Therefore, RF power levels are
thermally limited at low microwave frequencies, where
long active regions need to be used. As a further conse-
quence, InP TEDs are more likely to benet from reduced
heat-ow resistances [11]. Figure 12 compares the esti-
mated [20] heat-ow resistances of W-band (75110 GHz)
and D-band (110170 GHz) InP Gunn devices on integral
and diamond heatsinks as well as some measured values
for devices on integral heatsinks [2123]. Examples of how
significantly diamond heatsinks improve the RF perfor-
mance of both GaAs and InP Gunn devices are provided in
the sections on device structures.
Many different circuit congurations for oscillators
with TEDs have been investigated. At millimeter-wave
frequencies, waveguide circuits are quite common. Al-
though excellent results were reported from a few trans-
ferred-electron oscillators (TEO) in microstrip circuits
[10,11,2426], the vast majority of the state-of-the-art re-
sults were obtained in waveguide circuits. These results,
summarized in Fig. 13, include the performance of different
Plated gold
n contact
(Ni/Ge/Au/Ti/Au)
n
+
InP
n
+
InP
n
InP
n
+
InP
n
+
InGaAs
InP substrate
(a)
n contact
(Ni/Ge/Au/Ti/Au)
(d)
(b)
Photoresist
(c)
Plated gold
Figure 10. Steps in the fabrication of InP TEDs on integral heatsinks: (a) Island definition,
N-ohmic evaporation and gold plating of heatsink (E20mm); (b) substrate thinning, etchstop layer
removal, and second N-ohmic evaporation; (c) gold plating of ohmic contacts; (d) nal devices after
annealing and mesa etch. (After Kamoua et al. [16].)
device structures as illustrated in Fig. 7. Examples for the
RF performance of individual device structures are given
in the subsequent sections. An overview of typical cong-
urations for waveguide circuits [27,28] is shown in Fig. 14.
Examples of oscillator circuits using coaxial lines at mi-
crowave frequencies can be found in Ref. 1.
5. DEVICE STRUCTURES
5.1. Ohmic Cathode Contacts
TEDs with ohmic contacts on both heavily n-doped regions
of the structure of Fig. 7a (at-doping prole) are the sim-
plest structure, easy to fabricate, but characterized by low
DC-to-RF efciencies. These devices are typically operated
in a full-height waveguide cavity with a resonant cap on
top of the device package, and this conguration is illus-
trated in Fig. 14e. Modications of this conguration in-
clude the use of a reduced-height waveguide or a
mechanism for adjusting the position of the resonant cap
and the device package with respect to the bottom of the
waveguide. Fundamental-mode operation of Gunn devices
in a reduced-height postcoupled waveguide cavity was re-
ported up to millimeter-wave frequencies, for instance, for
a GaAs Gunn device at 84 GHz [29] and an InP Gunn de-
vice at 126GHz [30]. RF power levels (and corresponding
DC-to-RF conversion efciencies) of 420mW (6%) at
35 GHz [19], 280mW at 45 GHz [19], 150 mW at 60 GHz,
and 110 mW (2.8%) at 70GHz [31] were reported from at-
prole GaAs Gunn devices in the fundamental mode.
A sharp decline in the DC-to-RF conversion efciencies
of devices operating in the fundamental-mode presages
the abovementioned frequency limits for GaAs or InP
Gunn devices. However, this frequency limit can be ex-
tended by the extraction of higher harmonics from the in-
herently nonlinear Gunn device. Second-harmonic power
extraction has proved most successful in a slightly
0.600.89
Ribbon
Ring
Device
Heat sink
Pedestal
Lid
0.350.52
Ribbon
configurations
Ring
Ring
Ring
C
p
= 0.04 0.22 pF
L
p
= 0.03 0.20 nH
Package
C
p
L
p
(b)
Device
C
d
R
d
Equivalent circuit
(a)
348 UNC2A
0.5
0
.
7
6
3
.
3
0
.
2
0
.
5
0
.
5

0
.
0
5
2.92
thread
All dimensions are in millimeter
General tolerances are 0.15
Figure 11. (a) Hermetically sealed package for millimeter-wave
TEDs; (b) equivalent circuit of the parasitic elements.
Diamond heat sink
Integral heat sink
on copper
Measurements
R
t
h

(
K
/
W
)
100
80
60
40
20
0
30 40 50 60 70 80 90
d (m)
Figure 12. Estimated and measured [2123] heat-ow resistanc-
es R
th
of InP TEDs on integral heatsinks and estimated R
th
of InP
TEDs on diamond heatsinks against device diameter d; estimates
based on the spreading approximation [20].
July 2003
15
10
4.0
2.6
2.2
1.4
0.7
0.58
6
13
4.5
4
3.5
4
10
3.5
3.5
4
3
5
7
4.5
4 3.1
2.8
2.5
3.8
3
2.7
2.5
2.6
2.3
1.6
0.6
0.31
0.3
0.32
0.35
0.2
<0.07
0.09
0.21
0.29
0.19
0.16
<0.09
GaAs
InP
InP (U. of Michigan)
1000
100
10
1
0.1
30 100 400
Frequency (GHz)
R
F

o
u
t
p
u
t

p
o
w
e
r

(
m
W
)
3.6
1
1
f
3
f
f
2
Figure 13. State-of-the-art RF power levels from TEDs under

CW operation in the frequency range of 30400GHz. Numbers
next to the symbols denote DC-to-RF conversion efciencies in
percent.
modied version of a resonant-cap, full-height waveguide
cavity. The size of the waveguide is appropriate for the
second-harmonic frequency, but impedes wave propaga-
tion at the fundamental frequency. If in such a circuit the
fringe capacitance of the cap and the device capacitance
together resonate with the inductance of the bias post (see
Fig. 14e) at half the output frequency, this signal cannot
propagate, and the termination of the device is almost
purely reactive at the fundamental frequency. This reac-
tive termination causes a large voltage swing in the device
and, as a result, strong nonlinear operation. A cap of
appropriate size together with the coaxial post provides
impedance matching into the waveguide at the second-
harmonic frequency, and a waveguide backshort at one
side of the cavity provides power tuning. The resonant
circuit at the fundamental frequency is decoupled from
the load, which corresponds to Q values typically higher
than those in fundamental-mode operation. As results, re-
duced frequency pulling with load changes or improved
frequency stability may be observed. However, more com-
plicated circuits with precise mechanical dimensions are
necessary if wide-range frequency tuning is to be imple-
mented. As an example for second-harmonic power ex-
traction, RF power levels (and DC-to-RF conversion
efciencies) of 123mW (3.1%) at 83 GHz, 96 mW (2.7%)
at 94 GHz were measured with GaAs Gunn devices [31].
The advantages of InP can be clearly seen at millime-
ter-wave frequencies with short active regions where low-
er inertial energy time constants lead to a much higher
fundamental frequency limit. RF power levels (and corre-
sponding DC-to-RF conversion efciencies) of 200mW
(5%) at 80 GHz and 150mW (3.5%) at 94 GHz [22] in the
fundamental mode as well as 7 mW at 180GHz and
3.2 mW at 206 GHz [32] in a second-harmonic mode were
reported from similar at-doping InP Gunn devices on
integral heatsinks.
RF power stability against package temperature typi-
cally ranges from 0.02 to 0.06dB/1C as quoted by var-
ious manufacturers for commercially available GaAs
Gunn devices (see also Ref. 11). Conversely, values of
around 0.013 dB/1C [11] or as low as 0.005 dB/1C
[21,22] were reported from InP Gunn devices.
The more complex structures of Figs. 7c7e also employ
ohmic cathode contacts at the anode and cathode, but de-
vice characteristics and RF performance are discussed
separately in subsequent sections.
5.2. Current-Limiting Cathode Contacts
A partially annealed ohmic contact significantly reduces
the typical Schottky barrier height of metals on the semi-
conductors GaAs and InP (0.60.9 eV), but still leaves a
small barrier (o200meV). If such a contact is formed on
the cathode side of the two-zone structure (see Fig. 7b) and
is reverse-biased, this barrier causes a high-eld region at
the cathode contact of the device under bias. Electrons in-
jected over this barrier enter the active region at a higher
energy and, under this high electric eld, transfer faster
into the upper valleys. This faster transfer reduces the
dead space. The shallow Schottky barrier also limits the
current ow into the active region at the cathode. Therm-
ionic emission and thermionic eld emission contribute to
the current ow, and, in this case, the current density J
c
as a function of the voltage V
c
across the barrier can be
approximated by
J
c
V
c
J
r
exp
qV
c
rkT
_ _
exp
1 rqV
c
kT
_ _ _ _
11
where J
r
A
T
2
exp( qf
Bn
/kT) is the saturation current;
k, T, r, A
, and qf
Bn
denote the Boltzmann constant, ab-
solute temperature, ideality factor, effective Richardson
constant, and effective barrier height on n-type material,
respectively [1]. Current limiting as a boundary condition
at the cathode causes the electrons in the active region of
the device to approximate the current valley condition
[15,33] for a saturated electron velocity v
s
and a doping
concentration N
D
_
T
0
Jt dt J
0
N
D
qv
s
12
Coaxial
section
(b)
Coaxial
section
(c)
Cap resonator
(e)
Impedance
transformer
Bias
Insulator
To
load
Device
Tuning
short
(a)
(d)
Coaxial
section
Iris
Figure 14. Examples of waveguide circuits for TED oscillators.
(After Kuno [27].)
with large space charge waves superimposed on an almost
constant electric eld throughout the active region [15].
This mode of operation yields high RF power levels in the
fundamental mode as well as a second-harmonic mode.
Corresponding DC-to-RF conversion efciencies are typi-
cally the highest reported to date. RF power levels (and
DC-to-RF conversion efciencies) of more than 500mW
(15%) at 35 GHz, more than 350mW (13%) at 44 GHz and
380mW (10.6%) at 57 GHz [19,34] in the fundamental
mode as well as 175mW (7%) at 94 GHz and 65 mW (2.6%)
at 138 GHz [23,35] in a second-harmonic mode were
achieved using this technology. These devices are on inte-
gral heatsinks, and still higher RF power levels are ex-
pected from devices on diamond heatsinks. Devices on
integral heatsinks with high RF power levels are also
commercially available.
Equation (11) expresses the strong temperature depen-
dence that is inherent in the current ow through a shal-
low Schottky barrier in the reverse direction [1]. However,
the high-efciency mode reduces DC input requirements
and provides higher RF impedance levels. As a conse-
quence, larger device diameters can be used, which have
lower heat-ow resistances (see Fig. 12). In turn, lower
heat-ow resistances R
th
entail lower active-layer temper-
atures T
op
[see Eq. (10)]. The typical temperature increase
DT
op
remains below 100 K at maximum RF output power
[23,34], and low operating active-layer temperatures en-
sure reliability and excellent temperature stability over
wide temperature ranges of 30 to 701C for devices at
56 GHz [34] and 94 GHz [34,35] as well as of 0501C for
devices at 140GHz [23]. The temperature-dependent per-
formance of a D-band InP Gunn device in Fig. 15 serves as
an example.
5.3. Graded Active Region
A doping prole with a lower doping concentration N
D
at
the cathode and a linear grading toward a higher N
D
at
the anode as shown in Fig. 7d decreases the peak electric
eld near the anode to a large extent and increases the
electric eld near the cathode [16]. Both effects are bene-
cial to the device operation, and enhance the DC-to-RF
conversion efciency as well as the RF output power of the
device. A lower electric eld near the anode reduces the
power dissipation in this region and also allows higher
bias voltages without the onset of impact ionization and
avalanche breakdown. A higher electric eld near the
cathode causes a larger fraction of the electrons to trans-
fer to the upper valleys over a shorter distance, which is
equivalent to a shorter dead-space region. This is illus-
trated in Fig. 16, which compares domain formation in a
at-prole (Fig. 16a) and a graded-prole (Fig. 16b) TED
structure, both with a 1.0-mm-long active region. Accumu-
lation domains form in the at-prole structure, whereas
dipole domains form in the graded-prole structure. A
higher fraction of electrons in the upper valleys slightly
lowers the average electron velocity throughout the active
region; as a consequence, the graded-prole structure op-
erates at lower current densities compared to a at-prole
structure of a similar (average) doping concentration. In
fact, a lower average electron velocity and shorter dead
space may somewhat decrease the optimum operating fre-
quency f
op
for the same device length l. However, more
efcient device operation extends the upper frequency
limit and allows shorter active regions. Similar to the
three zone structures of Fig. 7a, the structures with grad-
ed-doping proles have ohmic contacts at anode and cath-
ode; therefore, they are easy to fabricate. They were
investigated in both InP and GaAs material systems.
RF power levels (and DC-to-RF conversion efciencies)
of 345mW (6.8%) at 31.2GHz and 325 mW (6.6%) at
34.9 GHz from devices on integral heatsinks [36] and
210 mW (3.5%) around 60 GHz from devices on diamond
heatsinks [37] were achieved in the GaAs material sys-
tem. The devices on diamond heatsinks showed operating
active-layer temperatures below 1501C and fundamental-
mode operation up to 84 GHz (33 mW and 1.7%) [38]. De-
vices from InP material with properly designed [16,39]
graded-doping proles yielded the highest RF power levels
reported for any Gunn device to date. Fundamental-mode
operation was demonstrated up to 165 GHz, and RF power
levels that exceeded 310 mW at 82 GHz [40], 200mW at
103 GHz, 130 mW around 132GHz, 80 mW at 152GHz,
and 25 mWat 163GHz [41] were obtained from devices on
diamond heatsinks [30]. As an example, DC-to-RF conver-
sion efciencies exceeded 4% at 82 GHz [39,40] and 2.3%
between 102 and 132 GHz [42].
As illustrated in Fig. 17, the InP Gunn devices on di-
amond heatsinks allow single-mode operation over a wide
range of DC input power levels. Excellent tuning behavior
was also observed, which can be expected from operation
in the fundamental mode. This tuning behavior over a
range of more than 4.5 GHz is shown in Fig. 18 for the
device of Fig. 17 near maximum DC bias. The oscillation
frequency changes almost linearly with the position of the
backshort, which is the only tuning element in this full-
height waveguide resonant cap cavity (see Fig. 14e for a
schematic). Improved DC-to-RF conversion efciencies re-
duce the DC input power requirements for the same RF
power levels, and, similar to devices with current-limiting
50
40
30
20
10
0
P
o
w
e
r

(
m
W
)
0 10 20 30 40 50
0
1
2
3
4
138.0
138.5
139.0
139.5
140.0
F
r
e
q
u
e
n
c
y

(
G
H
z
)
E
f
f
i
c
i
e
n
c
y

(
%
)
Temperature (C)
Figure 15. Measured RF performance of a D-band InP Gunn
device against ambient temperature. (After Crowley et al. [23].)
cathode contacts, lower the operating active-layer temper-
atures, in particular on diamond heatsinks [30,38].
InP Gunn devices with a graded-doping prole yielded
the highest RF power levels from any Gunn device not
only in the fundamental mode but, for the same graded-
doping prole, also in a second-harmonic mode. Examples
are the RF power levels of 1.8 and 1.1 mW at second-har-
monic frequencies of 280GHz and 315GHz, respectively
[41,43] (see inset of Fig. 17 for the doping prole) as well
as 34 mW at 193GHz and 26 mW at 199GHz [40,44].
When the graded-doping prole is optimized just for sec-
ond-harmonic power extraction, even higher RF power
levels of, for example, more than 3.5 mW around 300 GHz
are generated and, as indicated in Fig. 13, RF power gen-
eration is observed up to at least 325GHz [40,45].
5.4. Injection over a Homo- or Heterojunction Barrier
Injection of electrons with an energy higher than that at
thermal equilibrium (i.e., hot electrons) over a barrier
reduces the dead space in the active region. Several con-
cepts (e.g., planar-doped barrier, camel cathode, and he-
terojunction barriers) have been investigated in the GaAs/
AlGaAs material system to improve efciency, but also to
eliminate cold-start problems in GaAs Gunn devices. In-
jection over a heterojunction barrier has proved the most
D
e
n
s
i
t
y

