04b

polarization and increased bandwidth for cylindrical dielectric
resonator antenna, Electron. Lett. 37:406408 (March 2001).

21. K. P. Esselle, Circularly polarized higher-order rectangular
dielectric resonator antenna, Electron. Lett. 32:17421743
(Sept. 1996).
22. C. Y. Huang, J. Y. Wu, and K. L. Wong, Cross-slot-coupled
microstrip antenna and dielectric resonator antenna for
polarization, IEEE Trans. Anten. Propag. AP-47:605609
(1999).
23. M. T. Lee, K. M. Luk, E. K. N. Yung, and K. W. Leung,
Microstripline feed circularly polarized cylindrical dielectric
resonator antenna, Microwave Opt. Technol. Lett. 24:206207
(March 2000).
24. K. W. Leung and H. K. Ng, Theory and experiment of
circularly polarized dielectric resonator antenna with a para-
sitic patch, IEEE Trans. Anten. Propag. AP-51:405412
(March 2003).
25. R. K. Mongia, A. Ittipiboon, M. Cuhacim, and D. Rosecoe,
Circularly polarized dielectric resonator antenna, Electron.
Lett. 30:13611362 (Aug. 1994).
26. K. W. Leung, W. C. Wong, K. M. Luk, and E. K. N. Yung,
Circular-polarised dielectric resonator antenna excited by
dual conformal strips, Electron. Lett. 36:484486 (March
2000).
27. A. Petosa, A. Ittipiboon, and M. Cuhaci, Array of circular-
polarised cross dielectric resonator antennas, Electron. Lett.
32:17421743 (Sept. 1996).
28. Z. Li, C. Wu, and J. Litva, Adjustable frequency dielectric
resonator antenna, Electron. Lett. 32:606607 (March 1996).
29. Z. N. Chen, K. W. Leung, K. M. Luk, and E. K. N. Yung, Effect
of parasitic disk on a coaxial probe-fed dielectric resonator
antenna, Microwave Opt. Technol. Lett. 15:166168 (June
1997).
30. H. K. Ng and K. W. Leung, Excitation of CP aperture-coupled
dielectric resonator antenna with a parasitic patch, IEEE
Antennas and Propagation Soc. Int. Symp. Digest, Boston,
July 2001, Vol. 4, pp. 202205.
31. K. W. Leung, W. C. Wong, and H. K. Ng, Circularly polarized
slot-coupled dielectric resonator antenna with a parasitic
patch, IEEE Anten. Wireless Propag. Lett. 1:5759 (2002).
32. K. W. Leung and H. K. Ng, Dielectric resonator antenna fed
by a displaced conformal strip, Microwave Opt. Technol. Lett.
29:185187 (May 2001).
33. R. D. Nevels, The annular aperture antenna with a hemi-
spherical center conductor extension, IEEE Trans. Anten.
Propag. AP-35:4145 (Jan. 1987).
34. R. D. Nevels and J. E. Wheeler, Radiation from a dielectric
coated hemispherical conductor fed by a coaxial transmission
line, IEEE Trans. Electromagn. Compat. 31:1620 (Feb.
1989).
35. K. W. Leung, K. M. Luk, K. Y. A. Lai, and D. Lin, Theory and
experiment of coaxial probe fed dielectric resonator antenna,
IEEE Trans. Anten. Propag. AP-41:13901398 (1993).
36. K. W. Leung, K. M. Luk, K. Y. A. Lai, and D. Lin, Theory and
experiment of an aperture-coupled hemispherical dielectric
resonator antenna, IEEE Trans. Anten. Propag. AP-43:1192
1198 (1995).
37. K. W. Leung, Analysis of zonal and rectangular slot on a
conducting spherical cavity, IEEE Trans. Anten. Propag.
49:17391745 (Dec. 2001).
38. K. M. Luk and K. W. Leung, eds., Dielectric Resonator
Antennas, Baldock, Hertfordshire, England; Philadelphia:
Research Studies Press Ltd., 2003.
DIELECTRIC RESONATOR FILTERS
S. BILA
D. BAILLARGEAT
S. VERDEYME
P. GUILLON
IRCOM
Limoges, France
Because microwave ltering is an important function re-
quired to keep a number of different systems in working
order, the specications of a lter are varied. We can,
however, try to classify some of them as presented in
Table 1. The constrains are electrical, mechanical, ther-
mal, and commercial.
The objective in this article is to show the advantages of
the dielectric resonator (DR) technique to satisfy some of
these functions, along with its disadvantages, in compar-
ison with some other well-known solutions.
DRs are suitable for bandpass ltering. DR lters are
classied as three-dimensional (3D) devices, in opposition
to two-dimensional (2D) planar ones.
The main advantages of 2D solutions are their
relative bycompact dimensions, their easier integration
in circuit or module environment, and their well-known
design and manufacturing procedures. They are, however,
limited in their applications to the processing of low pow-
er, sizable relative bandwidth signal, in relation to the
poor unloaded quality factor of localized microwave ele-
ments or planar resonators. Some solutions are proposed
to restrict losses, such as applying supraconductors or ac-
tive-element techniques, but they remain inadequate to
replace 3D devices, in particular for high-power require-
ments.
In the class of 3D devices, designers have rst chosen
waveguides or metallic empty cavities to satisfy their very
narrow bandwidth ltering requirements. However, since
the mid-1980s, high-dielectric-constant materials, having
low loss tangent and good thermal stability, have become
available. The DR solution has been preferred for a num-
ber of applications, in particular spatial ones. This tech-
nique allows us to reduce significantly the cavities and
waveguide device sizes, for equivalent electrical and in-
creased thermal performances. Some average ratios can
be given for dual-mode resonators (DR compared with
cavity):
1 : 4 in volume
1 : 2 in mass
Moreover, the DR shape and the mode in which it is
excited can be chosen to give a response to particular re-
quirements, as we will see later in this article. A number
of DR shapes and lter topologies have, however, been
proposed. Our work here is limited to the presentation of
the most popular ones.
In this article, we present some characteristic
parameters of DR lters, which are generally introduced
974 DIELECTRIC RESONATOR FILTERS
Previous Page
during the synthesis procedure. These definitions are
helpful in explaining the choices of lter designers, in
particular the DR shapes and arrangements in multipole
devices.
In Section 1, the class of devices loaded by cylindrical
DRs is investigated. This is a common shape for the DR. It
can be excited on a symmetric mode (TE
0n
or TM
0m
) or on
a rst hybrid mode (HEM
11
). Different DR arrangements
in the lter are presented and discussed.
For particular applications in microwave ltering,
we can t the DR shape or the DR mode. For high-power
applications, it is important to put the high-dielectric-
constant material in contact with a metallic enclosure, to
improve the thermal dissipation. DRs of a quarter-cut
cylinder, a cylindrical rod, or a dielectric plate shielded in
a metallic cavity are investigated. Solutions are also given
to optimize the isolation of the lter response on the
frequency axis, or to apply DR to the millimeter wave
ltering. These particular applications are included in
Section 2.
Approaches developed to design DR lters are dis-
cussed in Section 4. A four-pole DR lter synthesis is pro-
posed as an example.
1. ELECTRICAL CHARACTERISTIC PARAMETERS OF
A DR FILTER
Applying conventional methods, the design of microwave
lters starts with the selection of an ideal transfer func-
tion that fullls the electrical objectives of the specica-
tions. The synthesis of this ideal transfer function leads to
an equivalent lumped-element circuit. This circuit is char-
acterized by a coupling matrix that depends on the
lumped-element values. A number of studies have been
devoted to this task [1,2]. Different circuit topologies can
be chosen. However, for the DR lters presented in this
article, the equivalent circuit is close to the one presented
in Fig. 1 (canonical symmetric design). Nevertheless,
if another solution is chosen, the same characteristic
parameters may have to be computed.
C C C
R
R R
C R C R C
R
L /3 L /3
L /3 L /3
L /3
L /3
L /2
L /2
L /2
L/2
L /2
L /2
L /2
L /2
1 1 2
n n 1 n/2 1 . . .
. . .
P
1
M
1,2
M
n 1,n
M
2,n 1
M
n/2,n/2 1
M
1,n
P
2
1:n
2
1:n
1
Figure 1. Equivalent circuit of a symmetric n-pole lter.
Table 1. Microwave Filter Specications
Electrical Mechanical Thermal Commercial
Central frequency Mass and volume Temperature range of use Components and materials costs
Passband width Vibration resistance Maximum dissipated power Machining cost
Design cost
Passband ripple Machining tolerances Sensitivity of electrical
response to temperature
variations
Delay for design and realization of lter
Out-of-band selectivity and
rejection
Manufacturing difculties
Response isolation
Insertion losses
SWR
Group delay
Power capabilities
DIELECTRIC RESONATOR FILTERS 975
Fromthe circuit presented Fig. 1, we dene the following:
*
The central frequency of the lter:
f
0
1
2p
LC
p 1
*
The unloaded quality factor of each resonator:
Q
0
Lo
0
R
; o
0
2pf
0
2
*
The input and output coupling coefcients. In the
electrical scheme, the coupling levels between the ex-
citation access (P
1
and P
2
) and the rst and last res-
onators (1 and n) are characterized, respectively, by
the ratios 1/n
1
and 1/n
2
. It is however generally pre-
ferred to dene external quality factors Q
ei
[1] at port
i, i 2 f1; 2g, by
Q
ei
2pLf
0
R
ei
with 3
R
ei
R
0
n
2
i
4
as the external loading resistance at port P
i
. Coupling
coefcients a
i
are also dened at port i by
a
i
Q
0
Q
ei
5
*
The coupling coefcient between resonators. The res-
onators are intercoupled, longitudinally and cross-
wise, in the general case. We dene a coupling
coefcient K
ij
between resonators i and j by
K
ij
M
ij
L
6
The crosswise coupling coefcients (except K
n=2;n=21
)
may cancel to obtain conventional Butterworth or
Chebyshev responses. Some of them are different
from zero and can be negative for elliptic bandpass
function realizations, including transmission zeros in
the out-of-band part of the transmission response.
These parameters help the designer perform the third
synthesis step, the computation of the device dimensions.
The topology of the lter, in particular the DR arrange-
ment, is easily directly deduced from the equivalent-cir-
cuit one, which, in fact, justies this approach. In Section
3, we explain how the dimensions of the lter are com-
puted from knowledge of the parameters f
0
, M
ij
, and Q
ei
.
2. FILTERS COMPOSED OF CYLINDRICAL DRs
Cylindrical DRs are more often used to realize multipole
lters. We discuss this class of solutions here. The DR is
generally shielded in a metallic box, to avoid radiation
losses. It can then be excited on transverse electric TE
0n
modes, transverse magnetic TM
0m
modes, or hybrid
HEM
nm
modes. The natures of the transmission lines or
waveguides used to couple the lter, the nature of the DRs
arrangement in the device, and the nature of the electri-
cal, mechanical, and thermal characteristics of the lter
depend on the choice of the DR mode. Table 2 compares
the performance levels of the TE
01
, TM
01
, and HEM
11
modes. The dimensions are optimized to obtain a resonant
frequency equal to 4GHz. We notice that the DR acting on
the TE
01
mode is the less bulky one. Even if the electrical
performances are comparable, the TM
01
mode is more ra-
diative in the cavity, which increases the metallic losses on
the enclosure. The HEM
11
mode is in fact very interesting
for reduction of lter size. The lter performance of each
particular TE, TM, or hybrid mode is discussed below.
2.1. Monomode Filter
The DRs are excited on the symmetric TM
01
or TE
01
modes. An n-pole monomode lter is then composed of
n DRs. The dimensions of each DR and of the metallic en-
closure are generally chosen to optimize, at the lter cen-
ter frequency, the dielectric and metallic losses and the
device sizes, as well as to avoid spurious responses around
the lter bandpass.
Different topologies of the TE
01
mode, have been re-
ported in the literature. The DRs can be placed side by
side on the same plane. To obtain good isolation on this
mode, the ratio between diameter and height of each DR
is generally chosen to be 2. Then the radial radiation of
each DR is small and the DRs can be coupled directly; a
metallic iris does not have to be placed systematically be-
tween the DRs to limit the lter size.
Two examples of realization are presented in Figs. 2
and 3. Different techniques can be employed to couple the
lter. Propagative rectangular waveguides can be connect-
ed to both ends of a waveguide section above cutoff, which
contains the DR. The TE
01
mode of the DRs is excited if
their axis are positioned along the wide dimension of the
monomode waveguide (Fig. 2). The distances between the
rst (and respectively last) DR and the junctions between
Table 2. Comparison between TE
01
, TM
01
, HEM
11
, and DR
a
Mode Performance Levels
f
0
(GHz) Q
0
D
RD
(mm) H
RD
(mm) D
C
(mm) H
c
(mm)
TE
01
4 8600 14 5.5 28 19.5
TM
01
4 7500 28 8.3 52 56
HEM
11
4 9600 19.3 6 40 20
a
Cylindrical DR (diameter D
RD
, height H
RD
, permittivity e
r
36, loss tangent 10
4
) enclosed in a cylindrical metallic cavity (diameter D
C
, height H
C
, metallic
conductivity 1.7 10
7
S/m).
the waveguide sections enable us to tune the level of the
input (respectively output) coupling coefcient. This tech-
nique is suitable for high-power applications.
To improve the integration of the lter in its environ-
ment, we must couple the rst and last DRs to microstrip
lines (Fig. 3). Nevertheless, the metallic losses of such a
structure increases, as the DR must be placed near the
metallic strip and ground plane to obtain the required
coupling levels. The evolution of the coupling coefcient as
a function of the distance between the line and the DR is
given in Ref. 3.
Dielectric resonators might also be coupled through co-
axial probes [4] or loops [5]. Particular attention to the
positioning of the excitation systems around the DR en-
able us to obtain a good isolation of the bandpass response
on the frequency axis.
From the devices shown in Figs. 2a and 2b [46], we
obtain narrow-bandpass Chebyshev or Butterworth re-
sponses. Stopband lters can also be realized easily using
these DR coupling techniques, coupling the DR to a prop-
agative waveguide or a transmission line [7].
Coupling the DRs side by side enables us to maintain
them easily, such as on a dielectric substrate, whose ma-
terial is chosen to improve the temperature stability of the
lter. Moreover, some tuning elements can be integrated
around the DRs; some metallic or dielectric screws are
generally placed along the DRs axis to tune the lter res-
onant frequency, as well as between the DRs to adjust the
coupling coefcients.
On the TE
01
, TM
01
, and TM
02
modes, DRs have also
been mounted axially on a cylindrical dielectric rod. To
obtain an elliptic response on these symmetric modes, in
the same device we can combine axial and side-by-side
mounting congurations. A two-stage device is construct-
ed. The transversal coupling M
ij
can then be achieved, as
it has been done for empty metallic cavities. A negative
coupling is obtained by setting some upper and lower cav-
ity axis; the resulting transmission zeros placed around
the passband response increase the lter selectivity [8].
Different techniques, including those mentioned previ-
ously, have been developed for the mounting of DRs in
their enclosure; for instance, the DRs can be glued on a
dielectric support. This technique may, however, be
critical because of the generation of parasitic gaps
between the glued materials and the poor glue loss tan-
gent. A mounting based on a differential dilatation
phenomenon between each DR and it environment is
more suitable for obtaining high electrical performance.
The capability of the lter to withstand vibration is
fundamental for space applications. Some test measure-
ments are given, for example, in Ref. 9.
2.2. Dual-Mode Filters
Dual-mode lters [10] are now widely used because they
offer equivalent electrical performance levels, smaller size,
and less mass than do classical fundamental TE or TM-
mode lters. A metallic screw, or another perturbation is
placed around the DR to break the rotational symmetry.
Then, on the rst or second hybrid mode, the two polar-
izations sections are imposed, and their frequencies differ
in relation to the perturbation dimension. Figure 4 pre-
sents a two-pole dual-mode lter, composed of only one
DR. Two monomode DRs would be coupled to obtain
the same electrical response. The DR is excited through
coaxial probes. Two tuning screws are generally added
in the excitation probe axis to tune the central resonant
frequency.
The coupling screw is placed at an angle of 451 from the
excitation axis. In this case, the symmetry of the structure
sufces to x the direction of the two polarizations. The
electromagnetic environment of each mode differs; thus
the resonant frequencies f
1
and f
2
of the two polarizations
differ. The power combining is constructive between f
1
and
f
2
at the output access, which explains the bandpass re-
sponse obtained from this device. Some transmission zeros
are also observed, due to the combination of opposite phases
between the two polarizations, and between these phases
and a higher-order mode of the DR [11].
y
z
X
1 2 n n
a
b
(a) (b)
Evanescent waveguide
Propagative waveguide
Figure 2. DRs excited on a TE
01
mode
through rectangular waveguides.
Microstrip line
Metallic cavity
RD1 RD3 RD2
Figure 3. DR lines excited on a TE
01
mode through microstrip
lines.
For n-pole elliptic realization, with n42 and n as an
odd number, we need to couple parallel modes in adjacent
DRs (longitudinal coupling) and to avoid extra coupling
between orthogonal modes of different DRs. Three screws
are placed around each DR as shown in Fig. 4, and adja-
cent DRs are generally iris-coupled. Two orthogonal rect-
angular apertures machined in a metallic plate enable us
to impose the required coupling level between each set of
parallel polarizations. To obtain a negative sign on some
crosswise coupling coefcients, the different screws are not
positioned at the same angle with the excitations in the
different cavities. This technique is well known for metal-
lic cavity realizations [12]. An example of four-pole topol-
ogy is presented in Fig. 5, where the DRs are coupled to
inputoutput coaxial probes.
The excitation can also take the form of rectangular
waveguides, coupled to the inputoutput cavities through
rectangular irises, for power applications. These irises are
then generally placed in a plane perpendicular to the DR
axis.
Adjacent DRs can also be coupled directly, rather than-
iris coupled [13]. Then realizations of small coupling levels
may require large separation between adjacent DRs, and
may consequently require significant sizes. However, the
introduction of evanescent waveguide sections between
DRs enable us to reduce the device dimensions [14]. The
drawbacks of this method are the lter spurious charac-
teristics and the dependent coupling level between the
mode sets of adjacent DRs. In the same way as monomode
realizations, dualmode DRs can be mounted in a planar
relationship to one another. Each DR is again enclosed in a
metallic cavity. The DR intercouplings are controlled in-
dependently by a metallic iris that contains two rectan-
gular noncrossing apertures placed in an appropriate
manner [15]. This solution is interesting for its exibility
in the arrangement of the DR cavities.
More than two modes have also been coupled in the
same DR cavity, to conserve weight and size in comparison
with the previous solutions. The two polarizations of the
HEM
11
modes and the TM
01
mode have been simulta-
neously excited in a planar DR-mounted cavity. Two of
these three-pole modules have been coupled through an
iris composed of two separate T-shaped apertures [16].
3. PARTICULAR APPLICATIONS OF DRS IN
MICROWAVE FILTERING
3.1. DR for High-Power Applications
The dual-mode devices we have presented above are in-
teresting for their high electrical performance levels and
limited sizes. But even if the lter dissipated power is
small, the resulting thermal dissipation remains critical
for certain applications, such as space applications, be-
cause the thermal conductivity of most dielectric material
is poor.
A solution consists of positioning the high permittivity
resonators in contact with the metallic enclosure, to im-
prove the thermal dissipation efciency, and then to im-
prove the power-handling capability of the lter.
We can rst take advantage of the electromagnetic eld
symmetry of a cylindrical DR. It can be divided into two or
more parts, without modifying the resonant frequency and
eld repartition of some modes, if the physical metallic
walls are placed in planes in which electrical wall condi-
tions are naturally veried [17,18]. In this way a cylindri-
cal DR excited on a TE
01
mode can be split, for example,
into four parts. If each of the cut planes are in contact with
a metallic wall (Fig. 6), all quarter-cut DRs will resonate
at the same frequency.
The improvement of the power-handling capability is
not the sole purpose of this technique. A number of cylin-
drical DR modes do not satisfy the electrical wall condition
in the planes where they are imposed on the quarter-cut
DR. They are then suppressed in the image DR. Hence
the out-of-band rejection performances of the lter is
improved, suppressing spurious responses.
D
Coaxial
probe 1
Coaxial
probe 2
Metallic
screw
Metallic
enclosure
Figure 4. Dual-mode DR excited by two coaxial probes.
D
Coupling
screw
Tuning
screw
Crossing
iris
Coaxial
probe
Figure 5. Example of four-pole dual-mode DR lter.
This technique also provides very compact structures,
reducing the size not only of the DR but also of the metallic
enclosure. This is an important advantage, particularly
for 900-MHz3-GHz applications. Nevertheless, the
metallic losses on the metallic plane in contact with the
DR increase dramatically, resulting in a poor unloaded
quality factor Q
0
of the DR. We can note, however,
that the image resonator might be preferred to coaxial
dielectric eld resonators, considering the sizes and losses
at the same frequency. Figure 7 presents a possible
arrangement of the DRs to realize an elliptic ve-pole
function [17].
Different techniques have been proposed to increase
the unloaded quality factor of the split DR. TM dual-mode
DRs have been developed for use in the L and C frequency
bands for mobile communication applications. A cross is
formed as shown in Fig. 8, by two parallelepipedic DRs
excited on a TM
01
mode [19]. The tuning element required
to couple the two degenerated modes is not a metallic
screw, but a perturbation directly machined near the cen-
ter of the cross. An unloaded quality factor equal to 9000
has been obtained at 1.9GHz, for a dielectric loss tangent
equal to 5 10
5
[20]. A high-permittivity dielectric plate
has also been placed in a metallic enclosure to provide a
good compromise between the unloaded quality factor lev-
el and the thermal dissipation capability [21]. The corners
of a thin parallelepipedic plate have been cut to provide
a good contact between the resonator and a cylindrical
metallic cavity (Fig. 9).
The dimensions are optimized to limit the metallic and
dielectric losses on the rst TE mode that has degenerat-
ed. It has been shown that the electrical performance
of this resonator is not far from that of the cylindrical
(a)
(b)
Metallic
enclosure
Metallic
enclosure
DR
Quarter
cut DR
Figure 6. Equivalent cylindrical (a) and quarter-cut (b) DRs
excited on a TE
01
mode ( electric eld lines).
1
2
3
4
5
Iris coupling between DRs
Waveguide output
Half cylindrical DRs Transversal probe coupling
Figure 7. Five-pole elliptic lter using the
split-DR technique.
DR
Metallic
enclosure
Slot machined to couple
the DR two polarizations
Figure 8. TM dual-mode resonator.
dual-mode one, and this solution is more suitable for
power applications. Metallic screws are generally
placed around the dielectric plate to couple the rst
TE polarizations. Slots are then machined in the plate,
both to couple these polarizations and to optimize the
out-of-band rejection of the lter. Superimposed cavities
are coupled through metallic cross irises. The topology of
an eight-pole autocorrected quasielliptic lter is presented
in Fig. 10. The transmission and reection responses,
along with the group delay of the lter, are presented in
Fig. 11.
Some other studies have been performed to optimize
the DR shape to increase the unloaded quality factor of the
resonators, in particular by choosing adequate forms of
the DR dielectric support in contact with the metallic
enclosure [22].
3.2. Filter Congurations for Optimization of
Out-of-Band Rejection
Conventional DR lters have relative poor stopband re-
jection performance, due to the excitation of higher-order
mode resonances. A lowpass lter, placed at the output of
the bandpass DR lter, might solve this problem, but the
bulk and the electrical performance of the cascaded lter
suffer from this solution. Different lter congurations
have, however, been proposed to increase the out-of-band
rejection.
We have already underlined that the quarter-cut DR,
or the TM
01
DR, which are of interest for power applica-
tions, are also efcient techniques for elimination of some
of the spurious responses.
Coupling structures have been designed to suppress
the excitation of some modes. A single-mode TE
01
lter
realization is, for example, presented in Ref. 23. The
diameter: height ratio of the DR is generally chosen to
optimize the mode isolation. A hole can be machined in the
cylindrical DR, along its axis [24,25], and the DR shape
can be matched [26] to increase this isolation. A more
sophisticated solution consists of mixing DRs excited on
different modes in a same lter. To obtain part of the ring
DR isolation, and part of the compactness of dual-mode
devices, TE
01
ring DRs and a HEM
11
dual-mode DR have
been coupled to realize six-pole elliptic lters, combining
four DRs. DRs can be mounted axially [27] or side by side
[28] in their metallic enclosures. As an example, the center
frequency of the realized lter is equal to 1.23 GHz; its
passband is 20 MHz, and the out-of-band rejection is bet-
ter than 40 dB in the 11.9 GHz frequency band [27].
3.3. DR for High-Frequency Applications
When the frequency increases, the dimensions of cylindri-
cal DRs excited on the rst TE
01
, TM
01
, and HEM
11
modes
become too small. The limit of conventional DR applica-
tions can be set around 20 GHz. To solve the manufactur-
ing problem, spherical DRs have been proposed. But the
critical mechanical stability of the devices, along with the
spurious modes around the bandpass limits their appli-
cations.
H E
M
D
Figure 9. High-permittivity dielectric plate resonator technique.
Cross iris
Coaxial bore
Cross screw
Regulating screw
Figure 10. Dielectric plate eight-pole connect-
ed elliptic lter.
We can, however, use cylindrical DRs on whispering
gallery modes [29]. For important azimuthal variation
number, the DR can be used easily up to 100GHz. More-
over, the eld is very effectively stored in with the DR, and
the unloaded quality factor, which thus depends only on
the material loss tangent, is important. Examples of lter
realizations are given in Ref. 30.
4. THEORETICAL DESIGN OF MICROWAVE DR FILTERS
4.1. Introduction
DRs are used for the realization of narrowband lters up
to 0.01% relative bandwidth. The electrical responses of
such devices are then very sensitive to their geometric and
physical characteristics, and particular attention must be
paid to their design.
The purpose of the theoretical design can be not only to
optimize the lter performances but also to reduce the cost
of the product. If the design is not efcient, the time re-
quired for the tuning can be important; different devices
are manufactured, and even in the phase when the di-
mensions are known, an experimenter must spend time
and effort to tune each lter.
However, DR structures are difcult to analyze, be-
cause their geometries are very complex. We have seen
that the lter topologies are diverse; DRs can be intercou-
pled or coupled through a metallic iris; they may be ex-
cited by coaxial probes, metallic waveguides, and
microstrip lines; their shapes are not systematically
(a)
Center 3 930,000 000 MHz Span 20,000 000 MHz
2: 21.215 dB 10 dB/ Ref .92 dB CH
2
S
21
/M log MAG
35.000 000 MHz
Ref = 1
Marker 2-1
35 MHz
x2
2
3
1
(b)
3: 5.9921 dB 5 dB/ Ref 0 dB CH
2
S
22
log MAG
25.000 000 MHz
Ref = 1
Marker 3-1
25 MHz
x2
2
3
1
C2
(c)
3: 5.9921 dB 5 dB/ Ref 0 dB CH
2
S
22
log MAG
25.000 000 MHz
Ref = 1
Marker 3-1
25 MHz
x2
2
3
1
C2
Figure 11. (a,b) Transmission and reection coefcient variations as a function of the frequency;
(c) group delay variation in lter passband.
cylindrical; and they can be maintained in their metallic
enclosures through a wide variety of systems. Moreover,
the structures are very compact, and if we can dene dif-
ferent segments in its composition, strong couplings are
generated between the DRs through high-order modes.
Thus the classical approach that is applied in the circuit
software, namely, the segmentation method, is not ef-
cient here; we cannot characterize each segment indepen-
dently from the others, and we cannot connect the
different contributions to obtain the device response.
Analytical-approach models were initially developed to
assist designers. These methods are described in Ref. 31.
Since the late 1980s, rigorous analyses have been per-
formed, rst on some parts of the DR lter. Examples
include application of modal methods [32,33], the nite-
element method [34], or the method of lines [35] to char-
acterize axisymmetric dielectric-loaded metallic cavities,
computing the resonant frequencies of these devices. Some
other studies have been performed on the design of
three-dimensional resonator, which is nonsymmetrical in
structure. The resonant frequency of a DR shielded in a
parallelepipedic enclosure has been computed applying
the modal method [36] and the nite-element method
[37,38]. From these computations we can easily deduce
the coupling coefcient between two DRs for symmetric
structures. The coupling coefcient between a DR and a
waveguide or a transmission line has been computed by
applying the nite-element method [39,40]. Now, with the
evolution of computer capabilities, a number of research
teams are interested in the electromagnetic optimization
of DR devices applying numerical simulation. A number of
articles deal with the rigorous design of multipole lters.
In this section we will describe a solution for the rigorous
design of a multipole DR lter using the nite-element
method [40].
4.2. Method for Optimized Design of a DR Filter
The procedure generally applied to determine the geomet-
ric dimensions of a multipole lter is deduced from the
lumped-element synthesis presented in Section 1. To ex-
plain this approach, we have chosen here to design a dual-
mode four-pole DR lter because this design groups to-
gether some problems found in a large variety of DR lter
topologies. In the dual-mode DR presented in Section 1,
the metallic screws are replaced by slots directly ma-
chined in the DR. The four-pole lter, shown in Fig. 12,
consists of a metallic cavity, two input/output coaxial
probes, and two slotted DRs coupled through metallic
cross irises. Because the experimental lter will not be
tuned, the synthesis procedure has to be performed rigor-
ously. To compute the lter dimensions that satisfy given
electrical characteristics, we develop the approach pre-
sented in Fig. 13.
Metallic cavities :
Slotted
resonator
Dielectric resonators :
Dielectric
support
Crossed iris :
Crossed
iris
Notch 3-4
Notch R2
Notch 1-2
Notch R1
Iris 2-3
Iris 1-4
x
y
z
Tuning notches : Coupling notches : Notch-DR distances :
H
C
= 14.1 mm H
DR
= 3.4 mm
d
DR1
= 1.10 mm D
DR1
= 0.62 mm
D
DR2
= 0.62 mm d
DR2
= 1.10 mm
L
14
= 5.20 mm
D
12
= 0.93 mm
D
34
= 0.93 mm
w = 1.00 mm w = 1.00 mm
L
23
= 7.70 mm
w = 1.00 mm
t = 1.00 mm
R
DR
= 7.6 mm
DR
= 37
R
C
= 15.5 mm
Figure 12. Four-pole slotted DR lter.
In the rst stage, an equivalent lumped-element circuit
is synthesized from the ideal transfer function. The syn-
thesis leads to a coupling matrix that characterizes the
ideal equivalent circuit. This objective coupling matrix
gives all the information about the ideal electrical char-
acteristic parameters of the lter.
Applying the 3D nite-element method [40], we can
then compute the initial dimensions of the structure with
respect to the previous electrical parameters. An electro-
magnetic synthesis allows us to determine the following:
1. The DR and metallic cavity dimensions required to
satisfy the center frequency f value.
2. The probe depth penetration, computed to obtain the
required input/output coupling coefcient levels.
3. The dimension of the cross iris to obtain the required
coupling coefcient between the parallel polariza-
tions of the two DRs. The theoretical synthesis
method is generally interrupted here for classical
applications. Screws are then placed around the l-
ter to couple the polarizations and to account for the
resonant frequencies of the lter. Then a set of irises
are manufactured, and each experimental device
has to be tuned. Choosing the slotted DR solution,
we can continue with the synthesis computations.
4. The coupling notch dimensions, which impose the
coupling coefcient between the two polarizations of
each DR.
5. The dimensions of a second notch, which are intro-
duced to compensate for the inuence of the probes
and the iris on the resonant frequency of the excited
polarization.
Then, all the dimensions of the device presented in Fig. 12
are known, but only approximately because of the seg-
mentation approach applied in these initial steps, which
does not account for the indirect dependence between the
different elements.
In the third stage, an electromagnetic optimization loop
is performed applying the following procedure:
1. The 3D nite-element method is applied in order to
compute the scattering parameters between the ac-
cess ports of the whole structure.
2. The scattering parameters are approximated as ra-
tional functions in the frequency domain.
3. From the approximated rational functions, an equiv-
alent circuit, namely, a coupling matrix, of the sim-
ulated lter is synthesized.
4. By comparing the extracted coupling matrix and the
ideal one, the lter dimensions are corrected accord-
ing to the dimension sensitivities from the electro-
magnetic synthesis.
The loop is performed as long as the electrical objective
is not attained.
This procedure is detailed in Refs. 40 and 41. The ideal
transfer function and the electromagnetic response at the
Ok
Ideal transfer function
Lumped element
synthesis
Ideal coupling matrix
Couplings
comparison
Equivalent coupling matrix
Lumped element
synthesis
Characteristic polynomials
Rational
approximation
Scattering parameters
Electromagnetic
analysis
Geometrical dimensions
Electromagnetic
synthesis
Figure 13. Design method of multipole lter.
|S
11
| (dB)
0
10
20
30
40
50
60
5.46 5.48 5.5 5.52 5.54 5.56 5.58 5.6
f (GHz)
|S
12
| (dB)
0
10
20
30
40
50
60
5.46 5.48 5.5 5.52 5.54 5.56 5.58 5.6
f (GHz)
Ideal TF
FEM
Ideal TF
FEM
Figure 14. Comparison of ideal and electromagnetic results.
end of the optimization are compared in Fig. 14. We can
note the good convergence of the electromagnetic model
response with a return loss of 430 dB in the passband. In
Ref. 40, the lter is designed with a slightly different l-
tering pattern. The electromagnetic response along with
the experimental results are presented in Fig. 15. The l-
ter is realized and measured without tuning elements.
The difference between the computed and measured cen-
ter frequencies is less than 0.2%, and the difference be-
tween the bandwidths is less than 5%. The theoretical and
experimental standing-wave ratios are in good agreement.
5. CONCLUSION
DRs are currently placed in a number of devices, especial-
ly in microwave lters. The topologies of these lters are
diverse, and they are a function of the system in which
they are placed. Their main applications are found in the
treatment of power signals and very narrow bandwidth
ltering.
In this article, we have described different topologies
proposed in recent years (as of 2004). A number of re-
search teams are still working to increase the performance
of DR lters, and to widen their domain of applications.
Some of the topics under consideration are the following:
*
The realization of lters in the millimeter wavelength
band
*
The work performed to decrease the volume and
weight of DR devices
*
The realization of DR lters without mechanical tun-
ing for mass production
*
The design of recongurable DR lters for multifre-
quency band applications
*
The development of advanced synthesis techniques
for simplied design of DR lters [42], for example,
with a minimum number of elements
BIBLIOGRAPHY
1. G. Matthaei, L. Young, and E. M. T. Jones, Microwave Filters,
Impedance Matching and Coupling Structures, Artech House,
Dedham, MA, 1963.
2. A. E. Atia, A. E. Williams, and M. W. Newconb, Narrowband
multiple coupled cavity synthesis, IEEE Trans. Circ. Syst.
CAS-21(5):649655 (Sept. 1974).
3. D. Chaimbault, S. Verdeyme, and P. Guillon, Rigorous design
of the coupling between a dielectric resonator and a micro-
strip line, Proc. 24th European Microwave Conf. (EMC),
Cannes, Sept. 58, 1994.
4. Y. Kobayashi and M. Minegishi, A low loss band pass lter
using electrically coupled high Q TM
01d
dielectric rod resona-
tors, IEEE Trans. Microwave Theory Tech. 36(12):17271732
(Dec. 1988).
5. T. Nishikawa et al., Microwave Bandpass Filter Provided
with Dielectric Resonators, U.S. Patent 4,143,344 (March 6,
1979).
6. S. B. Cohn, Microwave band pass lters containing high Q
dielectric resonators, IEEE Trans. Microwave Theory Tech.
16:218277 (April 1968).
7. C. V. Ren, Waveguides band stop lter utilizing Ba
2
Ti
9
Oz
0
resonators, Proc. 1978 Int. Microwave Symp. June 1978,
p. 227.
8. A. E. Atia and A. E. Williams, General TE
011
mode waveguide
band pass lters, IEEE Trans. Microwave Theory Tech.
24(10):640 (Oct. 1976).
9. S. Vigneron, B. Theron, V. Madrangeas, F. Tounsi, and P. Gu-
illon, Dielectric resonators lters for space applications, Proc.
ESA-ESTEC Workshop on Microwave Filters for Space
Applications, June 1990, pp. 7180.
10. S. J. Fiedziusko, Dual mode dielectric resonator loaded cavity
lters, IEEE Trans. Microwave Theory Tech. 30(9):13111316
(Sept. 1982).
11. D. Baillargeat, S. Verdeyme, and P. Guillon, Symmetrical
quasi elliptic two pole dielectric resonator lter, Electron.
Lett. 31(11):893895 (May 25, 1995).
12. A. E. Atia and A. E. Williams, New types of waveguide band
pass lters satellite transponders, Comsat Tech. Rev. 1(1):
2143 (1971).
13. K. A. Zaki, C. Chen, and A. E. Atia, Canonical and
longitudinal dual mode dielectric resonator without iris,
|S
12
| (dB)
0
10
20
30
40
5.45 5.50 5.55 5.60 5.65
f (GHz)
: Measure
: FEM
|S
11
| (dB)
0
10
20
30
40
5.45 5.50 5.55 5.60 5.65
f (GHz)
: Measure
: FEM
Figure 15. Comparison of theoretical and measured results.
IEEE Trans. Microwave Theory Tech. 35(12):11301135 (Dec.
1987.)
14. S. C. Chen and K. A. Zaki, A novel coupling method for
dual mode dielectric resonators and waveguide lters,
IEEE Trans. Microwave Theory Tech. 38(12):18851892
(Dec. 1990).
15. W. C. Tang, D. Siu, B. C. Beggs, and J. Sferazza, Planar Dual
Mode Cavity Filters including Dielectric Resonators, U.S.
Patent 4,652,843, March 24, 1987.
16. W. C. Tang, D. Siu, B.C. Beggs, and J. Sferazza, Triple Mode
Dielectric Loaded Band Pass Filter, U.S. Patent, 4,675,630
(June 23, 1987).
17. T. Nishikawa, K. Wakino, K. Tsunoda, and Y. Ishikawa,
Dielectric high power band pass lter using quarter cut
TE
01d
image resonator for cellular base structures, IEEE
MTT-S Digest, 1987, pp. 133136.
18. R. J. Cameron, W. C. Tang, and L. M. Kudsia, Advances in
dielectric loaded lters and multiplexers for communications
satellites, Proc. ESA/ESTEC Workshop on Microwave Filters
for Space Applications, Noordwijk June 1990.
19. K. Wakino, T. Nishikawa, and Y. Ishikawa, Miniaturization
technologies of dielectric resonator lters for mobile commu-
nications, Proc. MTT Workshop on Filters and Multiplexers
for Mobile Communications, June 17, 1993.
20. Y. Ishikawa, J. Hattori, S. Abe, E. Kobayashi, and T. Nishi-
yawa, 1.9 GHz compact low loss dielectric diplexer designer
by dual mode waveguide transmission line method, Proc.
24th European Microwave Conf. (EMC), Sept. 1994, Vol. 2,
pp. 11791184.
21. S. Moraud, S. Verdeyme, P. Guillon, Y. Latouche, S. Vigneron,
and B. Theron, A new dielectric loaded cavity for high power
microwave ltering, IEEE MTT-S Digest, 1996, Vol. 2,
pp. 615618.
22. S. Gendraud, S. Moraud, S. Verdeyme, P. Guillon, S. Vigne-
ron, and B. Theron, New miniaturized dielectric loaded cavity
microwave resonators, Proc. 25th European Microwave Conf.
(EMC), Bologne, Sept. 1995, pp. 691695.
23. R. V. Snyder, Dielectric resonator lter with wide stop bands,
IEEE Trans. Microwave Theory Tech. 40:21002102 (Nov.
1992).
24. Y. Kobayashi and M. Miura, Optimum design of shielded
dielectric rod and ring resonators for obtaining the best
mode separation, IEEE MTT-S Digest, 1984, Vol. 1, pp.
184186.
25. S. W. Chen and K. A. Zaki, Dielectric ring resonators loaded
in waveguide and on substrate, IEEE Trans. Microwave The-
ory Tech. 39:20692076 (Dec. 1991).
26. R. M. Mansour, Dual mode dielectric resonator lter with
improved spurious performance, IEEE MTT-S Digest, 1993,
pp. 443446.
27. J. F. Liang, K. A. Zaki, and A. E. Atia, Mixed modes dielectric
resonator loaded cavity lters, IEEE MTT-S Digest, 1994,
pp. 731734.
28. C. Wang, H. W. Yao, and K. A. Zaki, Mixed modes cylindrical
planar dielectric resonator lters with rectangular enclosure,
IEEE MTT-S Digest, 1995, Vol. 2, pp. 501504.
29. X. H. Jiao, P. Guillon, L. A. Bermudez, and P. Auxemery,
Whispering gallery modes of dielectric structures: Application
to millimeter wave bandstop lters, IEEE Trans. Microwave
Theory Tech. 35(12):11691175 (Dec. 1987).
30. D. Cros, F. Nigon, P. Besnier, M. Aubourg, and P. Guillon,
Whispering gallery dielectric resonator lters, IEEE MTT-S
Digest, 1996, Vol. 2, pp. 603606.
31. D. Kajfez and P. Guillon, Dielectric Resonators, Artech House,
Dedham, MA, 1986.
32. W. K. Hui and I. Wolff, A multicomposite, multilayered cylin-
drical dielectric resonator for applications in MMICs. IEEE
Trans. Microwave Theory Tech. 2(3):415423 (March 1994).
33. K. A. Zaki, Electromagnetic simulation of dielectric resonator
devices using mode matching technique, Proc. MTT-S Work-
shop WFFA, Dielectric Resonators in Microwave Active and
Passive Circuits, San Francisco, 1996.
34. P. S. Kooi et al., Finite element analysis of the shielded cy-
lindrical dielectric resonators, IEE Proc. 132(Part H)(1):716
(Feb. 1995).
35. D. Kremer and R. Pregla, The method of lines for the hybrid
analysis of multilayered dielectric resonators, IEEE MTT-S
Digest, Orlando, 1995, Vol. 2, pp. 491494.
36. X. P. Liang and K. A. Zaki, Modeling of cylindrical dielectric
resonators in rectangular waveguides and cavities, IEEE
Trans. Microwave Theory Tech. 41(12):21742181 (Dec. 1993).
37. R. Ferrari and G. Maille, Three dimensional nite element
method for solving electromagnetic problems, Electron. Lett.
14(15):467468 (July 1978).
38. S. Verdeyme, M. Aubourg, and P. Guillon, Three dimensional
nite element method applied to dielectric resonators devices,
IEEE MTT-S, Los Angeles, 1989.
39. J. P. Cousty, S. Verdeyme, M. Aubourg, and P. Guillon, Finite
element for microwave device simulations: Applications to
microwave DRs lters, IEEE Trans. Microwave Theory Tech.
40(5):925932 (May 1992).
40. D. Baillargeat, S. Verdeyme, P. Guillon, and M. Guglielmi,
New dual mode coupling using slotted dielectric resonator,
Microwave Opt. Technol. Lett. (Jan. 1993).
41. S. Bila, D. Baillargeat, S. Verdeyme, M. Aubourg, P. Guillon,
F. Seyfert, J. Grimm, L. Baratchart, C. Zanchi, and J. So-
mbrin, Direct electromagnetic optimization of microwave l-
ters, IEEE Microwave Mag. 2(1):4651 (March 2001).
42. S. Bila, D. Baillargeat, S. Verdeyme, F. Seyfert, L. Baratchart,
C. Zanchi, and J. Sombrin, Simplied design of microwave
lters with a symmetric transfer functions, Proc. 33rd Euro-
pean Microwave Conf. (EuMW 03), Munich, Germany,
Oct. 2003.
DIELECTRIC RESONATOR OSCILLATORS
VINCENT F. FUSCO
Queens University of Belfast
Belfast, Northern Ireland
ANDREW DEARN
Plextek Ltd.
Grear Chesterford, United
Kingdom
Oscillators producing energy at microwave frequencies
are an essential component in most microwave communi-
cation systems, such as communication links, radar, and
frequency synthesizers. There are significant commercial
pressures to improve the performance of oscillators with
respect to giving them lower noise characteristics, higher
DC-to-RF efciency, better temperature and frequency
stability, and so on. The dielectric resonator when coupled
with two/three-terminal active devices provides a vehicle
for producing high-quality xed-frequency or narrowband
DIELECTRIC RESONATOR OSCILLATORS 985
tunable oscillators. Low-loss temperature-stable dielectric
materials with high-quality Q factors mean that minia-
ture resonators can be formed that have physical compat-
ibility with microwave integrated circuits [1]. A dielectric
resonator (DR) can be used to form the stabilizing element
in an oscillator. Dielectric resonator oscillators (DROs) are
characterized by the following properties: high-frequency
stability, high efciency, and low manufacturing cost. They
also can be made to provide good temperature stability.
Two- and three-terminal active devices exhibit phase
noise characteristics at microwave frequencies. Oscillators
constructed using these devices can have their phase noise
behavior improved by the addition of a DR element.
For a three-terminal device the dielectric resonator for
rst-order feedback topologies [2] can occupy one of sev-
eral positions within the circuit, as illustrated in Fig. 1.
In Fig. 1a the DR is mounted in order to provide par-
allel feedback to the three-terminal active device. Here the
coupling action is between device ports [3].
In Fig. 1b the DR is located so that energy produced as
a result of negative resistance obtained at the device port
is reected back with correct phasing into the active de-
vice by the DR, which presents a high impedance at the
device port. The DR acts as a bandstop lter, and in this
way an oscillation is set up. This conguration has good
phase noise characteristics. However, it is sensitive to load
variation and tends to mode-jump in the presence of two
resonant circuits [4]. The conguration in Fig. 1c gener-
ally provides superior performance compared to the other
two methods and is compatible with the off-chip bonding
processes required for stabilizing monolithic microwave
integrated circuit (MMIC) oscillators. Here the DR acts as
a very high-Q resonator element for the series feedback
oscillator, giving good frequency stability and low phase
noise. In a eld-effect transistor (FET) realization, where
the DR is coupled to the gate circuit and the feedback is a
common source, the operating point is very insensitive to
load variation, due to intrinsic isolation between the input
and output provided by the low gate-to-drain capacitance
of the active device. A varactor diode can be added for
tuning purposes. Higher-order feedback implementations
[5], for example, shunt feedback DRO congurations such
as those given in Fig. 2, are also possible [6]. Multiple DR
and pushpush frequency doubling oscillator congura-
tions are also possible [7].
In Fig. 2 the DR behaves as a high-Q lter in the pos-
itive feedback path. In Fig. 2a part of the output signal is
coupled back to the input port. By adjusting lengths
1
and
2
, the Barkhausen oscillation condition can be satised.
In Fig. 2b the DR is coupled between two ports and the
output is taken from the third. Here the position of the DR
is adjusted in order to maximize the negative resistance at
the output port of the active device.
In the shunt feedback congurations the two coupling
coefcients in the parallel feedback case cannot be adjust-
ed separately. Also, since the coupling is to an open-circuit
line, these types of circuits tend to be very sensitive to the
lateral position of the DR.
The selection of the active device for oscillator applica-
tions depends on the frequency of operation. Below 12GHz,
bipolar junction transistors (BJTs) and HBT nd application
due to their superior icker 1/f and noise levels [8]. A 4-GHz,
21-dBmseries feedback bipolar transistor DRoscillator with
a phase noise spectral density of 130dBc/Hz at 10kHz
from the carrier was reported [9]. Values of 89dBc/Hz at
10kHz offset have been reported at 21.4GHz for a reection
type DRO producing 10dBm output power [10]. Stabilized
Gunn oscillators operating at 35GHz [11] and HBT DROs at
25GHz have also been reported [12].
It is significant to note that generally as frequency in-
creases, output power can be maintained only at the ex-
pense of increased phase noise, since the DR must be
coupled more tightly to the circuit.
The key elements to be optimized with respect to oscil-
lator performance are
Output power
Startup stability
Phase noise
DC-RF efciency
Tuning range
Sensitivity to DR placement
Frequency pushing/pulling
For a very high-frequency operation into the millimeter-
wave region, or for very high Q-factor operation in the
centimeter wavelength region, DRs operated in whisper-
ing gallery mode [13] are employed [1416]. Here the im-
portant dimension is the circumference rather than the
diameter of the DR. This leads to more practical-sized DRs
for higher-frequency circuits.
DR
DR
DR
O/P
O/P O/P
A
OR
B
B A
Z
0
(a) (b) (c)
Figure 1. Typical DRO topologies.
Z
L
Z
L
1
/4
DR
DR
2
(a) (b)
Figure 2. Shunt feedback topologies.
986 DIELECTRIC RESONATOR OSCILLATORS
1. DIELECTRIC RESONATORS
A dielectric resonator (DR) is a high-dielectric-constant
ceramic material formed into a regular geometric shape,
usually a solid or hollow cylindrical or cuboid shape. The
material usually has a relative dielectric constant of
around 3040 (but can be as high as 92) and exhibits
low-loss characteristics. The high permittivity of the ma-
terial means that energy can be stored within the DR. The
DR can resonate in various modes that are governed by
the dimensions, geometry, and electrical properties of the
DR itself and by the physical environment in which it is to
operate. The development of temperature-stable high-
dielectric-constant dielectric materials means that it is
now possible to construct resonators that exhibit mini-
mum change in resonant frequency with temperature.
The most commonly used DR shape is a solid cylinder.
Here the most common mode of operation is denoted the
TE
01d
or dipole mode (Fig. 3).
For a dielectric constant of 40, only 5% of the electric
eld and 40% of the magnetic eld exist outside the DR.
This energy decays rapidly as the distance from the DR
surface increases [17]. The resonant frequency of an
isolated cylindrical DR is given approximately by Gisson
[18] as
f
GHz
34
a