(
c
m
3
)
610
17
410
17
210
17
110
17
2.50
16
0 0.2 0.4 0.6 0.8 1.0 1.2
x (m)
0
2
4
6
(a)
(b)
D
e
n
s
i
t
y

(
c
m
3
)
610
17
410
17
210
17
110
17
210
16
810
15
0 0.2 0.4 0.6 0.8 1.0 1.2
0
2
4
6
x (m)
Figure 16. Evolution of electron density against position x (active region 0.11.1mm) for a at-
prole (a) and graded-prole (b) TED structure at f 130GHz and ot n
p/4
, n0,2,4,6, during one
RF cycle.
F
r
e
q
u
e
n
c
y

(
G
H
z
)
P
R
F

(
m
W
)
P
DC
(W)
0
20
40
60
80
2.5 3.5 3.0
0.3 m 1.0 m 0.3 m
4.0 4.5
1.5%
2.0%
2.5%
100
120
132.0
132.5
133.0
n
+
n
n
+
210
18
cm
3 210
18
cm
3
2 10
16
cm
3
7.5 10
15
cm
3
Figure 17. Bias-dependent RF characteristics of a D-band InP
Gunn device (K: output power, : oscillation frequency, :
lines of constant efciency). Inset: Nominal doping prole. (After
Eisele and Haddad [30].)
Frequency (GHz)
R
F

o
u
t
p
u
t

p
o
w
e
r

(
m
W
)
B
a
c
k

s
h
o
r
t

p
o
s
i
t
i
o
n

(
m
m
)
0
20
40
60
80
100
120
130 132 134 136
1.2
1.3
1.4
1.5
1.6
Figure 18. Mechanical tuning characteristic for the D-band InP
Gunn device of Fig. 17 close to maximum applied bias.
successful and was experimentally investigated in the Al-
GaAs/GaAs system with improved efciencies and upper
frequency limits [29,46]. Figure 19 shows the band dia-
grams of isotype heterojunctions in the lattice-matched
GaAs/Al
x
Ga
1x
As and InP/In
x
Ga
1x
As
y
P
1 y
material
systems. In both material systems, layers can be grown
lattice-matched over a wide composition range and, thus,
bandgap or conduction-band offset can be tailored suitably.
Linear composition grading from the GaAs to the wid-
er-bandgap AlGaAs eliminates the rst barrier, and the
doping spike as shown in Fig. 7e at the beginning of
the active region (GaAs) reduces or eliminates the notch
at the interface from the AlGaAs layer to the GaAs region.
At the proper bias level, hot electrons are now ballisti-
cally launched into the active region from an approxi-
mately 200-meV-high step in the conduction band. The
optimization of such a design requires advanced simula-
tion schemes, such as ensemble Monte Carlo techniques.
Using a reduced-height post-coupled waveguide cavity
(see Fig. 14b for a schematic) for the oscillator, an RF
output power of 71mW with a corresponding DC-to-RF
conversion efciency of 2.8% was measured in the funda-
mental mode at 77.6GHz. Low-noise operation of such a
Gunn device was achieved at least up to 84GHz [29]. As a
characteristic of fundamental-mode operation [30], a wide
tuning bandwidth of more than 6GHz was observed by
simply adjusting the position of the backshort [29].
In contrast to the GaAs/AlGaAs material system, lat-
tice-matched InGaAsP has a smaller bandgap than does
InP as illustrated in Fig. 19. Therefore, electrons cannot
be ballistically launched into the active region. However, a
wide composition range in the InGaAsP material system
can be grown lattice-matched to InP. This wide composi-
tion range allows a wide range in the bandgap and con-
duction-band offset to be implemented. As a consequence,
the proper current-limiting injection at the cathode can be
designed. Theoretical investigations predict a significant
improvement in DC-to-RF conversion efciencies at W-
band and D-band frequencies [33,47] while preserving the
higher frequency limit of InP compared to GaAs.
In summary, different schemes for accelerating elec-
trons and transferring them faster to the upper valleys
can be incorporated into a device design. Shorter transfer
times reduce the dead space, improve the DC-to-RF con-
version efciency, and allow for a shorter active region to
achieve a higher operating frequency. However, a reduced
dead space effectively increases the transit time and ac-
tually lowers the optimum operating frequency f
op
for an
active region of the same length l. Finite valley transfer
times still impose some physical upper frequency limit.
Additionally, if electrons gain too much energy and/or the
active region is too short, the long energy relaxation time
prevents the electrons from losing enough energy to trans-
fer back to the lower valley. No domains can form in such a
structure, the dynamic resistance between the two device
terminals remains positive for all frequencies, and no RF
power is generated.
6. NOISE
The noise in the output spectrum of an oscillator consists
of uctuations in RF amplitude (AM noise) and oscillation
frequency (FM noise). However, FM noise generally dom-
inates in Gunn devices and can be described as effective
frequency modulation Df
rms
versus frequency f
m
off the
oscillation frequency f
o
. It corresponds to the noise-to-
carrier ratio N/C|
FM
as, for example, seen on a spectrum
analyzer
N
C
FM
Df
2
rms
2f
2
m
13
To compare the noise performance of different oscillators,
including those with other two-terminal devices on a more
equitable basis, the FM noise measure M [48] is more
appropriate
M
Df
2
rms
Q
2
f
2
o
kT
0
B
P
RF
14
(a) (b)
E
v
E
F
E
v
E
c
E
c
E
c E
c
E
F
V
0
V
0
+
+
n
+
n
n
+
n
+
n
n
+
GaAIAs GaAs
GaAs or
GaAIAs
GalnAs or
GalnAsP InP InP
Figure 19. Band diagram of isotype heterojunctions in GaAs/AlGaAs and InP/InGaAsP at zero
bias. (After Friscourt et al. [33].)
where T
0
is the absolute temperature and B is the mea-
surement bandwidth. The loaded Q factor of the oscillator
circuit is determined in a waveguide setup as illustrated
in Fig. 20a. The sweep oscillator injects a signal at the
power level P
i
into the oscillator under test (OUT), and the
maximum continuous frequency range Df
s
over which the
OUT remains injection-locked with the sweep oscillator is
determined [48]:
Q
2f
o
Df
s
P
i
P
RF
_ _
1=2
15
An alternative is also shown in Fig. 20b. The signal from
the OUT is reected at a tunable low-loss short and in-
jected back as P
i
through the coupler into the OUT. If the
position of the short is moved by more than half of the
guide wavelength l
g
, the oscillation frequency continuous-
ly changes from a lower to an upper limit. This maximum
tuning range (i.e., the self-injection locking range) is now
Df
s
in Eq. (15).
Thermal noise of electrons is the dominant effect in
Gunn devices. If a TED is designed (using subcritical
N
D
lo1 10
12
cm
2
) for and operated in the amplier
mode [1], the small-signal noise measure M approaches
the asymptotic limit M
0
M
0
qD
kjm
diff
jT
0
16
where the ratio of the diffusion coefcient D and the dif-
ferential mobility m
diff
is the crucial factor. InP shows an
advantage of 142/72 (approximately 3 dB; see Table 2) over
GaAs. If an equivalent differential mobility m
eff
is intro-
duced for the oscillator mode (critical N
D
l41 10
12
cm
2
)
with the device conductance per unit area G
D
m
eff
jG
D
j
qN
D
17
the large-signal noise measure M can be dened as
M
qD
km
eff
T
0
18
The noise performance of Gunn devices near the carrier is
dominated by icker noise components with typical corner
frequencies in the range of 100kHz1 MHz. As predicted
by Eq. (13), the phase noise decreases by 20 dB per de-
cade at higher off-carrier frequencies f
m
. Table 3 summa-
rizes typical results from GaAs and InP Gunn devices
[24,29,30,39,41,44,49].
Although Eq. (18) predicts lower values of M for oscil-
lators with InP Gunn devices, experimental results indi-
cate little difference between the two. Figure 21 [50]
compares the FM noise measure M of Gunn devices with
that of other two-terminal devices in the frequency range
of 75155 GHz. This gure highlights the low-noise
Bias supply
Sweep oscillator
Isolator
Precision attenuator
Power meter
Precision
attenuator
Frequency
meter
10 dB coupler
10 dB coupler 20 dB coupler
20 dB coupler
Oscillator under test (OUT)
Tunable
back short
Tunable short
Power meter
Variable attenuator
Harmonic mixer
Spectrum analyzer
Waveguide
Coaxial cable
(a)
(b)
P
RF
P
i
+
Figure 20. Waveguide test setup to determine the injection locking range Df
s
and Q factor of a
transferred-electron oscillator: (a) using injection locking with a sweep oscillator; (b) using self-
injection locking.
characteristics of TEDs, where the large-signal FM noise
measure of both GaAs and InP Gunn devices typically re-
mains below 25 dB [11,24,30,49,51]. Some InP devices
with current-limiting contacts show excess icker noise
components near the carrier frequency f
o
. Oscillators with
Gunn devices in a second-harmonic mode yield lower
values for the phase noise, but they yield similar values
for the noise measure because much higher Q values
are achieved in a circuit without a resistive load at the
fundamental frequency.
7. FUTURE TRENDS
7.1. More Power and Higher Operating Frequencies with
INP TEDs
Fundamental-mode operation up to 165 GHz was demon-
strated with InP Gunn devices. Therefore, InP or InP with
heterojunction barriers are promising material systems
not only for device structures that exhibit improved per-
formance in the fundamental mode at frequencies around
or above 100GHz but also for other device structures that
generate significant RF power levels up to submillimeter-
wave frequencies in a second-harmonic mode. In all these
devices, proper heat management is one of the important
factors, and employment of the appropriate diamond heat-
sink technologies is mandatory for maximum RF output
power as well as reliable long-term operation. Power-com-
bining techniques [28,52] can be employed to increase the
available RF power levels. As examples, four InP Gunn
devices in a power-combiner circuit delivered an RF power
of 260 mW to the load in CWoperation at 98.6GHz [11,53],
and two devices each on diamond heatsinks yielded more
than 300mW at 103GHz [42] or more than 125mW at
152 GHz [41].
Structures with a at-doping prole, ohmic contacts
at cathode and anode, but with a doping notch near the
cathode, were originally investigated for low-noise micro-
wave and millimeter-wave amplier applications [11], but
have now emerged as quite promising candidates for ef-
cient second-harmonic power extraction at J-band (220
325 GHz) frequencies and above [39,45]. Figure 22a com-
pares the predicted RF performance of two different dop-
ing proles in the active region of InP Gunn devices in a
second-harmonic mode as shown in Fig. 22b [45]. In all
simulations, the predicted operating temperatures were
kept similar and below 420K to ensure a fair comparison
and, much more importantly, reliable long-term operation
of fabricated devices on diamond heatsinks. The improve-
ments are even more pronounced at frequencies above
300 GHz, where initial Monte Carlo simulation results in-
dicate an improvement in RF output power by a factor of
23 compared to structures with a graded-doping prole.
RF power levels of more than 7 mW at 360GHz and
4.7 mW at 500 GHz are predicted for structures with a
doping notch as opposed to 5 and 2.6 mW, respectively, for
structures with a graded-doping prole at the same fre-
quencies [45]. These performance improvements are at-
tributed to a faster transfer of electrons to the upper
valleys near the cathode and, as a comparison in Fig. 23
indicates, a rather at electric eld prole throughout
most of the active region [39,45] of the structure with a
notch in the doping prole.
7.2. New Material Systems
Wide-bandgap materials such as GaN and AlGaN are also
regarded as promising for high-power Gunn devices.
Favorable material parameters, for example, are higher
critical electric elds for avalanche breakdown, higher
thermal conductivities, higher permissible operating
temperatures, and expected higher carrier drift velo-
cities. However, high threshold electric elds E
th
and
Table 3. Phase Noise of Free-Running Oscillators using
GaAs or InP Millimeter-Wave Gunn Devices
Material
System
Phase
Noise
(dBc/Hz)
Off-Carrier
Frequency
(kHz)
Oscillation
Frequency
(GHz)
RF Output
Power
(mW) Ref.
GaAs o80
a
100 77 440
a
29
GaAs 70 100 80 55 24
GaAs 100 1,000 80 55 24
GaAs 120 10,000 80 55 24
GaAs 80 100 94 10 49
GaAs 105 1,000 94 10 49
InP 75 100 94 20 49
InP 100 1,000 94 20 49
InP o110 500 81 126 39
InP o110 500 103 180 41
InP o108 500 132 120 30
InP o103 500 151 58 30
InP o94 500 199 19 44
a
Reported as typical value, corresponding RF output power not mentioned.
75 95 115 135 155
10
20
30
40
50
Frequency (GHz)
N
o
i
s
e

m
e
a
s
u
r
e

(
d
B
)
Small signal
Large signal
Si IMPATT diodes
GaAs IMPATT
& MITATT diodes
GaAs TUNNETT
diodes
InP Gunn devices
Figure 21. Comparison of the FM noise measure M in free-run-
ning oscillators with different two-terminal devices at millimeter-
wave frequencies of 75155GHz. (After Eisele and Haddad [50].)
consequently higher DC bias voltages may result in severe
thermal limitations of the device performance [54].
Furthermore, major improvements in material quality or
availability and in fabrication technologies are needed
before devices for system applications can be developed.
7.3. Circuit Integration
Waveguide circuits are, in general, rugged, can dissipate
heat easily, and offer high Q values. Therefore, they are
the preferred circuits to obtain the maximum RF output
power. However, they are bulky and in most cases must be
machined, assembled, and tested individually, which pro-
hibits low-cost mass production. Therefore, high-volume
production for system applications such as wireless com-
munication (high-speed data transmission) or collision
avoidance radar in automobiles must be based on quite
different approaches. Both hybrid and monolithic integra-
tion were attempted with Gunn devices. Hybrid micro-
strip-line oscillators with InP Gunn devices exhibited
excellent performance. RF power levels (with correspond-
ing DC-to-RF conversion efciencies) of 52 mW (3.5%) at
94 GHz [25], but also 40 mW (1.4%) at 81 GHz [11,24] and
4200 mW (47.5%) around 35.5GHz [10,35] were mea-
sured and are considered comparable to values from sim-
ilar Gunn devices in waveguide circuits. Hybrid
integration of a GaAs Gunn device for automotive appli-
cations at 77 GHz was also demonstrated [26]. The ther-
mal conductivity of the semiconductor materials InP and
GaAs is rather low (0.68 and 0.46Wcm
1
K
1
, respec-
tively) when compared to metals or diamond
(20 Wcm
1
K
1
at room temperature). Therefore, fully
monolithic integration of low-efciency Gunn devices at
millimeter-wave frequencies encounters severe thermal
Graded doping
profile
Doping profile
with notch
R
F

o
u
t
p
u
t

p
o
w
e
r

(
m
W
)
60
50
40
30
20
10
0
160 180 200 220 240 260 280 300 320
Frequency (GHz)
(a)
>2 10
18
cm
_
3
>2 10
18
cm
_
3
2.3 10
16
cm
_
3
>2 10
18
cm
_
3
>2 10
18
cm
_
3
3 10
16
cm
_
3
1 10
16
cm
_
3
1 10
15
cm
_
3
1.1 m
n
+
n
+
n
+
n
n
+
n
Graded doping profile

1.1 m
0.1 m
Doping profile with notch
(b)
+
+
_
_
Figure 22. Comparison of predicted RF power level from InP Gunn devices with doping proles
optimized for second-harmonic power extraction.
30
10
10
30
50
70
90
110
130
150
0.0 0.3 0.7 1.0 1.3
Graded structure
2
4
6
0
Position (m)
(a)
E
l
e
c
t
r
i
c

f
i
e
l
d

(
k
V
.
c
m
1
)
E
l
e
c
t
r
i
c

f
i
e
l
d

(
k
V
.
c
m
1
)
30
10
10
30
50
70
90
110
130
0.0 0.3 0.7 1.0 1.3
Position (m)
(b)
4
6
2
0
Notch structure
Figure 23. Evolution of electric eld against position x (active region 0.11.2mm) for a graded-
prole (a) and a at-prole with notch (b) TED structure at the fundamental frequency f 120GHz
and ot np/4, n0,2,4,6, during one RF cycle.
limitations. A summary of reports on integration of Gunn
devices at frequencies up to approximately 68 GHz can be
found elsewhere [28].
BIBLIOGRAPHY
1. S. M. Sze, Physics of Semiconductor Devices, 2nd ed., Wiley,
New York, 1981.
2. U. V. Bhapkar and M. S. Shur, Monte Carlo calculation of ve-
locity-eld characteristics of wurtzite GaN, J. Appl. Phys.
82(4):16491655 (1997).
3. S. Krishnamurthy, M. van Schilfgaarde, A. Sher, and
A.-B. Chen, Bandstructure effect on high-eld transport in
GaN and GaAlN, Appl. Phys. Lett. 71(14):19992001 (1997).
4. B. K. Ridley and T. B. Watkins, The possibility of negative
resistance effects in semiconductors, Proc. Phys. Soc. Lond.
78:293304 (1961).
5. C. Hilsum, Transferred electron ampliers and oscillators,
Proc. Inst. Radio Eng. 50:185189 (1962).
6. J. B. Gunn, Microwave oscillation of current in III-V semi-
conductors, Solid-State Commun. 1:8891 (1963).
7. J. B. Gunn, Instabilities of current in III-V semiconductors,
IBM J. Res. Devel. 8:141159 (1964).
8. H. Kroemer, Theory of the Gunn effect, Proc. IEEE 52:1736
(1964).
9. L. Wandinger, mm-Wave InP Gunn devices: Status and
trends, Microwave J. 24(3):7178 (1981).
10. B. Fank, J. Crowley, D. Tringali, and L. Wandinger, Basics
and recent applications of high-efciency millimeter wave InP
Gunn diodes, Proc. 1st Int. Conf. Indium Phosphide and Re-
lated Materials for Advanced Electronic and Optical Devices,
Norman, OK, March 2023, 1989; SPIE 1144:534546 (1989).
11. I. G. Eddison, Indium phosphide and gallium arsenide trans-
ferred-electron devices, in Infrared and Millimeter Waves,
Vol. 11, Millimeter Components and Techniques, Part III,
Academic Press, Orlando, FL, 1984, pp. 159.
12. J. A. Copeland, LSA oscillator-diode theory, J. Appl. Phys.
38:30963101 (1967).
13. P. N. Butcher, Theory of stable domain propagation in the
Gunn effect, Phys. Lett. 19:546547 (1965).
14. J. F. Caldwell and F. E. Rosztoczy, Gallium arsenide Gunn
diodes for millimeter-wave and microwave frequencies, Proc.
4th Int. Symp. GaAs and Related Compounds, Denver, CO,
1972.
15. P. A. Rolland, M. R. Friscourt, D. Lippens, C. Dalle, and
J. L. Nieruchalski, Millimeter wave solid-state power sources,
Proc. Int. Workshop on Millimeter Waves, Rome, April 24,
1986, pp. 125177.
16. R. Kamoua, H. Eisele, and G. I. Haddad, D-band (110
170GHz) InP Gunn devices, Solid-State Electron. 36:1547
1555 (1993).
17. H. Eisele, Selective etching technology for 94-GHz GaAs IMP-
ATT diodes on diamond heatsinks, Solid-State Electron.
32:253257 (1989).
18. H. Eisele and G. I. Haddad, D-band InP Gunn devices with
second-harmonic power extraction up to 290GHz, Electron.
Lett. 30:19501951 (1994).
19. Y. E. Ma, Millimeter-wave active solid-state devices, Millime-
ter Wave Technol. III; SPIE 544:95102 (1985).
20. L. W. Holway and M. G. Adlerstein, Approximate formulas for
the thermal resistance of IMPATT diodes compared with com-
puter calculations, IEEE Trans. Electron Devices ED-24:
156159 (1977).
21. M. A. di Forte-Poisson, C. Brylinski, G. Colomer, D. Osselin,
S. Hersee, J. P. Duchemin, F. Azan, D. Lechevallier, and
J. Lacombe, High-power high-efciency LP-MOCVD InP
Gunn diodes for 94GHz, Electron. Lett. 20:10611062 (1984).
22. M. A. di Forte-Poisson, C. Brylinski, N. Proust, D. Pons, M.
Secoue, P. Arse`ne Henry, M. Calligaro, and J. Lacombe, LP-
MOCVD InP Gunn devices developed for 94GHz millimeter
range operation, Proc. 1st Int. Conf. Indium Phosphide and
Related Materials for Advanced Electronic and Optical De-
vices, Norman, OK, March 2023, 1989; SPIE 1144:551560
(1989).
23. J. D. Crowley, C. Hang, R. E. Dalrymple, D. R. Tringali, F. B.
Fank, L. Wandinger, and H. B. Wallace, 140GHz indium
phosphide Gunn diode, Electron. Lett. 30:499500 (1994).
24. D. C. Smith, T. J. Simmons, and M. R. B. Jones, A comparison
of the performance of millimeter-wave semiconductor oscilla-
tor devices and circuits, Digest of the 1983 IEEE MTT-S Int.
Microwave Symp., May 31June 3, 1983, Boston, pp. 127
129.
25. C. Kim, C. Dunnrowicz, J. Crowley, B. Fank, and L. Wand-
inger, Millimeter-wave tunable microstrip InP Gunn oscilla-
tors, Microwave J. 32(4):91102 (1989).
26. N. Priestley, K. Newsome, I. Dale, and P. Norton, A Gunn di-
ode based surface mount 77GHz oscillator for automotive ap-
plications, 2002 IEEE MTT-S Int. Microwave Symp. Digest,
June 27, 2002, Seattle, pp. 18631866.
27. H. J. Kuno, IMPATT devices for generation of millimeter
waves, in K. Button, ed., Infrared and Millimeter Waves,
Academic Press, New York, Vol. 1, 1979, Chapter 2.
28. K. Chang, Handbook of Microwave and Optical Components,
Vol. 2, Wiley, New York, 1990.
29. I. Dale, J. R. P. Stephens, and J. Bird, Fundamental-mode
graded-gap Gunn diode operation at 77 and 84GHz, Proc.
Microwaves 94, London, Oct. 2527, 1994, pp. 248251.
30. H. Eisele and G. I. Haddad, High-performance InP Gunn de-
vices for fundamental-mode operation in D-band (110
170GHz), IEEE Microwave Guided Wave Lett. MGWL-
5:385387 (1995).
31. S. J. J. Teng and R. E. Goldwasser, High-performance second-
harmonic operation W-band GaAs Gunn diodes, IEEE Elec-
tron Device Lett. EDL-10:412414 (1989).
32. A. Rydberg, High efciency and output power from second-
and third-harmonic millimeter-wave InP-TED oscillators at
frequencies above 170GHz, IEEE Electron Device Lett. EDL-
11:439441 (1990).
33. M.-R. Friscourt, P.-A. Rolland, and M. Pernisek, Heterojunc-
tion cathode contact transferred-electron oscillators, IEEE
Electron Device Lett. EDL-6:497499 (1985).
34. B. Fank, J. Crowley, and C. Hang, InP Gunn diode sources,
Millimeter Wave Technol. III; SPIE 544:2228 (1985).
35. J. D. Crowley, R. E. Dalrymple, C. Hang, D. R. Tringali,
F. B. Fank, and L. Wandinger, InP Gunn diodes serve milli-
meter-wave applications, Microwaves RF 33(3):143146
(1994).
36. J. Ondria and R. L. Ross, Improved performance of funda-
mental and second-harmonic mmw oscillators through active
doping concentration contouring, 1987 IEEE MTT-S Digest,
1987, pp. 977980.
37. T. Hisatsugu, S. Yamamura, S. Yokogawa, Y. Hirachi, A.
Miura, and A. Shibatomi, A 160-mW 65-GHz Gunn diode,
Technical Digest 1976 Int. Electron Device Metting, Washing-
ton, DC, Dec. 68, 1976, pp. 9497.
38. K. Akamatsu, A. Yokohata, S. Kato, N. Ohkuba, and M. Ohm-
ori, High-efciency millimeter-wave GaAs Gunn diodes oper-
ating in the fundamental mode, Digest 19th Int. Conf.
Infrared and Millimeter Waves, Sendai, Japan, Oct. 1720,
1994, pp. 8990.
39. H. Eisele and R. Kamoua, InP Gunn devices for low-noise
and high-performance oscillators in the 80400GHz
frequency range, 2001 IEEE 19th Int. Conf. Terahertz
Electronics Proc., Charlottesville, Virginia, VA, Oct. 1516,
2001.
40. H. Eisele, Conventional and novel approaches to RF
power generation with two-terminal devices at terahertz
frequencies, invited paper, 2002 IEEE 10th Int. Conf. Tera-
hertz Electronics Proc., Cambridge, UK, Sept. 910, 2002,
pp. 1318.
41. H. Eisele, A. Rydberg, and G. I. Haddad, Recent advances in
the performance of InP Gunn devices and GaAs TUNNETT
diodes for the 100300-GHz frequency range and above, Spe-
cial Issue on Terahertz Electronics, IEEE Trans. Microwave
Theory Tech. MTT-48(4):626631 (2000).
42. H. Eisele and G. I. Haddad, Efcient power combining with D-
band (110170GHz) InP Gunn devices in fundamental-mode
operation, IEEE Microwave Guided Wave Lett. MGWL-8:24
26 (1998).
43. H. Eisele, Second-harmonic power extraction from InP
Gunn devices with more than 1 mW in 200320GHz
frequency range, Electron. Lett. 34(25):24122413
(1998).
44. H. Eisele, High performance InP Gunn devices with 34mWat
193GHz, Electron. Lett. 38(16):923924 (2002).
45. R. Kamoua and H. Eisele, Theoretical and experimental com-
parison of optimized doping proles for high-performance
InP Gunn devices at 220500GHz, 2003 IEEE MTT-S Int.
Microwave Symp. Digest, Philadelphia, June 813, 2003,
pp. 907910.
46. N. R. Couch, H. Spooner, P. H. Beton, M. J. Kelly, M. E. Lee,
P. K. Rees, and T. M. Kerr, High-performance, graded AlGaAs
injector, GaAs Gunn diodes at 94 GHz, IEEE Electron Device
Lett. EDL-10:288290 (1989).
47. R. Kamoua, Heterojunction D-band (110170GHz) InP Gunn
devices, Solid-State Electron. 37:269274 (1994).
48. K. Kurokawa, Noise in synchronized oscillators, IEEE Trans.
Microwave Theory Tech. MTT-16:234240 (1968).
49. C. Dalle, P. A. Rolland, and G. Lleti, Flat doping prole dou-
ble-drift silicon IMPATT for reliable cw high-power high-ef-
ciency generation in the 94-GHz-window, IEEE Trans.
Electron Devices ED-37:227236 (1990).
50. H. Eisele and G. I. Haddad, Two-terminal millimeter-wave
sources, IEEE Trans. Microwave Theory Tech. MTT-46:739
746 (1998).
51. D. M. Brookbanks, A. M. Howard, and M. R. B. Jones, Si
IMPATTs exhibit low noise at mm-waves, Microwaves RF
22(2):6872 (1983).
52. K. Chang and C. Sun, Millimeter-wave power-combining
techniques, IEEE Trans. Microwave Theory Tech. MTT-
31:91107 (1983).
53. J. J. Sowers, J. D. Crowley, and F. B. Fank, CW InP
Gunn diode power combining at 90 GHz, Digest 1982 IEEE
MTT-S Int. Microwave Symp., Dallas, TX, June 1517, 1982,
pp. 503505.
54. R. Kamoua, Y. Zhu, and Y. Corcoran, Potential of GaN Gunn
devices for high power generation above 200GHz, Proc. Conf.
Wide-Bandgap Electronic Devices, April 2427, 2000,
Vol. MRS 622, pp. 261266.
FURTHER READING
E. L. Holzman and R. S. Roberston, Solid-State Microwave Power
Oscillator Design, Artech House, Boston, 1992.
S. M. Sze, High-Speed Semiconductor Devices, Wiley, New York,
1990.
S. M. Sze, Modern Semiconductor Device Physics, Wiley, New
York, 1998.
K. S. Yngvesson, Microwave Semiconductor Devices, Kluwer
Academic Publishers, Boston, 1991.
J. G. Webster, ed., Encyclopedia of Electrical and Electronics
Engineering, Wiley, New York, 1999.
R. E. Miles, P. Harrison, and D. Lippens, Terahertz Sources and
Systems, NATO Science Series, Series II: Mathematics, Phys-
ics, and Chemistry, Vol. 27, Kluwer Academic Publishers,
Dordrecht, 2001.
T. Pearsall, InP: Properties, Processing, and Applications, IEE,
Stevenage, UK, 1999.
GYRATORS
DOUGLAS R. FREY
Lehigh University
1. DEFINITION
A gyrator is a nonreciprocal electrical network. It is capa-
ble of transforming signals or energy represented in terms
of one electrical quantity, such as voltage or magnetic
eld, to another electrical quantity that may be of similar
type or of a complementary type, such as current or elec-
tric eld. Such networks are quite useful in electronic sys-
tems, since one often wishes to design systems with a
limited set of component types or with restrictions regard-
ing certain physical parameters.
2. HISTORICAL USAGE
Tellegen rst proposed the idea of a gyrator in his original
work in 1948 [1]. In this paper he explained that resistors,
capacitors, inductors, and ideal transformers were the
four basic circuit building blocks. However, these elements
are all reciprocal and could, therefore, only be expected to
go into the creation of reciprocal networks. Reciprocal net-
works are those networks whose impedance (or admit-
tance) matrices are symmetrical. In order to realize
nonreciprocal networks, one would need a nonreciprocal
building block. Tellegen proposed such a network, calling
it a gyrator. This name was given because the equations
produced for an electrical network with a gyrator were
identical to those of a mechanical gyrostatic network. As
time went on other researchers [25] picked up on the idea
and began to look for circuit realizations for this abstract
functional block. In addition, over the years other systems,
such as microwave circulators, have been recognized as
being analogous to gyrators, which helps in the under-
standing of these systems.
1874 GYRATORS
3. THE BASICS OF TWO-PORT GYRATORS
3.1. Introduction to Two-Ports and Passivity
A gyrator is a special type of electrical two-port network.
Electrical two-port networks are any circuits where one
can identify two ports, or simply, two pairs of nodes to
which one might consider the connection of two pairs of
wires. One of the nodes at each port may be in common
for example, ground may be common to both ports. Addi-
tionally, a bipolar transistor may be considered a two-port,
where port 1 is the baseemitter node pair, and port 2 is
the collector-emitter node pair.
Except for trivial cases, every two-port possesses a
mathematical description relating the port voltages, V
1
and V
2
, and associated port currents, I
1
and I
2
. Figure 1
shows the standard reference labeling for the voltages and
currents of a two-port. Notice that the port currents are
dened as owing into the voltage reference for each
port. These sign conventions make the definition of power
regarding a two-port more precise. Specifically, the power
as a function of time P(t) delivered to a two-port is given by
P(t) V
1
(t)I
1
(t) V
2
(t)I
2
(t), which is analogous to the def-
inition in a one-portthat is, a two-terminal element. A
two-port network is lossless when the average power P(t)
delivered to the network is zero. A network is called pas-
sive if the average power delivered to the network is pos-
itive. Active networks are those networks where P(t) is
negative on average. In general, the power delivered to a
network can be positive or negative instantaneously, re-
gardless of its passivity. For example, a capacitor in a res-
onant circuit alternately sinks and sources power
instantaneously, despite its lossless average power con-
sumption. Power can be dened just as easily in the fre-
quency domain, again in a way analogous to one-ports.
The Fourier transform of power P(o) as a function of fre-
quency is given by P(o) V
1
(o)I
1
(o)* V
2
(o)I
2
(o)*, where
the * indicates complex conjugation.
The general description for a two-port is given by a re-
lation between its port voltages and currents. One such
description is the so-called y-parameter model given by
I
1
I
2
y
11
y
12
y
21
y
22
V
1
V
2
)I YV 1
Equation (1) denes the port current and port voltage
vectors, I and V, respectively, in addition to the y-param-
eter matrix, Y. By inverting this vector equation, one re-
lates the port voltages to the port currents with z
parameters. Specifically
V
1
V
2
z
11
z
12
z
21
z
22
I
1
I
2
)V ZI 2
where Z is the z-parameter matrix for the two-port. Using
models of the type shown in Eqs. (1) and (2), two-ports can
be compared by their two-port parametersfor example, y
parameters. Two-ports characterized by symmetric y-pa-
rameter matrices (equivalently z-parameter matrices) are
called reciprocal. Networks possessing this reciprocity are
not necessarily lossless, but can always be realized with
reciprocal physical elements. Note that one-ports (two ter-
minal elements) are always reciprocal. The concept of a
two-port can be extended in an obvious way to N-ports by
considering the voltage-current relationship measured us-
ing N pairs of terminals or, equivalently, N-ports.
3.2. Mathematical Two-Port Definition of a Gyrator
In the context of this discussion, a gyrator is simply a
special case of a linear two-port. While there are many
possible two-port descriptions, the most common way of
writing the basic equations relating the port parameters
in a gyrator is as follows:
I
2
gV
1
; I
1
gV
2
3
Using these equations it is simple to write the y-parameter
two-port description for a gyrator as
I
1
I
2
0 g
g 0
V
1
V
2
)I YV 4
This suggests that a gyrator can be implemented with
voltage-controlled current sources, having gains of g and
g, respectively. By inverting the relations in Eq. (3), one
obtains a gyrator formulation based on current-controlled
voltage sources. Specifically
V
V
1
V
2
0 r
r 0
I
1
I
2
ZI; r 1=g 5
where the z-parameter matrix Z is just the inverse of the
y-parameter matrix Y. Since it is most convenient to real-
ize practical voltage-controlled current source networks,
as opposed to current-controlled voltage source networks,
the formulation in Eq. (4) is generally preferred. For the-
oretical purposes, of course, both formulations are useful.
Using Eq. (4) it is simple to show that a gyrator is a loss-
less electrical network. Specifically, we obtain
PV
T
I V
1
V
2
I
1
I
2
V
1
V
2
0 g
g 0
V
1
V
2
V
2
V
1
V
1
V
2
0
6
The fact that gyrators are, in theory, lossless makes them
attractive in lter synthesis, and this will become clear
later.
I
1
V
1
+
V
2
+
Two-port
network
I
2
Figure 1. Basic two-port model.
GYRATORS 1875
3.3. Properties of Gyrators
Gyrators possess several properties that make these cir-
cuits interesting for use in electronics. A rst property is
that these two-ports are not reciprocal networks, since
their y-parameter matrices are not symmetric. In fact,
these matrices are skew symmetric. It is well known to
circuit theorists that nonreciprocal networks cannot be
realized with only passive componentsthat is resistors,
capacitors, and inductors. This means that gyrators are
strictly active networks that must, therefore, be realized
with active components, such as transistors or operational
ampliers. Two-port gyrators have been given their own
circuit symbol, which is shown in Fig. 2. The gyration
constant, g is built into the symbol.
Perhaps the most important property of a gyrator is its
ability to transform admittances into impedances. Specif-
ically, when an admittance is connected to one port of a
gyrator, the impedance looking into the other port is ex-
actly a scaled version of that admittance. The derivation
can be accomplished with the help of Fig. 3, where Z
load
is
the impedance attached to port 2, Y
load
is its reciprocal
that is, the admittance attached to port 2and Z
input
is
the impedance seen looking into port 1. We have
Z
load
V
2
I
2
;
Z
input
V
1
I
1
I
2
=g
gV
2
1
g
2
1
Z
load
1
g
2
Y
load
7
The gyration constant g determines the scale factor, but
the nature of the input impedance is determined by the
admittance attached to port 2. Therefore, if a capacitor is
attached to port 2, then we have
Z
input
1
g
2
sCsL
eq
; L
eq
C
g
2
8
This simple relation explains the vast majority of the gy-
rators popularity in electronic design. It shows that a ca-
pacitor can be used to replace an inductor in a circuit with
the help of a gyrator. Since inductors are rarely desirable
in electronic circuits operating below about 1GH
z
, this
idea is quite appealing. Capacitors and gyrators are con-
veniently realized within integrated circuits.
Filter synthesis based on the inductor simulation al-
ready described is usually done by starting with an RLC
(resistor, inductor, capacitor) prototype, and replacing the
inductors with capacitorgyrator combinations. Consider
the following example of a very simple second-order band-
pass lter shown in Fig. 4. After replacing the inductor
with a gyrator/capacitor combination, the lter is realized
solely using RC (resistor, capacitor) passive elements, as
shown in Fig. 5. Some original related work appears in
Refs. 6 and 7. Furthermore, the lter can be tuned elec-
tronically if the gyration constant can be varied electron-
ically. An electronically tunable lter using gyrators will
be shown later.
A byproduct of the property under discussion is that
series and parallel circuits may be interchanged with the
help of a gyrator. Suppose one port, say, port 2, of a gyrator
is loaded with a parallel combination of elements. The ad-
mittance of this combination is the sum of the admittances
of each of the elements. At the other port, port 1, the input
impedance will be a scaled version of this admittance;
hence, a sum of impedances. Since the composite input
impedance seen at port 1 is given by a sum of impedances,
it must be equivalent to a series combination of elements.
Therefore, the gyrator converts a parallel network into a
series network. Using similar logic, it becomes clear that
a series network connected to port 2 will be reected as a
parallel network looking into port 1 (these results are
summarized next):
Y
load
N
k1
Y
k
)Z
input
Y
load
g
2
N
k1
Y
k
g
2
N
k 1
Z
ink
Z
load
N
k1
Z
k
)Y
input
1
Y
load
=g
2
g
2
Z
load
N
k1
g
2
Z
k
N
k 1
Y
ink
9
4. CIRCUIT REALIZATIONS FOR GYRATORS
4.1. The 2G
m
Cell Realization
The realization of gyrators in electronic form is quite sim-
ple; however, as usual, different circuit realizations are
I
2
I
1
g
V
2
+
V
1
+
Figure 2. Electrical circuit symbol for a gyrator.