e
r
p
a
3:45
_ _
1
where a, are in millimeters and 0:5oa=o2, while
30oe
r
o50.
For operation in fundamental mode the dimensions of
the DR are approximately one guide wavelength:
lg l
0
=

e
r
p
.
The lowest frequency of operation is limited by tolera-
ble DR size; typically this is 1 GHz, while the highest fre-
quency, about 100GHz, is governed by internal losses and
minimum resonator dimensions [19]. In order to avoid
spurious modes, the length of the DR is usually restricted
to lie between 0:175a0:225a.
2. DIELECTRIC RESONATOR MATERIAL PROPERTIES
The Q factor and temperature stability provided by di-
electric resonators are invariably impaired by imperfect
material parameters. The material from which the reso-
nator is constructed will have losses produced by its nite
conductivity and also by polarization-induced damping
under radiofrequency (RF) excitation conditions [20].
If dielectric loss within the resonator is denoted Q
d
,
then
Q
d
W
W
d
tand
2
where tan d is the loss tangent of the resonator material, W
is the total energy stored in the cavity, and W
d
is the
energy stored inthe dielectric resonator. Q
d
is oftenquoted as
Q
d
C
f
3
where C is a constant quoted by the resonator manufac-
turer.
Sometimes the loss tangent is given as
tand ABf 4
where A and B are constants quoted by the manufacturer.
A more complete description of the losses in the reso-
nator yields the total dissipated power in the resonator
P
tot
to be
P
tot
P
d
P
c
P
r
P
ext
5
where
P
d
power dissipated in the dielectric material
P
c
power dissipated in the surrounding enclosed metal
P
r
radiation loss
P
ext
power coupled to the external circuit
From this, Q factors for each term can be related to
Dielectric loss
Q
d
oW
e
P
d
6
Conductor loss
Q
c
oW
e
P
c
7
Radiation loss
Q
r
oW
e
P
r
8
External Q factor
Q
ext
oW
e
P
ext
9
where W
e
is the total electric energy stored in the cavity
dened by the shield enclosing the dielectric resonator.
E Field
H Field
a
L
Figure 3. Cylindrical dielectric resonator T
o1d
mode.
The total loaded Q factor for the enclosed DR is
1
Q
L
1
Q
d
1
Q
c
1
Q
r
1
Q
ext
10
while the unloaded Q factor is
1
Q
u
1
Q
d
1
Q
c
1
Q
r
11
Techniques for measuring these quantities have been sug-
gested in Refs. 2123.
Temperature-dependent effects that cause a change in
the DRs stabilizing function within an oscillator also need
to be quantied.
The linear coefcient of expansion for a material is de-
ned as the change in length of a rod of the material D,
divided by its length L such that
D
L
aT 12
where a (ppm/1C) is the linear expansion coefcient. As a
DR puck expands or contracts, its resonant frequency will
vary such that
Df
f
o
aDT 13
The negative sign indicates that as the DR puck becomes
longer, its resonant frequency decreases.
With a DR, its relative permittivity e
r
is also a function
of temperature. This variation is expressed very approx-
imately as
De
r
e
r
t
E
DT 14
where t
E
is the temperature coefcient of the dielectric
resonator (ppm/1C).
By combining these relationships together, we obtain
an approximate equation for the temperature stability of a
DR:
Df
f

@f
@L
DL
f

@f
@e
r
De
r
f
15
By denoting the temperature coefcient of the resonant
frequency of the DR as t
DR
, we obtain
t
DR
a
T
e
2
16
This equation implies that by making t have twice the
magnitude of a and giving it an opposite sense of opera-
tion, t
e
can be reduced to zero; that is, the DR can be tem-
perature-compensated. Note, however, that since the DR
will expand as temperature is increased, t must always be
negative for temperature compensation to occur.
Now consider the effect that these temperature changes
have on the stability of a DRO. If an unstabilized oscillator
has a negative frequency drift with temperature, then a
DR with a positive temperature coefcient is required so
that temperaturefrequency stabilization can be achieved.
The frequency stability t
f
of a DRO has been modeled in
Ref. 24 as
t
f
t
DR
k 2
4Q
u
_ _
@f
@t
17
where @f/@t is the temperature-induced phase variation of
the active-circuit one-port measurement
t
DR
1
f
o
df
o
dT
The second term just stated indicates that for a free-run-
ning oscillator the frequency drift is amplied by an
amount proportional to the coupling coefcient of the DR.
By arranging Eq. (17), it is possible to nd the desired
value for t
DR
that yields zero t
f
t
DR

k 2
4Q
u
_ _
@f
@t
18
Here @f/@t and t
DR
are unknown. The in situ value for t
DR
when placed in its operating conguration can be found
using a load-pull technique.
Alternatively if @f/@t is assumed to be constant for a
small change in coupling coefcient k, then Eq. (17) can be
used to form two simultaneous equations from which the
in situ t
DR
can be obtained directly. This second method
should be used with caution since @f/@t is nonlinear.
From this discussion it is clear that judicious selection
of t
DR
and coupling coefcient can be used to compensate
for temperature-induced DRO frequency drift [24].
3. QUALITY FACTOR
Quality or Q factor relates energy stored to average power
loss. In a DR-based circuit, this is of critical importance
since it is a measure of the resonator bandwidth, which is
inversely proportional to the Q factor. Temperature sta-
bility and AM/FM noise performance of dielectric resona-
tor oscillator circuits also depend on the Q factor:
Qo
0
energy stored
average power loss
o
0
resonant frequency rad=s
19
A second equation for quality factor that relates to group
delay through a resonant circuit is useful for oscillator
work [25]. This is the loaded quality factor Q
L
:
Q
L
o
0
t
2

o
0
720
df
df
20
Here, t group delay(s) and fphase of the open-loop
voltage transfer function (degrees). The loaded Q is used
in oscillator design to express the width of the phase slope
and resonance curve, including the effects of external com-
ponents. As a consequence, Q
L
is dominated by compo-
nents external to the DR that have their own Q factor,
called, Q
ext
.
The unloaded Q factor Q
u
is used when the Q of the
resonant circuit is determined only by dissipation losses
in the resonator. These various Q factors are related as
follows:
1
Q
L
1
Q
u
1
Q
ext
21
For oscillator design Q
L
can be used in the determination
of single-sideband noise prediction.
4. LUMPED-ELEMENT COUPLING MODELS
By assuming that the aspect ratio of the DR and the fre-
quency of operation have been selected such that only a
single mode is excited then an isolated DR can be repre-
sented as a series resonant circuit with equivalent-circuit
components R
r
, C
r
, and L
r
[26].
Figure 4 shows a DR coupled to a short section of
microstrip line, at reference plane pp
1
, which is repre-
sented by R
1
, C
1
, and L
1
. When the resonator is excited in
T
o1d
mode, the coupling can be represented by a mutual
inductance term L
m
, which is proportional to the separa-
tion distance d, between the microstrip line and the DR
(Fig. 4) [27].
If the microstrip line is assumed to have zero loss in the
coupling region, R
1
can be neglected. By neglecting the
microstrip-line capacitance, the resulting simplied
equivalent circuit can be expressed close to resonance as
a parallel resonant circuit (Fig. 5) where [28]
R
o
2
L
2
m
R
r
c
L
r
o
2
L
2
m
Lo
2
L
2
m
C
r
The general expression for a parallel resonant circuit is
Z
R
1j2Q
0
d
22
Hence
Q
u
o
0
L
r
R
r
o
0
RC 23
o
2
0
1
L
r
C
r
1
LC
24
d
o o
0
o
25
Thus the equivalent circuit impedance (Fig. 4) can be
written as
Z
T
joL
1
R
1 j2Q
u
d
26
These results are valid close to the fundamental resonance
TE
01d
mode of the DR where d
2
tends to zero. Outside this
frequency range other DR modes exist, which can be mod-
eled by a Foster-type equivalent circuit consisting of a se-
ries cascade of parallel tuned circuits [29]. With this
approach characterization of the ith resonant circuit re-
quires the ith resonant frequency, unloaded Q factor, and
effective coupled resistance to be determined. In addition,
when the resonant frequencies of several modes occur in
close proximity the individual modal performances cannot
be easily established because of mode interaction [30,31].
It is useful to note that TE
01d
mode can be well separated
from other modes by correct selection of DR and enclosure
dimensions.
When integrating the DR into an oscillator by means of
a microstrip connecting line, the parallel tuned circuit in
Fig. 5 becomes externally loaded as shown in Fig. 6.
Here the DR is loaded by the internal impedance of the
generator and the load. If Z
g
and Z
L
are assumed to be real
with line lengths
1
and
2
reduced to zero at resonance,
the parallel structure reduces to R. From Eq. (27) the val-
ue of R depends on the amount of coupling between the DR
and the line. A coupling coefcient term observed at the
input port is dened as the ratio of the resonator coupled
resistance k at the resonator frequency to the resistance
external to the resonator [32]
k
R
R
ex
27
P
R
R
R
1
C
1
L
R
C
R
L
M
P
Figure 4. Equivalent circuit of DR coupled to microstrip line.
P P
R
C
L
L
1
Figure 5. Simplied model of DR coupled to microstrip line.
or
k
Q
u
Q
e
28
for a DR coupled to a matched line R
g
R
L
Z
0
, here
k
R
2Z
0
29
for a DR coupled to a short-circuited load R
L
0, here
k
R
Z
0
30
When a reactive termination loads the resonant circuit,
this presents itself in series with the parallel equivalent
circuit. This will cause an imperfect match and will also
cause an additional reactive load to appear in series with
the equivalent circuit [33].
If x
1
is the normalized reactance appearing in series
with the parallel resonant circuit, then the definition for
the coupling coefcient observed at the input port becomes
k
R
Z
N
1 x
2
1
31
and for a short circuit
k
R
Z
0
1 x
2
1
32
here
x
1
oL
1
Z
N
33
where Z
N
is the total impedance loading the DR.
Also under these conditions the resonant frequency
shifts to a new frequency f
L
where
f
L
f
0
1
x
1
k
2Q
u
_ _
34
and
Q
u
Q
L
1k 35
When coupled to a matched microstrip line, x
1
is usually
much less than one.
When the microstrip-line lengths
1
,
2
are not equal to
zero, then the DR is coupled to the generator and load ter-
minations Z
L
,Z
g
via microstrip-line segments
1
;
2
(Fig. 6).
For a matched line Eq. (28) is valid. For a short-circuit
load and assuming lossless line
k
R
Z
0
1 tan
2
b
2
36
Here maximum coupling occurs at
nl
g
4
; n0; 2; 4 37
while minimum coupling occurs at
ml
g
4
; m1; 3; 5 38
where l
g
effective wavelength of the microstrip line.
The same result occurs when considering the imped-
ance presented at the input port under similar conditions.
Terminating the line in an arbitrary reactance shifts
the maxima and minima of the magnetic eld along the
line. Since the TE
01d
mode couples to the magnetic eld,
line coupling will be maximized at peaks in the magnetic
eld standing wave and minimized at troughs. Coupling to
the maximum point on the standing wave will present the
coupling coefcient given by Eq. (29) at n0,2,4, and so
on. The maximum coupling coefcient is reported experi-
mentally to consistently occur at a distance d0.7a,
where a is the radius of the DR puck [33].
5. MEASUREMENT OF DR COUPLING COEFFICIENT
The most convenient method for experimentally estimat-
ing coupling coefcient k involves a measurement of load-
ed quality factor Q
L
, which is related to k as
Q
L
Q
u
1 k
39
However, Q
u
is not normally known and has to be deter-
mined by the losses at o
0
. From Ginzton [34] and Khanna
Z
g
P
P P
1
1
Z0
Z
g
Z
L
Z
in
Z
R
Z
L