I
2
I
1
Z
load
Z
input
V
2
+
V
1
+
g
Figure 3. Reecting load impedance with a gyrator.
R
+
L V
in
+
V
out
C
Figure 4. Bandpass lter.
1876 GYRATORS
preferable to others, depending on the application. To be-
gin, consider the simplest generic realization consisting of
a pair of transconductance ampliers, as shown in Fig. 6.
Each transconductance amplier is assumed to have in-
nite input and output impedance, with an output current
equal to the transconductance G
m
g, times the input
voltage applied to the and terminals. The circuit
shown in Fig. 6 satises the basic two-port relations for a
gyrator, given by Eq. (3).
The circuit of Fig. 6 does not implement the most gen-
eral form of a gyrator, since both ports of the gyrator re-
alization in the gure have ground in common. Therefore,
only ground-referenced impedances may be transformed
as described elsewhere in this article. This limitation
stems from the fact that the transconductors in Fig. 6
have single-ended outputs. If differential input/differential
output transconductors are used, then a general oating
gyrator realization is createdthat is, a gyrator whose
ports need not be referenced in any way to ground.
Unfortunately, the realization of fully differential gyra-
tors is not easy. In general, this realization requires more
circuitry and the management of common-mode signals.
Figure 7 shows how a oating gyrator can be realized us-
ing single-ended transconductors; however, this circuit
suffers from common-mode problems. An analysis of this
structure yields the following results:
I
2
gV
1
; I
2
gV
1
; I
1
gV
2
; I
1
gV
2
)I
2
I
2
I
2
gV
1
V
1
gV
1
I
1
I
1
I
1
gV
2
V
2
gV
2
10
The crucial assumption embodied in Eq. (10) is that
andcurrents are equal and opposite one another. This
can happen only if the andinput voltages at the ports
are exactly equal and opposite. Since this special case
cannot be relied on in practice, additional circuitry must
be added to deal with any common-mode current compo-
nent. To do this compensation, more transconductors can
be added to process the average voltage at each port. As
one might expect, this additional circuitry is an unwel-
come addition to the design. As a result, this idea is rarely
found in practical designs.
There is an alternative for the simulation of a oating
inductor using gyrators. A pair of gyrators is used with a
grounded impedance Z
load
as shown in Fig. 8. The equa-
tions describing this system are given by
I
2
gV
1
; I
1
gV
2
; I
4
gV
3
; I
3
gV
4
V
load
V
2
V
3
; I
load
I
2
I
3
V
input
V
1
V
4
; I
input
I
1
I
4
Z
input
V
input
I
input
V
1
V
4
I
1
I
2
=g I
3
=g
gV
2
1
g
2
I
2
I
3
V
2
1
g
2
I
load
V
load
1
g
2
Z
load
1
g
2
Y
load
11
Clearly, if a grounded capacitor is used as the grounded
load in Fig. 8, then a oating simulated inductor is real-
ized. The obvious benet of this realization is that it re-
quires only single-ended transconductance ampliers,
congured as in Fig. 6, and a grounded internal load to
obtain a oating input port. Observe that no common-
mode problems exist with this realization since common-
mode signals at the input port cause canceling currents at
the grounded port. Even in practice, with unmatched gy-
rators, there is a negative-feedback effect regarding com-
mon-mode errors that is highly desirable. The network of
Fig. 8 is the preferred realization of oating inductors us-
ing capacitors and gyrators. Reference 8 represents some
original work in this area.
R
V
in
+
V
out
+
C
L
C
g
Figure 5. Bandpass lter with inductor replaced by gyrator/
capacitor.
+
+

G
m
V
1
V
2
+
+
G
m
G
m
e
e
Figure 6. Realization of gyrator using transconductance ampli-
ers.
+
G
m
I
1
I
1+
V
1+
V
1
V
2
V
1
+
+
G
m
I
2
I
2+
V
2+
V
2
+
G
m
I
1
G
m
I
2
Figure 7. Floating-gyrator implementation.
V
1
V
2
I
2
I
3
I
1
I
input
V
4
I
4
V
3
Z
load
+ V
input
g g
Figure 8. Floating-gyrator realization.

GYRATORS 1877
4.2. Realization with Operational Ampliers
Gyrators may be realized with operational ampliers;
however, modications must be made to account for the
fact that these are voltage-controlled voltage sources. A
voltage-controlled current source (VCCS) may be created
using an op amp, as is well known, using the circuit of
Fig. 9. A gyrator can then be realized with a second VCCS
preceded by an inverter, recalling that the gains in differ-
ent directions have opposite signs. A more clever variation
of this idea is shown in Fig. 10, where only two op amps
are required to implement the entire gyrator. Of course,
this gyrator is ground-referenced as, for example, is the
one in Fig. 6. Floating-gyrator structures can be imple-
mented using the ideas stated, and similarly, a oating
inductor may be synthesized via a pair of ground-refer-
enced gyrators implemented with op amps and a grounded
capacitor. References 6 and 7 give more discussion of the
topics in the last two sections.
4.3. Other Realizations
Gyrators may be realized with any active circuitry that
can implement either a VCCS or a CCVS (current-con-
trolled voltage source). For example, a simple transistor-
level realization for a gyrator appears in Fig. 11. Q
1
Q
3
create a rst transconductance amplier, and Q
4
imple-
ments an inverting transconductance amplier. The sig-
nal levels must be restricted with such an implementation
due to the nonlinearity of the transistor junctions.
Alternatively, the transconductance ampliers consti-
tuting the gyrator may be implemented using operational
transconductance ampliers (OTAs). OTAs are essentially
bipolar differential pairs loaded with current sources in
such a way as to create a nearly ideal transconductance
amplier. An example of an integrated version of an OTA
is the LM3080 integrated circuit manufactured by Na-
tional Semiconductor Corp. of Santa Clara, CA. An impor-
tant feature of OTAs is that their transconductance can be
tuned over a wide range by varying a control current. As a
result, gyrators made using OTAs are electronically tun-
able. This is quite desirable in applications where one
wishes to electronically tune a lter. An example of this
capability is given later.
Of course, there are limitations imposed by the use
of OTAs since these circuits are not ideal in practice.
Specifically, they suffer from nite input and output im-
pedance, and these impedances vary as a function of the
transconductance. As a result, the tuning range of OTA
tuned lters can be limited. Furthermore, at higher fre-
quencies the complexity of OTAs introduces an unwanted
phase shift, which degrades the behavior of the gyrator, as
well as compromising the usable tuning range. Finally,
these circuits become quite nonlinear for inputs above a
few tens of millivolts, which limits the dynamic range of
the resulting lters.
5. FILTER REALIZATION USING GYRATORS
This section considers the general problem of lter syn-
thesis based on gyrators. Basically, synthesis with gyra-
tors involves either the substitution of inductors, in the
practical case, or the partition of state equations.
5.1. Replacement of Inductors in Ladder Networks
The replacement of inductors in lters has already been
implied. In this section, the idea is generalized. Consider
the case where a prototype passive RLC lter has been
specied. This is usually done, starting from a lter spec-
ication, using lter design tables or software to produce
one of a variety of passive lter structures. Special cases
will be considered in the discussion that follows. All other
cases are obvious variations.
R
R R
I = V
1
/R
+
V
1
+
R
Figure 9. VCCS realization using an op amp.
V
2
R
R
R
R
R R R
I
2
I
1
+
+
V
2
+
Figure 10. Gyrator realization using op amps.