Z0
2
2
Microstrip line
P
P P
Z
L
DR
Figure 6. Oscillator DR coupling.
[35], Q
L
can be established from the phase of the input
impedance locus obtained by a one-port measurement
made on the DR
Q
L
f
0
f
4
f
3
40
where f
0
is the center frequency, f
3
is the frequency at
which the phase is lagging that at f
0
by 451, and f
4
is the
frequency at which the phase is leading that at f
0
by 451.
The results obtained by this method are typically 10%
greater than those obtained by the method of deembedded
S parameters shown next.
By assuming zero radiation loss at the DR microstrip
junction, it can be shown from scattering parameter the-
ory that for a matched system
k
1 S
11
2
S
21
2
2S
21
2
41
A number of useful computer source codes for computing
the resonant frequency and unloaded Q factor for shielded
DRs are given in Ref. 36.
6. DIELECTRIC RESONATOR COUPLED TO A
MICROSTRIP LINE
The most frequently used method of integrating a DR into
a microwave integrated circuit is to couple it to a micro-
strip line. Here the problem is conned to the TE
01d
mode
of a cylindrical DR (Fig. 1). Here the sufxes refer to the
standing-wave pattern in the azimuthal, radial, and axial
directions, respectively, for the fundamental TE mode
d 1.
A typical coupling conguration is shown in Fig. 7. An
approximate representation of the H-eld lines is given in
order to illustrate the inductive nature of the coupling be-
tween a DR in TE
01d
mode and the quasi-TEM mode of
microstrip line.
Here the magnetic eld lines match each other princi-
pally on the radial direction of the DR eld directly under
the microstrip line. The coupling between the DR and the
line is inversely proportional to the separation and, be-
tween them, is dened by a mutual inductance coupling
coefcient term k. With this type of coupling a proportion
of energy is radiated away from the DR. Therefore the ef-
fect of losses in the microstrip substrate, signal line, and
enclosure act to perturb the electromagnetic eld and
alter the Q factor of the DR [37,38]. For oscillator design
an equivalent circuit model for the line DR coupled
arrangement is required. The DR may be xed to the
line using a low-loss adhesive [39,40]. If the adhesive has a
slow setting time, then reworking of the DR position for
tuning can be made.
7. DIELECTRIC RESONATOR TUNING
For optimum phase noise designs the oscillator center fre-
quency should be equal to that of the dielectric resonator.
For many circuits it is useful to have a DR tuning facility.
One way of achieving this is to use mechanical tuning
(Fig. 8).
Here a metal [41], dielectric [42], or DR [43] plunger is
inserted in either the topwall or the sidewall of the struc-
ture. The presence of the tuning element causes a local-
ized distortion of the electromagnetic elds. With a metal
plunger, the additional losses in the plunger lower the Q
u
of the resonator as the coupling between DR and plunger
becomes tighter, that is, as L is reduced. For a metal
plunger as L is reduced, the frequency of operation is in-
creased, while the reverse is true for a dielectric plunger.
Generally a DRO used with the metal plunger tuning
arrangement also produces a reduced output power as L
decreases. With this arrangement, about a 3% tuning
range can be made.
With a second DR replacing the plunger topcap in
Fig. 8, the resultant change in Q
u
is smaller than for the
metal tuner case. This arrangement yields about an 8%
tuning range.
Electronic tuning of the DR is also possible using a
varactor diode as the tuning element. This can be mount-
ed internally to the DR [44] or more usually as the termi-
nation on a line coupled to the DR (Fig. 1). A large tuning
range requires tight coupling between the DR and the
varactor, which inevitably leads to a reduction in Q
u
.
If Q
u
is the unloaded Q of the varactor, then [45]
Q
ut
Q
u
2
f
0
Df
_ _
42
Metal
enclosure
DR
Elevation Plan
Tuning plate
d
Microstrip
Microstrip line
DR
P
P
H field
R
d + a
Figure 7. DR coupled to microstrip line,
physical and electrical equivalent.
where Q
ut
is the unloaded Q of the varactor tuned DR and
Df is the change in resonance frequency. According to
Ref. 45, an approximately 0.5% tuning bandwidth can be
obtained with this arrangement.
Other techniques such as optical tuning [46], magnetic
tuning [47], and segmented disk tuning also exist [48].
8. OSCILLATOR BASIC THEORY
8.1. Negative Resistance
In order for microwave oscillation to begin, a means to
overcome the resistive losses in the circuit has to be pro-
vided. These losses include undesired stray and parasitic
resistances, and also the load into which the oscillator
must operate (usually 50 O). To do this, the concept of
negative resistance is introduced. To illustrate the concept,
consider the simple equivalent circuit of Fig. 9. If the
amount of negative resistance exactly cancels the sum of
the positive resistances, then the circuit will have a res-
onant or oscillation frequency where the inductive reac-
tance is equal to the capacitive reactance; that is
2pfL
1
2pfC
43
where f resonant frequency, Linductance, and C
capacitance. This condition is known as the steady-state
oscillation condition. In order for oscillation to build up,
an excess of negative resistance is required. Then, any
small perturbation in the circuit, such as electrical noise,
will kickstart a resonance. As oscillation builds, the
amount of negative resistance decreases until the
steady-state condition is met. A negative resistor can be
formed from an unstable active device with suitable feed-
back applied. Consider a transistor with feedback such
that the magnitude of the input reection coefcient (S
11
)
is greater than one. This means that when an AC signal is
incident on the port, more energy is reected than is ac-
tually incident. This can be regarded as a negative resis-
tance.
8.2. Oscillator Equation
The one-port negative-resistance oscillator schematic is
illustrated in Fig. 10. The oscillator can be considered as a
two-port negative-resistance circuit and a one-port reso-
nator. The resulting oscillator then operates into a
one-port load. The large-signal steady-state oscillation
condition is given by
G
r
.
S
0
11
1 44
where G
r
is the reection coefcient of the resonator and
S
0
11
is the large-signal input reection coefcient of the
negative-resistance circuit, when terminated in the load.
The reection coefcients stated immediately above are
complex numbers, and so the equation can be expanded
into its magnitude and angle parts:
jG
r
j
.
jS
0
11
j 1 45
AngG
r
AngS
0
11
0 46
If the equations are now expanded into impedance forms,
then we obtain
RR
n
0 47
X
L
X
C
0 48
where R is the sum of positive resistances, R
n
is the neg-
ative resistance, X
L
is the inductive reactance, and X
C
is
the capacitive reactance. Equation (48) simply expands to
Eq. (43) and determines the resonant frequency.
We have already stated that before steady-state oscil-
lation can be achieved, there must be an excess negative
resistance to enable resonance to start. At the buildup of
DR
Gnd
R2
R1
L
Figure 8. Mechanical DR tuning arrangement.
L R R
C
Figure 9. Simple equivalent circuit of a microwave oscillator.
1-Port oscillator
Resonator 2-Port NRC
S11 R L S22
Load
Figure 10. Negative-resistance oscillator schematic showing
port definitions.
oscillation, Eqs. (47) and (48) can be approximated to
r r
n
o0 49
x
L
x
C
0 50
where the resistances and reactances are now small-sig-
nal values. In terms of reection coefcients, the small-
signal approximation to the oscillation condition becomes
jG
r
j
.
jS
0
11
j > 1 51
AngG
r
AngS
0
11
0 52
where the reection coefcients are now the more easily
measurable small-signal values. It should be noted here
that G
r
depends on the value of characteristic impedance
Z
0
used to calculate it. Hence it is possible with this ap-
proach to nd a startup condition in terms of G in one port
but not in another for a given Z
0
.
8.3. Oscillator Graphical Analysis
The solving of the small-signal oscillation condition is il-
lustrated graphically in Fig. 11 with reference to Fig. 10.
Here the Z (complex) represents the overall impedance of
the closed-loop oscillator, including the load. The frequency
at which the imaginary part goes through zero (i.e., res-
onance) is clear, and at this frequency there is an excess
negative resistance.
8.4. N-Port Oscillation Condition
It can be shown that the condition for oscillation is also
present at the output port of the previously discussed neg-
ative-resistance circuit. This means that we could also
solve
G
L
.
S
0
22
1 53
where G
L
is the reection coefcient of the load and S
0
22
is
the large-signal output reection coefcient of the nega-
tive resistance circuit, when terminated in the resonator.
This condition can be shown to hold at all ports of an
N-port oscillator [49].
8.5. Oscillator Figures of Merit
8.5.1. Frequency Pulling. As the reection coefcient of
the load G
L
forms a vital part of Eq. (53) and inuences the
input reection coefcient S
0
11
in Eq. (44), it is obvious that
any change in its phase will affect the oscillation frequen-
cy. The change in oscillation frequency due to a change in
the load reection coefcient is known as frequency pull-
ing. This is usually determined by rst measuring the fre-
quency of an oscillator into a load of known reection
coefcient (often 12dB). The phase of the load is then
varied from 0 to 3601 by means of a phase shifter or sliding
load. The maximum deviation from the nominal frequency
is the pulling gure. Pulling can be greatly reduced by
isolating the oscillators output from the load. This is usu-
ally achieved by use of a buffer amplier with high reverse
isolation.
8.5.2. Frequency Pushing. The basic function of the
DRO is to convert DC energy to RF. The scattering pa-
rameters of the active device are dependent on the applied
bias voltages and currents. Even regulated power supply
voltages can experience uctuations which lead to minute
changes in output frequency. The change in oscillation
frequency with respect to DC supply voltage is known as
frequency pushing. The parameter is measured by rst
noting the nominal oscillator frequency. The applied DC
supply voltage is then varied (by one volt, for example)
and the frequency deviation measured. Frequency push-
ing is then expressed in units of frequency per volt.
In practical circuits pushing is minimized by using well-
regulated supply voltages.
8.6. Oscillator Stability
An oscillator is said to be stable if the output frequency
(and power) do not vary with temperature and time. Di-
electric resonators are good for producing stable oscilla-
tors because of their high Q factors. Dielectric resonators
may be used in oscillators in two distinct ways:
1. As a high-Q passive element coupled to a free-run-
ning oscillator. Here the DR is not used as the oscil-
lators main resonator, but is locked to this
resonance. Such an oscillator is known as a dielec-
trically stabilized oscillator (DSO).
2. As a circuit element in the oscillator, whereby the
DR actually determines the oscillation frequency.
It is known that free-running oscillators generally have a
negative temperature coefcient. Thus the oscillators fre-
quency falls as temperature is increased. The temperature
coefcient of a DR can be made positive by careful choice of
material composition. Thus the DR can be made to com-
pensate for free-running oscillator drift. The temperature
stability of a complete DRO is dependent on [50]
*
The coupling coefcient between the DR and the rest
of the circuit
*
The Q of the oscillator
100.0
12.0 8.0
Frequency = 10.8 GHz
80.0
60.0
40.0
20.0
0.0
20.0
40.0
60.0
80.0
100.0
Re [Z11]
I m [Z11]
GHz
Figure 11. Graphical oscillator solution showing one-port oscil-

lation condition.
*
The rate of change of the active devices reection co-
efcient phase with temperature
Typical variations in oscillator frequency with tempera-
ture in DROs are of the order of 4 ppm/1C. This value can
be reduced, to values as low as 1ppm/1C, by several tech-
niques, including
*
Phase Locking. A high-frequency voltage-controllable
DRO is phase-locked to a low-frequency crystal oscil-
lator of superior phase noise and stability; this im-
proves long-term frequency stability.
*
Digital Compensation. Here the performance of an
oscillator with temperature is measured and the in-
formation stored in programmable read-only memo-
ry. A temperature sensor circuit is then used, in
conjunction with a lookup table, to apply a correction
voltage to the oscillator in order to hold it at constant
frequency.
*
By stabilizing the DRO in a temperature-controlled
localized oven [51].
The primary properties of a dielectric resonator oscillator,
that is, frequency and output power, can be measured us-
ing the test setup illustrated in Fig. 12. The frequency of
oscillation is measured on a high-frequency spectrum an-
alyzer, or RF counter. The power can also be read from the
analyzer display, but is often more accurately determined
using a separate RF power meter.
Frequency pushing can now be determined by observ-
ing the shift in oscillation frequency as the supply voltage
is varied. Pulling is measured by replacing the power
meter with a tunable load of known return loss (usually
12 dB). The phase of this load is then varied through 3601,
and the resulting maximum excursion from the nominal
frequency are noted.
Phase noise can be measured on the analyzer display at
suitable offset frequencies. A correction has to be made for
the lter (resolution) bandwidth of the measuring instru-
ment. More accurate phase noise measurement (especially
close to the carrier) requires a dedicated phase noise
testset.
The abovementioned oscillator parameters are usually
determined over a range of temperatures. This enables the
variation of the frequency and output power with temper-
ature to be determined.
9. OSCILLATOR PHASE NOISE
The phase noise of an oscillator is an important quantity.
Ultimately phase noise limits adjacent-channel selectivity
in a receiver. The output signal from a physical oscillator
is not monochromatic. In a real oscillator, noise sidebands
arise since the frequency of the signal can vary with time
as a result of phase noise created by phase modulation of
the signal [52]. The usual method for characterizing the
noise is to determine the single-sideband (SSB) phase
noise power spectral density at a given frequency offset
from the carrier (dBc/Hz). A frequency offset gure of
10 kHz is often quoted.
The phase noise comes from the various noise sources
available in the circuit as well as the active-device intrin-
sic noise sources. These noise sources induce phase noise
by nonlinear device mechanisms causing upconversion of
the baseband noise to the oscillator frequency. Noise
sources that are due to white noise generally contribute
1/f
2
to the spectrumof the phase noise, while 1/f noise adds
1/f
3
to the phase noise spectrum. Individual noise sources
are to a rst approximation considered uncorrelated. In
practice, up/downconversion can occur, leading to correlat-
ed amplitude-modulated (AM) and frequency-modulated
(FM) noise effects. The 1/f noise acts to alter the frequency
of the oscillation.
When designing a DRO with low phase noise, the un-
loaded Q factor of the resonator should be as high as pos-
sible. The 1/f noise of the active device should be as low as
possible. JFETs, then bipolar transistors and HBTs, have
the lowest 1/f noise with GaAs FETs having the worst 1/f
performance. Also the center frequency of oscillation of the
DR is usually equal to that of the oscillator. Here the DR
acts to select the frequency of oscillation. Under these
conditions the center frequency of the DR will have a
maximum rate of change of phase with respect to frequen-
cy; that is, the resonator will have its highest Q. Noise
power density is split between AM and PM noise by equal
amounts. In addition, AM noise is usually much more
dominant than PM noise at offset frequencies far removed
from the carrier [53]. Close to the carrier PM noise is the
dominant factor. Methods for simulating oscillator phase
noise are given in Refs. 54 and 55.
Leesons Eq. (52) and Eq. (54) describe the expected
single-sideband (SSB) phase noise power density at a fre-
quency f
m
offset from the carrier for an oscillator using a
single resonator; here
L
pm
% 10 log
10
NRkT
A
1
8Q
2
L
f
0
f
m
_ _
2
_ _
dBc=Hz 54
where NR is the device noise ratio, Aoscillator output
power, Q
L
loaded Q, f
0
oscillator center frequency (Hz),
and f
m
carrier offset (Hz).
Equation (54) shows that the oscillator SSB phase noise
is affected by the loaded Q squared, namely, 6 dB improve-
ment per Q
L
doubling, hence the incentive for using a DR
with as high a Q as possible. Leesons equation shows that
the SSB phase noise reduces at 6 dB/octave over the range
it applies. This range is for f
m
greater than frequency f
1
,
Stable
power supply
DRO
Optional
tuning supply
Spectrum
analyzer
6 dB pad
Power meter Directional coupler
Figure 12. Typical DRO test setup for measuring power output
and frequency spectrum.
the 1/f icker noise corner frequency, and less than below
the frequency f
2
, where f
2
f
0
/2Q
L
. Above f
2
the oscillator
upconverted white noise dominates. To correctly apply
this equation, the 1/f noise for the device must be known
a priori. In addition, the device noise ratio at its operating
power level (assumed to be in the linear range of opera-
tion) and loaded Q for the circuit are also required. A de-
tailed discussion of low-noise oscillator design and
measurement methods can be found in Ref. 56.
10. DESIGN EXAMPLE10.8-GHz DRO
Consider the schematic of Fig. 13. The circuit consists of a
common-source MESFET-based negative-resistance cir-
cuit (NRC), with series capacitive feedback. The DR is
coupled to the gate via a 50-O-terminated microstrip
transmission line. The substrate for the transmission
line is chosen to be 254-mm-thick alumina (e
r
9.9). The
width of the line is 238mm, which corresponds to a char-
acteristic impedance of 50 O.
The rst step is to choose a suitable DR size. As a prac-
tical example, consider the Murata TE
01d
DRD series [57].
The most suitable dimensions for a 10.8-GHz circuit are a
diameter of 6.5mm and a thickness of 2.9 mm. It can be
shown that this puck will resonate at approximately
10.34 GHz when enclosed in a well-spaced metallic envi-
ronment. In a practical circuit the resonator will operate
at a higher frequency, due to the presence of the thin sub-
strate. The shift in frequency can be approximated by the
analytical formula in Ref. 50. Final tuning to 10.89 Hz by
use of a metal tuning screw can then be achieved.
Step 2 involves choice of a suitable transistor and bias
condition. For this example a GMMT 0.5-mm-gate-length/
300-mm-gate-width GaAs MESFET is used. When biased
at 5 V V
ds
and I
ds
50% I
dss
, the transistor exhibits
some 12 dB of available gain at this frequency. This is
more than adequate for this type of oscillator design.
Next the two-port NRC is designed. To do this, the series
capacitive feedback is varied until the magnitude of the
input reection coefcient (S
11
) is greater than one, indi-
cating negative resistance. In practice a value greater than
1.2 should be used. This is generally enough to ensure suf-
cient excess negative resistance to kickstart an oscilla-
tion. For this transistor a value of 0.24pF results in |S
11
|
being of the order of 1.4 at 11GHz, as illustrated in Fig. 14.
Next the complete oscillator is simulated to solve the
small-signal approximation to the complex condition for
oscillation, that is, G
.
S
11
41. A model for the coupled DR
can be developed using the mutually coupled parallel LCR
model and coupling factor k, as discussed previously. In
practice the equivalent circuit is merely an RFopen circuit
at resonance. Therefore the circuit can be simplied to the
circuit shown in Fig. 15. To complete the oscillator design,
the transmission-line length l
1
, at which point the center
of the DR is placed, is calculated. For this example a value
of 4.427 mm is determined. The resulting oscillation
frequency, in terms of real and imaginary parts of the
overall impedance, is shown in Fig. 16. The angle goes
through zero at 10.8 GHz, with the resulting magnitude
L = 1000
R = 20
C = 1000
CMP8
C
C

=

C
s
L = 1000
R

=

1
0
0
0
C

=

1
0
0
0
MESFET
f20 fet3
S
W = 238 m
L = L1 + L2
R

=

5
0
R

=

5
0
DR
D
G
Figure 13. Circuit schematic for
10.8-GHz DRO example design
(e
r
10.0, h254mm).
simultaneously exceeding 1.2. A large-signal simulation
can be performed using a harmonic balance technique to
predict output power and other important oscillator pa-
rameters. For this example 11dBm output power was
predicted, with a DC-to-RF conversion efciency of ap-
proximately 13%. In addition the phase noise was predic-
ted as 80 dBc/Hz at 10 kHz offset. This is achieved by
including important noise parameter values in the non-
linear device model.
In practice the circuit is now fabricated by using an
arbitrary length for l
2
, and suitably terminating the line
in the characteristic impedance, 50 O. The circuit is biased
as required and the DR placed on the alumina in approx-
imately the correct position. The puck is then moved
around until oscillation of suitable magnitude, Q factor,
and phase noise is obtained. The resonator is then xed
into position, and mechanically tuned to the exact
required frequency.
1.4
1.3
1.2
1.1
1.0
0.9
0.0
Frequency 5.0 GHz/DIV
20.0
|S11|
Figure 14. One-port input reection plot.
L = 1000
R = 20
C = 1000
D
G
C

=

C
s
L = 1000
R

=

1
0
0
0
C

=

1
0
0
0
MESFET
f20 fet3
S
W = 238 m
L1
R

=

5
0
Figure 15. Simplied equivalent
circuit for 10.8GHz DRO example
design.
100.0
50.0
0.0
50.0
100.0
18.0 2.0
Frequency 10 GHz/DIV
Z11
I m [Z11]
Re [Z11]
Figure 16. One-port oscillation condition.
Electronic tuning of the circuit to enable phase locking
to a crystal reference is possible. To achieve this, a l/4
length of transmission line incorporating a tuning varac-
tor is also coupled to the resonator. Tuning the varactor
capacitance changes the phase of the coupled line, result-
ing in pulling of the DR frequency.
11. DRO CAD TECHNIQUES
The practical realization of dielectric resonator oscillators
DROs can be assisted by the use of linear and nonlinear
CAD tools. Linear scattering parameters can be used
to represent the active device and feedback network
[3,5861]. Important DRO performance features such as
output power, efciency, steady state stability and tuning
range, have been simulated [6264].
Series feedback DROs have been designed using S-pa-
rameter methods whereby a topology is selected to yield
maximum negative resistance at one port while maintain-
ing small-signal startup conditions [50,65]. When parallel
feedback topology is used, a more complex design proce-
dure is required [66]. Here, when the DR is coupled be-
tween two of the device terminals, the reection coefcient
at the third terminal G
3
is expressed in terms of the DR
and active device scattering parameters. The position of
the DR and hence the coupling coefcient are optimized to
maximize G
3
and a matching circuit designed to maximize
the startup condition [61].
Small-signal techniques do not necessarily lead to max-
imum power oscillators, nor do they ensure stability of
operation. Nonlinear large signal techniques are needed to
do this. Broadly two classes exist here: time domain [67]
and harmonic balance [68]. The former method uses nu-
merical integration of state space equations describing the
active device and the embedding circuit. All circuit ele-
ments must be represented in the time variable as volt-
age-dependent current equations. There are no
constraints on the harmonic content; hence highly non-
linear circuits can be examined. The time-domain ap-
proach is well suited to oscillator simulation because
oscillation startup, transient, and steady response can
all be obtained on a single analysis run. However, each
timestep must be small enough to ensure numerical con-
vergence of the integration routine. This is a severe lim-
itation when high-Q resonators such as DRs are used [69].
With the harmonic balance approach, nonlinear circuit
elements are analyzed in the time domain and trans-
formed into the frequency domain, where they are repre-
sented by a nite number of harmonic currents. Linear
elements are solved directly in the frequency domain.
Where linear and nonlinear components are joined, cur-
rent continuity forms the boundary condition for the har-
monic balance algorithm. With this approach the high-Q
circuits are solved by numerically efcient algorithms di-
rectly in the frequency domain. However, in order to sim-
ulate circuits with strong nonlinearities, many harmonics
are needed for accurate representation of the nonlinear
waveforms, thus increasing computational time. In addi-
tion, the harmonic balance approach yields only steady-
state information about the oscillators behavior.
In a DRO the DR dampens higher harmonics; thus the
harmonic balance approach is attractive here [63]. In Ref.
63 the device topology was xed a priori and the circuit
element values were numerically optimized to satisfy a
prescribed set of electrical specications. An alternative
approach is to synthesize from the maximum added power
condition the general device terminations [70]. This ap-
proach has been adapted for DRO synthesis [71,72]. Most
commercially available industrial circuit simulators for
microwave use such as a Hewlett-Packard model [73] will
handle time-domain and harmonic balance simulation
with embedded large-signal active-device models.
BIBLIOGRAPHY
1. C. L. Ren and J. K. Ploude, Application of dielectric resonators
in microwave components, IEEE Trans. Microwave Theory
Tech. MTT-29:754759 (1981).
2. S. J. Fiedziuszko, Microwave dielectric resonators, Microwave
J. 189 (Sept. 1986).
3. A. Khanna, Parallel feedback FET DRO using 3-port S-pa-
rameters, IEEE Int. Microwave Symp. Digest, San Francisco,
1984, pp. 181183.
4. H. Abe et al., A highly stabilized low-noise GaAs FET inte-
grated oscillator with a dielectric resonator in the C-band,
IEEE Trans. Microwave Theory Tech. MTT-26(3):156162
(1978).
5. R. S. Pengelly, Microwave FETsTheory, Design and Appli-
cations, 2nd ed., Wiley, New York, 1986.
6. O. Ishikora et al., A highly stabilized GaAs FEToscillator us-
ing a dielectric resonator feedback circuit in the 914GHz
range, IEEE Trans. Microwave Theory Tech. MTT-28(8):
817824 1980.
7. K. W. Lee et al., Push-push, frequency doubling, MMIC
Oscillators, IEEE GaAs IC Symp., 1987, pp. 227230.
8. S. Chen et al., U-Band MMIC HBT DRO, IEEE Microwave
Guided Wave Lett. 4(2):5052 (1994).
9. K. Borst, N. Loechel, and K. Wacker, Low noise synthesized
microwave local oscillator for high capacity digital radio sys-
tems using a dielectric resonator and a SAW reference, Proc.
19th European Microwave Conf., London, 1989, pp. 555560.
10. K. Imai and H. Nakakita, A 22GHz band low noise down
converter for satellite broadcast receivers, Proc. 19th Europe-
an Microwave Conf., London, 1989, pp. 549554.
11. C. D. Prasad, S. S. Serin, and D. Singh, 35 GHz dielectric res-
onator stabilized gunn oscillator, Int. J. Infrared Millimeter
Waves 17(2):393402 (1996).
12. K. Ogawa et al., 25GHz dielectric resonator oscillator using
an Al GaAs/GaAs HBT, Electron. Lett., 26(18):15141516
(1990).
13. D. Cros and P. Guillon, Whispering gallery dielectric resona-
tor modes for W-band devices, IEEE Microwave Theory Tech.
38(11):1036 (1990).
14. A. Bermadez et al., A 94GHz low noise GaAs FET oscillator
using whispering gallery dielectric resonator modes and a
new push-push conguration reducing 1/f converted noise,
IEEE MTT-S Int. Microwave Symp. Digest, 1988, pp. 481484.
15. S. Kharkovsky, A. Kirichenko, and A. Kogut, Solid-state os-
cillators with whispering-gallery mode dielectric resonators,
Microwave Opt. Technol. Lett. 12(4):210-213 (1996).
16. I. U. Khairuddin and I. C. Hunter, Whispering-gallery dielec-
tric resonator mode GaAs FET oscillator, IEE Proc. Micro-
waves Anten. Propag. 143(5):441443 (1996).
17. D. Kajfez, Simple models, in D. Kajfez and P. Guillon, eds.,
Dielectric Resonators, Artech House, Dedham, MA, 1986.
18. A. W. Gisson, Integrated equation methods, in D. Kajfez and
P. Guillon, eds., Dielectric Resonators, Artech House, Ded-
ham, MA, 1986.
19. M. R. Stieglitz, Dielectric resonators: Past, present and
future, Microwave J. 121126 (April 1989).
20. R. E. Collin, Foundations for Microwave Engineering,
McGraw-Hill, New York, 1966, pp. 1720.
21. B. W. Hakki and P. D. Coleman, A dielectric resonator method
of measuring inductive capacities in the millimetre range,
IRE Trans. Microwave Theory Tech. MTT-8:402410 (1960).
22. W. E. Courtney, Analysis and evaluation of a method of mea-
suring the complex permittivity and permeability of micro-
wave insulators, IEEE Trans. Microwave Theory Tech. MTT-
18:476485 (1970).
23. C. Tsironis, Highly stable dielectric resonator FEToscillators,
(1985).
24. M. Massias, M. J. Howes, and V. Postoyalko, A novel analysis
and design technique for temperature stable DROs, Proc.
19th European Microwave Conf., London, 1989, pp. 418423.
25. K. Kurokawa, Microwave solid state oscillator circuits, in M.
J. Howes and O. V. Morgan, eds., Microwave DevicesDevice
Circuit Interactions, Wiley, New York, 1976.
26. R. R. Bonetti and A. E. Atia, Generalized dielectric resonator
lters, COMSAT Tech. Rev. 11(2):321343 (1981).
27. P. Guillon and S. Mekerta, A bandstop dielectric resonator
lter, IEEE MTT-S Int. Microwave Symp. Digest, 1981, pp.
170173.
28. B. Ginzton, Microwave Measurements, McGraw Hill, New
York, 1957, pp. 391427.
29. D. Kajfez and E. J. Hwan, Q-Factor measurement with a net-
work analyser, IEEE Trans. Microwave Theory Tech. MTT-
32(7):666670 (1984).
30. T. Hammersley, J. Richardson, and V. Postoyalko, Varactor
tuning of dielectric resonator oscillators, Microwave Eng.
Europe 2528 (May/June 1990).
31. W. P. Wheless and D. Kajfez, Experimental characterization
of multi-moded microwave resonators using an automated
network analyser, IEEE Trans. Microwave Theory Tech.
MTT-35(12):12631270 (1987).
32. A. Khanna and Y. Gorault, Determination of loaded, unloaded
and external quality factors of a dielectric resonator coupled
to a microstrip line, IEEE Trans. Microwave Theory Tech.
MTT-31(3):261264 (1983).
33. Alpha Industries, The tuning and exciting of DR modes,
Mickrowelter Mag. 11(Part 6):550552 (1985).
34. B. Ginzton, Microwave Measurements, McGraw Hill, New
York, 1957, pp. 391427.
35. A. P. S. Khanna, Q measurement of microstrip coupled
dielectric resonators, Microwaves RF 23(1):8186 (1984).
36. L. A. Trinogga, G. Kaizhou, and F. C. Hunter, Practical
Microstrip Circuit Design, Ellis Horwood, Chichester, UK,
1991.
37. P. Guillon, B. Bysery, and M. Chaubet, Coupling parameters
between a dielectric resonator and a microstrip line,
(1985).
38. P. Guillon and Y. Garault, Coupling between a microstrip
transmission line and a dielectric resonator, IEEE MTT-S Int.
Microwave Symp. Digest, 1976, pp. 200202.
39. Trans-Bond, Trans-Tech., Inc., Dielectric Resonator Mounting
Adhesive, Alpha Industries, 19/21 Chapel St., Marlow, Bucks,
SL7 3HN, UK.
40. Araldite Epoxy Resin, Ciba-Geigy, Plastics, Duxford,
Cambridge, CB2 4QA, UK.
41. D. Kajfez, Simple models, in D. Kajfez and P. Guillon, eds.,
Dielectric Resonators, Artech House, Dedham, MA, 1986.
42. Alpha Industries, A Designers Guide to Microwave Dielectric
Ceramics, Trans-Tech, Inc., Marlow, Bucks, UK, 1990, Publi-
cation 5008040, Rev. 2.
43. K. Wada, E. Nagata, and I. Haga, Wideband tunable DRVCO,
Proc. 15th European Microwave Conf. 1985, pp. 407412.
44. L. A. Trinogga and A. J. Fox, Inclusive varactor tuning of
dielectric resonator lters, Proc. 26th European Microwave
Conf., Prague, Sept. 1996, pp. 162164.
45. O. Y. Chan and S. Kazeminejad, Voltage controlled oscillator
using dielectric resonator, Electron. Lett. 24(13):407412
(1988).
46. P. R. Herczfeld and A. S. Daryoush, Optically tuned and FM
modulated x-band dielectric resonator oscillator, Proc. 14th
European Microwave Conf., Liege, Belgium, 1984, pp. 268
273.
47. J. Krupka, Magnetic tuning of cyclindrical Ho1d-mode dielec-
tric resonators, IEEE Trans. Microwave Theory Tech. MTT-
37:743747 (1989).
48. A. Nesic, A new method for frequency modulation of dielectric
resonator oscillators, Proc. 15th European Microwave Conf.,
1985, pp. 403406.
49. K. Kurokawa, Microwave solid state oscillator circuits, in
M. J. Howes and O. V. Morgan, eds., Microwave Devices
Device Circuit Interactions, Wiley, New York, 1976.
50. D. Kajfez and P. Guillon, eds., Dielectric Resonators, Artech
House, Dedham, MA, 1986.
51. Fundamentals of Quartz Oscillators, Hewlett-Packard Appli-
cation Note 200-2, Electronic Counter Series, Santa Rosa, CA,
1980.
52. D. B. Leeson, A simple model of feedback oscillation noise
spectrum (Letters), Proc. IEEE 329330 (Feb. 1996).
53. S. Hamilton, FM and AM noise in microwave oscillators,
Microwave J. 105109 (June 1978).
54. Hewlett-Packard, Simulating Noise in Nonlinear Circuits
Using the HP Microwave and RF Design Systems, Santa
Rosa, CA, 1993, Product Note 85150-4.
55. A. Howard, Simulate oscillator phase noise, Microwaves RF
6470. (Nov. 1993).
56. G. D. Vendelin, Design of Ampliers and Oscillators by the
S-parameter Method, Wiley, New York, 1982, pp. 145162.
57. Data Sheet, Microwave Ceramic Resonator Catalogue,
Murata-Erie Publication 58-04C, Tokyo, Japan, 1994.
58. K. Borst, N. Loechel, and K. Wacker, Low noise synthesized
microwave local oscillator for high capacity digital radio
systems using a dielectric resonator and a SAW reference,
Proc. 19th European Microwave Conf., London, 1989,
pp. 555560.
59. P. G. Wilson and R. D. Carver, An easy-to-use FET DRO
design procedure suited to most CAD programs, IEEE MTT-S
Int. Microwave Symp. Digest, 1989, pp. 10331036.
60. C. Y. Ho and T. Kajiba, DRO state of the art, Appl. Microwave,
Microwave Symp. Digest, Vol. 3, Long Beach, CA, spring 1990.
61. A. C. Murphy and P. J. Murphy, Computer program aids di-
electric resonator feedback oscillator design, Microwave J.
131 (Sept. 1988).
62. A. M. Pavio and M. A. Smith, A 20-40GHz push-push dielec-
tric resonator oscillator, IEEE Trans. Microwave Theory Tech.
MTT-33(12):13461349 (1985).
63. V. Rizzoli, A. Costanzo, and A. Neri, Analysis and optimiza-
tion of DROs using a general-purpose CAD program, Alta
Freq. 57(7):389398 (1988).
64. F. Filicori, V. A. Moncaco, and G. Vannini, Adesign method for
parallel feedback dielectric oscillators, 19th European Micro-
wave Conf., London, 1989, pp. 412417.
65. A. Sweet, MIC and MMIC Amplier and Oscillator Circuit
Design, Artech House, Dedham, MA, 1990.
66. A. Podocomeni and L. Conrodo, Design of microwave oscilla-
tors and lters using transmission mode dielectric resonators
coupled to microstrip lines, IEEE Trans. Microwave Theory
Tech. MTT-33:13241332 (1985).
67. M. I. Sobhy and A. K. Jastrezebski, Computer-aided design of
microwave integrated circuits, Proc. 14th European Micro-
wave Conf., Liege, Belgium, 1984, pp. 705710.
68. V. Rizzoli, A. Lipperini, and E. Menazzi, A general purpose
program for non-linear microwave circuit design, IEEE
Trans. Microwave Theory Tech. MTT-31:762770 (1983).
69. I. Kipinis and A. S. Khanna, Large signal computer-aided
analysis and design of silicon bipolar MMIC oscillators and
mixers, IEEE Trans. Microwave Theory Tech. MIT-37:(3)
558564 (1989).
70. Y. Xuan and C. Snowden, A generalized approach to the de-
sign of microwave oscillators, IEEE Trans. Microwave Theory
Tech. MTT-35(12):13401347 (1987).
71. S. W. J. Seawright et al., Optimum DRO synthesis, Proc. 19th
European Microwave Conf., London, 1989, pp. 406411.
72. S. W. J. Seawright et al., Computer-aided design of a micro-
wave MESFET DRO, Proc. 19th European Conf. Circuit The-
ory and Design, 1989, pp. 561565.
73. Hewlett-Packard, Microwave and RF Design System, MDS,
Vols. 18, Brighton, England, 1989.
DIELECTRIC RESONATORS
ANDREW R. WEILY
KARU P. ESSELLE
Macquarie University
Sydney, New South Wales,
Australia
ANANDA SANAGAVARAPU
MOHAN
University of Technology
Sydney, New South Wales,
Australia
1. INTRODUCTION
The widespread use of the dielectric resonator (DR) in mi-
crowave engineering is due to four main attributes: price,
performance, size, and versatility. The price of a DR is low,
particularly when manufactured in large quantities. As a
resonant element in microwave circuits they give very
high performance in terms of Q factor and temperature
stability. The DR is smaller and lighter than conventional
air-lled cavity resonators, enabling significant miniatur-
ization; the reduction in size being proportional to