+
+

I
1
I
2
V
1 V
2
V
ref Q
2
Q
1
Q
3
Q
4
Figure 11. Transistor realization of a gyrator.
1878 GYRATORS
The rst case considered is that of a doubly terminated
highpass RLC ladder network, shown schematically in
Fig. 12. It is desirable to replace the grounded inductors
with gyrator/capacitor combinations. Given the prototype
design values for the inductors L
k
, for k 1 through N, one
simply replaces each grounded inductor with one gyrator
terminated with a capacitance C
k
, given by the formula
C
k
g
2
k
L
k
12
where g
k
represents the gyration constant for the kth gy-
rator. In practice, all gyration constants might be chosen
to be equal for reasons of simplicity in the circuit design,
and possibly for the purpose of optimizing noise and dis-
tortion performance. The resulting gyrator-based imple-
mentation is now an active lter containing 2N capacitors
and N gyrators.
The doubly terminated lowpass lter structure ob-
tained by swapping the positions of the inductors and
the capacitors in the highpass lter of Fig. 12 is the dual-
lter network to that shown previously. Again, each of the
inductors can be replaced by a gyrator loaded by a capac-
itor, whose value is computed using Eq. (12). However,
this time the gyrator structures will have to be the oating
implementation described later. The complexity, in prin-
ciple, of the nal realization will equal that of the highpass
example; however, each of the gyrators will require twice
as much circuitry for its realization, therefore, the nal
circuit implementation will require considerably more cir-
cuitry as N gets larger.
A more complex lter variation is that of a bandpass
RLC lter. A doubly terminated bandpass lter can be
created starting from the highpass prototype of Fig. 12 by
replacing each grounded inductor with a grounded paral-
lel combination of a capacitor and inductor, and each se-
ries capacitor with a series combination of a capacitor and
an inductor. The resulting lter is of order 4N, instead of
2N, as it must be to realize a bandpass equivalent. Again,
gyrators may be used to replace each inductor, using
methods like Eq. (12). Half of the inductors may be re-
placed using simple ground referenced gyrators, while the
rest must be realized using the oating version.
An interesting nal variation of this idea is demon-
strated using the notch (bandstop) lter of Fig. 13. Here,
the grounded inductor can be replaced by a gyrator and a
grounded capacitor as outlined above. An alternative ap-
proach is to replace the grounded series LC (inductor, ca-
pacitor) combination using a gyrator loaded by a grounded
parallel LC combination as shown in Fig. 14. Then replace
the grounded inductor with another gyrator and a grounded
capacitor. This is shown in Fig. 15. While this idea may
seem to be wasteful in terms of component count, it shows
how to exploit the property described elsewhere in this
article regarding the conversion of series impedances com-
binations to parallel impedance combinations. In addition,
this circuit has practical value since the pair of gyrators
now allows two parameters to be tuned in this active lter
realization. Specifically, the lter design equations are
given as follows:
V
out
V
e
Hs
s
2
o
2
0
s
2
o
0
_
Qs o
2
0
;
o
0
LC
p ; Q
1
R
L
C
_
Cg
2
1
L
p
g
2
1
g
2
2
C
L
;
L
1
g
2
1
C
p
)o
0
g
2
C
L
C
p
_
Q
g
2
Rg
2
1
C
p
C
L
13
Using this result, it is easy to see that both the notch fre-
quency o
0
and the sharpness of the notch, proportional to
Q, can be controlled independently using only the gyration
constants, g
1
and g
2
. Using an OTA implementation of the
gyrators as discussed elsewhere, this tuning is relatively
simple.
5.2. Synthesis Based on Gyrators
Clearly, the inductor replacement strategy described
above could be applied in reverse to replace capacitors
with inductors; however, this is not a good option in prac-
tical cases. This is because capacitors are easier to realize
at frequencies below 1 GH
z
, and capacitors are of gener-
ally higher quality than inductors in this range of
C
1
R
0
V
S
C
N
+
R
0
L
N
L
1
V
out
Figure 12. Doubly terminated passive ladder highpass lter.
C
V
in
V
out
R
L
Figure 13. RLC notch lter.
R
V
in
V
out
+
L
p
C
p
g
1
Figure 14. Gyrator-based parallel-to-series conversion.
GYRATORS 1879
frequencies. Although inductors may be of greater interest
at very high frequencies, the use of gyrators at very high
frequencies is limited by the nonideal behavior of the ac-
tive circuitry used to realize them. It is of interest to note
here that, as with any active circuit, there is excess phase
shift introduced by the transconductors at high frequen-
cies, causing the gyrators to take on complex gyration
constants at high frequencies, as suggested later in the
article. Furthermore, the nite output impedance of the
transconductors limits there available DC gain. The net
result of these effects is to cause simulated inductors to
exhibit a reduced Q at both the low and high end of the
frequency spectrum. In some situations this may intro-
duce instability; however, this can be compensated for by
careful design.
Another option exists for the generic design of active
lters using gyrators. This option can be exercised by
casting the equations for a given lter in the G
m
C format.
This is done by writing the state equations for the desired
lter in the standard form
d
dt
xAxbu; y c
T
xdu;
Hs
y
u
c
T
sI A
1
bd
where xx
1
; x
2
; . . . ; x
N
T
14
where the input u and the output y are assumed scalars, x
is the (N1)-state vector, A is the N by N state matrix, b
and c
T
are N dimensional vectors, and d is a scalar. Now
assume that the input, u, is a voltage, and let each of the
state variables x
k
be equated to the voltage v
k
on some
grounded capacitor C
k
. The derivative of this voltage,
times the capacitance value, is equal to the current in
the respective capacitor. Using this idea the state equa-
tions may be converted into current equations of the form
C
d
dt
vi
C
CAv CbuG
m
v g
m0
u
)
C
1
yy
1
C
2
yy
2
.
.
.
C
N
yy
N
i
C1
i
C2
.
.
.
i
CN
g
m11
g
m12
g
m1N
g
m21
g
m22
g
m2N
.
.
.
.
.
.
.
.
.
.
.
.
g
mN1
g
mN2
g
mNN
v
1
v
2
.
.
.
v
N
g
m01
g
m02
.
.
.
g
m0N
u
where CdiagC
1
; C
2
; . . . ; C
N
15
where the dot above a variable denotes time differentia-
tion. The realization of a lter based on these equations
produces a lter composed of grounded capacitors, with
transconductance ampliers bridging between the capac-
itor nodes and the input. The class of G
m
C lters, some-
times referred to as OTAC lters, is based exclusively
on this formulation.
A gyrator-based synthesis is possible by partitioning
the G
m
matrix into symmetric and skew symmetric ma-
trices. The idea is best described by an example. Suppose
that a second-order version of the g
m
C formulation
is given. The G
m
matrix can always be decomposed as
follows
G
m
g
m11
g
m12
g
m21
g
m22
G
m1
G
m2
g
m11
g
m12
g
g
m21
g g
m22
0 g
g 0
16
where the off-diagonal elements of G
m1
are equal, making
this a symmetric matrix. Clearly, G
m2
is a skew symmetric
matrix. With this partitioning of the transconductance
matrix, it is possible to realize the system in Eq. (16) using
one reciprocal two-port, characterized by G
m1
, and a sec-
ond two-port, characterized by G
m2
, that is a gyrator.
Figure 16 shows the realization associated with this de-
composition assuming a single input term in Eq. (15) and
a special case for the outputthat is, g
m02
0 and y
x
1
v
1
. The reciprocal two-port can often be realized only
with resistors, but in general may require active circuitry.
This special case considered can be further explained
with a specific example. Suppose the second order system
of Eq. (16) is the bandpass lter described with the
R
V
in
V
out
+
C
L
C
p
g
2
g
1
Figure 15. Replacement of series LC with two gyrators and two
capacitors.
+
+
+
g
m01
g
m2
u
g
y
v
1
v
2
C
Reciprocal
two-port
G
m1
C
Figure 16. Generic synthesis of a second order G
m
C lter.
1880 GYRATORS
following state space description:
x
1
:
x
2
:
o
0
=Q o
0
o
0
0
x
1
x
2
o
0
0
u; y 1 0
x
1
x
2
Hs 1 0
s o
0
=Q o
0
o
0
s
1
o
0
0
so
0
s
2
so
0
=Qo
2
0
17
Let us nowassume that the input u is a voltage denoted by
v
in
. By equating the state variables, x
1
and x
2
, to respec-
tive voltages, v
1
and v
2
, and scaling each equation by the
same capacitance value C for convenience, one obtains
C v.v
1
C v.v
2
Co
0
=Q Co
0
Co
0
0
v
1
v
2
Co
0
0
v
in
Co
0
=Q 0
0 0
v
1
v
2
0 Co
0
Co
0
0
v
1
v
2
Co
0
0
v
in
G
m1
v G
m2
vg
m0
v
in
18
where v
v
1
v
2
Recognizing that Co
0
has units of conductance, one may
easily realize this bandpass lter with only a resistor, im-
plementing G
m1
, a gyrator, implementing G
m2
, and a
transconductance amplier realizing the nonzero term in
g
m0
. This realization is shown in Fig. 17, where g
m01
Co
0
and RQ/Co
0
. Observe that this realization is essentially
the same as that obtained by replacing the grounded in-
ductor in the bandpass lter of Fig. 4 with a gyrator/
capacitor combination.
The generalization of this synthesis technique to arbi-
trary-order systems is straightforward, although cumber-
some. In this case, the G
m
matrix is again partitioned into
symmetric and skew symmetric matrices; however, each
related off-diagonal pair of elements in the skew symmet-
ric G
m2
matrix must be realized with a separate gyrator.
This will not be much of a problem if the matrix is sparse,
which can often be arranged in setting up the state equa-
tions. The N-port gyrator described in, for example, Refs. 9
and 10 can be used to realize the entire G
m2
matrix at one
time.
6. ADVANCED TOPICS
6.1. Energy and Initial Conditions
Gyrators have already been shown to be lossless two-
ports. This idea can be extended to show a duality be-
tween the energy stored on a capacitor and the energy
stored in an inductor. Suppose that a capacitor, of value C,
is connected to one port of a gyrator. Further suppose that
this capacitor is charged to a voltage V. Then the energy
stored in this capacitor is given by
1
2
CV
2
. As described
earlier, the impedance seen looking into the other port of
the gyrator is an equivalent inductor. Given the lossless
nature of the gyrator, this equivalent inductor should be
expected to have the same apparent stored energy. How-
ever, in the case of an inductor the energy stored is
1
2
LI
2
,
where I is the current owing in the inductor. A capacitor
is in equilibrium with an open circuit across it, and an
inductor is in equilibrium with a short circuit across it.
Hence, by short-circuiting the port of the gyrator opposite
the capacitor, a current ows that will be equal to
the equilibrium current in the equivalent inductor. The
following analysis shows that the stored energy in
the equivalent inductor equals that actually stored on
the capacitor:
1
2
LI
2
1
2
C
g
2
gV
2
1
2
CV
2
19
The natural consequence of this energy relationship is
that the current at the inductive port cannot change in-
stantaneously, since the voltage at the capacitive port can-
not change instantaneously. Hence, initial conditions can
be readily translated from one port to another. These facts
demonstrate that the gyrator is truly an energy conser-
vative two-port, satisfying any intuition take one might
have regarding its operation.
6.2. Nonideal Effects
Gyrators, in practice, cannot be made to be ideal. There-
fore, practical gyrators are not lossless. Instead they in-
troduce small losses into the system. This is explained by
modifying the y-parameter matrix for the gyrator to in-
clude diagonal terms. With these terms the two-port is no
longer lossless, as is clear from this analysis:
I YV
g
11
g
g g
22
V GV
)V
T
I V
T
GV g
11
V
2
1
g
22
V
2
2
20
If the diagonal elements, g
11
and g
22
, are both positive,
then the two-port described in the equation in lossy, since
+
+
+
G
m01 u
g
C
C
R
v
1
v
2
y
Figure 17. General synthesis realization for a bandpass lter.
GYRATORS 1881
the power delivered to this two-port must be positive. In
practice, the loss terms arise naturally from the fact that
the transconductors comprising the gyrator are nonideal.
For example, the input/output impedance of the transcon-
ductors will not be innite in a practical device. In this
case, g
11
and g
22
are the nonzero input admittance of the
transconductors.
Furthermore, the transfer characteristics will not in
general be ideal. As an example, consider the case of the
gyrator realized using op amps as in Fig. 10. Suppose that
the op amps have a nite gain A. Then the y-parameter
matrix can be derived, and is found to be
Y
g
0
G
GdG g
0
where G
1
R
; g
0
21 KG
dG21 K
2
G; K
1
1 2=A
where A is the open-loop voltage gain of the operational
ampliers. Notice that this Y matrix corresponds to an
ideal gyrator when A becomes innite. Also observe that
the nite gain of the op amps has caused the y-parameter
matrix to no longer be skew symmetric, which in itself
adds loss to the system. Hence, in general, practical gy-
rators exhibit loss and asymmetrythat is, they lack
skew symmetryin their transfer characteristics.
Another possibility, considering Eq. (20), is for the di-
agonal elements, g
11
and g
22
, to be purely imaginary. In
this case, the power computed in Eq. (20) is imaginary,
which translates to purely reactive power. When dissipat-
ed power is purely reactive, no average power is dissipat-
ed. Hence, a gyrator with purely imaginary diagonal
elements is still lossless. Such a device could be synthe-
sized by adding reactive elements in series or parallel with
the ports of the gyrator, since the diagonal elements, g
11
and g
22
, amount to the input admittance looking into the
respective ports of the gyrator. Furthermore, stray
capacitance or inductance associated with the inputs
or the active circuitry making up the gyrator does not
contribute to loss.
6.3. The Hall Effect Device and Isolators
It has been observed that Hall effect devices implement a
lossy gyrator. This is because the physics of these
devices is such that the two electric eld controlled ports
behave as a pair as if they were a gyrator with lossthat
is, g
11
and g
22
in Eq. (20) are nonzero and not
purely imaginary. The physics of such devices is explained
in Ref. 11.
Figure 18 shows an interesting usage for a Hall effect
gyrator, and in fact any lossy gyrator. In the gure, a gy-
rator, assumed to have the y-parameter matrix of Eq. (20),
has bridging components, R
P1
and R
P2
, added around it.
Then a pair of sources, V
S1
and V
S2
, with respective source
resistance, R
S1
and R
S2
, are attached as shown. With a
little effort the response of this circuit from the sources to
the port voltages of the gyrator can be found to be
V
1
V
2
a
G
S1
g
22
G
S2
G
P
G
S2
G
P
g
G
S1
G
P
g G
S2
g
11
G
S1
G
P
V
1
V
2
where a g
11
G
S1
G
P
g
22
G
S2
G
P
G
2
P
g
2
G
P
G
P1
G
P2
22
By choosing the sum of the bridging elements equal to
gyration constant g, the response at port 2 can be made
totally independent of V
S1
, as opposed to the response at
port 1, which will depend on both sources. This creates a
circuit called an isolator, which can be found in various
applications, especially microwaves and optics.
6.4. Multiport and Gyrators
The concept of a gyrator need not be restricted solely to
two-port networks. In fact, an N-port gyrator can be con-
trived as a natural extension of the two-port gyrator. As
one might expect, the N-port gyrator must inherit the key
properties of the two-port type. It must be a nonreciprocal
lossless network; it must also reect impedances in a way
similar to the two-port gyrator. In general, N-port gyrators
have not found use in electrical systems. For a detailed
discussion of such networks, refer to Refs. 9 and 10.
There is one special case, however, of an N-port com-
plex gyrator, for N3, which has found extensive use in
microwave systemsnamely, the circulator. While practi-
cal circulators are quite complex structures, electrically
speaking, they can be viewed over a certain range of fre-
quency to be an approximately lossless three-port complex
gyrator. Reference 12 describes the three-port y-parame-
ter matrix for a circulator. Specifically
Y
a b g
b
a b
g
23
where the superscript * denotes complex conjugation. It is
referred to as being complex since the lower triangular
matrix part of Y is the negative of the conjugate transpose
of the upper triangular part. The power P delivered to a
I
1
V
1
R
S1
R
p1
R
p2
V
S1
+
+
Two-port
network
I
2
V
2
R
S2
V
S2
+
+
Figure 18. Connection for a lossy gyrator to implement an iso-