e
r
p
,
where e
r
is the relative permittivity of the DR material.
DRs are also extremely versatile components, nding im-
portant applications over a wide range of frequency bands
in microwave lters, oscillators, and antennas in both
microwave integrated circuit (MIC) and monolithic micro-
wave integrated circuit (MMIC) environments. They are
also used in measurement techniques for determining the
relative permittivity and loss tangent of dielectric mate-
rials and the surface resistance of superconductors.
The most common shape for the DR is a cylinder with
its diameter greater than its length, where the TE
01d
(transverse electric) mode is dominant. This mode is sim-
ilar to a magnetic dipole that is aligned with the axis of the
cylinder, with the azimuthal E elds forming concentric
circles about the axis. Other DR shapes that also nd use
are dielectric ring resonators, spheres, and parallelo-
pipeds. The resonator is usually placed in a cylindrical
or rectangular cavity, or on top of a substrate within a
cavity as shown in Fig. 1. The use of the substrate allows
external coupling of the DR mode to a nearby microstrip
line, while the cavity enclosure prevents radiation losses
and shields the resonator from external elds. The air
dielectric interfaces of the DR form the walls of a cavity
that support resonant modes, similar to the conguration
for a conventional cavity resonator. However, in a DR ev-
anescent elds exist in the air region outside the dielectric
material, due to imperfect connement of the eld by the
DR walls. For large dielectric constants the DR walls can
be approximated as perfect magnetic conductors (PMC), or
open circuits, that require the magnetic eld tangential to
the wall to be zero. The accuracy of this approximation
improves as the dielectric constant increases in value.
Richtmeyr pioneered the concept of the dielectric res-
onator in 1939 showing that suitably shaped dielectric
objects such as spheres and circular rings could act as
high-frequency resonators [1]. This seminal work received
little further attention until 1962, when Okaya and Ba-
rash described theory and experimental results on rect-
angular DRs made from TiO
2
(rutile) and SrTiO
3
(strontium titanate) [2]. They introduced an approximate
second-order analysis of the modes in rectangular DRs
that gave reasonably accurate results when compared
with measurements, and observed the poor temperature
stability of the available high-dielectric-constant materi-
als. In 1968 Cohn published a second-order analysis of the
cylindrical DR operating in the TE
01d
mode and equations
for DR interresonator coupling [3], which enabled high-Q
bandpass lters to be designed using high-purity TiO
2
(e
r
E100) [4]. Unfortunately the high sensitivity of TiO
2
to
changes in temperature made these DR lters impracti-
cal, but led Cohn to throw down a challenge to material
scientists to develop high-dielectric-constant materials
with improved temperature sensitivity.
This challenge was met when Raytheon developed the
rst barium tetratitanate ceramics that were both low loss
and temperature stable [5], which were further improved
on by researchers at Bell Laboratories [6]. At around the
same time it was shown that a dielectric resonator placed
DIELECTRIC RESONATORS 999
between two parallel conducting plates could be used to
accurately measure complex permittivity and permeabil-
ity [7]. The next advance was the development of ceramic
materials with adjustable temperature coefcients [8], en-
abling the value of temperature coefcient to be chosen to
minimize or cancel out the material thermal expansion,
leading to DR components with high temperature stabil-
ity. These breakthroughs in material science technology
paved the way for a large number of new temperature-
stable lter and oscillator circuits based on DRs [9].
The rst dual-mode axial-mounted DR-loaded cavity
lter, introduced in 1982 [10], used hybrid-mode DRs and
was able to match the performance of conventional dual-
mode waveguide lters. Dual-mode lters have the ad-
vantage of reduced size and weight over monomode devic-
es. Because of these advantages, the lter became popular
for space applications. Dual-mode lters in planar cavities
[11] and triple-mode lters were introduced soon after
[12]. Advances in the mode-matching technique also led to
the precise design of TE
01d
[13], TM
01d
[14], and EH
11d
mode lters [15,16].
In recent years, more emphasis has been placed on
techniques for improving the spurious performance of DRs
and DR lters, to overcome limitations imposed by the
crowded mode spectrum of DRs. Several methods have
arisen to achieve this goal. The simplest method involves
optimization of the DR dimensions [17]. Mode suppression
devices also can provide a small improvement. In TE
01d
-
mode DR lters the use of resonant irises [18] or quarter-
cut resonators [19] has shown a good suppression of high-
er-order modes. Quarter-cut resonators can also give ex-
tremely good heat dissipation and are suitable for use in
high-power lters.
More effective methods involve using composite res-
onators [20,21], exciting combinations of different modes
of the same type of resonator [22,23], or mixing two com-
pletely different types of resonators, for example, combline
and TE
01d
DRs [2426], or conductor- and dielectric-loaded
resonators [27]. Since the higher-order modes of the dif-
ferent resonators occur at different frequencies, the spu-
rious performance is enhanced considerably. Improved
performance can also be achieved with lters containing
rectangular DRs [28], dielectric comblines [29], conductor-
loaded DRs [30], and sandwiched conductor DRs [31]. The
development of the hybrid dielectric/high-temperature su-
perconductor (HTS) resonator is a further innovation that
has the benet of extremely low losses and reduced size
[32]. Another advance was the introduction of a resonator
that uses slots in the DR for tuning and coupling
dual-mode DR lters, thus replacing the tuning and cou-
pling screws traditionally used [33]. Filters constructed
using slotted DRs have the advantage that they require
very little tuning. An excellent reference on the topic of
DRs, DR lters, and DR oscillators is the book edited by
Kajfez and Guillon [34].
2. ANALYSIS OF DIELECTRIC RESONATORS
As mentioned previously, early analysis methods for DRs
modeled the resonator walls as perfect magnetic conductor
(PMC) surfaces [2]. Cohn [3] improved on this approach by
setting only the curved DR surfaces to be PMC walls, with
the at surfaces modeled as airdielectric interfaces, al-
lowing elds to leak out into an innite circular waveguide
with magnetic walls. Kobayashi derived analytical solu-
tions for dielectric rod resonators short-circuited at both
ends [35,36]. Regions not accounted for in the PMC wall
model were gradually included through the use of pertur-
bation and variational techniques and can be described as
approximate techniques [3751]. The resonant frequen-
cies calculated using perturbational techniques or varia-
tional methods are more accurate than the simple PMC
wall models, but are usually limited to the axially sym-
metric modes. The approximate methods have limitations
since in some of the regions they may not satisfy Maxwells
equations or exact boundary conditions. To increase the
accuracy of prediction and for applicability to more gen-
eral structures, rigorous analysis and numerical methods
are required.
The nite-difference frequency-domain (FDFD) method
is one numerical technique that has been successfully ap-
plied to rotationally symmetric DRs [5255]. This tech-
nique discretizes the elds inside the DR and then, by
imposing boundary conditions, formulates a matrix eigen-
value problem that is solved to obtain the resonant fre-
quency, eld distribution, and unloaded Q for any given
mode. The powerful nite-element method (FEM) has also
been applied to DRs [5659]. In this method the solution
region is discretized into subregions or elements. Govern-
ing equations are then derived, assembled over the ele-
ments of the solution region and the resulting system of
equations solved. FEM is a versatile method as it can
handle complex geometries and inhomogeneous media
much better than can the nite-difference method owing
to the use of highly exible adaptive meshing schemes.
Line
(a) (b)
DR
DR
Cavity
Substrate
Microstrip
Cavity
Tuning
plunger
Ground
Figure 1. (a) Conguration of a cylindrical dielec-
tric resonator loaded in a cavity; (b) dielectric res-
onator on a substrate coupled to a microstrip line.
1000 DIELECTRIC RESONATORS
Early FEM formulation solutions were plagued by the ap-
pearance of nonphysical spurious modes, but by using
mixed elements these modes may be eliminated [59]. In-
tegral equation (IE) formulations [60,61] and the method
of moments (MoM) have also been applied to DRs, to ac-
curately compute resonant frequencies, unloaded Q, and
eld distributions. IE formulations and MoM use Greens
functions and hence are ideally suited to modeling reso-
nators in open regions; their use can become quite com-
plicated for cavity-type applications.
Mode matching combines analytical solutions and com-
putational models for closed-region problems, and has
been extensively applied to analyze DRs in cylindrical
and rectangular cavities [6274]. This technique divides
the DR into subregions and may be classied as either
radial or longitudinal mode-matching, depending on the
way the subregions of the DR are divided and matched.
The elds in each subregion are expressed as a linear
combination of the appropriate eigenmode elds of an in-
nite waveguide with unknown coefcients. Fields of the
adjacent regions are then matched to satisfy boundary
conditions, from which an innite system of linear equa-
tions is obtained in terms of the unknown coefcients. The
zeros of the determinant of the system yield the resonant
frequencies of the DR. The number of modes used in the
generation of the system must be increased until relative
convergence is achieved. Solution of the resonant frequen-
cies also gives the modal eld distributions, and hence the
unloaded Q may be calculated for any given mode. Be-
cause of the appearance of complex modes in the longitu-
dinal mode-matching technique [70] as well as
computational difculties, the radial mode-matching tech-
nique has emerged as the more popular method. Radial
mode matching has been successfully applied to analyze
DRs in rectangular cavities by using a BesselFourier se-
ries to match the elds between two different coordinate
systems at a ctitious boundary [72]. Resonant frequen-
cies of uniaxial anisotropic DRs [73] and generalized mul-
tilayer anisotropic DRs [74] have also been analyzed using
radial mode matching. For the multilayer anisotropic DRs,
whispering gallery modes and unloaded Q were also cal-
culated. Although mode matching is very accurate and
computationally efcient, it is limited in the number of
inhomogeneous regions and resonator shapes it can mod-
el. The problem size may also be limited when using FEM
and MoM by the excessive time it takes to ll and solve a
matrix, as the number of unknowns increases.
The nite-difference time-domain (FDTD) method [75]
has been widely used to solve a broad range of electro-
magnetic problems. In this method Maxwells time-depen-
dent curl equations are solved directly. The time-domain
elds are calculated using explicit update equations. This
approach has several advantages over the frequency-do-
main methods mentioned previously. Broadband results
may be obtained from a single simulation, arbitrary struc-
tures may be analyzed, and the problem size is not limited
by the size of the matrix to be inverted, a limitation en-
countered in frequency-domain methods. Frequency-
domain parameters may be obtained by Fourier-
transforming time-domain values obtained from the
FDTD algorithm. Nonlinear materials and devices may
also be modeled. FDTD has been applied to dielectric res-
onators using the 2D rotationally symmetric (RS) FDTD
algorithm coupled with the discrete Fourier transform
(DFT) [76,77]. To improve computational efciency as
well as computing Q values, digital ltering coupled
with Pronys method [78] or the matrixpencil technique
[79] has been used instead of the DFT. Computational ef-
ciency has also been improved by using digital ltering
and the MUSIC (multiple signal classication) algorithm
[80], and the Pade interpolation method [81]. These tech-
niques reduce the number of timesteps required to obtain
accurate resonant frequencies from time-domain samples.
A nonorthogonal FDTD method has also been used to
model DRs [82]. An effective dielectric constant was used
with a 3D rectangular FDTD algorithm [83] to increase
the accuracy in modeling the curved surfaces of a DR for
hybrid modes in rectangular enclosures. A locally confor-
mal FDTD method has also been used to model both open
and closed DRs [84]. Resonant frequencies for selected
higher-order modes have been calculated using FDTD
[85,86], as have unloaded Qusing perturbation techniques
[87,88].
3. MODES AND MODE CHARTS
The modes of a cylindrical DR of the type shown in Fig. 1a
may be classied as transverse electric (TE), transverse
magnetic (TM), or hybrid electromagnetic (HEM). Three
subscripts are used to differentiate between the different
modes of a given classication. The rst subscript is an
integer m denoting the azimuthal eld variation of the
mode, while the second integer n refers to the number of
radial variations. Usually, the third subscript p of a res-
onator mode refers to the integer number of half-wave-
length variations in the axial or z direction. However, the
imperfect walls of the DR leads to a eld variation smaller
than half a wavelength and so is denoted by [34]
pl d for l 0; 1; 2; 3; . . . 1
A further property of cylindrical DRs is the circular sym-
metry of the TE and TM modes, that is, m0 for all TE
and TM modes. The hybrid modes are further classied
into EHand HE modes according to the function P, dened
as the ratio of the H
z
to E
z
eld components of the hybrid
mode [89]. For |P|o1, the mode is classied as HE and is
TM-like in its characteristics. When |P|41, the mode is
classied as EH and is TE-like. The modes that exist in
the cylindrical DR and their appropriate subscripts are
summarized in Table 1.
A number of authors have used different notations in
the literature to designate hybrid modes. Notations such
as HEM
mnp
[60] or simply HE
mn
[64] have been used but
do not differentiate between the TE-like and TM-like na-
ture of the hybrid modes. This distinction is important
when determining the appropriate method for (1) external
coupling to the mode, (2) interresonator coupling, and
(3) spurious mode suppression. The HE/EH notation also
includes more information about the eld distribution of
the mode.
To understand the nature of the modes more clearly,
the eld distributions of several lower-order modes have
been calculated for the generic structure of Fig. 2 using
the FDTD method. This gure shows the parameters used
to characterize the DR. These parameters were set to D
x
0, d/D2.0, L/D0.4, ML, e

r
38.0, and e
3
1 for the
eld calculations. The results of these calculations are
plotted for TE
01d
, TM
01d
, EH
11d
, and HE
11d
modes in Fig. 3a
and the EH
21d
, HE
21d
, EH
31d
, and HE
31d
modes in Fig. 3b.
Each plot represents the transverse elds of the resonator
taken as a slice through the center of the DR. It should be
noted that the eld patterns within the dielectric material
for the HE
mnp
and EH
mnp
modes have a strong resem-
blance to the respective TM
mnp
and TE
mnp
modes in a
PMC wall cylindrical resonator. This is further conrma-
tion of the TM-like nature of the HE modes and TE-like
characteristics of the EH modes.
Each mode of the cylindrical DR can be characterized
by either an electric (P) or magnetic dipole (M) moment.
These dipole moments are shown for some of the lower-
order modes in Fig. 4. For a given mode, a comparison
between the dipole moment and the eld plot of Fig. 3
conrms that the E eld (or H eld) does indeed align with
the electric (or magnetic) dipole moment. This dipole mo-
ment characterization is useful for determining the best
way to externally couple energy to a particular mode. It is
usually best to couple to electric dipole modes via the E
eld using an electric eld probe. For magnetic dipole
modes, it is best to couple through the H eld using a
loop-type probe that intercepts the magnetic eld lines.
These recommendations serve as a rough guide for cou-
pling to the modes of the DR, but are not the only methods
possible.
Mode charts for the resonator conguration of Fig. 2
have been calculated using the commercial software CST
Microwave Studio [90], and are presented in Figs. 58.
Mode charts are important for determining the spurious
performance of a DR, enabling the frequency spacing be-
tween adjacent modes to be analyzed. This knowledge
allows the dimensions of the DR and cavity to be designed
to give optimal spurious separation. The charts also
show the effect that the proximity of the cavity to
the DR has on the resonant frequency of each mode.
This information is useful for determining the appropri-
ate location of tuning elements in the cavity. Figure 5
shows the effect of the cavity height on several lower-order
modes. Over the range plotted almost all the mode reso-
nant frequencies decrease as the cavity height increases;
only the TM
0np
modes increase in frequency. For
M/D40.7, the cavity height has little or no effect on the
modes, and the resonant frequencies approach a constant
value.
A plot of normalized frequency versus DR aspect ratio
(D/L) is given in Fig. 6. This shows that the TE
01d
mode is
dominant (lowest in frequency) for D/L41, while the
HE
11d
mode is dominant for D/Lo1. In practice D/L is
typically set to approximately 2.5, so that the TE
01d
mode
is dominant and there is good separation between the
TE
01d
and HE
11d
modes. Figure 7 is a mode chart for the
ratio between the cavity diameter and DR diameter (d/D).
Similar to the conguration shown in Fig. 5, once the
cavity diameter is greater than a certain value, this
ratio no longer inuences the resonant frequency of the
modes, and they converge to a constant value. Kobayashi
showed that by using a dielectric ring resonator, a DR
with a concentric hole in its center, the mode separation is
improved [17]. This property is illustrated in the mode
chart of Fig. 8. It shows that increasing the diameter
of the hole (D
x
) has little effect on the TE
01d
mode,
while the higher-order modes tend to increase in
frequency.
4. SIMPLIFIED ANALYSIS OF THE TE
01d
MODE
This section examines a simplied method for calculating
the resonant frequency of a TE
01d
mode in a DR. It is more
accurate than the magnetic wall models and provides in-
tuition into the nature of the elds in and around the DR.
Although the accuracy is less than that of variational [46]
and mode-matching solutions [62], the formulation has
the advantage of low complexity. This method was rst
presented by Itoh and Rudokas in 1977 [41] and was
expanded on by Kajfez [34].
Table 1. Mode Classications for Cylindrical DRs
Mode Mode Classication m n p
TE
0np
Transverse electric (TE) m0 n1,2,3,y pd, 1 d, 2 d, 3 d,y
TM
0np
Transverse magnetic (TM) m0 n1,2,3,y pd, 1 d, 2 d, 3 d,y
EH
mnp
Hybrid (TE-like) m1,2,3,y n1,2,3,y pd, 1 d, 2 d, 3 d,y
HE
mnp
Hybrid (TM-like) m1,2,3,y n1,2,3,y pd, 1 d, 2 d, 3 d,y
M M L
h
d
D
D
X
tan
c
tan
c
3
1
Figure 2. Conguration of a dielectric ring resonator (DRR) in a
cylindrical cavity. The DRR is supported in the cavity by a low
dielectric constant material (e
3
). (FromKobayashi and Miura [17],
r 1984 IEEE.)
The conguration under analysis is shown in Fig. 9 and
consists of a cylindrical DR with radius a and height L,
separated by M
1
and M
2
from the lower and upper ground
planes, respectively. This structure is analyzed using a
cylindrical coordinate system (r,f,z) and is divided into six
regions as shown in the gure. At resonance most of the
energy of the DR is stored in region 1, while the elds de-
cay exponentially in regions 24. Regions 5 and 6 are ig-
nored since very little energy is stored there. This
approximation introduces a small error, but greatly sim-
plies the solution of the equations that dene the struc-
ture. Since the TE
01d
mode is circularly symmetric, only
the E
f
, H
r
, and H
z
elds exist. To satisfy the boundary
conditions and ensure the continuity of the tangential
elds, E
f
in each region is
E
f1
A
1
J
1
k
r1
r cosbz y 2
E
f2
A
2
J
1
k
r1
r sinha
2
z M
1
3
E
f3
A
3
J
1
k
r1
r sinh a
3
z L M
2
4
E
f4
A
4
K
1
k
r4
r cosbz y 5
where
A
2
A
1
cosy
sinha
2
M
1
; A
3
A
1
cosbL y
sinha
3
M
2
;
A
4
A
1
J
1
k
r1
a
K
1
k
r4
a
and
b
2
k
2
0
e
r1
k
2
r1
k
2
0
e
r4
k
2
r4
; a
2
2
k
2
r1
k
2
0
e
r2
;
a
2
3
k
2
r1
k
2
0
e
r3
6
E-field
TE
01
E-field
EH
11
H-field
TM
01
H-field
HE
11
E-field
(a)
H-field
H-field E-field
Figure 3. Field distributions for the (a) TE
01d
, TM
01d
, EH
11d
, and HE
11d
modes and (b) EH
21d
,
HE
21d
, EH
31d
, and HE
31d
modes.
In Eqs. (2)(5) and those that follow J
n
k
r
r is the Bessel
function of the rst kind, order n, while K
n
k
r
r is the
modied Bessel function of the second kind, order n.
From Maxwells equations H
r
is derived from E
f
as
follows
H
r1
jb
om
0
A
1
J
1
k
r1
r sinbz y 7
H
r2

ja
2
om
0
A
2
J
1
k
r1
r cosha
2
z M
1
8
H
r3

ja
3
om
0
A
3
J
1
k
r1
r cosha
3
z L M
2
9
H
r4
jb
om
0
A
4
K
1
k
r4
r sinbz y 10
while H
z
is obtained in a similar manner:
H
z1
jk
r1
om
0
A
1
J
0
k
r1
r cosbz y 11
H
z2
jk
r1
om
0
A
2
J
0
k
r1
r sinha
2
z M
1
12
H
z3
jk
r1
om
0
A
3
J
0
k
r1
r sinha
3
z L M
2
13
H
z4

jk
r4
om
0
A
4
K
0
k
r4
r cosbz y 14
E-field
EH
21
E-field
EH
31
H-field
HE
21
H-field
HE
31
E-field H-field
H-field H-field
(b)
Figure 3. (Continued).
Applying continuity conditions to H
z1
and H
z4
at r a
gives the rst eigenvalue equation
J
1
k
r1
a
k
r1
J
0
k
r1
a

K
1
k
r4
a
k
r4
K
0
k
r4
a
15
where
k
2
r4
k
2
0
e
r1
e
r4
k
2
r1
16
The continuity conditions for H
r1
and H
r2
at z 0 give
tany
a
2
b
cotha
2
M
1
17
while the continuity conditions for H
r1
and H
r3
at z L give
tanbL y
a
3
b
cotha
3
M
2
18
M
TE
01

P
TM
01
M
HE
11
P
EH
11

M
M
HE
21
P
P
EH
21
M
M
HE
31

P
P
EH
31
M
M
HE
41

P
P
EH
41

Figure 4. Magnetic (M) and electric (P) dipole
moments for several modes of the cylindrical DR.
(From Kobayashi and Miura [17], r 1984 IEEE.)
Combining Eqs. (17) and (18) by eliminating y yields the
second eigenvalue equation
bL tan
1
a
2
b
cotha
2
M
1
_ _
tan
1
a
3
b
cotha
3
M
2
_ _
pp;
p0; 1; 2; 3; . . .
19
For the TE
01d
mode p0. Solving the coupled Eqs. (6),
(15), and (19) gives k
0
, from which the resonant frequency
is calculated
f
0
k
0
c
2p
20
where c is the speed of light in a vacuum, or 2.99792458
10
8
m/s. In practice, Eqs. (6), (15), and (19) are solved nu-
merically using a zero-nding computer program or sub-
routine such as Broydens method. Several example
congurations from the literature as well as their com-
puted and measured resonant frequencies are given in
Table 2. These results show that when M
1
and M
2
are
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1
1.5
2
2.5
3
3.5
4
4.5
M/D
r
(
D
/
o
)
2
TE
01
EH
11
HE
11
TM
01
HE
21
EH
21
TE
011+
TM
011+
Figure 5. Mode chart of normalized frequency versus M/D (with
D/L2.5, d/D2, D
x
0).
0 1 2 3 4 5 6 7 8 9 10
0
1
2
3
4
5
(D/L)
2
TM
011+
TE
01
EH
11
HE
11
TM
01
HE
21
EH
21
r
(
D
/
o
)
2
Figure 6. Mode chart of normalized frequency versus (D/L)
2
(with d/D2, M/D0.4, D
x
0).
1 2 3 4 5
0
1
2
3
4
d/D
r
(
D
/
o
)
2
TE
01
EH
11
HE
11
TM
01
HE
21
EH
21
Figure 7. Mode chart of normalized frequency versus d/D (with
D/L2.5, M/D0.4, D
x
0).
0 0.1 0.2 0.3 0.4 0.5 0.6
0
1
2
3
4
5
D
X
/D
r
(
D
/
o
)
2
TM
011+
TE
01
EH
11
HE
11
TM
01
HE
21
EH
21
Figure 8. Mode chart of normalized frequency versus D
x
/D
(with D/L2.5, d/D2, M/D0.4).
r
z
3
4
6
5
r3
L
M
2
M
1
1
2
z = L+M
2
z = L
z = 0
z = M
1
r = 0 r = a
r6
r1
r2

r5
r4
Figure 9. Conguration of the DR used in the simplied analysis
of a TE
01d
mode. (From Itoh and Rudokas [41], r 1977 IEEE.)
large, which corresponds to an open DR, the accuracy of
the method worsens. The accuracy increases as the PEC
walls get closer to the resonator. When M
1
M
2
0, the
analysis is exact. This leads to the conclusion that the
method is best suited to DRs placed on thin substrates with
the PEC cavity enclosure close to the topface of the DR,
such as would be seen in an MIC or MMIC environment. It
also works well for DRs sandwiched between two PEC
plates, a conguration typically used to measure relative
permittivity and loss tangent of dielectric materials [7].
5. Q FACTOR
The quality factor is an important gure of merit for
assessing the performance of a DR, or any microwave res-
onator. It characterizes both the dielectric and conductor
losses for closed resonators, as well as radiation losses for
open resonators. In this section, only closed resonators are
considered. Filter insertion loss is inversely proportional
to the resonator unloaded Q. Design objectives require a
low insertion loss, which in turn necessitates high-Q res-
onators. Narrow bandwidth lters also require high-
Q resonators, particularly in todays crowded electromag-
netic spectrum where communications receivers need
high cochannel and spurious signal rejection.
General Q calculations will be described for the struc-
ture shown in Fig. 2. It consists of a cylindrical cavity
made out of metal, a high-permittivity dielectric ring res-
onator, and its dielectric supporting structure. The un-
loaded quality factor Q
u
of a resonator is dened as [91]
Q
u
o
0
storedenergy
average power dissipated
o
0
W
P
c
P
d
P
sup
21
where o
0
is the resonant frequency of the DR; W is the
stored energy; and P
c
, P
d
, and P
sup
are the power dissi-
pated due to losses in the conductor, dielectric resonator,
and dielectric support, respectively. The unloaded Q may
be expressed in terms of the individual Q factors of the
conductor Q
c
, the dielectric resonator Q
d
, and dielectric
support Q
sup
as follows:
1
Q
u
1
Q
c
1
Q
d
1
Q
sup
22
The stored energy, W may be written as [92]
WW
e
W
m
23
where W
e
and W
m
denote the energies stored in the elec-
tric and magnetic elds, respectively. The stored electric
energy is dened as
W
e
1
4
_
V
ejEj
2
dV 24
and the stored magnetic energy is
W
m
1
4
_
V
mjHj
2
dV 25
However, at resonance, the stored magnetic energy equals
the stored electric energy, or W
e
W
m
; hence
W2W
e
2W
m
1
2
_
V
ejEj
2
dV
1
2
_
V
mjHj
2
dV 26
So, from Eq. (26) it is clear that the stored energy may be
calculated from either the electric eld or the magnetic
eld.
The conductor power dissipation is calculated from the
tangential magnetic elds at the surface of the conducting
enclosure as
P
c
1
2
_
S
R
S
H
tan
j j
2
dS 27
where R
S
is the surface resistance of the conductor, H
tan
is
the magnetic eld tangential to the enclosure walls, and S
denotes the surface area of the enclosure walls. Introduc-
ing the skin depth, d 1=
pf
0
m" sss
0
_
, where f
0
is the reso-
nant frequency, s
0
is the conductivity of the enclosure, and
" ss is a relative conductivity, which accounts for the surface
roughness of the conductor, the power dissipation due to
the conducting walls may be redened as
P
c
o
0
d
4
_
S
mjH
tan
j
2
dS 28
Then from Eqs. (26) and (28) we can now dene the con-
ductor Q as [85]
Q
c
o
0
W
P
c
2
d
_
V
mjHj
2
dV
_
S
mjH
tan
j
2
dS
29
To model both damping and ohmic losses within
the dielectric resonator material, a complex permittivity
Table 2. Comparison of Simplied Method with Measured Values for TE
01d
Mode
e
r1
e
r2
e
r3
e
r4
a (mm) L (mm) M
1
(mm) M
2
(mm) f
0
(GHz) Measured f
0
(GHz) Computed
36.2 1.0 1.0 1.0 3.015 4.16 3.41 3.41 7.94 [46] 8.311 ( 4.7%)
34.19 9.6 1.0 1.0 7.49 7.48 0.7 0.72 4.348 [34] 4.355 ( 0.2%)
63.7 1.0 1.0 1.0 14.29 11.25 0.0 0.0 2.131 [7] 2.131 (0%)
is introduced:
e
c
e j
s
o
30
The loss tangent is the ratio between the imaginary and
real parts of the complex permittivity and is a measure of
the power loss in the dielectric. The loss tangent is dened
as
tan d
c
s
oe
31
Rearranging Eq. (31), we can obtain the effective conduc-
tivity to account for the losses in the dielectric resonator:
s
eff
2pf
0
e
r
e
0
tan d
c
32
Then the dielectric resonator power dissipation may be
calculated as
P
d
1
2
_
V
0
s
eff
jEj
2
dV
0
33
where V
0
denotes the volume of the dielectric resonator
only. From Eqs. (26) and (33) the dielectric resonator Q
is [85]
Q
d
o
0
W
P
d
o
0
_
V
ejEj
2
dV
_
V
0
s
eff
jEj
2
dV
0
34
where V is the volume of the entire resonator and e is a
function of position for inhomogeneous resonators. Using
a similar approach we may dene the Q factor for the
dielectric support
Q
sup
o
0
W
P
sup
o
0
_
V
ejEj
2
dV
_
V
00
s
eff
3
jEj
2
dV
00
35
where s
eff
3
is the effective conductivity of the dielectric
support and V
00
is the volume of the dielectric support only.
The effective conductivity of the dielectric support may be
found using Eq. (32).
Q factors as a function of M/D for the resonator con-
guration of Fig. 2 have been calculated using the com-
mercial software CST Microwave Studio [90], and are
presented in Figs. 1013 for the TE
01d
, EH
11d
, HE
11d
,
and TM
01d
modes. The metallic walls of the cavity are
modeled as copper, with s 5.810
7
S/m, while the loss
tangent of the dielectric material is set to tand 5 10
5
.
Each of these gures show the effect of the cavity height
on Q
c
, Q
d
, and Q
u
. In general, as the cavity height in-
creases so does the value of Q
c
, which leads to a greater
value of Q
u
. This is to be expected since the elds outside
the DR walls are evanescent, so increasing the height of
the cavity reduces the size of the tangential elds induced
on the metallic walls. In practice, the cavity size cannot be
increased indefinitely to increase the Q factor, because the
overall volume of the cavity needs to be kept small. For
lter applications, the cavity size cannot be made too large
0
10.000
20.000
30.000
40.000
50.000
60.000
70.000
80.000
90.000
100.000
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
M/D
Q

f
a
c
t
o
r
Q
c
Q
d
Q
u
Figure 10. Q factor versus M/Dfor the TE
01d
mode (with e
r
38,
e
3
1, D/L2.5, d/D2, D
x
0).
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
20.000
40.000
60.000
80.000
100.000
120.000
M/D
Q

f
a
c
t
o
r
Q
c
Q
d
Q
u
Figure 11. Q factor versus M/Dfor the EH
11d
mode (with e
r
38,
e
3
1, D/L2.5, d/D2, D
x
0).
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
20.000
40.000
60.000
80.000
100.000
120.000
140.000
160.000
180.000
M/D
Q

f
a
c
t
o
r
Q
c
Q
d
Q
u
Figure 12. Qfactor versus M/Dfor the HE
11d
mode (with e
r
38,
e
3
1, D/L2.5, d/D2, D
x
0).
because substantial eld strength is required on the cavity
walls to generate interresonator coupling. Cavity design
should take into consideration Q
u
, cavity volume, and
spurious performance; it usually requires a compromise
between these parameters and is further inuenced by the
intended DR application.
6. MATERIAL PARAMETERS
An important parameter for evaluating the temperature
stability of a dielectric resonator is the temperature coef-
cient of resonant frequency t
f
. This parameter takes into
account the thermal linear expansion of the DR material
and its enclosure, as well as the variation in the relative
permittivity e
r
of the DR material as a function of tem-
perature. The temperature coefcient is expressed as the
ratio of the change in frequency to the original frequency
divided by the change in temperature:
t
f

Df
0
f
0
DT
36
Units of the temperature coefcient are parts per million
per degree Celsius (ppm/1C). The temperature coefcient
for the structure in Fig. 2, which includes the effect of the
dielectric support, is [13]
t
f
A
r
t
r
A
3
t
3
A
D
A
x
A
L
t
a
A
C
t
C
37
where
t
r
De
r
e
r
DT
; t
3
De
3
e
3
DT
;
t
a
DD
DDT

DD
x
D
x
DT

DL
LDT
;
t
C
Dd
dDT
38
and
A
r
e
r
f
0
Df
0
De
r
; A
3
e
3
f
0
Df
0
De
3
; A
D
D
f
0
Df
0
DD
;
A
x
D
x
f
0
Df
0
DD
x
; A
L
L
f
0
Df
0
DL
; A
c
d
f
0
Df
0
Dd
39
In Eq. (38) t
r
and t
3
represent the temperature coefcients
of the dielectric constants of the resonator and support,
while t
a
and t
c
are the coefcients of thermal linear ex-
pansion for the dielectric material and conductor enclo-
sure, respectively. The constants A
r
, A
3
, A
D
, A
x
, A
L
, and A
c
can be determined accurately by calculating the small
resonant frequency shifts Df
0
caused by the small changes
De
r
, De
3
, DD, DD
x
, DL, or Dd, respectively. In practice, tem-
perature compensation of the dielectric resonator cong-
uration can be obtained by choosing t
r
to cancel out t
a
and
t
c
. Manufacturers usually supply materials with a range
of t
r
values to choose from. Some typical values of dielec-
tric constant, loss tangent, temperature coefcient, and
thermal expansion for DR materials from two different
manufacturers are given in Tables 3 and 4.
7. EXTERNAL AND INTERRESONATOR COUPLING
For a DR to be useful in a microwave circuit, it must be
coupled to an external transmission line, another resona-
tor, or both. One of the simplest ways to externally couple
the TE
01d
mode in an MIC environment is to place it next
to a microstrip line, as shown in Fig. 1b. Since the TE
01d
mode behaves like a magnetic dipole, its magnetic eld
loops link with the transverse eld of the microstrip line.
The coupling structure and eld lines of the DR and mi-
crostrip are illustrated in Fig. 14a [93]. By simply chang-
ing the distance d between the DR and microstrip line, the
coupling value is varied. The equivalent circuit of the cou-
pling structure is a parallel resonant circuit placed in se-
ries with the microstrip line [94], as shown in Fig. 14b.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
M/D
20.000
40.000
60.000
80.000
100.000
120.000
140.000
160.000
180.000
Q

f
a
c
t
o
r
Q
c
Q
d
Q
u
Figure 13. Q factor versus M/D for the TM
01d
mode (with e
r
38, e
3
1, D/L2.5, d/D2, D
x
0).
Table 3. Typical Parameters of DR Materials Available from Trans-Tech
Series Dielectric Constant Q (1/tand) Available t
f
(ppm/1C) t
a
(ppm/1C) Composition
2900 2930.7 450,000 at 2 GHz 4/2/0/ 2 410 BaZnTa oxide
3500 34.536.5 435,000 at 2 GHz 6/3/0/ 3 10 BaZnCoNb
4300 43 49500 at 4.3 GHz 6/3/0/ 3/ 6 6.5 Zrtitaniumbased
4500 44.746.2 49500 at 4.3 GHz 6/3/0/ 3/ 6 6.5 Zrtitaniumbased
8300 3536.5 49500 at 4.3 GHz 9/6/3/0/ 3 10 Titanatebased
8700 29.531 410,000 at 10GHz 4/2/0 10 BaZnTa oxide
The external coupling, characterized by the external cou-
pling factor Q
e
, may be calculated using a magnetic eld
ux method. This procedure relates Q
e
to the energy
stored in the resonator and the magnetic ux linking the
microstrip and DR together [93]
Q
e
4
Z
c
o
0
m
2
0
W
_ _
S
H
.
dS
_ _
2
40
where
W
1
4
e
0
e
i
_
V
i
jE
i
j
2
dV
i
41
In these equations, Z
c
is the characteristic impedance of
the microstrip, His the magnetic eld vector, S is the cross
section of the microstrip substrate, E
i
is the electric eld
vector in medium i, e
i
is the relative permittivity of medi-
um i, and V
i
is the volume of medium i. Details of an an-
alytical expression for Q
e
that gives good agreement with
measured results are presented in Ref. 93.
The analysis of interresonator coupling coefcients be-
tween two DRs in a waveguide will now be examined. A
typical coupling structure for this situation is shown in
Fig. 15. It consists of two identical DRRs placed coaxially
within a cutoff circular waveguide of diameter d. The
spacing between the two DRRs is 2 M, and the spacing
to the PEC walls enclosing the cylindrical waveguide at
each end is M
1
. This structure supports many different
modes, but for this analysis we are only interested in the
modes that couple magnetically through a dominant TE
evanescent mode in the cutoff waveguide such as the TE
01d
and EH
11d
modes. The derivation of the coupling coef-
cient presented follows the approach of Zaki and Chen
[95]. The equivalent circuit for a magnetically coupled
mode near resonance, shown in Fig. 16a, consists of two
series-resonant circuits coupled by a mutual inductance
M. Dening the coupling coefcient k in terms of the
equivalent-circuit parameters gives
k
M
L
42
Since this analysis will exploit the structural symmetry of
the conguration given in Fig. 15, we dene an equivalent
circuit with a symmetry plane as shown in Fig. 16b. Re-
placing the symmetry plane with a PEC yields a single
resonant circuit. The resonant frequency of the circuit
f
e
may be related to its equivalent-circuit parameters to
obtain
f
e
1
2p
L MC
_ 43
Repeating the procedure for the PMC wall yields a circuit
with a resonant frequency f
m
that may be given in terms of
the equivalent-circuit parameters as
f
m
1
2p
LMC
_ 44
Solving Eqs. (42)(44) for the coupling coefcient k yields
k
M
L

f
2
e
f
2
m
f
2
e
f
2
m
45
Table 4. Typical Parameters of Resomics DR Materials Available from Murata
Material Code Dielectric Constant Q (1/tand) Available t
f
(ppm/1C) t
a
(ppm/1C)
U 3640 46000 at 7 GHz 410 67
M 3740 47000 at 7 GHz 06 67
V 3336 410,000 at 10GHz 08 1213
R 2931 412,000 at 10GHz 06 10.7
B 2729 415,000 at 10GHz 06 11
E 2425 420,000 at 10GHz 06 10.7
F 2324 35,000 at 10GHz 04 11
DR
E
Substrate
H H
Microstrip Line
Cavity
(a)
Z
0
L
R
Z
eq
(b)
Z
0

d
C
Figure 14. (a) Conguration of TE
01d
mode DR coupled to a mi-
crostrip line (from Guillon et al. [93], r 1981 IEEE); (b) equiva-
lent circuit of the coupled DR (from Khanna and Garault [94],
r 1983 IEEE).
The computation of inter-resonator coupling has been re-
duced to the problem of calculating the resonant frequen-
cies of two different structures that have a PEC or PMC
wall at the plane of structural symmetry. A similar equa-
tion may be derived for electric coupling, only the signs of
the numerator values are reversed [14]. A more general
equation for both electric and magnetic coupling is
k
jf
2
e
f
2
m
j
f
2
e
f
2
m
46
The resonant frequencies f
e
and f
m
may be calculated us-
ing an approximate analysis method, mode matching, or
numerical techniques such as FDTD, FEM, or FIT. Each
method has been used in the literature to accurately cal-
culate DR interresonator coupling coefcients. This tech-
nique may also be used to calculate interresonator
coupling through a slot in a conducting wall, as long as
the conguration has a structurally symmetry plane sim-
ilar to that shown in Fig. 15. Measured and computed
coupling coefcients for two EH
11d
DRRs coupled via a slot
are compared in Fig. 17 [96]. The inset of the gure shows
the conguration of the coupling structure.
BIBLIOGRAPHY
1. R. D. Richtmeyr, Dielectric resonators, J. Appl. Phys. 10:
391398 (June 1939).
2. A. Okaya and L. F. Barash, The dielectric microwave resona-
tor, Proc. IRE 50:20812092 (Oct. 1962).
3. S. B. Cohn, Microwave lters containing high-Q dielectric
resonators, IEEE Trans. Microwave Theory Tech. 16:218227
(April 1968).
4. W. H. Harrison, A miniature high-Q bandpass lter employ-
ing dielectric resonators, IEEE Trans. Microwave Theory
Tech. 16:210218 (April 1968).
5. D. J. Masse et al., A new low-loss high- kappa temperature-
compensated dielectric for microwave applications, Proc.
IEEE 59:16281629 (Nov. 1971).
6. J. K. Plourde, D. F. Linn, H. M. OBryan Jr., and J. Thomson
Jr., Ba
2
Ti
9
O
20
as a microwave dielectric resonator, J. Am.
Ceramic Soc. 58:418420 (Sept.Oct. 1975).
7. W. E. Courtney, Analysis and evaluation of a method of mea-
suring the complex permittivity and permeability of micro-
wave insulators, IEEE Trans. Microwave Theory Tech.
18:476485 (Aug. 1970).
M
(a)
M C L L C M
(b)
L L
C C
2M 2M
Structurally Symmetric Plane
Figure 16. Equivalent circuit of two magnetically coupled reso-
nators: (a) in conventional form; (b) including a structural sym-
metry plane. (From Zaki and Chen [95], r 1987 IEEE.)
2 4 6 8 10 12 14 16 18 20
0
0.005
0.01
0.015
0.02
0.025
Slot Length (mm)
C
o
u
p
l
i
n
g
A
A AA
FDTD
Measured
Figure 17. Measured and calculated slot coupling coefcients of
an EH
11d
DRR for a slot width of 1.6 mm and a slot thickness of
1 mm (dimensions d37.4mm, D21mm, D
x
8.4 mm, LM
7.4 mm, e
r
35, f
0
4.030GHz). Inset shows conguration of the
structure. (From Weily et al. [96]; this material is used by per-
mission of John Wiley and Sons, Inc. Copyright r 2001 by John
Wiley and Sons, Inc.)
M M
1
L
d
D
D
X
M
1
M L
M M
1
L
d
D
D
X
M
1
M L
1
Figure 15. Conguration of coupled dielectric ring
resonators (DRRs) showing the position of the struc-
turally symmetric plane. (From Kobayashi and
Minegishi [13], r 1987 IEEE.)
8. K. Wakino, T. Nishikawa, S. Tamura, and Y. Ishikawa,
Microwave bandpass lters containing dielectric resonators
with improved temperature stability and spurious
response, IEEE MTT-S Int. Microwave Symp. Digest, 1975,
pp. 6365.
9. J. K. Plourde and C. L. Ren, Application of dielectric resona-
tors in microwave components, IEEE Trans. Microwave
Theory Tech. 29:754770 (Aug. 1981).
10. S. J. Fiedziuszko, Dual-mode dielectric resonator loaded ca-
vity lters, IEEE Trans. Microwave Theory Tech. 30:1311
1316 (Sept. 1982).
11. W. C. Tang et al., Planar Dual-Mode Cavity Filters Including
Dielectric Resonators, U.S. Patent 4,652,843 (March 24, 1987).
12. W. C. Tang, An 8-pole quasi-elliptic function lter realized in
3 dielectric resonator cavities, IEEE MTT-S Int. Microwave
Symp. Digest, 1986, pp. 349351.
13. Y. Kobayashi and M. Minegishi, Precise design of a bandpass
lter using high-Q dielectric ring resonators, IEEE Trans.
Microwave Theory Tech. 35:11561160 (Dec. 1987).
14. Y. Kobayashi and M. Minegishi, A low-loss bandpass lter
using electrically coupled high-Q TM
01d
dielectric rod resona-
tors, IEEE Trans. Microwave Theory Tech. 36:17271732
(Dec. 1988).
15. K. A. Zaki, C. Chen, and A. E. Atia, Canonical and longitu-
dinal dual-mode dielectric resonator lters without iris, IEEE
Trans. Microwave Theory Tech. 35:11301135 (Dec. 1987).
16. Y. Kobayashi and K. Kubo, Canonical bandpass lters using
dual-mode dielectric resonators, IEEE MTT-S Int. Microwave
Symp. Digest, 1987, pp. 137140.
17. Y. Kobayashi and M. Miura, Optimum design of shielded
dielectric rod and ring resonators for obtaining best mode
separation, IEEE MTT-S Int. Microwave Symp. Digest, 1984,
pp. 184187.
18. R. V. Snyder, Dielectric resonator lters with wide stopbands,
IEEE Trans. Microwave Theory Tech. 40:21002103 (Nov.
1992).
19. T. Nishikawa, K. Wakino, K. Tsunoda, and Y. Ishikawa, Di-
electric high-power bandpass lter using quarter-cut TE
01d
image resonator for cellular base stations, IEEE Trans.
20. K. Wakino, T. Nishikawa, H. Matsumoto, and Y. Ishikawa,
Miniaturized bandpass lters using half wave dielectric res-
onators with improved spurious response, IEEE MTT-S Int.
21. Y. Kobayashi and C. Inoue, Bandpass and bandstop lters
using dominant TM
01d
mode dielectric rod resonators, IEEE
MTT-S Int. Microwave Symp. Digest, 1997, pp. 793796.
22. J.-F. Liang, K. A. Zaki, and A. E. Atia, Mixed modes dielectric
resonators lters, IEEE Trans. Microwave Theory Tech.
42:24492454 (Dec. 1994).
23. C. Wang, H.-W. Yao, K. A. Zaki, and R. R. Mansour, Mixed
modes cylindrical dielectric resonator lters with rectan-
gular enclosure, IEEE Trans. Microwave Theory Tech. 43:
28172823 (Dec. 1995).
24. H. Y. Hwang, N. S. Park, H. Y. Cho, S. W. Yun, and I. S. Chang,
The design of bandpass lters made of both dielectric and co-
axial resonators, IEEE MTT-S Int. Microwave Symp. Digest,
1997, pp. 805808.
25. J. F. Liang and W. D. Blair, High-Q TE
01
mode DR lters for
PCS wireless base stations, IEEE Trans. Microwave Theory
Tech. 46:24932500 (Dec. 1998).
26. A. R. Weily and A. S. Mohan, Design and wideband spurious
performance prediction for a mixed resonator combline/TE
01
-
mode DR lter using FDTD, Microwave Opt. Technol. Lett.
23(5):266273 (Dec. 5, 1999).
27. C. Wang, K. A. Zaki, and A. E. Atia, Dual-mode combined di-
electric and conductor loaded cavity lters, IEEE MTT-S Int.
28. A. Abdelmonem, J.-F. Liang, H.-W. Yao, and K. A. Zaki, Full
wave design of spurious free DR TE mode band pass lter,
IEEE Trans. Microwave Theory Tech. 43:744752 (April
1995).
29. C. Wang, K. A. Zaki, A. E. Atia, and T. G. Dolan, Dielectric
combline resonators and lters, IEEE Trans. Microwave The-
ory Tech. 46:25012506 (Dec. 1998).
30. I. C. Hunter, J. D. Rhodes, and V. Dassonville, Dual-mode l-
ters with conductor loaded dielectric resonators, IEEE Trans.
31. A. R. Weily and A. S. Mohan, Microwave lters with improved
spurious performance based on sandwiched conductor dielec-
tric resonators, IEEE Trans. Microwave Theory Tech.
49:15011507 (Aug. 2001).
32. J. A. Curtis, S. J. Fiedziuszko, and S. C. Holme, Hybrid di-
electric/HTS resonators and their applications, IEEE MTT-S
Int. Microwave Symp. Digest, 1991, pp. 447450.
33. D. Baillargeat, S. Verdeyme, M. Aubourg, and P. Guillon, CAD
applying the nite-element method for dielectric-resonator
lters, IEEE Trans. Microwave Theory Tech. 46:1017 (Jan.
1998).
34. D. Kajfez and P. Guillon, eds., Dielectric Resonators, 2nd ed.,
Noble Publishing, Tucker, GA, 1998.
35. Y. Kobayashi and S. Tanaka, Resonant modes of a dielectric
rod resonator short-circuited at both ends by parallel con-
ducting plates, IEEE Trans. Microwave Theory Tech.
28:10771085 (Oct. 1980).
36. Y. Kobayashi, K. Kojima, and S. Yoshida, Shielded TM
010
di-
electric rod resonator, Electron. Commun. Japan 64-B:6571
(Feb. 1981).
37. J. S. Fiedziuszko and A. Jelenski, The inuence of conducting
walls on resonant frequencies of the dielectric microwave res-
onator, IEEE Trans. Microwave Theory Tech. 19:778780
(Sept. 1971).
38. J. Van Bladel, On the resonances of a dielectric resonator of
very high permittivity, IEEE Trans. Microwave Theory Tech.
23:199208 (Feb. 1975).
39. Y. Konishi, N. Hoshino, and Y. Utsumi, Resonant frequency of
a TE
01d
dielectric resonator, IEEE Trans. Microwave Theory
Tech. 24:112114 (Feb. 1976).
40. M. Verplanken and J. Van Bladel, The electric-dipole reso-
nances of ring resonators of very high permittivity, IEEE
Trans. Microwave Theory Tech. 24:108112 (Feb. 1976).
41. T. Itoh and R. S. Rudokas, New method for computing the
resonant frequencies of dielectric resonators, IEEE Trans.
Microwave Theory Tech. 25:5254 (Jan. 1977).
42. M. W. Pospieszalski, On the theory and application of the
dielectric post resonator, IEEE Trans. Microwave Theory
Tech. 25:228231 (March 1977).
43. P. Guillon and Y. Garault, Accurate resonant frequencies of
dielectric resonators, IEEE Trans. Microwave Theory Tech.
25:916922 (Nov. 1977).
44. M. W. Pospieszalski, Cylindrical dielectric resonators and
their applications in TEM line microwave circuits, IEEE
Trans. Microwave Theory Tech. 27:233238 (March 1979).
45. M. Verplanken and J. Van Bladel, The magnetic-dipole reso-
nances of ring resonators of very high permittivity, IEEE
Trans. Microwave Theory Tech. 27:328333 (April 1979).
46. M. Jaworski and M. W. Pospieszalski, An accurate solution of
the cylindrical dielectric resonator problem, IEEE Trans.
Microwave Theory Tech. 27:639643 (March 1979).
47. R. R. Bonetti and A. E. Atia, Design of cylindrical dielectric
resonators in inhomogeneous media, IEEE Trans. Microwave
Theory Tech. 29:323326 (April 1981).
48. J. Krupta, Optimization of an electrodynamic basis for deter-
mination of the resonant frequencies of microwave cavities
partially lled with a dielectric, IEEE Trans. Microwave The-
ory Tech. 31:302305 (March 1983).
49. D. Maystre, P. Vincent, and J. C. Mage, Theoretical and
experimental study of the resonant frequency of cylindrical
dielectric resonator, IEEE Trans. Microwave Theory Tech.
31:844848 (Oct. 1983).
50. M. Tsuji, H. Shigesawa, and K. Takiyama, Analytical and ex-
perimental investigations on several resonant modes in open
the dielectric resonators, IEEE Trans. Microwave Theory
Tech. 32:628633 (June 1984).
51. J. Krupta, Computations of frequencies and intrinsic Q fac-
tors of TE
0nm
modes of a dielectric resonators, IEEE Trans.
Microwave Theory Tech. 33:274277 (March 1985).
52. P. Guillon, J. P. Balabaud, and Y. Garault, TM
01p
tubular and
cylindrical are dielectric resonator mode, IEEE MTT-S Int.
53. J. E. Lebaric and D. Kajfez, Analysis of dielectric resonator
cavities using the nite integration technique, IEEE Trans.
Microwave Theory Tech. 37:17401747 (Nov. 1989).
54. C. C. Su and J. M. Guan, Finite difference analysis of dielec-
tric-loaded cavities using the simultaneous iteration of the
power method with the Chebyshev acceleration technique,
IEEE Trans. Microwave Theory Tech. 42:19982006 (Oct.
1994).
55. J. Mielewski, A. Cwikla, and M. Mrozowski, Accelerated FD
analysis of dielectric resonators, IEEE Microwave Guided
Wave Lett. 8:375377 (Nov. 1998).
56. F. H. Gil and J. P. Martinez, Analysis of dielectric resonators
with tuning screw and supporting structure, IEEE Trans.
57. P. S. Kooi, M. S. Leong, and A. L. S. Parakash, Finite element
analysis of the shielded dielectric resonator, Proc. IEE
132(Part H):716 (Feb. 1985).
58. M. Taheri and D. Mirshekar-Syahkal, Accurate determina-
tion of modes in dielectric-loaded cylindrical cavities using a
one-dimensional nite element method, IEEE Trans. Micro-
wave Theory Tech. 37:15361541 (Oct. 1989).
59. V. Madrangeas, M. Aubourg, P. Guillon, S. Vigneron, B. Ther-
on, and D. Parise, A new nite element method formulation
applied to D. R. microwave lter design, IEEE MTT-S Int.
60. A. W. Glisson, D. Kajfez, and J. James, Evaluation of modes in
dielectric resonators using a surface integral equation formu-
lation, IEEE Trans. Microwave Theory Tech. 31:10231029
(Dec. 1983).
61. D. Kajfez, A. W. Glisson, and J. James, Computed modal eld
distributions for isolated dielectric resonators, IEEE Trans.
62. Y. Kobayashi, K. Kojima, and S. Yoshida, Resonant modes for
a shielded dielectric resonator, Electron. Commun. Japan
64-B:4451 (Oct. 1981).
63. U. S. Hong and R. H. Jansen, Numerical analysis of shielded
dielectric resonators including substrate, support disk,
and tuning post, Electron. Lett. 18(23):10001002 (Nov. 11,
1982).
64. K. A. Zaki and A. E. Atia, Modes in dielectric-loaded wave-
guides and resonators, IEEE Trans. Microwave Theory Tech.
31:10391045 (Dec. 1983).
65. Y. Kobayashi, T. Aoki, and Y. Kabe, Inuence of conductor
shields on the Q-factors of a TE
0
dielectric resonator, IEEE
66. K. A. Zaki and C. Chen, Intensity and distribution of hybrid-
mode elds in dielectric-loaded waveguides, IEEE Trans.
67. K. A. Zaki and C. Chen, Loss mechanisms in dielectric-loaded
resonators, IEEE Trans. Microwave Theory Tech. 33:1448
1452 (Dec. 1985).
68. K. A. Zaki and C. Chen, New results in dielectric-loaded res-
onators, IEEE Trans. Microwave Theory Tech. 34:815824
(July 1986).
69. Y. Kobayashi and S. Nakayama, Design charts for shielded
dielectric rod and ring resonators, IEEE MTT-S Int. Micro-
wave Symp. Digest, 1986, pp. 241244.
70. C. Chen and K. A. Zaki, Resonant frequencies of dielectric
resonators containing guided complex modes, IEEE Trans.
Microwave Theory Tech. 36:14551457 (Oct. 1988).
71. S.-W. Chen and K. A. Zaki, Dielectric ring resonators loaded
in waveguide and on substrate, IEEE Trans. Microwave
Theory Tech. 39:20692076 (Dec. 1991).
72. X.-P. Liang and K. A. Zaki, Modeling of cylindrical dielectric
resonators in rectangular waveguides and cavities, IEEE
73. Y. Kobayashi and T. Senju, Resonant modes of shielded
uniaxial-anisotropic dielectric rod resonators, IEEE Trans.
74. C. Wang and K. A. Zaki, Generalized multilayer anisotropic
dielectric resonators, IEEE MTT-S Int. Microwave Symp.
Digest, 1998, pp. 233236.
75. K. S. Yee, Numerical solution of initial boundary value prob-
lems involving Maxwells equations in isotropic media, IEEE
Trans. Antennas Propag. 14:302307 (May 1966).
76. A. Navarro, M. J. Nun ez, and E. Martin, Study of TE
0
and
TM
0
modes in dielectric resonators by a nite difference time-
domain method coupled with the discrete Fourier transform,
IEEE Trans. Microwave Theory Tech. 39:1417 (Jan. 1991).
77. Y. Chen, R. Mittra, and P. Harms, Finite-difference time-
domain algorithm for solving Maxwells equations in rota-
tionally symmetric geometries, IEEE Trans. Microwave The-
ory Tech. 44:832839 (June 1996).
78. J. A. Pereda, L. A. Vielva, A. Vegas, and A. Prieto, Computa-
tion of resonant frequencies and quality factors of open
dielectric resonators by a combination of the nite-difference
time-domain (FDTD) and Pronys methods, IEEE Microwave
Guided Wave Lett. 2:431433 (Nov. 1992).
79. A. R. Weily and A. S. Mohan, Mode-chart calculation for
dielectric- and conductor-loaded resonators using modal
extraction and FDTD, Microwave Opt. Technol. Lett. 21(6):
405411 (June 20, 1999).
80. Z. Bi, Y. Shen, K. Wu, and J. Litva, Fast nite-difference time-
domain analysis of resonators using digital ltering and spec-
trum estimation techniques, IEEE Trans. Microwave Theory
Tech. 40:16111619 (Aug. 1992).
81. Z. Ma and Y. Kobayashi, Analysis of dielectric resonators us-
ing the FDTD method combined with the Pade interpolation
technique, IEICE Trans. Electron. E-84C:973976 (July
2001).
82. P. Harms, J.-F. Lee, and R. Mittra, A study of the non-orthog-
onal FDTD method versus the conventional FDTD technique
for computing resonant frequencies of cylindrical cavities,
IEEE Trans. Microwave Theory Tech. 40:741746 (April
1992).
83. N. Kaneda, B. Houshmand, and T. Itoh, FDTD analysis of di-
electric resonators with curved surfaces, IEEE Trans. Micro-
wave Theory Tech. 45:16451649 (Sept. 1997).
84. S. Dey and R. Mittra, A conformal nite-difference time-do-
main technique for modeling cylindrical dielectric resonators,
IEEE Trans. Microwave Theory Tech. 47:17371739 (Sept.
1999).
85. C. Wang, H.-W. Yao, and K. A. Zaki, An improved FDTD
method for analysis of higher order modes high Q dielectric
resonators, IEEE AP-S Int. Symp. Digest, 1995, pp. 236239.
86. D. Kajfez, A. Elsherbeni, and A. Mokaddem, Higher order
modes in dielectric resonators, IEEE AP-S Int. Symp. Digest,
1996, pp. 306309.
87. C. Wang, B.-Q. Gao, and C.-P. Deng, Q factor of a resonator by
the nite-difference time-domain method incorporating per-
turbation techniques, Electron. Lett. 29(21):18661867 (Oct.
14, 1993).
88. C. Wang, B.-Q. Gao, and C.-P. Deng, Accurate study of Q-fac-
tor of a resonator by a nite-difference time-domain method,
IEEE Trans. Microwave Theory Tech. 43:15241529 (July
1995).
89. E. Snitzer, Cylindrical dielectric waveguide modes, J. Opt.
Soc. Am. 51:491498 (May 1961).
90. CST Microwave Studio 5.0 User Manual, CST Darmstadt,
Germany, 2004.
91. R. F. Harrington, Time Harmonic Electromagnetic Fields,
McGraw-Hill, New York, 1961.
92. D. K. Cheng, Field and Wave Electromagnetics, Addison-Wes-
ley, Reading, MA, 1989.
93. P. Guillon, B. Byzery, and M. Chaubet, Coupling parameters
between a dielectric resonator and a microstripline, IEEE
Trans. Microwave Theory Tech. 33:222226 (March 1985).
94. A. P. S. Khanna and Y. Garault, Determination of loaded, un-
loaded, and external quality factors of a dielectric resonator
coupled to a microstrip line, IEEE Trans. Microwave Theory
Tech. 31:261264 (March 1983).
95. K. A. Zaki and C. Chen, Coupling of non-axially symmetric
hybrid modes in dielectric resonators, IEEE Trans. Micro-
wave Theory Tech. 35:11361142 (Dec. 1987).
96. A. R. Weily and A. S. Mohan, Time domain CAD of dual-mode
HE
11
dielectric resonator microwave lters, Microwave Opt.
Technol. Lett. 29(1):611 (April 5, 2001).
DIGITAL MICROWAVE RECEIVERS
JOHANN F. LUY
TH. MUELLER
DaimlerChrysler Research
Ulm, Germany
1. INTRODUCTION
The general tasks of a receiver are amplication, channel
selection, and demodulation of a signal. The kind of chan-
nel selection may be used to distinguish between analog
and digital receiver architectures. In a superheterodyne
receiver (Fig. 1), the problem of image frequency suppres-
sion can be solved quite well by proper selection of the IF
frequencies and bandpass lters. Channel selection is de-
termined by the analog lters, and reconguration by
software is not supported.
Quadrature direct-conversion architecture is obtained
if the RFand rst IF mixer stages are dropped and the RF
signal is fed directly to the analog in-phase and quad-
rature downconversion mixers (Fig. 2). The reduced com-
plexity of this zero-IF receiver is the advantage of this
concept. The disadvantages are the problems with static
and dynamic offsets and corresponding errors in ampli-
tude and phase of the received signals.
The simplied principle of a digital receiver is shown in
Fig. 3, where the analog signal passes an antialiasing l-
ter (AAF) and a sample-and-hold (S/H) unit. The sampled
values of the signal are then converted into the digital
domain by an analog-to-digital converter (ADC) and fed
into an accumulator (ACCW) for further digital signal
processing. Digital signal processing of the received signal
provides high exibility and opens new possibilities re-
garding diversity techniques. By broadband sampling of
the FM band between 87 and 108 MHz, it has been shown
that coherent superposition of different channels contain-
ing the same information can lead to significantly im-
proved signal quality [1]. Channel diversity and MIMO
signal processing techniques become possible with digital
front ends [2].
2. RF AND IF SAMPLING ARCHITECTURE
By sampling the radio frequency (RF) or intermediate fre-
quency (IF), the number of input stages and therefore the
hardware effort can be reduced. The architectures of both
receiver types are shown in Fig. 4. The performance of the
ADC with respect to bandwidth and resolution [3] is de-
cisive for the choice of the receiver architecture. The ADC
performs a quantization in time and in value of the input
RF or IF signal.
Figure 1. Analog receiver: superheterodyne principle.
Figure 2. Analog receiver: direct-conversion architecture.
1014 DIGITAL MICROWAVE RECEIVERS
The sampling process with the sampling frequency f
S
can be understood in frequency domain by folding the in-
put signal with a series of diracs with distance f
S
(see
Fig. 5). The input signal at frequency f
IF
is shifted by the
sampling process to f
IF
kf
S
, where k is an integer value.
A lter has to select a frequency band so that only the de-
sired signal is sampled and all the other frequency com-
ponents are suppressed to avoid aliasing [4]. The
requirements on the slope of the lter are minimal if the
input frequency is an odd multiple of f
S
/4 because the dis-
tance to the next possible interferer becomes maximal.
The complex downconversion is usually realized in the
digital domain. This enables the use of high-order modu-
lation schemes, and ensures exact gain and phase match-
ing in in-phase and quadrature branches. Especially for a
xed IF frequency, the digital downconversion can be re-
alized by multiplying the sampled signal with 1 s and
1 s (see Fig. 4b).
By choosing a high sampling rate, the achievable sig-
nal-to-noise ratio in the desired channel can be calculated
by the formula
SNR
CH
SNR
ADC
10 log
10
f
S
2B
CH
_ _
with the channel bandwidth B
CH
, the sampling frequency
f
S
, and the signal-to-noise ratio of the ADC SNR
ADC
in
decibels. It can be seen that with increasing sampling rate
the SNR also increases. The state of the art is about 72 dB
SNR
ADC
at a sampling rate of 105MHz [5].
The highest exibility can be reached with broadband
RF sampling receivers [6]. Here all channels are available
in parallel, and the channel bandwidth and demodulation
are dened by digital signal processing so that the radio is
fully software-dened. In contrast to IF sampling receiv-
ers, the dynamic range that has to be handled by the ADC
becomes critical. To calculate the necessary dynamic
range, we have to accumulate the power of all interferers
and noise in the digitized frequency band. As an example,
the specication of GSM
1
denes a maximum interferer
power of 23 dBm, whereas noise and interferers in the
desired channel with a bandwidth of 200kHz have to keep
below 112 dBm. If we perform IF sampling, the IF lter
will be able to suppress most of the interferers so that the
interfering power will be below 43 dBm.
The second important factor when dening receivers is
intermodulation. The state of the art is about 90dB spu-
rious-free dynamic range [5]. To compare this with tradi-
tional downconverting architectures, we should remember
that the spurious-free dynamic range (SFDR) of an am-
plier or mixer can be calculated out of the third-order
interception point ICP3
out
and the output noise power
N
out
:
SFDR
2
3
ICP3
out
N
out
:
A standard monolithic integrated mixer [7] will show an
SFDR of about 82dB referring to a channel bandwidth of
200 kHz.
The selection of the sampling frequency also depends
on the technical feasibility of the antialiasing lter. Choos-
ing a frequency plan where the input signal with band-
width B is centered on f
SIG
an odd multiple of f
S
/4, a lter
margin on each side of f
S
/2 B is allowed to reach the
specied suppression of interferers. In practice [8], band-
width B should be maintained below f
S
/4, which corre-
sponds to a minimum oversampling ratio of 2.
An important issue for the implementation of RF or IF
sampling ADCs is sampling jitter [9]. The sources of sam-
pling jitter are mainly the clock generator phase noise and
the aperture jitter of the sampling unit. The slope of the
input signal will transform the error Dt of the sampling
time in an amplitude error Dy as shown in Fig. 6:
Dy Dt
.
@y
@t
t
S
In broadband sampling receivers the input signal consists
of the desired signal and a couple of interferers. The slope
of the signal can be calculated by the sum of the slopes of
each partial signal, weighted with its amplitude. If one
signal is dominant, it will dene the slope of the input
signal. If the signal amplitude is much lower than the am-
plitude of the interferer, then the slope of the input signal
will be dened by the interferer. If the sampled signal is
independent of the sampling frequency, then we have to
look on the statistical properties of the input signal.
A/D
d
i
g
i
t
a
l
m
i
x
e
r
dig.
filter
further
signal-
proces-
sing
(a)
(b)
Clk
Q
A/D
I
Clk
0,1,0,-1
1,0,-1,0
LP filter
LP filter
LO
Figure 4. (a) RF sampling receiver; (b) IF sampling receiver.
AAF S/ H ADC ACCU
(tt
0
)
Figure 3. Digital receiver illustrating the bandpass-limited sam-
pling concept.
1
European Telecommunication Standard, ETS 300-577, FTS 300
577, Radio Transmission and Reception, Dec. 1999, European
Telecommunication Standards Institute, Valbonne, France.
DIGITAL MICROWAVE RECEIVERS 1015
For a sinusoidal input signal y with amplitude A
sin
, we
obtain the variance of the slope
Var
@y
@t
_ _
_
p
slope
yy
2
dy
2A
2
sin
f
2
p
2
Finally we can calculate the achievable signal to noise ra-
tio in decibels when sampling a sinusoid signal with a
nonideal sampling circuit:
SNR
dB
10 log
P
SIG
P
noise
_ _
10 log
1
2
A
2
sin
2A
2
sin
f
2
p
2
Dt
2
_ _
20 log2f pDt
3. SIX-PORT ARCHITECTURE
The six-port direct-conversion technique is another prom-
ising approach toward software congurable receivers at
millimeter-wave carrier frequencies. The crucial advan-
tage over the sampling architecture is its simple construc-
tion requiring only the reference oscillator as an active RF
component, making this technology particularly attractive
for future low-cost and broadband RF front ends operating
at millimeter-wave frequencies.
The six-port technology traces back to the 1970s, when
it was developed as an alternative network analyzer [10].
By means of a six-port reectometer, it is possible to mea-
sure the complex reection coefcient of a device under
test (DUT). Furthermore, by the application of 2 six-
portsone on each side of the unknown DUTit is pos-
sible to determine the complete set of the S-parameters
[11]. Starting in the mid-1990s the six-port principle was
investigated for use as a direct-conversion receiver
[12,13]. Complete receivers were realized, and signals
that were modulated by quadrature phase shift keying
(QPSK) and quadrature amplitude modulation (QAM)
schemes have since been received and demodulated suc-
cessfully. Sony has veried the concept for a multicarrier
system based on HIPERLAN/2 (High-Performance Local
Area Network/2) and UMTS-FDD (Universal Mobile
Telecommunication SystemFrequency-division duplex)
standards [14,15], and Nokia has found the six-port di-
rect-conversion technique in a receiver a viable concept for
mobile terminals [16]. Other applications include anticol-
lision radars [17], very accurate short-range distance mea-
surement [18], and direction nding systems [19].
The functional principle of six-port devices is based on
the measurement of four independent output powers cor-
responding to the complex ratio between two superposed
electromagnetic waves under different phase angles.
While in reectometers used for the measurement of the
complex reection coefcient, a leakage from the LO port
to the test port of the DUT is essential, six-port circuits for
receivers require a very large isolation of the LO and the
RF ports. Carrier recovery (Fig. 7) is a possible approach
to avoid the need for an independent LO in a six-port
direct digital receiver [22].
To describe this principle, the six-port in Fig. 8 is con-
sidered together with two input signals, one from a local
oscillator (LO)
a
LO
t ja
0;LO
j
.
e
j2pf
LO
t j
LO
and the other from the RF port

a
RF
t ja
0;RF
j
.
e
j2pf
RF
t j
RF
which is the RF signal to be detected, that is, measured by

the six-port receiver.
Ideally, the six-port circuit represents a nondissipative
network and the frequencies of the RF and LO signal are
identical. For example, at port 3, the RF signal is attenu-
ated twice by 3dB and 901 phase-shifted. This signal is
added to the LO signal, which is also attenuated twice by
3dB although it does not receive a phase shift. For zero
phase shift between the LOand the RF signal and the same
amplitudes at the input ports, this leads to a resulting am-
plitude of
2
p
=2 for the superposed wave to be detected at
port 3. During a 3601 phase shift of the RF signal, the am-
plitudes resulting at the output ports are sine functions.
t
y = S(t)
S(t)
t
S

y
t
t
S
resulting
inaccuracy
of the measured
voltage
sampling error
sampling time n
.
T
S
U/V
Figure 6. Transformation of sampling error in amplitude error.