lator.
1882 GYRATORS
three-port having this y-parameter matrix is given by
P V
1
V
2
V
3
a b g
b
a b
g
V
1
V
2
V
3
aV
2
1
V
2
2
V
2
3
b b
V
1
V
2
V
2
V
3
g g
V
1
V
3
24
As suggested, this power can be made purely reactive
if all the coefcients multiplying the voltage products
are purely imaginary. This condition is always met
if a is imaginary, since the real parts of b and g cancel in
the nal result. Hence, the circulator described by
this equation is a lossless three-port given purely
imaginary values for a. The circulator is interesting in
that the transfer characteristics from port to port when
driven by sourcesfor example, V
S1
, V
S2
, and V
S3
is
similar to the isolator previously described. Specifically,
V
S1
does not affect the port 2 voltage, V
S2
does not affect
the port 3 voltage, and V
S3
does not affect the port 1 volt-
age. More details of the design and use of circulators is
given in Ref. 13.
BIBLIOGRAPHY
1. B. D. H. Tellegen, The gyrator, a new circuit network element,
Phillips Research Report, 1948, Vol. 3, pp. 81101.
2. A. Antonou, Realization of gyrators using operational ampli-
ers, and their use in RC-active network synthesis, Proc. IEE
116:18381850 (1969).
3. S. Singer, Loss-free gyrator realization, IEEE Trans. Circ.
Syst. 35:2634 (Jan. 1988).
4. Y. P. Tsividis and J. O.Voorman, Integrated Continuous-Time
Filters: Principles, Design, and Applications, IEEE Press,
Piscataway, NJ, 1993.
5. H. Y. Lam, Analog and Digital Filters: Design and Realiza-
tion, Prentice-Hall, Englewood Cliffs, NJ, 1979.
6. R. S. H. Riordan, Simulated inductors using differential am-
pliers, Electron. Lett. 3:5051 (1967).
7. D. F. Sheahan and H. J. Orchard, Bandpass lter realisation
using gyrators, Electron. Lett. 3(1):4042 (1967).
8. D. F. Sheahan, Gyrator-oatation circuit, Electron. Lett.
3(1):3940 (1967).
9. Synthesis of active RC systems with a multiport gyrator and a
dened structure, IEEE Trans. Circ. Syst. CAS-27:191199
(1980).
10. A. G. J. Holt and R. L. Linggard, The multiterminal gyrator,
Proc. IEEE 56:13541355 (1968).
11. A. G. Milnes, Semiconductor Devices and Integrated Electron-
ics, Van Nostrand-Reinhold, New York, 1980.
12. R. H. Knerr, A proposed lumped-element switching circulator
principle, IEEE Trans. Microwave Theory Tech. MTT-20:396-
401 (1972).
13. J. Helszajn, Synthesis of octave-band quarter-wave coupled
semitracking stripline junction circulators, IEEE Trans.
Microwave Theory Tech. 43:573581 (1995).
GYROTRONS
R. LAWRENCE IVES
Calabazas Creek Research, Inc.
Saratoga, California
1. INTRODUCTION
The gyrotron is a vacuum tube capable of delivering high
levels of radiofrequency (RF) power at frequencies from
several gigahertz (GHz) to more than 200GHz, which cov-
ers most of the microwave and millimeter-wave bands.
Because of the structure of the RF elds in the cavity, the
magnitude of the electric elds and the RF losses in the
cavity walls are much lower in gyrotrons than in most
microwave and millimeter-wave vacuum devices. As the
RF frequency increases, it is not necessary to reduce the
size of the cavity and output waveguide. For these reasons,
the gyrotron is the principal RF device for delivering high
levels of RF power in this frequency range.
Gyrotrons typically require a very high magnetic eld
to provide the electron cyclotron motion for power extrac-
tion, and most gyrotrons above 30 GHz require a super-
conducting magnet. Until recently (as of 2003), this
significantly increased system cost and complexity, limit-
ing applications for these devices. Recent advances in cry-
ogen-free magnets are eliminating the requirement for
liquid cryogens and allowing additional applications. Gy-
rotrons also have a narrow RF bandwidth, typically
around 0.1%. Historically, narrow bandwidth and the
overmoded nature of the gyrotron limited their use as
ampliers, so almost all high-power gyrotrons are oscilla-
tors. Research on high-power gyroampliers is in progress
at several locations around the world, and bandwidths of
7% and higher have been achieved [1], increasing their
potential use for high-power, high-resolution radar [2].
Gyroklystrons are also being developed to drive high-pow-
er accelerators, which do not require large bandwidths
[3,4].
Gyrotrons are used for electron cyclotron resonance
heating (ECRH), electron cyclotron current drive (ECCD),
and diagnostic measurements in fusion plasma devices.
They are also being used for industrial heating applica-
tions such as ceramic sintering. Figure 1 shows a photo-
graph of Dr. Howard Jory, one of the pioneers in gyrotron
development, holding a 28-GHz, 10-kW continuous-wave
(CW) harmonic gyrotron used for industrial heating. Be-
hind him is a 110-GHz gyrotron, rated at approximately
450 kW CW, used for electron cyclotron resonance heating.
Communications and Power Industries, Inc. in Palo Alto,
CA manufactures both devices.
Research on gyro-type devices began in the 1950s,
when the astrophysicist R. Q. Twiss described an ampli-
fying mechanism for monochromatic radiation of angular
frequency o from stimulated emission of an ensemble of
electrons [5]. Twiss formula predicted amplication for
Cerenkov radiation and for cyclotron radiation. Working
independently, Schneider described the stimulated emis-
sion of radiation from electrons in a magnetic eld in 1959
using a quantum-mechanical model [6]. Also in 1959,
GYROTRONS 1883
Gapanov described this mechanism using a classical
approach [7].
The rst experimental results describing a fast-wave
cyclotron interaction were reported by R.H. Pantell in
1959 [8]. His device produced radiation between 2.5 and
4.0 GHz from a 1-kV, 3-mA electron beam. A number of
additional experiments were reported during the early
1960s [912], but the experiment that conrmed the cy-
clotron maser interaction was performed by Hirsheld
and Wachtel in 1964 [13].
During the mid to late 1960s, major advances were
made by scientists in the Soviet Union, although much of
this work went unreported in the Western world because
of the political climate at the time [14]. Soviet scientists
embarked on a major program to develop megawatt gy-
rotrons and made major advances in electron mode selec-
tion, open resonators and waveguides [15], ray tracing to
optimize the interaction between the electron beam and
the RF wave in the circuit, and high-cross-section cavities
utilizing whispering gallery modes [16]. Whispering gal-
lery modes are transverse electric modes where the num-
ber of azimuthal eld variations significantly exceeds the
number of radial eld variations. Use of these modes was
further facilitated by the development by Vlasov et al. of a
quasioptical device for converting the whispering gallery
waveguide mode into a Gaussian mode that can be prop-
agated in a narrow wave beam without waveguides [17].
An equally important Soviet development was the mag-
netron injection gun, which generates the required elec-
tron beam for efcient gyrotron operation.
For a more complete historical description of gyrotron
development, the reader is referred to three publications
that cover the subject in more detail [1820]. The reader is
also referred to a listing of journals in the bibliography
that contain most of the published work in this area.
2. BASIC THEORY OF OPERATION
Gyrotrons exploit the negative mass instability to achieve
azimuthal bunching of a cycloiding electron beam. A
transverse electric eld in the cavity modies the energy
of the electrons such that higher-energy electrons gyrate
more slowly around the magnetic ux lines than do lower-
energy electrons. A schematic cross section of a gyrotron
beam is shown in Fig. 2. The gyrotron beam is a hollow
electron beam with individual electrons rotating at the
cyclotron frequency around magnetic eld lines with orbit
diameters equal to twice the Larmour radius. The Lar-
mour radius is given by
R
l
gmv
?
eB
0
where g 1 v
2
?
=c
2
v
2
z
=c
2
1=2
is the relativistic factor,
m is the rest mass of the electron, v
>
is the velocity of the
electron in the plane perpendicular to the gyrotron axis, v
z
is the electron velocity in the axial direction, e is the
charge of the electron, and B
0
is the applied magnetic eld
parallel to the gyrotron axis.
Figure 3 more closely examines the effect of the electric
eld on the individual electrons in a cyclotron orbit when
the frequency of the electric eld is equal to the cyclotron
Figure 1. Dr. Howard Jory holds a 28-GHz, 10-kW CW harmonic
gyrotron used for industrial heating. Behind him is a 110-GHz,
450-kW CW, 1-MW pulsed, gyrotron used for electron cyclotron
heating of fusion plasmas.
Magnetic flux
lines
Beam radius
Electron cyclotron
orbits
Figure 2. The cylindrical electron beam of the gyrotron consists
of electrons orbiting around magnetic ux lines.
1884 GYROTRONS
frequency. The cyclotron frequency is given by
O
eB
0
gmc
The electric eld components have the form
E
y
E
0
J
0
m
k
?
r sink
z
z cos my
E
r
m
k
?
r
J
m
k
?
r sink
z
z sinmy
where E
y
and E
r
are the azimuthal and radial components
of the electric eld, respectively; J
m
and J
0
m
are the Bessel
function and Bessel function derivative, respectively; m is
the azimuthal mode number of the electric eld;
k
?
X
0
mn
=r
0
, where X
0
mn
is the nth root of the correspond-
ing Bessel function derivative and r
0
is the circuit radius;
and k
z
pp/L, where p is the axial mode number of the
circuit eld (typically 1) and L is the cavity length.
In Fig. 3, each successive image is an integral number
of RF periods later in time as the electrons traverse the
cavity. Electrons accelerated by the electric eld gain en-
ergy and, as a result of the relativistic mass increase, their
angular velocity decreases as g increases. Conversely, elec-
trons decelerated by the electric eld lose energy and gain
angular velocity. This causes azimuthal bunching of the
electrons, sometimes referred to as the cyclotron reso-
nance maser (CRM) instability.
If the frequency of the electric eld exceeds the cyclo-
tron frequency, the bunch will eventually fall back in
phase and more electrons will undergo deceleration than
acceleration, as shown in Fig. 4. This will result in trans-
fer of energy from the electrons to the electric eld. Opti-
mized performance of gyrotrons requires careful design of
the circuit geometry and control of the magnetic eld. The
circuit geometry affects the strength and prole of the
electric eld. The magnetic eld affects the location of the
beam and the amount of transverse energy available for
transfer to the electric eld. In many cases the magnetic
eld is tapered, that is, it is modied in strength through
the cavity, to optimize the efciency of power extraction.
Harmonic gyrotrons operate at a multiple of the cyclo-
tron frequency. As the harmonic number increases, beam
placement becomes more critical and the theoretical ef-
ciency decreases. The advantage of harmonic gyrotrons is
that the magnitude of the magnetic eld is 1/n times that
required for nonharmonic operation. This can eliminate
the need for a superconducting magnet or, at least, signif-
icantly reduce the cost.
For a more complete description of gyrotron theory,
both linear and nonlinear, the reader is referred to several
excellent publications on the subject [2022].
3. BASIC COMPONENTS
Figure 5 shows a schematic layout of a typical gyrotron
oscillator. The basic components are an electron gun, in-
put beam tunnel, circuit, output taper, collector, window,
and magnets. There are numerous variations of each of
these components, depending on the operating character-
istics of the gyrotron. The discussion that follows describes
the most common types of components, including their
purpose and performance characteristics.
3.1. Electron Gun
The function of the electron gun is to produce the electron
beam required for interaction with the desired operating
mode in the circuit. The gun must produce an electron
beam that will be located at the proper radius for efcient
interaction with the cavity electric elds and must contain
most of its energy in cyclotron motion. Gyrotrons employ a
magnetron injection gun (MIG) that emits electrons into a
region of crossed electric and magnetic elds. This geomet-
ric conguration causes the individual electrons to spiral
around the magnetic eld lines as they are accelerated by
Magnetic flux line
0