f
P P P
=
f
IF
f
S
f
S
f f f
IF
f
IF
-k.f
S
input signal sampling sampled signal Figure 5. General six-port architecture.
At each output port (ports 36) the signal passes
through detector circuits. A diode is used to rectify the
signal, and a subsequent lowpass lter (LPF) outputs the
amplitude of the signal, which is amplied and A/D-con-
verted. It becomes clear that the diode is the most critical
block in this receiver chain that has to be operated in its
quadratic region where the output voltage depends lin-
early on the power to be detected. However, minor devia-
tions from the square law can be compensated in the
digital domain by suitable calibration methods.
From at least three linearly independent output pow-
ers, the complex ratio between the two input signals
Gt j
at
0;RF
at
0;LO
j
.
e
j2p Df t Dj
can be determined. Furthermore, it can be shown that if
the leakage from the LO signal to the RF port is neglected,
the following linear relationship exists between the I and
Q components of the RF waveform and the power ratios
P
i
/P
3
(i 3y6):
It
6
i 3
A
i
P
i
t
P
3
t
Qt
6
i 3
B
i
P
i
t
P
3
t
where A and B are calibration coefcients to be deter-
mined by a suitable method. In the case of QPSK modu-
lation, this can be done by solving eight equations from
four known different states. As real six-ports are depen-
dent on the operating frequency, a calibration has to be
done for each frequency. Simulation and measurement re-
sults for QPSK modulation from 2 to 20GHz using the
same circuit and the described calibration method conrm
the broadband operation of the six-port technology [20].
Apart from the described simple method, the so-called
dual-tone calibration method is often used [21]. This meth-
od has to be applied if there is a relevant leakage from the
LO to the RF port as, for example, in reection coefcient
measurements or radar applications.
The advantages and disadvantages of the six-port tech-
nology, when compared to conventional heterodyne or
zero-IF (ZIF) architectures, strongly depend on the appli-
cation of interest. First to mention are the additional out-
put paths, which require more LNAs, lters, and ADCs. In
real communication standard environments, a DC offset is
caused by second-order nonlinearities, which are also
present in ZIF receivers and require an additional com-
pensation circuit in each path. In comparison to hetero-
dyne receivers, an image rejection lter and one LO can be
saved in a six-port architecture. It should also be men-
tioned that three output ports in suitable ve-port archi-
tecture are in principle sufcient to explicitly determine
the amplitude and phase. The additional sixth port serves
a way as a cyclic redundancy check and extends the
LNA
RF input
Reference

shift
Carrier
recovery
module
6
5
1
2
3
4 S
i
x
-
p
o
r
t
Six-port module Base-band module
D
e
c
o
d
e
r
I
Q
Figure 7. Architecture of hardware six-port
direct digital receiver with carrier recovery
from QPSK signal (from Ref. 22).
LO-Signal
1 Power
divider
9
0
0
9
0
0
9
0
0
Hybrid-Coupler
3
4
2
5
6
RF-Signal
Port 3 (P3)
Port 4 (P4)
Port 5 (P5)
Port 6 (P6)
Figure 8. Sampling process in the frequency
domain.
operation bandwidth. This can reduce the hardware effort
further with respect to cost of accuracy and bandwidth.
However, the main advantage of the six-port technolo-
gy is its broadband capability and the possibility for
software conguration and calibration. The six-port tech-
nology also becomes very attractive when extremely high
carrier frequencies at millimeter wavelengths are used.
The trend toward higher data rates will make use of new
bands: 17, 24, and 60 GHz have been considered. The in-
trinsic broadband property makes it in principle possible
to design RF front-ends operation over more than three
decades of bandwidth. Additionally, the same hardware
can in general be used for different standards simply by
loading suitable baseband processing algorithms. When
these software algorithms also become available in the fu-
ture, extremely broadband six-port RF front-ends might
be the technology of choice for SDR platforms. This can
help support a certain quality of service (QoS) as different
bands and standards can be probed and selected to trans-
mit information in harsh side conditions.
The six-port technique is very well suited for accurately
measuring the phase difference of two waves. This makes
the principle very attractive for determining the angle of
an incident wavefront, which is essential for multiple-
input/single-output (MISO) and multiple-input/multiple-
output (MIMO) systems. In a simple approach as
described in Ref. 19, the LO is replaced by a second
antenna and can lead to accurate results in the case where
a plane wave at a xed frequency has to be detected. Mod-
ications including an optimization of the number of input
antennas and output ports as well as frequency selection
mechanisms are considered and might offer smart six-
port-based antennas for digital beamforming, MISO,
MIMO, and space diversity compatibility.
4. HIGH-Q MILLIMETER-WAVE FILTERS WITH
INTEGRATION CAPABILITY
Important issues in digital microwave receivers are the
design and the realization of high-Q lters for planar
integration. Several approaches are known from the liter-
ature (Fig. 9):
*
Dielectric resonators attain the highest possible Q
values but have poor integration capability.
*
Use of multiple wafers that are glued or bonded may
lead to unloaded Q values up to several hundred. In-
tegration capability depends on the possible use of
multiple wafers.
*
Use of a single wafer with planar lters usually re-
sults in unloaded Q-values that are quite low. If the
substrate is 3D-structured, unloaded Q values up to
400 are feasible at center frequencies around 20 GHz.
Three-dimensional structuring of silicon substrates can
help provide enough volume to obtain high quality fac-
tors. In order to avoid multiple wafer bonding or gluing
techniques, a tilted waveguide resonator concept is ap-
plied with feeding structures of an inductive type. The
concept of a microelectromechanical system antialiasing
lter (MEMS AAF) is illustrated in Fig. 10. In one real-
ization scheme the silicon material itself is used as dielec-
tric ll material whereas the cavity surrounding is
realized by subsequent placing of metallized via holes
[23]. Q values of 100 are possible at center frequencies of
22 GHz if high-resistivity silicon is used as the dielectric
ll material for the resonator.
An air-lled cavity concept leads to still higher Q val-
ues: r400 are obtained from a microstrip-coupled resona-
tor operated in a H
101
resonance. Figure 11 gives
an impression from the layout and the realization of a
22.4-GHz microstrip-coupled lter [24].
5. APPLICATIONS: SOFTWARE-DEFINED MULTIPLE-
STANDARD TUNER PLATFORM
The number of different audio and video standards is ac-
tually increasing. Beside analog audio and video, we have
to cope with different digital audio standards such as DAB
[Digital Audio Broadcasting (European standard)], IBOC
4000
500
400
300
200
100
Q
u
a
l
i
t
y

f
a
c
t
o
r
monolithic substrate-Integrated
wafer-bonding
hybrid
Integration capability
MS, CPW,
spirals
System-on-chip
(60....360)
DC,2002
Katehi, Rebeiz 400..500
MEMSCAP 700
Dielectric resonator (DR)
cavities
TEMEX
Figure 9. Quality factor versus integration ca-
pability of several lter techniques.
[in-band/on-channel (radio)], DRM [Digital Radio Mondial
(international consortium)]. Sirius Satellite Radio, and
XM Satellite Radio, beside the new digital video standards
such as DVB-T (Digital Video Terrestrial Broadcasting). A
realization by parallel implementation of all standards is
no longer practical because of size, weight, and power con-
sumption [25]. Therefore a fully software recongurable
receiver has to handle these standards and be capable of
making updates during the receiver lifetime. The princi-
pal block diagram is shown in Fig. 12. Two receiver paths
are realized to enable antenna diversity for all frequency
ranges.
Two independent frontends amplify the signal and re-
alize, if necessary, the downconversion on an IF frequency.
In a rst realization step the front end is able to handle
the FM frequency range from 88 to 108MHz and the DVB-
T frequency range between 470 and 861 MHz. It is possi-
ble to extend it to AM/DRM (0.110 MHz), DAB (174
230 MHz), and the satellite based digital audio broadcast-
ing services Sirius/XM Sat Radio at 2.3GHz.
The sampling clock is about 76.8 MHz, so the FM spec-
trum can be sampled directly and channel selection is
performed in the digital domain. Because of limited signal
processing power for DVB-T and FM, two different sam-
pling rates are used. For FM the signal is directly sampled
so that the wanted signal between 88 and 108MHz ap-
pears at 11.231.2MHz after sampling. The DVB-T signal
is mixed down to an IF frequency of 155MHz and sampled
after IF ltering with a sample rate of 48 MHz. The signal
will nally appear at 11MHz.
Direct sampling of the FM signal requires that the ADC
be able to handle the whole dynamics of the FM band.
Slow fading of the input signal is adapted by digital pro-
grammable ampliers, whereas fast fading must be han-
dled in digital domain.
The main task of the input stages is to suppress out-of-
band interferers that may appear in the wanted signal
band after sampling. A four-stage lter based on low-cost
printed LC resonators is built up and realizes a suppres-
sion higher than 80 dB of out-of-band interferers. For FM
the highest signal levels can be expected in the wanted
signal band so that the ltering fullls the requirements.
The requirement for the ADC is a signal-to-noise ratio
of about 90 dB related to a signal bandwidth of 400kHz
and a spurious-free dynamic range (SFDR) of 95 dB. Par-
ticularly regarding the SFDR, care has to be taken when
designing the ADC driver stages.
For DVB-Treception, an analog downconverter stage is
implemented to achieve a xed intermediate frequency of
155 MHz. To avoid IF disturbance by the image frequency,
an adjustable analog input lter is used. It is based on LC
resonators with varactor diodes and can be controlled over
a digital to analog converter by the microcontroller. A
suppression of more than 80 dB of the image frequency
310 MHz above the used channel is realized. The frequen-
cy response is depicted in Fig. 5. After downconversion the
Figure 10. Concept of inductively coupled 901 tilted resonator (a)
and part of realized version with metallized via holes (b) in high
resistivity silicon.
Figure 11. Microstrip-coupled resonator. The length of each side
is 9.45mm.

reconfig.
frontend 1
reconfig.
frontend 2
ADC1
ADC2
clock
microcontroller
interfaces
F
P
G
A
D
S
P

C
l
u
s
t
e
r
Figure 12. Block diagram of a recongurable receiver for auto-
motive applications.
amplication can be directly set by the digital signal pro-
cessing unit to achieve an optimum amplitude at the input
of the analog to digital converter.
By using an IF bandwidth of about 10 MHz, all stan-
dards in this frequency range can be handled, so it is only
a question of digital signal processing to process them.
Because DVB-T uses an 8-channel OFDM modulation
scheme, the inuence of phase noise of the local oscillator
is critical [26,27]. At 100kHz offset a phase noise of
110 dBc is achieved, which conforms to simulations suf-
cient for DVB-T.
The complex downconversion and channel ltering is
realized in a FPGA (eld-programmable gated array).
Further signal processing tasks, including demodulation,
channel estimation, interleaving, and decoding, are
implemented partly in the FPGA and in an array of DSPs.
6. A MILLIMETER-WAVE SAMPLING HEAD
The basic components of a millimeter-wave sampling head
are shown in Fig. 13 [28]. The analog RF input signal
passes the high-Q MEMS AAF and is then converted to a
low impedance level (o10 O). The impedance transformer
and ampliers can be realized in commercial SiGe tech-
nology for operation frequencies up to 24 GHz [29]. As a
switch stage, a silicon millimeter wave integrated circuit
(SIMMWIC) Schottky diode quadruple bridge is used [30].
The Schottky diodes have a cutoff frequency of 1.5THz.
The impedance transformer II is again realized in SiGe
technology and has to provide a high input impedance
level.
7. CONCLUSIONS
Two kinds of digital receivers are discussed in this article:
the direct bandpass-limited sampling receiver and the six-
port conguration. The sampling architecture poses the
highest requirements for the analog-to-digital converter.
Bandpass limitation can be used to relax these require-
ments or to trade against higher resolution of the con-
verter. The bandpass lter makes this approach basically
a narrowband solution. The six-port digital receiver is
based on a scalar measurement technique that uses mul-
tiple detection channels at different points during the
period of a signal, which consists of the signal to be re-
ceived and a reference signal of the same frequency. As
this is an additive rather than a multiplicative mixing
process, the system dynamics of a six-port receiver are in-
ferior to, for example, those of a heterodyne or direct con-
version receiver. The six-port receiver, however, is
basically broadband-capable and does not have to cope
with image rejection problems.
Direct narrowband sampling of millimeter-wave sig-
nals becomes feasible with silicon-based high-speed inte-
gration technologies and high-Q MEMS approaches. If a
sampling and clock frequency of 10.66 GHz is applied to a
carrier frequency of 24 GHz and a signal with 250MHz
bandwidth is used a resolution of 3 bits is obtained if noise
shaping is not applied.
Acknowledgments
The authors acknowledge the contributions of Torsten
Mack, Dr. Karl Strohm, and Dr. Franz Schmu ckle.
BIBLIOGRAPHY
1. H. Schnepp, Th. Mu ller, and J.-F. Luy, Channel diversity in a
software autoradio, Proc. IEEE Vehic. Technol. Conf., Rhodos,
Greece, May 2001, pp. 190194.
2. J. Lindner, Advanced modelling of wireless digital transmis-
sion: From scalars to vectors, Proc. Frequenz 55, 2001, pp.
296300.
3. R. H. Walden, Analog to digital converter survey and analysis,
IEEE J. Select. Areas Commun. 17(4):539550 (April 1999).
4. J. H. Reed, Software Radio: A Modern Approach to Radio
Engineering, Prentice-Hall, Englewood Cliffs, NJ, 2002.
5. http://www.analog.com:AD6645 datasheet.
6. J. Tsui, Digital Techniques for Wideband Receivers, Artech
House, Norwood, MA, 1995.
7. http://www.analog.com:AD8345 datasheet.
8. T. Mu ller, J. Hao, and J.-F. Luy, A fully software congurable
receiver for automotive applications, Proc. Software Dened
Radio Technical Conf., San Diego, Nov. 11, 2002.
9. T. Mu ller, S. Herter, and J.-F. Luy, Clock generator phase
noise in RF sampling receivers, Proc. 2nd Karlsruhe Work-
shop on SDR, Karlsruhe, Germany, March 2003.
10. G. F. Engen and C. A. Hoer, Application of an arbitrary 6-port
junction to power measurement problems, IEEE Trans.
Instrum. Meas. IM-21:470474 (Nov. 1972).
In
Clock
Clock
R
out
=10
R
in
=200
Hold
capacitance
Out
Figure 13. High-frequency sample-and-hold
circuit.
11. G. F. Engen, The six-port reectometer: An alternative net-
work analyzer, IEEETrans. Microwave Theory Tech. MTT-25:
10751080 (Dec. 1977).
12. B. Huyart et al., A MMIC six-port reectometer, Proc. 35th
Midwest Symp. Circuits and Systems, Washington, DC, Aug.
12, 1992, pp. 14851488.
13. Ji. Li, R. G. Bosisio, and K. Wu, A six-port direct digital re-
ceiver, IEEE MTT Symp. Digest, San Diego, May 1994, Vol. 3,
pp. 16591662.
14. M. Ratni, D. Krupezevic, Z. Wang, and J.-U. Ju rgensen,
Broadband digital direct down conversion receiver suitable
for SDR, Proc. PIMRC, 2002.
15. J.-U. Ju rgensen, D. Krupezevic, M. Ratni, and Z. Wang, Base-
band aspects of a direct conversion receiver concept utilizing
ve-port technology, Proc. 2nd Karlsruhe Workshop on Soft-
ware Radio, 2002.
16. J. Hyyryla inen, L. Bogod, S. Kangasmaa, H.-O. Scheck,
and T. Yla murto, Six-port direct conversion receiver, Proc.
27th European Microwave Conf. and Exhibition, Jerusalem,
1997.
17. C. G. Migelez, B. Huyart, E. Bergeault, and L. P. Jallet, A new
automobile radar based on the six-port phase/frequency dis-
criminator, IEEE Trans. Vehic. Technol. 49(4):14161423
(July 2000).
18. A. Stelzer, C. G. Diskus, R. Weigel, and H. W. Thim, Using a
six-port device in an FM-CW radar, Proc. 8th Int. Symp. Mi-
crowave and Optical Technology ISMOT 2001, Montreal,
Quebec, Canada, June 1923, 2001, Polytechnic Internation-
al Press, pp. 323326.
19. B. Huyart, J.-J. Laurin, R. G. Bosisio, and D. Roscoe, A direc-
tion-nding antenna system using an integrated six-port cir-
cuit, IEEE Trans. Anten. Propag. 43(12):15081512 (Dec.
1995).
20. T. Mack, A. Honold, and J.-F. Luy, An extremely
broadband software congurable six-port receiver platform,
European Microwave Week 2003, Oct. 610, Munich, pp.
623626.
21. G. F. Engen, Calibrating the six-port reectometer by means
of sliding terminations, IEEE Trans. Microwave Theory Tech-
nol. MTT-26:951957 (1978).
22. R. G. Bosisio, Six-port receivers; Proc. Workshop on Software
Radios, RAWCON 2003, Boston.
23. K. M. Strohm, F.-J. Schmu ckle, B. Schauwecker, J.-F. Luy,
and W. Heinrich, Silicon micromachined RF MEMS resona-
tors, Proc. IEEE MTT-S, Seattle, 2002, pp. 12091212.
24. K. M. Strohm, F.-J. Schmu ckle, O. Yaglioglu, J.-F. Luy, and
W. Heinrich, 3D silicon micromachined resonators, Proc.
IEEE MTT-S, Philadelphia, 2003, pp. 18011804.
25. T. Mu ller, M. Sliskovic, J.-F. Luy, and H. Schopp, An automo-
tive broadband sampling receiver for analog and digital
broadcasting standards, Proc. 2003 Software Dened Radio
Technical Conf., Orlando, FL, Nov. 1719, 2003.
26. C. Muschallik, Inuence of RFoscillators on an OFDM signal,
IEEE Trans. Consum. Electron. 31(1):3 (Aug. 1995).
27. M. S. EI-Tanany and Y. Wu, L. Ha zy, Analytical modeling and
simulation of phase noise interference in OFDM-based digital
television terrestrial broadcasting systems, IEEE Trans.
Broadcast. 47(1) (March 2001).
28. E. Biebl, Sampling architectures, Proc. IEEE MTT-S, Phila-
delphia, 2003, pp. 10551058.
29. http://www.atmel.com/atmel/products/prod45.htm
30. Datasheet, SIMMWIC Schottky diodes, DaimlerChrysler,
2002.
DIGITAL RADIO
VALENTINE A. AALO
Florida Atlantic University
Boca Raton, Florida
This article presents an overview of digital radiocommu-
nications with emphasis on wireless personal communi-
cations. Digital radio consists of two main processes:
speech coding (and video coding for future systems) and
modulation. In speech coding, the continuous analog voice
signal is converted into a discrete digital form. In commu-
nication systems that provide digital voice services, it is
necessary to encode the analog speech into a digital
stream for transmission over the channel and at the re-
ceiver, to reconstruct the signal with acceptable delity.
Modulation, on the other hand, is the process of impress-
ing the discrete digital signal onto a radio signal, by vary-
ing some parameter or parameters in combination
(usually the amplitude, frequency, or phase) of the radio
signal. Digital transmission systems exist in a wide vari-
ety of forms, mainly determined by the nature of the chan-
nel over which the system operates. A process which is
common to many digital transmission systems is regener-
ation. Typically, the received signal at the input of a re-
peater (receiver) arrives attenuated and dispersed by the
channel and corrupted by noise. The rst operation of the
receiving repeater, therefore, is to preamplify and shape
the weakened signal to a level and form from which a re-
liable threshold detection may be performed. In a wireless
channel, the transmitted signal also undergoes multipath
fading and shadowing due to obstructions in the trans-
mission path. To combat the effect of multipath fading, the
receiver may employ diversity reception, channel coding,
and equalization.
The main objective of wireless personal communication
is to allow the user access to the capabilities of the global
network at any time and without regards to location and
mobility. There has been a tremendous growth in the
number of subscribers to wireless services in the last de-
cade. As a result, it became very difcult to operate and
maintain the quality of the original rst-generation ana-
log cellular systems. As the number of subscribers in-
creases in these systems, call quality diminishes. To
handle the increasing trafc, the cellular concept in which
identical frequencies at noninterfering distances are re-
used, was adopted giving rise to more interference in the
system. Digital radio, used extensively in second-genera-
tion and the evolving third-generation systems, consider-
ably improves the quality of the cellular system and
enhances the services available to the mobile subscribers.
Some advantages of digital cellular radio include [1,2]:
1. More efcient use of the limited radiofrequency (RF)
spectrum to improve system capacity
2. Improving the voice quality beyond what is possible
with analog cellular systems, especially maintaining
voice quality in heavy trafc conditions and use of
voice activity detection to save power and increase
throughput
DIGITAL RADIO 1021
3. Providing support for a wider array of services and
features
4. Simplifying the task of frequency planning, operat-
ing, and maintaining the cellular system
5. Providing a smooth transition from the analog sys-
tems to digital radio systems.
Cordless and cellular telephony have gained wide-
spread user acceptance. Cordless telephones are low-pow-
er, low-range phones that enable the user to move around
the home or ofce and still place and receive phone calls.
The handsets typically operate within 100 m of the users
base station, which is connected to the public switched
telephone network (PSTN). Cordless telephony has
evolved from being a simple home appliance to sophisti-
cated systems in applications for universal low-power
cordless and telepoint systems aimed at pedestrians, and
cordless private branch exchange (PBX). Digital cordless
telephone systems (such as CT2, DECT, PHS) are opti-
mized for low-complexity equipment and high-quality
speech in a quasistatic (with respect to user mobility) en-
vironment. They can support higher data rates and more
sophisticated applications.
On the other hand, digital cellular radio, originally tar-
geted at vehicular users in urban areas, was developed to
maximize bandwidth efciency and frequency reuse in a
macro-cellular, high-speed fading environment. The rst-
generation cellular systems are analog systems based
mostly on frequency-division multiple access (FDMA)
and have very limited capacity and poor to average speech
quality. The second-generation cellular systems (e.g., IS-
54, GSM, IS-95) are all-digital systems and use more ef-
cient multiple access techniques to share the available
spectrum among the users. Although personal communi-
cations services (PCS) may be regarded as a third-gener-
ation system, its implementation uses modied versions of
the cellular protocols used in the second-generation sys-
tems. While the rst-generation analog and second-gen-
eration digital systems are designed to support voice
communication with limited data communication capabil-
ities, third-generation systems will focus on providing a
wide variety of services which include wireless extensions
of integrated services digital network (ISDN) and broad-
band asynchronous transfer mode (ATM). These systems
will concentrate on service quality, system capacity, and
terminal and personal mobility issues. They will use a va-
riety of cell structures ranging from the conventional mac-
rocells to microcells for urban areas, picocells for indoor
applications, and supercells for satellite-based systems.
1. DIGITAL CELLULAR RADIO
1.1. The Cellular Concept
The continuous increase in the demand for telecommuni-
cations services and systems has resulted in spectral
congestion. Thus the original one cell system with a
high-power transmitter to provide good coverage in a
wide service area has quickly become limited in capacity.
The cellular concept in which the geographic service area
is divided into small regions (called cells), each of which is
served by a low-power transmitter (base station), has pro-
vided a solution to the spectral congestion problem. Adja-
cent cells are assigned different frequency channels so
that the frequencies can be reused throughout the cover-
age area, leading to a considerable improvement in system
capacity. Figure 1 shows a seven-cell frequency reuse pat-
tern. Although frequency reuse increases system capacity,
it also increases the amount of adjacent-channel and co-
channel interference present in the system. Therefore, for
the efcient utilization of the radio spectrum, the frequency
allocation scheme must be optimized to increase capacity
and minimize interference. As the trafc grows in the cov-
erage area, new cells and channels can be added to the
system. The hexagonal cell structure is usually employed
in the design of a cellular system; however, in practice the
actual cell coverage area (footprint) is irregular and de-
pends on the terrain and multipath characteristics of the
radio channel [3]. As such, there may be some regions
within the coverage region where there is exceptionally
high likelihood of deep signal fades (called blindspots) due
to shadowing, tunnels, and other obstructions in the sig-
nal path. Blindspots can be overcome by using repeaters
that receive the signals selectively in one direction, am-
plify them, and then retransmit exact replicas of the sig-
nals in the required direction. Three types of repeaters
may be identied, namely: broadband repeaters, frequency-
band-selective repeaters, and channel-selective repeaters
[4]. Signal degradation at the cell boundaries is handled
by handoff operation.
1.2. Handoff
Handoff encompasses a set of functions that are supported
between a mobile user and the cellular network that al-
lows the user to move from one cell to another or one radio
channel to another, within or between cells, while a call is
in progress. When a mobile user is engaged in a call, it will
frequently move out of the coverage area of the base sta-
tion with which it is communicating, and unless the call is
passed to another base station, the call will be lost. Thus,
the system continuously monitors the quality of the
1
2
3
4
5
6
7
1
2
3
4
5
6
7
1
2
3
4
5
6
7
Figure 1. Frequency reuse.
1022 DIGITAL RADIO
signals received from the active mobile users. When the
signal falls below a preset threshold the system checks
whether another base station can receive the mobile user
at a better signal level, and if so, the mobile user is com-
manded by a control signal to switch to the new frequency
(corresponding to the new base station). Although the pro-
cess of measuring the signal quality, channel allocation,
and handoff may take a few seconds, there should not
be any noticeable break in conversation of the mobile
user. Effective and reliable handoff is essential in control-
ling cochannel interference, especially for microcellular
systems.
1.3. The Cellular Network
The structure of all cellular networks is essentially simi-
lar. Being complete telephone networks, they have dedi-
cated exchanges within the interconnected system with
base stations connected to the exchanges. In practice,
however, there are different ways of planning the network
depending on the capacity requirement, implementation
cost, and capabilities of the chosen manufacturers equip-
ment. As shown in Fig. 2, a typical cellular system consists
of mobile stations, base stations, and a mobile switching
center (MSC). Each mobile unit contains a transceiver,
antenna, and control circuitry and may be handheld or
mounted in a vehicle. The communication between the
mobile station and the base station is dened by a stan-
dard air interface. Each cell in the coverage area is served
with one or more base stations that are connected to the
MSC. The MSC, in turn, coordinates the activities of the
base stations and connects them via microwave or ber
links to the public switched telephone network (PSTN)
which forms the global telecommunications network that
connects wireline telephone switching centers (called cen-
tral ofces) to MSCs throughout the world.
1.4. Multiple-Access Schemes
Multiple-access techniques are utilized to accommodate
multiple users in the available channels. The multiple ac-
cess scheme controls the allocation of the channel capacity
to the users. The allocation scheme is chosen to maximize
the spectral efciency and minimize transmission delay in
the system. Other desirable properties of the multiple-ac-
cess scheme include fairness of the allocation process, sta-
bility of the system, robustness with respect to equipment
failure and changing conditions of other users in the sys-
tem, and exibility in allowing the integration of voice and
data trafc. In addition, in a wireless mobile communica-
tion environment, the hidden terminal problem and near
far effect, the effects of multipath fading and shadowing,
as well as the effects of cochannel and adjacent channel
interference must also be considered. The commonly used
multiple-access schemes in cellular systems are frequen-
cy-division multiple access (FDMA), time-division multi-
ple access (TDMA), and code-division multiple access
(CDMA). In FDMA, the available bandwidth is divided
into channels that are assigned to the users on demand
and each user uses the channel for the entire duration of
the transmission. Frequency guardbands are provided at
the edges of the band to minimize crosstalk. Although
FDMA has relatively low complexity, requires few over-
head bits (for synchronization and framing), and usually
does not require equalization since the effect of intersym-
bol interference (ISI) is minimal, it is wasteful because
only a xed number of users (channels) can be supported
and when a channel is not being used it remains idle.
Furthermore, FDMA requires very tight ltering to min-
imize adjacent channel interference and intermodulation.
In TDMA systems, the usage of each frequency channel is
partitioned into multiple time slots, and each user is as-
signed a specific frequencytimeslot combination. Thus, in
a given cell only a single mobile user uses the entire fre-
quency at any given time. Although TDMA has the disad-
vantages of requiring synchronization (as well as
overhead for guard timeslots) and equalization and can
also be wasteful, it permits the use of exible bit rates and
may be used for bursty transmission to save power.
Another major advantage of TDMA (over FDMA) arises
from the fact that by transmitting and receiving in different
time slots it may be possible to eliminate the duplexer cir-
cuitry in the mobile unit, replacing it with a fast-switching
circuit that turns the transmitter and the receiver on and
off at the appropriate times, thus prolonging the battery
life of the handsets. Also, TDMA-based systems tend to be
more exible and more open to technological change.
Thus, with improvements in speech coding algorithms, a
TDMA channel is more easily recongurable to accept new
techniques supporting higher, lower, or variable bit rates,
without disrupting the frequency plan of the cellular
network. With CDMA (which uses direct-sequence
PTSN
Incoming
call
Interrogation
MCS1
HLR
MCS2
VLR
Reply
Call
routing
Interrogation
Reply
Registration
Figure 2. The cellular network.
DIGITAL RADIO 1023
spreading), a frequency channel is shared simultaneously
by multiple users in a given cell, and the signals are dis-
tinguished by spreading them with different codes. CDMA
has the advantage of offering multipath immunity and in-
terference rejection and provides a graceful performance
degradation as the number of users in the system increas-
es. However, CDMA is susceptible to the nearfar problem
and requires power control [3]. Figure 3 illustrates the
three commonly used multiple access schemes in cellular
networks.
In general, digital systems can support more users per
base station per megahertz of spectrum, allowing wireless
system operators to provide service in high-density areas
more economically. The use of TDMA or CDMA digital ar-
chitecture provides the additional advantage of sharing
the radio hardware in the base station among the multiple
users. It offers exibility for mixing voice/data communi-
cation and the support of new services. A potential for
further capacity increases is also possible with the use of
reduced-rate speech coders. Furthermore, reduced RF
transmit power (increasing battery life of handsets) and
the use of encryption for communication privacy, are pos-
sible. It offers a more natural integration with the evolving
digital wireline network and reduced system complexity
(mobile-assisted handoffs, fewer radio transceivers, etc.).
While the second-generation cellular systems are based on
digital transmission, some of them are designed to coexist
with their analog counterparts, while all the evolving
third-generation cellular and PCS systems use digital
transmission.
In wireless communications systems, it is usually de-
sirable for the subscriber to simultaneously send informa-
tion to the base station while receiving information from
the base station. The process whereby the subscriber can
transmit and receive information simultaneously is
known as duplexing. The two most commonly used du-
plexing techniques are frequency-division duplexing
(FDD) and time-division duplexing (TDD). In FDD, the
forward link (base-to-mobile station) and the reverse link
(mobile-to-base station) transmissions are done simulta-
neously on different frequency channels. In this case, a
device called a duplexer is used inside each subscriber unit
and base station to allow the simultaneous signal trans-
mission and reception on the duplex channel pair. In TDD
systems, the same frequency band is used in both the for-
ward and the reverse links but it is required that the
transmissions in different directions occur in different
time slots.
A performance measure that is commonly used to char-
acterize a digital radio system is the spectral efciency.
The spectral efciency of a digital radio system E
s
is
dened as [5]
E
s
Z
c
WCA
c
erlangs=MHz=km
2
1
where Z
c
is the carried trafc per channel in erlang,
W is the channel bandwidth in MHz, A
c
is the cell area
in km
2
, and C is the cluster size (number of cells in a reuse
cluster).
1.5. Speech Coding
In wireless systems that provide digital voice services,
there is the need to encode the analog speech signal into a
digital stream for transmission over the channel (air in-
terface). At the receiver, the signal is reconstructed with
acceptable delity. There are several major parameters to
consider in choosing a speech coding scheme for wireless
application. These include the transmitted bit rate (kbps),
the delivered speech quality, robustness to transmission
errors, and complexity of implementation of the chosen
scheme. Available speech coding techniques may be clas-
sied into three main categories: waveform coding, model-
based coding, and hybrid techniques. Waveform coding
techniques are usually the simplest to implement and
their implementation may be done in either the time do-
main or the frequency domain. At the transmitter, the
analog speech is sampled, quantized, and encoded into
digital stream for transmission. At the receiver, a decoder
reconstructs the original speech signal. The coderdecoder
combination is commonly referred to as a codec. Waveform
Channel 1
Channel 2
Channel 3
Channel N
Frequency
Time
Code
Channel 2
Channel 3
Channel N
Channel 1
Frequency
Time
Code
C
h
a
n
n
e
l