0

0

0

0
E
0
cos t E
0
cos t
E
0
cos t E
0
cos t
E
0
cos t
Bunched electrons
Decelerated electrons
Accelerated electrons
Figure 3. The cavity electric eld inter-
acts with the electrons orbiting around
the magnetic ux lines. Each diagram il-
lustrates an integral number of cyclotron
periods later in time as the electrons tra-
verse the cavity. The force on the elec-
trons is given by F
!
e E
!
:
0
t E
0
cos
Electron bunch
Figure 4. When o4O
0
, more electrons will be decelerated than
accelerated, resulting in transfer of energy to the electric eld.
GYROTRONS 1885
the electric eld toward the circuit. This development was
pioneered in the Soviet Union in the 1970s and made rapid
development of gyrotrons possible [23].
Gyrotron cathodes typically operate temperature-lim-
ited, which means that the amount of current emitted
from the cathode is determined primarily by the temper-
ature of the emitting surface. The reader is referred to the
excellent treatment of cathode emission mechanisms pre-
sented by Gilmour [24]. Typical cathode temperatures
range between 950 and 10001C.
Gyrotron guns come in two basic types, modulated
(mod) anode and diode. A mod anode gun contains an elec-
trode near the emitting surface to modulate the electron
beam; that is, it uses an applied voltage to modify the
characteristics of the beam. This electrode is maintained
at a voltage between the cathode and ground and can have
a significant effect on the beam characteristics and hence
the performance of the gyrotron. By varying the voltage on
this electrode, the user can control the output power by
varying the amount of transverse energy that is present in
the electron beam. Most commercially available gyrotron
oscillators using mod anode guns operate at cathode volt-
ages between 60 and 80 kV with mod anode voltages
ranging from 15 kV up to about 30 kV above cathode
potential.
A diode gun possesses no intermediate electrode, so the
cathode voltage or the magnetic eld controls gyrotron
operation. Because the cathode power supply must also
supply the current in the beam, it cannot be modied as
easily as a mod anode supply, so diode guns typically op-
erate where RF power is turned on or off by pulsing of the
cathode power supply. For applications where it is not
necessary to modify the operating characteristics of the
tube, this represents a lower cost gyrotron and power sup-
ply conguration.
3.2. Magnetic Solenoids
The circuit requires an electron beam where the electrons
are orbiting the magnetic eld lines at a frequency near
that of the RF frequency of the cavity. For gyrotrons op-
erating at frequencies above 30 GHz, this typically re-
quires a superconducting solenoid. An exception can occur
if the circuit is operating at a harmonic of the cyclotron fre-
quency. The requirement of a superconducting solenoid lim-
its applications for gyrotrons to those where the additional
weight and complexity do not present a severe burden. Re-
cent development of cryogen-free, superconducting magnets
is now allowing mobile or airborne applications [25].
In addition to the solenoids required to produce the
cavity magnetic eld, other solenoids can be present
around the electron gun and collector regions. Solenoids
around the electron gun modify the amount of transverse
energy in the beam or its size in the circuit. These coils are
typically operated at room temperature because the elds
in the electron gun are usually a few hundred Gauss. The
ratio of the magnetic eld magnitude in the circuit to that
in the electron gun is referred to as the beam compression.
This value plays an important role in the design of the
electron gun and the combination of electric and magnetic
elds required for efcient circuit interaction. Beam com-
pressions of 1020 are typical.
Solenoids around the collector are used to distribute
the power deposited by the electron beam exiting from the
circuit. The fringing magnetic elds from superconducting
coils impose a magnetic eld in the collector that is higher
than in other linear beam tubes. This eld prevents the
electrons from spreading as a result of space charge and
results in a relatively thin beam of electrons impacting a
small region of the collector. This can result in excessive
localized heating of the surface and possible destruction of
the gyrotron. The collector coils are used to buck the eld
from the main solenoid and spread the electron impact
area over a larger region of the collector. In some more
advanced applications, an oscillating power supply drives
one or more solenoids to sweep the beam back and forth
along the collector surface.
3.3. Beam Tunnel
The beam tunnel must perform two important functions.
First, it must prevent RF elds from the circuit from trav-
eling back toward the electron gun. RF elds in the elec-
tron gun can severely affect the electron beam and cause
excessive current on the body of the tube or the mod an-
ode. It can also result in heating of the cathode emitter,
Electron beam
Magnetron injection
gun
Gun solenoids
Input beam
t unnel RF cavity
Collector solenoids
Window
Main solenoids
Collector
Figure 5. The basic components of a typ-
ical gyrotron are the electron gun, beam
tunnel, cavity, output beam tunnel, collec-
tor, window, and magnet solenoids.
1886 GYROTRONS
which will affect the beam current. Either condition will
adversely affect tube operation.
The second function of the beam tunnel is to prevent
parasitic oscillations between gun and circuit. Parasitic
oscillations result in excessive body or mod anode current
and can prevent tube operation. In some cases, dielectric
material with a high RF loss is used to load out any elec-
tromagnetic elds present.
3.4. Circuit
The RF circuit typically consists of a right circular cylin-
der whose radius a is chosen such that
a
cX
0
mn
2pf
where X
0
mn
is the nth root of the Bessel function derivative
for the TE
mn
cavity mode, c is the speed of light, and f is
the operating frequency. The cavity mode depends pri-
marily on the desired RF frequency and output power lev-
el. The application for the tube can also play an important
role. The rst commercial gyrotrons used TE
01
, TE
02
, or
TE
03
modes because they were reasonably close to the
fundamental mode, did not couple to nonsymmetric modes
that could travel down the beam tunnel toward the elec-
tron gun, and had very low RF loss in the circuit walls.
More than 100 gyrotrons at frequencies between 28 and
100GHz and power levels up to 340kW CW were pro-
duced by Varian Associates, Inc. (now Communications
and Power Industries, Inc.) from 1978 through 1990.
These tubes were used for electron cyclotron resonance
heating (ECRH) of plasmas in magnetic connement fu-
sion research. Gyrotrons of this type were also developed
in Germany, China, Japan, France, and Russia [26].
As the demand for higher-power tubes increased, gy-
rotrons that used whispering gallery modes were devel-
oped. Whispering gallery modes allow the electron beam
to be larger and minimize mode competition from other
circuit modes. These modes also have reasonable RF pow-
er dissipation in the circuit.
Designers of high-power gyrotrons must balance a
large number of factors to achieve a circuit conguration
that will provide the power and frequency required with
reasonable efciency while avoiding parasitic mode com-
petition, instabilities, and excessive RF power densities on
the circuit walls. For additional information on the design
of high-power gyrotron circuits, the reader is referred to
the article by Kreischer et al. [27].
3.5. Output Taper
Most gyrotron circuits are open-ended toward the collec-
tor, and the RF power diffracts into the output taper. The
output taper is usually a shallow, tapered section of cir-
cular waveguide that terminates the circuit interaction
and transmits the RF power toward the collector. The
waveguide radial dimensions must be increased in such a
way that the purity of the output mode is not compro-
mised. For gyrotrons where the RF power is extracted
along the axis of the tube, the output taper continues to
increase in the radial dimension until the desired collector
radius is achieved. The requirement to maintain mode
purity often conicts with the necessity of achieving the
collector radius within a reasonable distance. The size of
the collector is driven by the necessity to dissipate the
spent electron beam without incurring excessive power
densities on the walls, which could lead to melting or loss
of vacuum integrity.
For gyrotrons with average RF power levels exceeding
200 kW, particularly those employing whispering gallery
circuit modes, it is more expedient to extract the RF power
radially to allow more exibility in the collector design.
For these tubes, the output taper transmits the circuit
power to an RF launcher/antenna for eventual extraction
from the vacuum envelope. This eliminates further re-
quirements on the output taper to maintain mode purity
and allows the designer more freedom in transitioning to
the collector region.
3.6. Collector
Typical linear beam devices rely on termination of the
magnetic eld and space charge depression to spread the
spent electron beam in the collector. Most of these devices
employ an iron polepiece to terminate the magnetic eld
at the entrance to the collector. While this works well for
solid electron beams, it does not apply to gyrotrons. Be-
cause of the high magnetic elds required in gyrotrons,
this is rarely practical, particularly when superconducting
solenoids are used.
Forces imposed by nearby iron on the superconducting
coils would dramatically increase the complexity and cost
of the magnet. As a result, sufcient magnetic elds exist
in the collector to limit spreading of the electron beam
from space charge forces. In addition, the beam used in
gyrotrons is typically a thin cylindrical beam with high
current density. This also exacerbates the problem of
beam dissipation in the collector. Power densities up to
500 W/cm
2
are typical; 1000W/cm
2
is considered the upper
limit.
For gyrotrons at lower average power levels, these com-
plications may not be a critical issue. For these devices,
the output taper conducts the RF power to the collector
with high mode purity, and the collector also serves as the
output waveguide. In some cases, a downtaper is em-
ployed at the end of the collector to reduce the tube diam-
eter before the output window. For these devices, the
designer must also be concerned about modes that can
be trapped in the collector between the uptaper from the
circuit and the downtaper to the window.
For gyrotrons exceeding 500kW of average power, ra-
dial RF power extraction is almost always employed. The
size of the collector can be based on the dissipation re-
quirements of the spent beam alone. As a result, collectors
for high-power gyrotrons often constitute a major portion
of the device. In Fig. 1, the collector of the 500-kW CW
gyrotron behind Dr. Jory begins above the output window.
The designer must balance low power densities with the
increasing size, weight, and cost of large collectors.
Considerable research is now in progress to implement
collector voltage depression to improve the total efciency
GYROTRONS 1887
of the gyrotron while simultaneously reducing the
thermal power dissipation. The efciency of the RF
circuit is typically 3040%; therefore, most of the incident
beam power is dissipated in the collector. For a 1-MW CW
gyrotron, approximately 2 MW of thermal power must
be dissipated by the collector. To reduce this thermal
power and improve the gyrotron efciency, a voltage
can be applied to the collector that is between that of the
electron gun and the RF circuit. This slows the electrons
and reduces the incident power. Since the total gyrotron
input power is proportional to the voltage difference
between the electron gun and the collector, this also
increases the net efciency. The amount of voltage depres-
sion that can be applied depends on the energy of the
electrons following RF power extraction in the circuit.
Voltage depression that exceeds the energy of the elec-
trons will lead to reection of the beam back toward the
electron gun or circuit. This will lead to excessive heating,
loading of the power supplies, or loss of interaction ef-
ciency. Most gyrotrons in development today implement
voltage depression to increase the total efciency to
approximately 50%.
Additional increases in efciency can be achieved by
introducing collectors with multiple electrodes at varying
voltages and sorting the energy of the incident electrons
using carefully designed magnetic elds. This allows ad-
ditional increase in depression voltage for the higher-en-
ergy electrons while still avoiding reection of slower
ones. A two-stage depressed collector is currently being
developed and is predicted to increase the overall efcien-
cy of gyrotrons to more than 55% [28,29].
3.7. RF Launcher System
The RF launcher system converts the circular cavity/
waveguide mode to a quasioptical Gaussian mode. This
development was pioneered in the Soviet Union and al-
lows transmission of the RF beam with very low loss using
a series of metallic mirrors. Modern computer codes and
integration with computer numerically controlled (CNC)
mills and lathes allow precise control of the RF beam.
Modern designs convert the power from the circuit into
the desired Gaussian mode with greater than 95% ef-
ciency. More advanced computer codes are in development
that could increase the efciency to more than 98% [30].
The conversion of the waveguide mode to a quasioptical
Gaussian mode also results in a more convenient RF mode
for most user applications.
Figure 6 shows a typical RF launcher conguration
consisting of a waveguide launcher and two mirrors in-
ternal to the gyrotron. The exit angle of the RF power from
the launcher is at the waveguide bounce angle, and the
mirrors concentrate the power into a small circular beam
and tailor the distribution of power within the beam as
required by the output window.
3.8. Output Window
Until recently, the output power of gyrotrons was limited
by the capabilities of the output window. The RF power
produced in the high-vacuum region inside the gyrotron is
extracted from the device through a dielectric window.
The window must provide a vacuum interface while
transmitting the RF power from the gyrotron into an ex-
ternal waveguide that may, or may not, be under vacuum.
Some of the RF power transmitted through the window is
absorbed and results in heating of the dielectric material.
The window structure must be sufciently cooled to pre-
vent destruction from thermal stresses.
For gyrotrons producing power levels less than 500kW
continuously, common window materials include alumina
and sapphire; however, beryllia oxide, boron nitrite, sili-
con nitride, and other materials have been used in special
devices. At higher average power levels, the thermal char-
acteristics of standard materials is such that the heat de-
posited in the ceramic cannot be adequately removed by
cooling around the edges as in most linear beam devices.
Figure 7 shows a schematic diagram of a double disk win-
dow that uses a dielectric uid to convectively face cool the
disks.
Development of chemical-vapor-deposited (CVD) dia-
mond windows eliminated thermal stresses as a limiting
factor in gyrotrons [31]. CVD diamond is significantly
stronger, and has a higher thermal conductivity and
much lower RF loss than do other available materials
RF cavity
Path of
RF beam
Output
window Internal mirrors
Collector
Launcher
Figure 6. The Gaussian mode launcher converts the waveguide
mode to a quasioptical beam, and the internal mirrors shape the
RF beam prole.
Air
Ceramic
disk
Dielectric fluid
Coolant
input
Coolant
output
Vacuum
Figure 7. In the double-disk window, a dielectric uid ows
between the ceramic disks for enhanced cooling.
1888 GYROTRONS
[32]. Consequently, it has become the window material
used in all high-power gyrotrons. Unfortunately, the
CVD manufacturing process is time-consuming and
expensive. Currently, window ceramics for gyrotrons
producing 1MW or more of continuous power cost in ex-
cess of $100,000. Efforts are underway to develop more
efcient and less expensive methods for producing these
windows.
4. AMPLIFIERS
While most gyro devices are oscillators, research is in pro-
gress to develop gyroampliers as drivers for linear accel-
erators [33,34] and for high-frequency, high-resolution
radar [35]. While oscillator design is relatively straight-
forward, amplier design presents several challenges.
Since the circuits are typically overmoded, it becomes
more difcult to prevent RF leakage through the drift re-
gions between cavities. Also, the high cavity Q factors
characteristic of gyrotron output cavities limit the band-
width that can be achieved. While several amplier con-
gurations have been studied, the gyroklystron and the
gyro-TWT (traveling-wave tube) have achieved the most
success.
The gyroklystron consists of two or more cavities sep-
arated by drift spaces too small to propagate the operating
mode. An azimuthal drift is imposed on the electrons by a
sinusoidal energy modulation in the rst cavity. These
drifts are enhanced by intermediate buncher cavities, and
the output cavity extracts the power from the azimuthally
bunched electron beam. Most gyroklystrons employ cavity
modes that are above the fundamental TE
11
circular wave-
guide mode. Extreme care must be used to ensure that any
RF power converted from the cavity mode to the TE
11
mode is not allowed to propagate through the drift spaces
and cause spurious oscillations. This also limits the circuit
modes that can be used to those close to the fundamental
(typically TE
01
or TE
02
) circuit modes. Gyroklystron band-
widths of 7% have been achieved with saturated gain of
2040 dB. Operating efciencies greater than 30% are
common. An advanced, high-power, millimeter-wave ra-
dar system is currently under development by the U.S.
Naval Research Laboratory that uses a 94-GHz gyroklyst-
ron with a 700 MHz bandwidth, 10% duty factor, and a