1
C
h
a
n
n
e
l

2
C
h
a
n
n
e
l

3
C
h
a
n
n
e
l

N
Frequency
Time
Code
(a) (b) (c)
Figure 3. Multiple-access techniques: (a) FDMA; (b) TDMA; (c) CDMA.
1024 DIGITAL RADIO
speech coding techniques implemented in the time domain
include pulse-code modulation (PCM), differential PCM
(DPCM), delta modulation (DM), and adaptive predictive
coding (APC). One form of delta modulation known as
digitally variable slope delta modulation (DVSDM) is used
in the second-generation UK cordless telephone system
(CT2) because of its implementational simplicity and low
cost, but to some extent sacricing voice quality. When
waveform coding is implemented in frequency-domain,
the speech signal is ltered into contiguous, nonoverlap-
ping sub-bands encoded independently using time-domain
techniques. Examples include subband coding (SBC) and
adaptive transform coding (ATC) schemes. In model-based
speech coding techniques, signal processing algorithms
are used to extract and transmit certain parameters from
the analog speech waveform that correspond to the actual
time-varying parameters of the speech production mech-
anism in the human vocal tract (modeled as an electric
lter). Thus the algorithms, which are usually called
vocoders, attempt to describe the speech production mech-
anism in terms of a few independent parameters used as
the information-bearing signals. At the receiver, the re-
ceived parameters are decoded and used to control a
speech synthesizer that is an algorithmic representation
of the speech generation model, thereby approximating
the original speech signal. Vocoders are medium complex-
ity systems and operate at low bit rates. Their poor speech
quality may be attributed to the oversimplied source
models used and the assumption that the source and the
lter are independent. The poor and synthetic quality of
speech vocoders has led to the speech coding approach
known as residual excitation. In this approach, the speech
is synthesized, but a small part is transmitted as a coded
waveform part and another as a vocoded part; hence the
name hybrid. The penalty is the higher bit rate of trans-
mission required, but now a very much improved speech
quality is realized. Examples of model-based speech cod-
ing schemes that are often used in wireless applications
include linear predictive coder (LPC) which usually re-
quires some form of error correction coding when used in
wireless channels, long term predictors that include mul-
tipulse excitation (MPE-LPT) and regular pulse excitation
(RPE-LPT), code-excited linear predictive (CELP) coder
and quadrature code-excited linear predictive (QCELP)
speech coding schemes, and residual-excited linear
predictive coding (RELPC). Table 1 summarizes the
speech coding specication for some cordless and cellular
systems [3].
1.6. Modulation Techniques
Many modern mobile communications systems use digital
modulation techniques as they offer the advantages of
greater noise immunity and robustness to channel impair-
ments, easier multiplexing and integration of different
types of information (e.g., voice, data, and video), and
greater security over their analog counterparts. Digital
modulation also allows the use of source coding, error con-
trol coding, encryption, and equalization to improve the
performance of the overall system. New advances in very-
large-scale integration (VLSI) and digital signal process-
ing (DSP) technology have also improved the effectiveness
of the digital modulation schemes used in wireless com-
munications systems. The choice of a digital modulation
scheme for use in a wireless environment is inuenced by
several factors. A desirable modulation scheme provides
low bit error rates at low received signal-to-noise ratios,
occupies a minimum bandwidth, performs well in a mul-
tipath fading environment, and is easy and cost-effective
to implement. Since the existing modulation techniques do
not simultaneously possess all of these qualities, tradeoffs
are often made in a modulation scheme for a particular
wireless application. The performance of a digital modu-
lation scheme is usually measured in terms of power ef-
ciency and bandwidth efciency. Power efciency describes
the ability of the modulation scheme to preserve the del-
ity of the digital message at low power levels. It is usually
expressed as the amount of power (usually given as the
ratio of the signal energy per bit to noise power spectral
density, E
b
/N
0
) required at the input of the receiver to
achieve a specied probability of error. Bandwidth ef-
ciency describes the ability of the modulation scheme to
accommodate data within a limited bandwidth. It is de-
ned as the ratio of the throughput data rate per hertz in a
given bandwidth and directly reects how effectively the
allocated bandwidth is utilized. While power and band-
width considerations are very important, other factors
must also be considered in choosing a modulation scheme
for a wireless application. For example, cellular systems
are usually interference-limited and the performance of
the modulation scheme in an interference environment is
also important. The sensitivity of the receiver to timing
Table 1. Speech Coder Used in Various Digital Radio Systems
Standard Service Type Speech Coder Type Used Bit Rate (kbps)
GSM Cellular RPE-LTP 13
CD-900 Cellular SBC 16
USDC (IS-54) Cellular VSELP 8
IS-95 Cellular CELP 1.2, 2.4, 4.8, 9.6
IS-95 PCS PCS CELP 14.4
PDC Cellular VSELP 4.5, 6.7, 11.2
CT2 Cordless ADPCM 32
DECT Cordless ADPCM 32
PHS Cordless ADPCM 32
DCS-1800 PCS RPE-LTP 13
PACS PCS ADCM 32
DIGITAL RADIO 1025
jitters caused by the time-varying channel is an important
consideration. For personal communications systems that
serve a large number of subscribers in the service area, a
modulation scheme that allows a simple yet efcient de-
tector is desirable in order to minimize the cost and com-
plexity of the subscriber receiver unit. Digital modulation
schemes may be classied as linear or nonlinear.
In a linear modulation scheme, the transmitted signal
may be expressed as
ft RefAf t expjo
c
tg
Af
a
t coso
c
t f
b
t sino
c
t
2
where A is the amplitude, o
c
2pf
c
is the angular carrier
frequency, f
c
is the carrier frequency, and f(t) f
a
(t) jf
b
(t)
is the complex envelope representation of the modulating
signal. Linear modulation techniques have good band-
width efciency and are attractive for wireless communi-
cation systems where there is increasing demand to
accommodate more and more subscribers within a limit-
ed bandwidth. However, linear modulation schemes are
usually transmitted using RF ampliers, which have poor
power efciency. The use of power efcient ampliers leads
to the regeneration of ltered sidelobes causing severe ad-
jacent-channel interference. A number of techniques have
been developed to handle this problem in practice. Exam-
ples of linear modulation schemes that are commonly used
in practical mobile applications include binary phase shift
keying (BPSK), quadrature phase shift keying (QPSK),
offset QPSK, and p/4 QPSK. Coherent detection uses the
carrier frequency and phase information to provide opti-
mum detection. It is well known that when coherent de-
tection is used at the receiver the bit error rate
performance of BPSK in an additive white Gaussian noise
(AWGN) channel is given by
P
E;BPSK
g Q
2g
b
_
3
where
Qt
1
2p
p
_
1
t
expx
2
=2 dx
g
b
g
2
(E
b
/N
0
) and g is the random attenuation factor due
to channel fading. The average bit error rate is then ob-
tained by averaging Eq. (3) over the probability density
function of g
2
, p(g):
_
1
0
P
E;BPSK
gpg dg 4
In a Rayleigh fading environment, we have
pg
1
g
e
g=" gg
5
and the average bit error rate can be shown to be given by
"
PP
E;BPSK
1
2
1
" gg
1 " gg
_ _
6
where " gg Eg
b
E
b
=N
0
Eg
2
is the average signal-to-
noise ratio. In practice, the carrier phase information
may not be known precisely or may be random (due to
channel uctuations). In such cases, differentially coher-
ent detection may be employed. The probability of error
for differential PSK in AWGN is
P
E;DPSK
1
2
expg
b
7
while the average bit error rate can be shown to be given by
"
PP
E;DPSK
1
21 " gg
8
The bit error rate performance of QPSK is similar to that
of BPSK. However, QPSK comprises two orthogonal BPSK
signals and thus has the advantage of providing twice the
spectral efciency of BPSK with the same energy. As such,
twice as much data can be transmitted in the same band-
width. QPSK ideally has a constant amplitude property
but occasional p-radian phase shifts momentarily cause
the signal envelope of ltered QPSK to pass through zero.
This causes serious problems when QPSK is used in mo-
bile/satellite applications with nonlinear amplication be-
cause of interference from the sidelobes. OQPSK is a
modied version of QPSK in which p-radian phase shifts
do not occur. Although OQPSK has the same signal con-
stellation and, therefore, same bit error performance as
QPSK, it is not susceptible to adjacent channel interfer-
ence caused by the regeneration of sidelobes. In p/4 QPSK
the maximum phase transition of 1351 is a compromise
between the 1801 phase transition of QPSK and 901 for
OQPSK. Noncoherent detection can be used to demodu-
late a p/4 QPSK signal and can provide better perfor-
mance in a multipath fading environment than OQPSK.
Thus p/4 QPSK has the same performance (bit error rate
and spectral efciency) as QPSK but has less amplitude
uctuation. p/4 QPSK has been adopted in the North
American digital standard (IS-54), the Japanese digital
cellular and the Trans European Trunked Radio [3].
Nonlinear modulation techniques have constant enve-
lope so that power efcient class C ampliers can be used
without introducing degradation in the spectrum occupied
by the transmitted signal, but they usually occupy larger
bandwidths than do linear modulation schemes. Examples
of constant-envelope modulation schemes that are fre-
quently used in mobile communication applications are
frequency shift keying (FSK), minimum shift keying
(MSK), and Gaussian minimum shift keying (GMSK). In
binary FSK, the transmitted signal of bit duration T
b
may
be expressed as
f
k
t
2E
b
T
b
_
cos o
k
t; 0 t T
b
9
1026 DIGITAL RADIO
when the binary digit k(k 0, 1) is transmitted, where
o
0
o
1
2np/T
b
and n is an integer. The average proba-
bility of error of the optimum coherent detector in a Ray-
leigh fading channel corrupted by AWGN can be shown to
be given by
"
P P
E;CFSK
1
2
1
" gg
2 " gg
_ _
10
while it is
P
E;NCFSK
1
2" gg
11
for a noncoherent detector. The phase information in FSK
signal is not properly utilized at the receiver except for
synchronization. MSK is a special case of continuous
phase FSK (CPFSK) in which the peak frequency devia-
tion is half the bit rate. Thus, MSK may be regarded as a
special case of OQPSK with the rectangular pulseshaping
replaced by half-sinusoidal pulse shaping. Thus, like
OQPSK, MSK has a constant envelope but the phase tran-
sitions are continuous. Also, an MSK signal (like an FSK
signal) can be demodulated coherently or noncoherently.
Finally, GMSK may be regarded as a special case of MSK
in which the sinusoidal weighting function is replaced by a
Gaussian shaped pulse. GMSK also has constant envelope
and excellent spectral efciency. The average bit error rate
of coherently demodulated MSK (and GMSK) in Rayleigh
fading channel may be shown to be given by [3]
P
E;MSK
1
2
1
Z" gg
1 Z" gg
_ _
12
where Z is a constant that depends on the product of the
demodulator 3 dB bandwidth and the symbol duration
(BT) and is given by
Z
0:68; for GMSK BT0:25
0:85; for MSK BT1
_
13
GMSK has been adopted for use in GSM, DECT, and U.S.
cellular packet data (CDPD).
2. CHANNEL PROPAGATION
The performance of a wireless communications system is
limited by the nature of the mobile radio channel. The
transmission path between the transmitter and the re-
ceiver usually varies as a result of obstructions from build-
ings, mountains, and foliage and also as a result of
variations in the atmosphere. Thus, electromagnetic
wave propagation is usually inuenced by the mecha-
nisms of reection, refraction, and scattering. Multiple
reections cause the transmitted signal to travel along
different paths of varying lengths and attenuations, and
the interactions of these waves at the receiver location
causes multipath fading. Notwithstanding the multipath
fading, the long-term average strength of the received sig-
nal decreases as the separation between the transmitter
and the receiver increases.
2.1. Channel Propagation Models
Propagation models to characterize the mobile channel
can usually be classied into two groups, depending on
whether they focus on predicting the average received sig-
nal strength (large-scale models) or the variability of the
received signal (small-scale models) at a given distance
from the transmitter. Large-scale propagation models
based on measurements of actual channels indicate that
the mobile radio channel may be characterized by the
fourth-power loss model [6]:
P
r
P
t
V
d
a
14
where Pr is the average received power, Pt is the average
transmitted power, d is the distance between the mobile
and the base station, a is the exponent of power attenu-
ation (aE4, for macrocells), and V is a random variable
whose decibel value can be modeled by a zero-mean
Gaussian variable (i.e., V is lognormal) with standard de-
viation in the range of 612dB. The propagation in an ur-
ban microcellular channel with a line-of-sight (LoS) may
be characterized by the following dual-slope path loss
model [6]
P
r
P
t
V
d
a
1
1
d
d
0
_ _
a
2
15
where a1 and a2 are the attenuation exponents, and
d
0
4h
t
h
r
/l, where l is the transmission wavelength and
h
t
and h
r
are the transmitting and receiving antenna
heights, respectively. The decibel standard deviation of
the lognormal random variable V is now on the order of
3 dB. Measurements in several urban environments indi-
cate that for small transmitting and receiving antenna
heights (h
t
h
r
E3.7m), a
1
E2.2 and a
2
E3.3. For medium
antenna heights (about 8.5 m), a
1
E2.2 and a
2
E3.4 and for
large antenna heights, a
1
E2.1 and a
2
E4.2 [3].
On the other hand, small-scale multipath propagation
causes rapid uctuations in signal strength (fading) over
small distances or time intervals. Small-scale propagation
is also inuenced by Doppler shifts caused by relative mo-
tion between the transmitter and the receiver, and by the
time dispersion caused by the multipath propagation de-
lays. Time dispersion due to multipath propagation causes
the transmitted signal to undergo either at or frequency-
selective fading. The Rayleigh and Rice probability density
functions are commonly used to model envelope uctua-
tions in a at-fading channel when there is no direct
transmission path between the transmitter and the re-
ceiver and when a LoS component is present, respectively.
The Nakagami m distribution is a more general model
that has been shown to provide a better match to envelope
measurements in different mobile radio environments
than the Rayleigh and Rice distributions. The Rayleigh
DIGITAL RADIO 1027
distribution is a special case of the Nakagami distribution,
while the Rice distribution can be approximated by the
Nakagami distribution for a large range of mean signal-to-
noise ratio values.
2.2. Combating Multipath Fading
Two major causes of performance degradation in wireless
systems are multipath fading and shadowing. There are
three ways to combat the effects of fading in these systems:
diversity reception, channel coding, and equalization. Di-
versity reception techniques are used extensively on mul-
tipath fading radio channels to reduce the effect of fading
on system performance, including both xed and mobile
terminals. In diversity reception, several replicas of the
transmitted signal, each carrying the same information
and undergoing independent fading, are combined at the
receiver. The diversity may be obtained in time, frequency,
or space. There are several ways that the receiver may
combine the received diversity signals for optimum per-
formance. Three of the commonly used linear combining
schemes are maximal-ratio combining (MRC), equal-gain
combining (EGC), and selection diversity combining
(SDC). Microscopic diversity reduces the effect of instan-
taneous short-term (or small-scale) fading by combining
several uncorrelated signals received at the radio port us-
ing any of the combining methods. Macroscopic diversity
mitigates the effect of long-term (or large-scale) shadowing
by using several geographically distributed base stations
to serve each cell. The base station with the largest aver-
age local mean signal power is usually selected [5,7].
Figure 4 shows the diversity gain obtained by using
MRC microscopic diversity reception in detecting BPSK
signals in a Rayleigh fading channel. We observe from the
gure that with a coherent detector at the receiver, in or-
der to obtain an error rate of 10
3
, the receiver requires a
signal-to-noise ratio (SNR) of 24 dB when there is no di-
versity (L1), but only 11dB with dual-branch diversity
(L2) and 4dB with fourth-order (L4) diversity. The
performance of DPSK is about 3dB inferior to that of co-
herent PSK. In Fig. 5, the effect of macroscopic selection
diversity is shown, with the lognormal shadowing as-
sumed to have a decibel standard deviation of 6dB. At a
bit error rate of 10
3
and with dual order microscopic di-
versity, macroscopic diversity of order 3 (N3) provides
about 5dB improvement over the system with no macro-
scopic diversity (N1).
In channel coding schemes, extra bits (with no mes-
sage) are added to the information bitstream before being
transmitted over the channel. At the receiver, the added
redundant bits are used to detect/correct errors that may
have occurred in the bitstream. Channel error control
techniques used in wireless channels may be classied
into three groups, namely: error detection coding (the
most commonly used error detection scheme is the cyclic
redundancy check (CRC) codes because they are very easy
to implement using shift registers), forward error correc-
tion (FEC) coding (using block or convolutional codes), and
automatic repeat request (ARQ) transmission protocols.
The amount of diversity gain introduced by channel cod-
ing depends on the minimum distance of the code. The
addition (and interleaving) of redundant bits into the data
bitstreams in the channel coding process gives rise to time
or frequency diversity and improves the resistance of the
system to multipath fading.
The time-variant multipath channel after exhibits bur-
sty error characteristics. By the process of interleaving,
the bursty channel can be transformed into a channel
having independent errors by spreading the coded data
over several timeslots. Interleaving is used extensively in
the second-generation digital cellular systems.
A very serious problem in high-data-rate transmission
systems is intersymbol interference (ISI) caused by fre-
quency-selective multipath fading. In this case, increasing
the transmission power worsens the problem because the
interference power increases. Signal processing tech-
niques (known as equalization) may be used to minimize
DPSK
CBPSK
L = 1
L = 2
L = 4
SNR (dB)
5 10 15 20 25 30
P
e
10
0
10
1
10
2
10
3
10
4
10
5
10
6
10
7
10
8
Figure 4. Average bit error rate for a BPSK system with
L-branch microscopic MRC diversity in a Rayleigh fading channel.
DPSK
CBPSK
L = 2
N = 1
N = 3
SNR (dB)
P
e
5 10 15 20 25 30
= 6dB
10
0
10
1
10
2
10
3
10
4
10
5
10
6
Figure 5. Average bit error rate for a BPSK system with mac-
roscopic selection diversity (and dual-branch microscopic MRC
diversity) in a shadowed Rayleigh fading channel.
1028 DIGITAL RADIO
the effect of ISI. In wireless applications, adaptive equal-
ization is used since the mobile channel is random and
time-varying. The operation of an equalizer usually in-
volves the transmission of a known, xed-length training
sequence to set the parameters of the equalizer at the re-
ceiver. New algorithms, called blind equalization, which
do not require training sequences, are currently under re-
search. During data transmission, the adaptive equalizer
uses recursive algorithms to evaluate the channel and es-
timate the lter coefcients that are used to compensate
for the channel distortions. Adaptive equalizers can be
classied as linear or nonlinear depending on how the
equalizer output is used for subsequent control (feedback)
of the equalizer.
3. WIRELESS TRANSCEIVER STRUCTURE
The complexity of radiocommunication systems is increas-
ing significantly with the application of more sophisticat-
ed multiple access and digital modulation techniques that
are necessary in order to accommodate the tremendous
growth in the number of subscribers of wireless commu-
nication services. Advances in wireless technology and
new applications for wireless systems and services have
given rise to a variety of portable voice, data, and mess-
aging systems. The development of low-rate digital speech
coding techniques and the continuous increase in the de-
vice density of integrated circuits have led to completely
digital second-generation cellular systems. Also, the
evolving third-generation cordless, cellular, and PCS
systems are all expected to be fully digital. Digital
signal processing (DSP) techniques traditionally used for
speech and channel codecs are presently being used
extensively for advanced digital communications trans-
ceiver design. In addition to speech and channel codecs,
these techniques are also being used for detection and
demodulation, equalization, frequency synthesis, and
channel ltering.
3.1. Radio Receiver Principles
A considerable amount of computing resources are neces-
sary to achieve the performance desired for personal com-
munication systems and the required power needed to
drive the constituent units of the system may be prohib-
itive for portable applications. Thus, the key requirements
for wireless portable terminals are performance, cost,
power consumption, and size. Low power consumption
may be achieved through technology and system-level
tradeoffs. The receiver power is consumed by the RF com-
ponents, baseband DSP, digital application-specific inte-
grated circuits (ASICs), and mixed signal devices. At the
system level, power consumption may be optimized by
proper choice of system operations such as time-division
multiplexing and voice-activity detection. For example,
many digital processors feature powerdown modes that
allow turning off peripheral and certain computational
units. One drawback of such method, however, is that it
does not always allow for fast rampups. In a wireless
environment, the receiver may have to process very low
desired signal levels in the presence of large levels of
unwanted signals. Therefore, the architecture used
in the receiver front end must satisfy requirements
such as sensitivity, dynamic range, selectivity, and
manufacturability. Radio receiver principles that may
be used include the superheterodyne principle, digital
receivers with DSP techniques, and direct-conversion
receivers [8].
In a direct-conversion receiver, the received RF signal
is ltered in a duplexer, and passed through a low-noise
amplier followed by a bandpass lter. The output of the
lter is then split and, along with two local oscillators in
phase quadrature at the carrier frequency, are fed to a
quadrature mixer. The outputs of the quadrature mixer
are then passed through a lowpass lter followed by an-
alogdigital (A/D) conversion to produce the in-phase (I)
and quadrature (Q) samples. Since the receiver processes
the full RF spectrum at baseband, this architecture re-
quires high dynamic range, high sensitivity, low noise, as
well as proper amplitude and phase balancing between
the I and Q branches. The main advantage of the direct-
conversion architecture is its simplicity as it has a low
component count. It also has a wide tuning range and high
selectivity. However, a number of challenges are present
in its realization. For example, a high-gain low-noise mix-
er is necessary to combat the 1/f amplier noise at base-
band as well as a technique to cancel the associated large
DC-offset. The direct-conversion receiver architecture is
used in a number of cellular and wireless products. For
example, a radio receiver that incorporates direct-conver-
sion into an integrated circuit in a way that avoids the
need for discrete intermediate frequency lters and is par-
ticularly suitable for use in wireless devices has been pro-
posed [9]. Also, FSK paging receivers at 450 MHz and
930 MHz as well as some 900MHz wireless LAN products
are available.
3.2. Miniaturization
Wireless personal communication devices such as pagers
and cellular and cordless telephones are becoming more
compact and more light weight as a result of improve-
ments in device-mounting technology and development of
different kinds of devices. By using advanced very large-
scale integration (VLSI) technology, the implementation of
complex algorithms is economically feasible. The use of
complementary metal-oxide semiconductor (CMOS) de-
vice scaling technology has facilitated the employment of
denser and faster memory chips as well as digital micro-
processors. Rapid advances in solid-state integrated cir-
cuit technology have fueled the growth of commercial
wireless communication systems with the desire to pro-
duce high-performance, low-power, small-size, low-cost,
and high-efciency devices. The increasing use of inte-
grated circuits in radio designs has resulted in significant
improvements in the reliability and performance of the
digital receivers. Rapid advances in packaging technology
resulting in compact designs of wireless terminals as well
as considerable drops in manufacturing costs resulting
from improved procedures in assembly and testing have
also increased the deployment of these devices.
DIGITAL RADIO 1029
3.3. Design Tools
The lifecycles of many cellular and cordless products are
very short. As such, in order to compete successfully, com-
panies must turn system concepts into silicon VLSI very
rapidly. High-quality computer-aided design tools are
therefore very important for efcient design, simulation,
and realization of the systems. The design methodology
used in these tools must be chosen to allow different levels
of abstraction at different points in the timescale. The
algorithms used in these design tools are usually compre-
hensively specied with block diagrams that are specied
hierarchically, with each block representing a signal pro-
cessing operation. The blocks are usually parameterized
so that automatic evaluation of the different simulations
can be done based on the block diagrams. Depending on
the digital communication system involved, multirate and
variable-rate processing may be supported. Two simula-
tion approaches that are available in practical design tools
are the data-ow-oriented approach (e.g., COSSAP from
Synopsys) and the time-driven approach (e.g., SPW from
Candence) [10]. In addition to the algorithmic simulations,
the architecture of the digital communication system also
needs to be simulated. The simulation of the architecture
may be software-based or hardware-based.
4. OVERVIEW OF DIGITAL RADIO SYSTEMS
4.1. Digital Cellular Systems
Digitization allows the use of TDMA and CDMA over
FDMA as multiple-access alternatives. The North Amer-
ican Digital Cellular systems have evolved into two Inter-
im Standards, one based on TDMA (IS-54) and the other
based on CDMA (IS-95). The Global System for Mobile
communication (GSM) as well as the Japanese Pacific
Digital Cellular (PDC) system (which is very similar to
the IS-54 system) are also based on TDMA while the
Broadband-CDMA (IS-665) system is a specialized
CDMA system [3,5,11].
4.1.1. North American Digital Cellular Systems. The de-
velopment of a digital cellular standard in North America
came as a result of tremendous increase in the demand for
cellular services. The capacity of the rst-generation an-
alog advanced mobile phone system (AMPS) was limited,
and there was no new spectrum available to meet the in-
creased demand. Therefore, the objective of the second-
generation systems was not only to increase the capacity
of the existing spectrum but also to provide additional
services. The Cellular Telecommunications Industry As-
sociation (CTIA) which consists mainly of cellular service
providers and the Telecommunication Industry Associa-
tion (TIA) consisting of equipment manufacturers estab-
lished a technical committee to develop a digital standard.
Finally, in 1989, the industry adopted the dual-mode
transmission standard, which is referred to as the Elec-
tronics Industry Association Interim Standard 54.
4.1.2. TDMA System (IS-54). This is an all-digital sec-
ond-generation cellular system that was designed to
coexist with and eventually replace the rst-generation
analog cellular system. On the forward link, the spectrum
allocation for IS-54 is 824849MHz while on the reverse
link it is 869894MHz. The modulation scheme used is
differential quadrature phase shift keying (DQPSK) with
p/4 radians phase shift between successive symbols, to
reduce amplitude uctuations in the signal envelope.
However, as it is a linear modulation scheme, it has poor
power efciency resulting in larger size and weight of the
handset. Each TDMA frame has 6 timeslots of 324 bits
each, with a framelength of 40 ms, giving a bit rate of
48.6 kbps. Since the channel spacing is 30kHz, the result-
ing bandwidth efciency of 1.62 bps/Hz is relatively high.
The speech coder is VSELP operating at 7.95kbps and
produces a speech frame every 20 ms (or 159 bits every
second). Of these, the leading 77 bits of each frame are
protected with error control coding and the remaining 82
bits are unprotected, resulting in 260 channel bits per
frame. Thus the full-rate coder results in a transmitted
data rate of 13 kbps.
4.1.3. CDMA System (IS-95). This system is based on
direct-sequence CDMA (DSCDMA) and was proposed by
Qualcomm in 1989 and adopted in 1993. IS-95 was also
designed to be compatible with the AMPS. The spectrum
allocation for IS-95 on the forward link is 824849MHz,
while it is 869894MHz on the reverse link. With an
allowable bandwidth of 1.25 MHz, it uses a direct-se-
quence spread-spectrum signal with chip rate
1.228 Mbps. The speech coder used is QCELP with vari-
able rates (ranging from 1200 to 9600 bps) determined
by the accompanying voice activity detector. Block
interleaving with duration 20 ms provides time diversity
while the wide bandwidth allows for frequency diversity
and multipath (rake) diversity, making the system
robust to multipath fading. Different modulation and
spreading techniques are employed on the forward and
reverse links. On the forward link, BPSK modulation is
used with QPSK spreading. For a single user, either form
of modulation yields the same performance, but in a mul-
tiple-access environment the use of QPSK spreading ran-
domizes the phase of the desired user relative to the other
users in the system giving rise to much less phase degra-
dation for the desired user. Although the 64 64 Hada-
mard matrix used may allow 64 users in a cell, only 61
Walsh codes are available since the remaining codes are
reserved for the pilot, synchronization, and paging chan-
nels. Also, on the forward channel, many user signals are
multiplexed and transmitted to multiple users, allowing a
common pilot signal to be inserted for all the users. There-
fore, coherent demodulation is possible on the forward
link. On the reverse link, on the other hand, since the us-
ers operate asynchronously and are power controlled, no
pilot signal is transmitted by the mobile users. Therefore,
noncoherent M-ary (M64) orthogonal modulation/
demodulation, which is power efcient, is employed on
the reverse link.
IS-95 is modied in the following ways in order to sup-
port higher data rates for better speech quality at PCS
1030 DIGITAL RADIO
frequencies:
1. On the reverse link, the convolutional code rate is
changed from
1
3
to
1
2
.
2. On the forward link, the convolutional code rate is
changed from
1
2
to
3
4
.
3. The standard QCELP speech coder is replaced by
QCELP13, which also has variable rate and is de-
signed to provide improvements in spectral quanti-
zation, voice activity detection, and pitch prediction.
It operates in several modes (which includes
QCELP).
4. The data rate is changed from (1200, 2400, 4800,
9600bps) to (1800, 3600, 7200, 14400 bps).
4.1.4. Broadband CDMA System (IS-665). The wideband
CDMA standard supports several bandwidths (5, 10, or
15 MHz) at PCS frequencies. The forward link is similar to
that of IS-95 with a few exceptions. There is a pilot signal,
a synchronization signal, and up to seven paging signals
and several trafc signals are supported as options. Also,
unlike IS-95, where the chip rate is 1.228 Mbps, in IS-665,
several chip rates (of 4.096, 8.192, and 12.288Mbps) may
be used. On the reverse link, the mobile users transmit
pilot signals to the base station. Therefore, coherent de-
tection (of the QPSK modulated signal) is possible. Both
the CDMA (IS-95) and the Broadband CDMA (IS-665)
systems are synchronized by the Global Position Satellite
(GPS) time. The speech coding scheme used is ADPCM.
4.1.5. European Digital CellularGSM (DCS 1800). The
GSM standard was developed as a joint initiative by mem-
bers of the Conference of European Posts and Telecom-
munications Administration (CEPT) with the initial
objective of building a unied pan-European network, giv-
ing the subscribers a uniform service and easy roaming
throughout all of Europe. The GSM technical standard
makes full use of currently available technology, incorpo-
rating features such as low bit rate speech, convolutional
channel coding with bit interleaving, and frequency hop-
ping. Services supported by GSM may be classied into
three types: telephone services, data services, and supple-
mentary ISDN services. The spectrum allocation for GSM
at 900MHz is categorized into the standard or the extend-
ed GSM band while the allocation for GSM at 1800MHz is
referred to as Digital Cellular System 1800 (DCS 1800)
band. The frequency assignments for these bands are as
follows: for forward link, 935960 MHz standard GSM,
925960MHz extended GSM, and 18051880 MHz DCS
1800; for reverse link, 890915MHz standard GSM, 880
915MHz extended GSM, and 17101785 MHz DCS 1800.
With a spacing of 200kHz, the standard GSM has 124
channels, the extended GSM has 174 channels, and DCS
1800 has 374 channels. Each GSM channel supports eight
simultaneous users using TDMA of framelength 4.615 ms.
The modulation is GMSK with BT0.3 and slow frequen-
cy-hopping every frame at 217 hops per second is used to
provide additional protection against frequency-selective
fading and cochannel interference. Interleaving is also
used to minimize the effect of deep fades. The speech coder
is a regular pulse-excited linear predictive coder (RPE-
LPC) with long-term prediction with voice detection capa-
bility (voice activity detection factor of 40%) and provides
a net bit rate of 13 kbps. It operates in discontinuous
transmission mode to prolong battery life. Presently,
GSM networks have been deployed in over 60 countries
in Europe, the Middle East, Asia, and Africa. In North
America, GSM is deployed as PCS 1900.
4.1.6. Japanese Personal Digital Cellular. Established in
1991, the Japanese Personal Digital Cellular (PDC) sys-
tem is very similar to the North American IS-54 system in
terms of their operational characteristics and in the re-
quirement that they replace an existing analog cellular
system. The frequency allocation for the PDC represents
the main difference between the two systems. PDC has
two small frequency bands in the 800/900 and the
1400 MHz band. On the forward link, the frequency as-
signments are 810826MHz and 14771501MHz, while
on the reverse link, they are 940956MHz and 1429
1453 MHz. With a channel spacing of 25 kHz to be com-
patible with the existing analog system, PDC uses TDMA
to multiplex three slots for three users in a 20-ms frame
onto a carrier. The modulation is p/4 DQPSK with a chan-
nel data rate of 422kbps, and the VSELP voice coder uses
error correction coding. Mobile-assisted handoff facilitates
the use of small cells, and with the use of space diversity,
reduces the required carrier-to-interference ratio. The sys-
tem provides high quality services, high security, and long
handset battery life.
4.2. Digital Cordless Telephony
Cordless telephones are low-power, low-range, full-duplex
communication systems that use radio to extend the hand-
set to a dedicated base station with a specific telephone
number that is connected to the public switched telephone
network. Cordless telephone systems provide the user
with limited mobility and it is seldom possible to main-
tain a call if the user travels outside the coverage range of
the base unit. In the rst-generation cordless telephone
systems, the handset typically operates with localized mo-
bility within a very limited range (on the order of 10 m) of
the base unit and is used in the home or in the ofce. They
use analog frequency modulation and operate mainly as
extension telephone to a transceiver connected to the pub-
lic wired network. Because of its analog nature and lim-
ited operating range, it has limited trafc-carrying
capacity, which in turn limits the full development poten-
tial of these systems. Second-generation cordless tele-
phone systems are based on digital transmission format
and provide wider coverage ranges, offer good speech qual-
ity, provide better security, are more resistant to interfer-
ence and noise, and use compact handsets with built-in
antenna [2].
4.2.1. Cordless TelephoneCT2. This is a second-gen-
eration cordless telephone standard introduced in Great
Britain in 1989 and designed for residential and ofce use.
It is also used to provide telepoint services. Telepoint is a
service that is provided to cordless handset owners from
DIGITAL RADIO 1031
cordless base stations located in public places, such as
railway stations and shopping centers. This is a basic
public communication service for the less migratory, more
localized sector of the traveling market and does not com-
pete directly with the wide roaming mobile cellular net-
work. Thus, the handset purchased for residential or ofce
use can also be used to access the telepoint service while
the user is in transit between the home and the ofce. In
CT2, speech waveforms are coded using ADPCM with a bit
rate of 32 kbps. Two-way full-duplex conversation is
achieved using time-division duplexing (TDD). The mod-
ulation used is Gaussian ltered FSK with bandwidthbit
period product BT0.3. A Canadian enhancement of CT2
is called CT2 and provides additional mobility manage-
ment functions.
4.2.2. Digital European Cordless Telecommunication
(DECT). DECT is a pan-European standard for cordless
telephone that was designed to provide cost-effective com-
munication to high user densities in picocells. Intended
applications of DECT include residential cordless telepho-
ny, telepoint services, and cordless PBX. Although DECT
is functionally closer to a cellular system that a standard
cordless telephone system, the interface of DECT to the
PSTN or ISDN network remains the same as for a corded
telephone. DECT uses TDMA with TDD and the base sta-
tion can support multiple handsets simultaneously with a
single transceiver. The modulation and speech coding
techniques used in DECT are similar to those in CT2.
4.2.3. Personal Handyphone System (PHS). PHS is a
Japanese air interface standard with the design objective
of providing service not only for home and ofce use but
also for public access capability. PHS uses TDMA and
TDD, with each TDMA frame of 5 ms duration. The speech
coding used is ADPCM with data rate of 32 kbps in con-
junction with CRC error detection (with no error correc-
tion) and the modulation used is p/4 DQPSK. Since PHS
uses dynamic channel assignment the base stations can
allocate channels based on the signal strength seen at
both the base station and the portable, and handoffs are
supported only at walking speeds as the system is de-
signed for microcell/indoor PCS use.
4.2.4. Personal Access Communication Systems (PACS).
PACS is a third-generation personal communications sys-
tem designed to support voice, data, and video images for
low-speed portable applications in microcell/indoor envi-
ronments. The PACS interface provides wireless connec-
tivity to a local exchange carrier (LEC) and it uses TDMA,
with frequency-division duplexing (FDD). The modulation
used is p/4 QPSK, with coherent demodulation, which
provides substantially better performance than do other
digital cordless telephone systems with discriminator-
based receivers. Two-branch polarization diversity with
feedback at both the base station, and the handset gives
an improvement that approaches a four-branch diversity
reception system. The subscriber unit uses adaptive power
control to minimize battery drains during transmission
and to reduce cochannel interference on the reverse link.
4.3. Paging Systems
A traditional paging system is a one-way, wireless com-
munication device that sends brief messages (usually a
numeric message, an alphanumeric message, or a voice
message) to notify a subscriber of the need to call a par-
ticular telephone number or to receive further instruction
from another location. There are two types of paging sys-
tems: the radio common carrier (or a subscriber system)
and the private paging systems. The subscriber paging
system is a licensed, public paging company providing
paging services to the public and the coverage area may be
local, statewide, nationwide, or international. The private
paging system involves a customer-owned transmission
system and paging receivers for private paging use. When
multiple transmitters broadcast a page (known as simul-
cast), the subscribers can roam from the home area to
anywhere the paging system is networked. The traditional
definition of paging has evolved from the one-way com-
munication device to a two-way device that sends and re-
ceives data with services including customized response
functions, connection to online information services,
e-mail messaging, etc. A number of signaling standards
for paging systems have also evolved. Examples include
POCSAG, which is a one-way paging system; ERMES,
which is the European paging system; Motorolas FLEX
family of paging products; and AT&Ts P-act, which is a
narrowband PCS paging system [12].
5. PERSONAL COMMUNICATION SERVICES
Wireless personal communication services (PCS) is a con-
cept that extends wireless communications beyond the
limitations of the current cellular system to provide users
with the ability to communicate with anyone, anywhere,
and at any time. A number of characteristics are generally
associated with PCS in order to provide reliable service on
demand anywhere, any time. The handsets must be por-
table and easy to use with a long battery life, each user
must have a single personalized number that can be
reached anywhere, and the system must provide an indi-
vidualized feature prole that follows the user and pro-
vides service at any location. PCS must support various
user mobility patterns and provide wide user roaming
ability to provide for universal accessibility. Consequently,
different cell sizes must be used depending on the type of
application and user density. In general, four types of cells
can be distinguished for PCS: the picocell (for low-power
indoor applications), the microcell (for low-power indoor or
outdoor pedestrian applications in high-population-densi-
ty areas), the macrocell (for high-power vehicular appli-
cations), and supermacro cells (for use with satellite
systems). Radio systems for PCS must have a variety of
operating power levels, and the users should be able to use
the service in diverse environments with a wide variation
in the radio propagation properties. The system must
allow easy integration of the wireless system with the
1032 DIGITAL RADIO
wireline system and ubiquitous deployment of the radio
systems.
The services that can be offered by PCS and cellular are
identical, except that the operating frequencies are differ-
ent. The subscriber is indifferent to the frequency band as
long as the services are not affected. Thus, the main forces
that inuence the PCS and the cellular industries are
similar. Some of these forces are the regulators, PCS op-
erators, equipment vendors, subscribers, and competing
products. The service area in the United States and its
territories are divided into 51 major trading areas (MTAs)
and 493 basic trading areas (BTAs) according to the Rand
McNally 1992 Commercial Atlas and Marketing Guide,
123rd ed. Based on FCC regulation, each area in the Unit-
ed States can be served by at least six PCS operators, in
addition to the existing two cellular operator and one spe-
cialized mobile radio (SMR) service operators [12].
One of the most important dening elements of PCS is
the FCCs allocation of 120MHz of spectrum around the
1200 MHz frequency band for licensed operation and an-
other 20 MHz for unlicensed operation, resulting in a total
of 140MHz for PCS. This is about three times the 50 MHz
spectrum currently used by the cellular system, indicating
the resolve to make PCS more widely available. Figure 6
shows the frequency allocation for PCS.
No specific technology has been mandated by the FCC
for the PCS market. As such, a number of competing tech-
nologies have been chosen by the major PCS carriers.
Presently, seven PCS standards can be identied. These
include GSM (PCS 1900), a derivative of the GSM/DCS
1800 standard; CDMA, a modication of IS-95; DAMPS;
PACS, air interface for pedestrian applications; CDMA/
TDMA, a composite hybrid that uses TDMA within cells
and CDMA between cells; DECT-based TDMA; and broad-
band CDMA [12]. Of these, the four most widely deployed
systems at the present time are GSM, CDMA, TDMA, and
PACS.
BIBLIOGRAPHY
1. T. B. Bursh et al., Digital radio for mobile applications, AT&T
Tech. J. 72:1926 (1993).
2. J. E. Padgett, C. G. Gunther, and T. Hattori, Overview of
wireless personal communications, IEEE Commun. Mag.
33:2841 (1995).
3. T. S. Rappaport, Wireless Communications: Principles &
Practice, Prentice-Hall, Upper Saddle River, NJ, 1996.
4. R. C. Macario, Cellular Radio: Principles and Design, 2nd ed.,
5. V. K. Garg and J. E. Wilkes, Wireless and Personal Commu-
nications Systems, Prentice-Hall, Upper Saddle River, NJ,
1996.
6. M. V. Clarke, V. Erceg, and L. J. Greenstein, Reuse efciency
in urban microcellular networks, IEEE Trans. Vehic. Technol.
46:279288 (1997).
7. A. Abu-Dayya and N. C. Beaulieu, Micro- and macrodiversity
NCFSK (DPSK) on shadowed Nakagami-fading channels,
IEEE Trans. Commun. 42:26932702 (1994).
8. H. Meyr and R. Subramanian, Advanced digital receiver prin-
ciples and technologies for PCS, IEEE Commun. Mag. 33:
6878 (1995).
9. T. Okanobu, D. Yamazaki, and C. Nishi, A new radio receiver
system for personal communications, IEEE Trans. Consum.
Electron. 41:795803 (1995).
10. K. S. Shanmugam, Simulation and implementation tools
for signal processing and communication systems, IEEE
Commun. Mag. 32:3640 (1994).
11. A. Fukasawa et al., Wideband CDMA system for personal
radio communications, IEEECommun. Mag. 34:116123 (1996).
12. R. Kuruppillai, M. Dontamsetti, and F. J. Cosentino, Wireless
PCS, McGraw-Hill, New York, 1997.
DIODES
KONSTANTINOS MISIAKOS
NCSR Demokritos
Athens, Greece
The word diode originates from the Greek word DiodoB,
meaning passage or way through. In electronics terminol-
ogy, in fact, diode refers to a two-terminal device that al-
lows current to ow in one direction while it blocks the
ow of current in the opposite direction. Such devices
usually employ semiconductor junctions or metalsemi-
conductor junctions. There are also diodes made of vacu-
um tubes or metalpurely ionic crystal contacts. This
article deals with semiconductor p-n-junction diodes be-
cause they are the most widely used in practice because of
their versatility, reproducibility, stability, and compatibil-
ity with integrated circuit technology. Additionally, an
A
15
MTA
B
15
MTA
1850 1900 1950
Frequency in MHz
1990
C
15
MTA
A D B E F C
Licensed handset transmit Licensed base transmit Unlicensed
A
s
y
n
c
h
r
o
n
o
u
s
(
p
a
c
k
e
t
-
s
w
i
t
c
h
e
d
)
I
s
o
c
h
r
o
n
o
u
s
(
c
i
r
c
u
i
t
-
s
w
i
t
c
h
e
d
)
D
5
B
T
A
E
5
B
T
A
F
5
B
T
A
Figure 6. The US PCS frequency plan.
DIODES 1033
insight into the operation of the p-n-junction diode is the
basis for understanding the device physics of other semi-
conductor devices, the majority of which use the p-n-junc-
tion as the building block. The semiconductor of our choice
will be silicon because almost all diodes, discrete or inte-
grated, are made of this element. Extensions to other
semiconductors will be made to generalize theoretical
results or to set limits to the validity of certain equations.
Figure 1a shows the electrical symbol of a diode. The
arrow-type symbol indicates the conduction direction. For
a diode to conduct an appreciable electric current, the
voltage on the left side of the symbol must be a little high-
er than the voltage on the right side. If this polarity is re-
versed, the current drops to negligible values even for a
large bias. The two previous polarity modes are known as
forward bias and reverse bias, respectively. In Fig. 1b the
very basic material structure of a diode is shown. The
starting material is a high-purity silicon crystal, the prop-
erties of which are properly modied by selectively intro-
ducing dopants (elements) from either the third or the
fth column of the periodic table. The third-column ele-
ments, when introduced into the silicon lattice, behave as
acceptors: they trap electrons from the valence band,
thereby creating positively charged holes in the valence
band and negatively charged immobile acceptor ions. The
acceptor-doped part of the diode is called the p side. On the
other hand, fth-column elements behave as donors; they
give up their fth electron, creating a population of con-
duction band electrons and positively charged immobile
donor ions. The donor-doped part of the device is the n
side. The introduction of acceptor and donor dopants into
silicon creates the two polarity sides of the diode, as shown
in Fig. 1a. Schematically speaking, when applying a for-
ward bias, the higher voltage on the p side makes the
electron and hole gases move into each other. Thus, an
electric current is created through electron-hole pair re-
combination. On the contrary, a lower voltage on the p side
moves the charge carriers away, thus preventing recom-
bination and eliminating the current. In terms of dopants,
the previous account of how the p-n diode is formed also
holds for germanium diodes, which also is a fourth-column
elemental semiconductor. For compound semiconductors
(e.g., GaAs, InP, CdTe), the chemical origin of donor and
acceptor dopants is more complex in relation to the semi-
conductor elements themselves.
In Fig. 2 the currentvoltage (IV) characteristic of an
ideal diode as well as a realistic one is shown. The ideal
diode would behave as a perfect switch when forward-
biased; unlimited current ows without any voltage drop
across the device. The same ideal diode would allow no
current in the reverse direction, no matter what the mag-
nitude of the reverse bias is. Now, a realistic semiconduc-
tor diode would exhibit a resistance to current ow in the
forward direction, whereas in reverse bias a small current
would always be present due to leakage mechanisms. The
disagreement between the ideal and the actual electrical
behavior is not restricted only to the static IV character-
istics shown in Fig. 2. It extends to the transient response
obtained when applying a time-dependent terminal exci-
tation. The response of a realistic diode cannot follow at
exactly the same speed as the terminal excitation of an
ideal diode would. When designing a diode to be used as a
switching device, care is taken to bring the device electri-
cal characteristics as close to the ideal ones as possible.
This is done by choosing both the geometric features and
the fabrication process steps in a way to suppress the par-
asitic components of the diode. As a result of the semicon-
ductor electronic band structure as well as technological
constraints, material limitations impose certain basic re-
strictions on the device performance and create the sub-
sequent deviation from the ideal performance. In the
following sections, these restrictions will be investigated,
and the deviation from the ideal performance will be
analyzed in terms of the basic device physics, material
constants, and geometry considerations.
Before considering the device physics of the diode, we
will briefly discuss the steps in the basic fabrication pro-
cess employed when making a silicon diode. These steps
determine its basic geometric and technological charac-
teristics, which in turn determine the device electrical be-
havior. Today, almost all silicon diodes are made through
the standard planar process of the silicon integrated cir-
cuit technology. A silicon wafer is rst oxidized at temper-
atures in the vicinity of 10001C. Such oxidation creates a
silicon dioxide (SiO
2
) cover layer with a thickness on the
order of a micrometer. This layer is used as a mask for the
subsequent technological steps. The SiO
2
lm is then pat-
terned by lithographic techniques and through etching,
which allows windows of exposed silicon to be opened.
Then, either by diffusion or by ion implantation, dopants
are introduced into the exposed areas. The dopants are of
a type opposite to the one already existing in the original
wafer. In this way, p-n junctions are created in the exposed
areas. In the rest of the wafer, the SiO
2
layer stops the ions
+
+
+ +
+
+ +