saturated gain of 33 dB [1]. The device will be used for
tactical radar applications, studies of cloud physics, prop-
agation, and forward and backscatter studies [2].
In the gyro-TWT, an azimuthally bunched helical beam
interacts with a traveling wave. If the periodicity and ve-
locity of the electron bunches are such that synchronism
occurs between the electrons and the traveling wave, cu-
mulative bunching and energy extraction can occur. These
ampliers have considerably larger bandwidths than do
gyroklystrons; however, they operate at lower efciency
(o20%) and generally produce less power.
Research has also been performed on gyro backward-
wave oscillators, gyrotwystrons, and gyropeniotrons. The
reader is referred to the Further Reading list at the end of
this article for information on these devices.
5. CURRENT STATE OF THE ART
Gyrotrons are currently in development in several coun-
tries around the world, including Russia, Germany,
France, Japan, and the United States. Currently, the high-
est-power gyrotron is a device jointly developed by Ger-
many, France, and Switzerland to produce 1MW of RF
power in pulses lasting several minutes. The device,
shown in Fig. 8, operates at 140GHz and will be used in
the stellerator Wendlestein 7-X currently under construc-
tion at IPP Greifswald, Germany. A total of 10 MW of RF
power will be required for electron cyclotron heating of the
stellerator plasma. Researchers recently reported achiev-
ing 850 kW in 3-min pulses, which were partially limited
by the available power supply [36]. The gyrotron has also
produced 970 kW for 11.8 s. Communications and Power
Industries, Inc. in the United States produced several
1-MW, 10-s gyrotrons at 110 GHz. These are used for elec-
tron cyclotron heating of the DIII-D tokamak at General
Atomics in San Diego, CA.
The current state of the art in gyroklystrons is the
94-GHz device built by Communications and Power In-
dustries, Inc. for the WARLOCK radar system. This device
produces a peak power of 100kW and 10 kW of average
power.
Figure 8. A 140-GHz, 1-MW gyrotron under development for the
stellerator Wendlestein 7-X currently under construction at IPP
Greifswald, Germany.
GYROTRONS 1889
6. FUTURE DEVELOPMENT
Most gyrotron research is focused on increasing the out-
put power, primarily for ECRH and ECCD of fusion plas-
mas. It is anticipated that several tens of megawatts of RF
power will be required for heating these plasmas. It is de-
sirable to maximize the output power of the individual
gyrotrons to reduce the number of gyrotrons required. The
current goal is to develop gyrotrons producing 12 MW of
CW power at frequencies between 100 and 170 GHz.
Researchers in the United States are developing a 1.5-
MW, long-pulse gyrotron operating at 110 GHz. This de-
vice will be used for electron cyclotron heating of the DIII-
D tokamak operated by General Atomics in San Diego,
CA. The gyrotron will use a conventional cylindrical cav-
ity, diode electron gun, and a depressed collector.
As the required output power increases, the size of the
cavity must increase to maintain and handle the RF cur-
rents in the walls. Increasing the cavity size increases the
number or RF modes that can be excited, thereby increas-
ing problems with mode competition. Above 1.5 MW, it be-
comes more attractive to utilize a coaxial cavity to
decrease the number of competing modes. Researchers
in Germany are developing a 2-MW coaxial gyrotron at
170GHz for anticipated heating requirements for the next
generation of fusion experiments. The device incorporates
a depressed collector and quasioptical output [37]. Addi-
tional complications are introduced by the center conduc-
tor, which must be supported, aligned, and cooled within
the vacuum envelope of the gyrotron.
Research is in progress to develop step-tunable gyrot-
rons for operation over increased frequency ranges. Since
the gyrotron frequency is partially determined by the
magnetic eld, it becomes possible to use this character-
istic to excite operating modes at different frequencies. As
the magnitude of the magnetic eld is varied, different
operating modes are excited. The frequency of the gyrot-
ron jumps to the resonant frequency of that particular
mode [38,39]. A problem arises, however, with designing a
ceramic window capable of operating over the required
range of frequencies. Several alternatives are under de-
velopment, including Brewster angle windows.
Amplier research is focused on developing gyroampli-
ers for high-resolution radar and for high-power linear
accelerators. Current research goals for radar applications
include bandwidth enhancement, increased efciency, in-
creased output power, and gains exceeding 30 dB. Accel-
erator applications do not require large bandwidths, but
peak powers exceeding 100 MWat frequencies between 11
and 20 GHz and 10 MW devices at W band will be re-
quired. Figure 9 shows a 10-MW gyroklystron designed to
operate at 91.392 GHz that is currently under develop-
ment. High efciency and reasonable gain will be major
goals. Advances in gyrotron components, such as windows,
electron guns, launcher systems, and depressed collectors,
will also be applicable to amplier development.
Research is in progress to develop gyrotrons for indus-
trial heating applications, particularly for sintering of
ne-grain ceramics. Studies indicate that microwave and
millimeter sintering can produce extremely high heating
rates or selective heating in multiphase systems, leading
to novel ceramic materials with compositions and micro-
structures not possible using standard techniques [40].
Several experiments are in progress to develop sources for
these applications [41,42]. It is anticipated that other ap-
plications will materialize when efcient, cost effective
sources are available.
BIBLIOGRAPHY
1. M. Blank, K. Felch, B. G. James, P. Borchard, P. Cahalan, T. S.
Chu, H. Jory, B. G. Danly, B. Levush, J. P. Calame, K. T. Ngu-
yen, and D. E. Pershing, Development and demonstration of
high-average power W-band gyro-ampliers for radar appli-
cations, IEEE Trans. Plasma Sci. 30(3) (June 2002).
2. B. G. Danly, J. Cheung, V. Gregers-Hansen, G. Linde, and
M. Ngo, WARLOC: A high-power millimeter wave radar,
Proc. 27th Int. Conf. Infrared and Millimeter Waves, San
Diego, CA, Sept. 2002, pp. 233234.
3. J. M. Neilson, R. L. Ives, M. Read, M. Mizuhara, T. Robinson,
D. Marsden, W. Lawson, and B. Hogan, Development status of
a 10MW, 91GHz gyroklystron, Proc. 27th Int. Conf.
Infrared and Millimeter Waves, San Diego, CA, Sept. 2002,
pp. 237238.
4. W. Lawson, B. Hogan, E. S. Gouveia, B. Huebschman, and V.
L. Granatstein, Improved design of a frequency-doubling KU-
Band gyroklystron experiment, Proc. 27th. Int. Conf.
Figure 9. A 10-MW, 91-GHz gyroklystron under development for
high-power accelerator research.
1890 GYROTRONS
pp. 203204
5. R. O. Twiss, Radiation transfer and the possibility of
negative absorption in radio astronomy, Austral. J. Phys. 11
(1958).
6. J. Schneider, Stimulated emission of radiation by relativistic
electrons in a magnetic eld, Phys. Rev. Lett. 2 (1959).
7. A. V. Gaponov, Interaction between electron uxes and elec-
tromagnetic waves in waveguides, Izv. Vyssh. Uchebn. Zaved.
Radiozika 2 (1959).
8. R. H. Pantell, Electron beam interaction with fast waves,
Proc. Symp. Millimeter Waves, Microwave Res. Inst. Symp.
Ser. 9 (1959).
9. K. K. Chow and R. H. Pantell, The cyclotron resonance back-
ward wave oscillator, Proc. IRE 48(11) (1960).
10. I. B. Bott, Tunable sources of millimeter and submillimeter
wave radiation, Proc. IEEE 52(3) (1964).
11. I. B. Bott, A powerful source of millimeter wavelength elec-
tromagnetic radiation, Phys. Lett. 14(4) (1965).
12. J. Feinsten, Research on electronic interaction with the elds
of mirror resonators, Proc. 5th Int. Congress on Microwave
Tubes, 1964.
13. J. L. Hirscheld and J. M. Wachtel, Electron cyclotron maser,
Pys. Rev. Lett. 12(19) (1964).
14. M. I. Petelin, Relativism in microwave electronics, Proc. 22nd
Int. Conf. Infrared and Millimeter Waves, Wintergreen, VA,
July 1997.
15. L. A. Vaynshteyn, Open resonators and open waveguides, So-
vetskoye Radio (1966).
16. Y. V. Bykov et al., paper presented at VII Inter-collegiate
Conf. SHF Electronics, Tomsk, 1972.
17. S. N. Vlasov, L. I. Zagryadskaya, and M. I. Petelin, Transfor-
mation of a whispering gallery mode, propagating in a circu-
lar waveguide, into a beam of waves. Radio Eng. Electron.
Phys. 20 (Oct. 1975).
18. V. A. Flyagin, A. V. Gaponov, M. I. Petelin, and V. K. Yulpatov,
IEEE Trans. Microwave Theory Tech. MTT-25:512521
(1977).
19. J. L. Hirsheld and V. L. Granatstein, The electron cyclotron
maser-an historical survey, IEEE Trans. Microwave Theory
Tech. MTT-25(6) (June 1977).
20. J. L. Hirsheld, Gyrotrons, Infrared and Millimeter Waves,
Vol. 1, Academic Press, 1979, Chapter 1.
21. R. S. Symons and H. R. Jory, Cyclotron resonance devices,
Adv. Electron. Electron Phys. 55 (1981).
22. J. M. Baird, Gyrotron theory, in V. L. Granatstein and Alexoff,
eds., High Power Microwave Sources, Artech House, Norwood,
MA, 1987.
23. V. A. Flyagin, A. V. Gaponov, M. I. Petelin, and V. K. Yulpatov,
The gyrotron, IEEE Trans. Microwave Theory Tech. MTT-
25:514521 (1977).
24. A. S. Gilmour, Principles of Traveling Wave Tubes, Artech
House, Boston, MA, 1994, pp. 103149.
25. K. Hackett and C. W. Beason, Active denial technology, AFRL
Technol. Horiz., 910 (Sept. 2001).
26. M. Thumm, State-of-the-Art of High Power Gyro-Devices and
Free Electron Masers, Update 1995, ForschungszentrumKarl-
sruhe, ITP, Assoc. (FZK-Euratom Postfach 3640, D-6\76021
Karlsruhe, Germany).
27. K. E. Kreischer, B. Danly, T. B. Schutter, and R. J. Temkin,
The design of megawatt gyrotrons, IEEE Trans. Plasma Sci.
PS-13(6) (Dec. 1985).
28. A. Singh, S. Rajapatirana, Y. Men, V. L. Granatstein, R. L.
Ives, and A. J. Anatolak, Design of a multistage depressed
collector system for 1-MW CW gyrotronsPart I: Trajectory
control of primary and secondary electrons in a two-stage de-
pressed collector, IEEE Trans. Plasma. Sci. 27(2): 490502
(April 1999).
29. R. L. Ives, A. Singh, M. Mizuhara, R. Schumacher, J. Neilson,
M. Gaudreau, J. Casey, and V. L. Granatstein, Design of a
multistage depressed collector system for 1MW CW gyrot-
ronsPart II: System considerations, IEEE Trans. Plasma
Sci. 27(2):503511 (April 1999).
30. L. Neilson and R. Bunger, Surface integral equation analysis
of quasi-optical launchers using the fast multipole method,
Proc. 27th Int. Conf. Infrared and Millimeter Waves, San Die-
go, CA, Sept. 2002, pp. 119120.
31. K. Sakamoto, A. Kasugai, M. Tsuneoka, K. Takahashi, and
T. Imai, High power 170GHz gyrotron with synthetic dia-
mond window, Rev. Sci. Instrum. 70(1):208212 (Jan. 1999).
32. M. Thumm, Development of output windows for high-power
long-pulse gyrotrons and EC wave applications, Int. J. Infra-
red Millimeter Waves 19:314 (1998).
33. V. L. Granatstein and W. Lawson, Gyro-ampliers as candi-
date RF drivers for TeV linear colliders, IEEE Trans. Plasma
Sci. 24(3) (June 1996).
34. J. Cheng, X. Xu, W. Lawson, J. P. Calame, M. Castle, B. P.
Hogan, V. L. Granatstein, G. S. Nusinovich, and M. Reiser,
Experimental studies of a high-power, x-band coaxial gyro-
klystron, IEEE Trans. Plasma Sci. PS-27:11751187 (1999).
35. B. G. Danly, M. Blank, J. P. Calame, B. Levush, K. Nguyen, D.
Pershing, J. Petillo, T. A. Hargraves, R. B. True, A. J. Theiss,
G. R. Good, K. Felch, T. S. Chu, H. Jory, P. Borchard, B. G.
James, W. G. Lawson, and T. M. Antonsen, Development of a
W-Band gyroklystron for radar applications, Conf. Digest
23rd Int. Conf. Infrared and Millimeter Waves, Colchester,
UK, Sept. 1998, pp. 3233.
36. G. Dammertz, S. Alberti, et al., Progress of the 1 MW,
140GHz, CW gyrotron for W7-X, Proc. 27th Int. Conf.
pp. 34.
37. B. Piosczyk, A. Arnold, H. Budig, G. Dammertz, O. Dumbrajs,
O. Drumm, M. V. Kartikeyan, M. Kuntz, M. Thumm, and
X. Yang, Experimental results and technical requirements for
a 2 MW CW 170GHz coaxial cavity gyrotron, Proc. 27th Int.
Conf. Infrared and Millimeter Waves, San Diego, CA, Sept.
2002, pp. 78.
38. T. Idehara, I. Ogawa, S. Mitsudo, M. Pereyaslavets, N. Nish-
ida, and K. Yoshida, Development of frequency tunable, me-
dium power gyrotrons as submillimeter wave radiation
sources, IEEE Trans. Plasma Sci. PS-27:340354 (1999).
39. V. E. Zapevalov, S. S. Bogdashov, A. V. Chirkov, G. G. Denisov,
A. N. Kufn, et al., Optimization of the frequency step tunable
105-170GHz 1 MW gyrotron prototype, Proc. 27th Int. Conf.
pp. 12.
40. J. P. Calame, Y. Carmel, D. Gershone, and E. Pert, The mi-
crowave sintering of ceramics: New insights, models, and ap-
plications based on realistic ceramic microstructures, Proc.
22nd Int. Conf. Infrared and Millimeter Waves, Wintergreen,
VA, July 1997.
41. I. I. Antakov, I. G. Facheve, V. I. Kurbatov, E. V. Sokolov,
E. A. Solujanova, and E. V. Zasypkin, A KA-Band and W-
Band 10kW CW high efciency gyrotrons for materials
processing, Proc. 3rd Int. Workshop on Strong Microwaves
in Plasmas, Moscow/St. Petersberg, August 1996.
GYROTRONS 1891
42. A. W. Fliflet, R. P. Fischer, A. K. Kinkead, and R.W. Bruce,
Pulsed 35GHz gyrotron with overmoded applicator for sinte-
ring experiments, Microwave Process. Mat. V. 430 (1996).
FURTHER READING
The following periodicals contain most of the publications
related to gyrotron research and development:
IEEE Transactions on Plasma Science
IEEE Transactions on Microwave Theory and Tech-
niques
International Journal of Electronics
International Journal of Infrared and Millimeter Wave
Proceedings of the International Conference on Infrared
and Millimeter Waves (yearly)
Proceedings of the International Conference on Plasma
Sciences (yearly)
Izv. Vyssh. Uchebn. Zaved. Izvestiaya Vysshikh Ucheb-
nykh Zavedni
Additional sources of information are listed below.
R. B. Miller, An Introduction to the Physics of Intense Charged
Particle Beams, Plenum Press, New York and London, 1982,
pp. 238249.
H. R. Jory, E. Lien, and R. S. Symons, Final Report of Millimeter
Wave Study Program, performed for Oak Ridge National Lab-
oratory on Order Y-12 11Y-499438V, Varian Associates, Inc.,
Palo Alto, CA, 1975.
K. R. Chu, Theory of Electron Cyclotron Maser Interactions in a
Cavity at the Harmonic Frequencies, NRL Memorandum Re-
port 3672, Naval Research Laboratory, Washington, DC, 1977.
R. S. Symons and H. R. Jory, Small-signal theory of gyrotrons and
gyroklystrons, Proc. 7th Symp. Engineering Problems in
Fusion Research, 1977.
H. R. Jory, F. I. Friedlander, S. J. Hegji, J. P. Shively, and R. S.
Symons, Gyrotrons for high-power millimeter wave genera-
tion, Proc. 7th Symp. Engineering Problems in Fusion Re-
search, 1977.
V. L. Bratman, M. A. Moiseev, M. I. Petelin, and R. E. Erm, The-
ory of gyrotrons with a nonxed structure of the high frequen-
cy eld, Izv. Vyssh. Uchebn. Zaved. Radioz. 16(4) (1973).
A. V. Gaponov, A. L. Goldenberg, D. P. Grigorev, I. M. Orlova, T. B.
Pankratova, and M. I. Petelin, Induced synchrotron radiation
of electrons in cavity resonators, Sov. Phys. JETP Lett. 2(9)
(1965).
M. Caplan and C. Thorington, Improved computer modeling of
magnetron injection guns for gyrotrons, Int. J. Electron. 51
(1998).
J. Neilson, P. E. Latham, M. Caplan, and W. Lawson, Determi-
nation of the resonant frequency in a complex cavity using
scattering matrix formulation, IEEE Trans. Microwave Theory
Tech. 37 (1989).
V. A. Flyagin and G. S. Nusinovich, Gyrotron oscillators, Proc.
IEEE 76 (1988).
K. Felch, H. Huey, and H. Jory, Gyrotrons for ECH applications,
J. Fusion Energy 9(1) (1990).
Y. V. Bykov, A. L. Goldenberg, L. V. Nikolaev, M. M. Ofitserov, and
M. I. Petelin, Experimental investigations of a gyrotron with
whispering gallery modes, Izv. Vyssh. Uchebn. Zaved. Radioz.
18(10):15441546 (1975).
J. B. Mead, Millimeter wave radars for atmospheric remote
sensing, Proc. 22nd Int. Conf. Infrared and Millimeter Waves,
Wintergreen, VA, July 1997.
PENIOTRONS
G. Dohler, Int. J. Electron. 56:617627 (1984).
P. Vitello, IEEE Trans. Microwave Theory Tech. 32:917921
(1984).
L. Zhou, M. Jiang, H. Guo, and Y. He, Int. J. Electron. 57:1065
1075 (1984).
AMPLIFIERS
K. R. Chu and A. T. Drobot, Theory and Single Wave Simulation of
the Gyrotron Traveling Wave Amplier Operating at Cyclotron
Harmonics, NRL Memorandum Report 3788, Naval Research
Laboratory, Washington, DC, 1978.
K. R. Chu, A. T. Drobot, V. L. Granatstein, and J. L. Seftor, Char-
acteristics and optimum operating parameters of a gyrotron
traveling wave amplier, IEEE Trans. Microwave Theory Tech.
27(2):178187 (1979).
G. G. Denisov, V. L. Bratman, A. D. R. Phelps, and V. Samsonov,
Gyro-TWT with a helical operating waveguide: New possibil-
ities to enhance efciency and frequency bandwidth, Proc.
22nd Int. Conf. Infrared and Millimeter Waves, Wintergreen,
VA, July 1997.
Q. S. Wang, D. B. McDermott, and N. C. Luhmann, Operation of a
stable 200kW second-harmonic gyro-TWT amplier, IEEE
Trans. Plasma Sci. 24(3):700706 (1996).
E. V. Zasypkin, B. Levush, M. Blank, E. V. Sokolav, and I. I. Ant-
akov, Study of X-band three stage gyrotwystron amplier,
Proc. 22nd Int. Conf. Infrared and Millimeter Waves, Winter-
green, VA, July 1997.
1892 GYROTRONS

07b

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

07b

Uploaded by

Copyright:

Available Formats

GROUND PENETRATING RADAR

At the boundary between two media, some energy will be

for total internal reection yields

Figure 5. (a) Dielectric waveguide of permittivity e

Figure 7. Plot of b versus radian frequency o. Note from the plot

0:28; 0:498; 0:8358; 1:1086; 1:373; 1:60910

, and the lowest-order mode

-doped active region sandwiched

regions for the ohmic con-

Figure 6. Equal-area rule for TEDs.

-doped layer is deposited. Excess

Figure 13. State-of-the-art RF power levels from TEDs under

Graded doping profile

Figure 2. Electrical circuit symbol for a gyrator.

Figure 8. Floating-gyrator realization.

Figure 10. Gyrator realization using op amps.

Figure 18. Connection for a lossy gyrator to implement an iso-

You might also like