n p
(b)
(a)
Figure 1. Electrical symbol of a diode (a) and illustration of a
semiconductor p-n junction (b). In (b) the large circles with the
minus and the plus signs are the acceptor and the donor ions,
whereas the small circles are the holes (empty) and the electrons
(dark).
I
V
1 2
Figure 2. The currentvoltage characteristics of an ideal diode
(curve 1) and a realistic one (curve 2).
1034 DIODES
and prevents diffusion into the silicon bulk. On the back
surface, another diffusion or implantation of the same do-
pants as in the bulk is usually applied for reasons that will
become clear in the next sections. At the end, metal con-
tacts are evaporated on the front and the back. Lithogra-
phy, again, on the frontside denes the contacts of the
individual diodes. The metal contacts are required for the
diodes to interact with the external world in terms of ter-
minal excitation (voltage or current) and terminal re-
sponse (current or voltage, respectively). Similar
methods are used for germanium diodes, whereas the
compound semiconductor devices are usually made
by epitaxial growth on proper substrates and by in situ
doping.
1. FUNDAMENTALS OF P-N JUNCTIONS
The basic p-n-junction device physics was proposed by
Shockley [1]. He derived the current-voltage characteris-
tics, considering the electron and hole current continuity
equations and the relationship between the carrier quasi-
Fermi levels and the externally applied potentials. Here,
we rederive the general currentvoltage relation of a p-n
junction based on Shockleys classic work [1] and its later
extension [2].
1.1. Basic Equations and Assumptions
To formulate the electronhole transport in a semiconduc-
tor device mathematically, we can always start by ex-
pressing the carrier densities and currents in terms of the
carrier quasi-Fermi potentials under uniform tempera-
ture conditions:
J
n
em
n
nrF
n
1a
J
n
em
n
nEeD
n
rn 1b
nn
i
exp
eF
n
E
i
kT
_ _
1c
J
p
em
p
prF
p
2a
J
p
em
p
pE eD
p
rp 2b
pn
i
exp
eF
p
E
i
kT
_ _
2c
In Eqs. (1a) and (2a), J
n
and J
p
are the electric current
densities of electrons and holes, respectively. Equations
(1b) and (2b) express the currents in terms of drift and
diffusion, where m
n
, D
n
, n and n
p
, D
p
, p are the mobilities,
diffusivities, and volume densities of electrons and holes,
respectively. Finally, F
n
and F
p
are the electron and hole
quasi-Fermi potentials, E
i
is the intrinsic energy level, and
E the electric eld density. Figure 3 shows the energy
band diagram of a p-n junction under forward bias and
illustrates the space dependence of the quasi-Fermi po-
tentials, of the bottom of the conduction band E
c
, and of
the top of the valence band E
v
. Equations (1c) and (2c)
hold provided that the differences E
c
(eF
n
) and (eF
p
) E
v
are positive and at least several times the thermal energy
kT. Equations (1a) and (2a) are borrowed from thermody-
namics and hold provided that the bias is such that per-
turbations from equilibrium are small. Small here
implies that the energy distribution of electrons and holes
in the conduction and the valence band, respectively, con-
tinue (within a good approximation) to follow the Boltzm-
ann statistics. Additionally, we assume that the mean free
paths of the carriers are negligible compared to the phys-
ical dimensions of the device. Finally, Eqs. (1) and (2) hold
provided quantum-mechanical tunneling of carriers
across potential barriers is not important. Such a con-
straint is relaxed in the last section of nonconventional
transport diodes.
p Space
charge
region
n
Ohmic
contact
Ohmic
contact
0
eV
1 eF
p
eF
p
eF
n
eF
n
eV
2
E
v
C
1
C
j
C
2
E
c
x
Figure 3. The band diagram of a forward-
biased p-n junction. Boundaries C
1
and C
2
are
the ohmic contacts, whereas C
j
is the base-
injecting boundary.
DIODES 1035
The second set of equations to be considered is the elec-
tron and hole continuity equations:
@n
@t

1
e
r
.
J
n
Un; p G 3
@p
@t

1
e
r
.
J
p
Un; p G 4
where U is the electronhole net recombination rate either
by band-to-band transitions or through traps, whereas G
is the band-to-band generation rate resulting from ioniz-
ing radiation or impact ionization processes. For the sake
of simplicity, we assume that U and G are the same for
both carriers. Equations (3) and (4) are more general than
the previous ones because they do the bookkeeping by
equating the increase in the rate of carrier density to
minus the carrier losses resulting from carrier out-uxing
(rJ
p
/e and rJ
n
/e) and recombination.
Next is the Poisson equation, which relates the electric
eld to the charge density caused by both mobile and im-
mobile charges:
r
.
E
e
e
npN
D
N
A
5
where e is the semiconductor dielectric constant and N
D
and N
A
are the donor and acceptor densities, respectively.
The charge density resulting from donors and acceptors is
not carrier density dependent, unless the temperature
drops to the cryogenic region.
The nal equation to be considered is the one that
equates the electric eld to the gradient of the electrostat-
ic potential:
E
1
e
rE
c
1
e
rE
v
1
e
rE
i
6
Equation (6) implies that the electrostatic potential is de-
termined by conduction and the valence band edges be-
cause the carriers there have only potential energy. The
last equation assumes that the separation in the energy
scale of the three levels (E
c
, E
v
, and E
i
) is space-indepen-
dent. So, in Eq. (6), as well as in Eqs. (1b,c) and (2b,c), we
neglect band distortion resulting from heavy doping or
other effects (e.g., mechanical strain). This is discussed in
a later section on heavy doping effects.
1.1.1. Boundary Conditions. Equations (1)(6) form a
system of six relations with six unknown variables: F
n
,
F
p
, E
i
, J
n
, J
p
, and E. They apply, within the range of their
validity, to any semiconductor device. In this sense, any
semiconductor device understanding, design, operation,
and performance is based on this set of six equations.
What distinguishes a device of a particular kind is its
boundary conditions, as well as the doping and trap den-
sity and type.
1.1.2. Ohmic Contacts. For a diode, a two-terminal de-
vice, the boundary conditions necessarily include two
ohmic contacts that will supply the charge to be trans-
ported through the device. The voltage across and the
current through the two ohmic contacts are, interchange-
ably, the excitation or the response of the device. The
ohmic contacts are realized by depositing metals (e.g., Ti
or Al) on heavily doped regions of the semiconductor. An
ideal ohmic contact should, by definition, establish ther-
modynamic equilibrium between the metal and the semi-
conductor at all the contact points. In analytical terms,
this is expressed by equating the carrier quasi-Fermi po-
tentials to the metal Fermi potential
F
n
C
1
F
p
C
1
V
1
7a
F
n
C
2
F
p
C
2
V
2
7b
where V
1
and V
2
are the voltages of the metal contacts
(V
1
V
2
is the terminal voltage), whereas C
1
and C
2
are
the contact areas of the rst and the second ohmic contact,
respectively. The pinning of the Fermi potentials at the
externally applied voltages is illustrated in Fig. 3. If V
1
V
2
and G in Eqs. (3) and (4) were zero, the device would be
in equilibrium. Then the solution of the previous system of
six equations would be zero currents and equal and at
Fermi potentials throughout the diode. When an external
voltage is applied, the splitting of the Fermi potential val-
ues between the two ohmic contacts drives the device out
of equilibrium. Such a boundary value split enforces a
separation of the electron and hole quasi-Fermi potentials
through the device, as shown in Fig. 3. The separation of
the two potentials implies that the nonequilibrium condi-
tions mainly refer to the interaction between electrons
and holes. At any point, excluding ohmic contacts, elec-
trons are out of equilibrium with respect to holes because
the relaxation time of interband transitions (recombina-
tion and generation mechanisms) required to bring them
into equilibrium are too slow (milliseconds or microsec-
onds for germanium and silicon and nanoseconds for most
compound semiconductors). On the contrary, the intra-
band transitions resulting from scattering have short re-
laxation times (picoseconds) so that electrons or holes are
nearly in equilibrium within their band. This is required
for the carrier Fermi potentials to have a meaning, as
mentioned in the discussion following Eqs. (1) and (2).
1.1.3. SemiconductorInsulator Interfaces. The surface
that bounds the device includes, in addition to the ohmic
contact, the semiconductorvacuum or semiconductor
insulator interface. The exposed semiconductor surface
is usually covered by an insulating lm (SiO
2
in silicon) to
reduce recombination. If we assume that there is no in-
jection in the insulator, then at the interface the boundary
conditions for Eqs. (3) and (4) are
@n
s
@t

1
e
J
n
n
U
s
8
@p
s
@t

1
e
J
n
p
U
s
9
where the subscript s refers to surface densities and the
superscript n refers to the normal component looking
1036 DIODES
into the insulator. The boundary conditions for Eq. (5) are
dictated by the lows of electrostatics. The discontinuity of
the normal component of the dielectric displacement vec-
tor must be equal to the surface charge density, whereas
the tangential component of the electric eld must be
continuous.
Because the boundary conditions have been set, the
system of six equations [Eqs. (1)(6)] can be solved, in
principle. As it turns out, the solution of such a nonlinear
system of coupled equations can be found only numerically
even for one-dimensional p-n junctions with uniform ac-
ceptor and donor densities. To derive analytical approxi-
mations, we need to make certain assumptions regarding
the physical makeup of the device and the degree of bias.
These analytical expressions help predict the device
response under reasonable bias, whereas the appreciation
of their validity range provides an insight into the diode
device physics.
2. DOPING CARRIER PROFILES IN EQUILIBRIUM AND
THE QUASINEUTRAL APPROXIMATION
As mentioned earlier, a p-n-junction diode consists of an
acceptor-doped p region in contact with a donor-doped n
region. The two-dimensional area where the donor and
acceptor densities are equal is called the metallurgical
junction. Lets assume, for the moment, equilibrium con-
ditions. In such a case, the currents are zero and the qua-
si-Fermi potentials are equal and spatially independent,
F
n
F
p
F. Therefore, from Eqs. (1c) and (2c), pnn
2
i
,
where n
i
, is the intrinsic carrier density. Now, the six
equations reduce to the Poisson equation, which, with the
help of Eqs. (1c), (2c), and (6), takes the form
r
2
E
i
eE
e
2
e
n
i
exp
eF E
i
kT
_ _ _
n
i
exp
E
i
eF
kT
_ _
N
D
N
A
_
10
The last equation is known as the BoltzmannPoisson
equation. Approximate analytical solutions are possible
when the donor and acceptor densities are uniform in the
n and p regions, respectively. In this case, the eld is zero,
and the bands are at everywhere except at and near the
metallurgical junction. The nite eld region around the
metallurgical junction is called the space charge region,
whereas the zero-eld regions are called quasineutral re-
gions, for reasons to be explained shortly. In the n and p
quasineutral regions, electrons and holes are the majority
carriers, respectively. The majority-carrier densities are
equal to the respective doping densities. With reference to
the metallurgical junction, the space charge region ex-
tends W
A
and W
D
within the p and the n regions. At zero
bias, and in one dimension, an approximate solution of
Eq. (10) gives
W
A
2e
e
V
bi
N
D
N
A
N
A
N
D
11a
W
D
2e
e
V
bi
N
A
N
D
N
A
N
D
11b
V
bi
kT
e
ln
N
A
N
D
n
2
i
_ _
11c
In Eq. (11c) V
bi
is the zero bias electrical potential differ-
ence, or barrier, between the p and the n sides reected in
the level differences of the at bands of each side. Such a
barrier prevents majority carriers from diffusing into the
other side. These approximations result by assuming that
the electron and hole densities are zero in the space-
charge region. This is the depletion approximation, which
reduces Eq. (10) to a linear second-order differential equa-
tion with constant terms and coefcients.
The zero-eld condition for the rest of the n and p sides,
outside the space charge region, apparently justies the
term quasi-neutral regions. This term also applies when
the n and p regions have gradually changing dopant pro-
les in the sense that the net space charge is much less
than the majority-carrier charge. Here, by gradually
changing we mean that the doping prole N(x) in the qua-
sineutral region must be such that [3]
ekT
e
2
r
2
ln
Nx
n
i
_ _
{Nx 12
In such regions, the zero-bias majority-carrier density
continues to be nearly the same as the net dopant densi-
ty, but the electric eld is not zero as in the uniform doping
case.
3. FORWARD AND REVERSE BIAS CONDITIONS
The quasineutrality condition of the n and p regions is
preserved even under bias, but now the boundaries with
the space charge region move appropriately to accommo-
date the new boundary conditions. This neutrality condi-
tion can be expressed as
n % pN
D
N
A
13
Under a small forward bias, the applied voltage changes
the electric eld preferentially at the space charge region,
because it is the region with the fewest carriers, has the
highest resistance, and is in series with more conductive n
and p regions. The equilibrium barrier height V
bi
lowers
under forward bias, and the majority electrons overcoming
the repulsive eld diffuse from the n side to the p side,
whereas the holes are doing the opposite. The carrier quasi-
Fermi potentials are no longer equal, as shown in Fig. 3.
The diffusion process, through the space charge region and
inside the quasineutral regions, increases dramatically
DIODES 1037
the minority-carrier population on either side and gives
rise to an appreciable electric current.
For forward voltages, the degree of bias denes three
injection-level regimes distinguished by how the minority-
carrier density compares to the majority one in the qua-
sineutral regions. These regimes are the low-level, the
moderate-level, and the high-level injection condition. In
the low-level injection regime, the minority-carrier densi-
ty is well below the majority-carrier density, and the elec-
tric eld in the quasineutral regions is practically
unaffected by the bias. As a result, the applied voltage
drops across the space charge region and reduces the bar-
rier height from V
bi
to V
b
V
bi
V. Provided the depletion
approximation still holds, Eq. (11) still applies and V
bi
is
replaced by V
b
. In low-level injection, the majority-carrier
density is the same as at zero bias, as Eq. (13) points out,
and is nearly equal to the net doping density. In the high-
level injection regime, the minority-carrier injection is so
intense that the injected carriers have densities far ex-
ceeding the dopant densities. Now, both carrier densities
are about the same, np, to preserve neutrality in the
quasineutral region. In other words, there is no real dis-
tinction between minority and majority carriers in terms
of concentrations, but we obtain an electronhole plasma
having densities well above those of the dopant densities
instead. In the moderate injection, the minority-carrier
density approaches the order of magnitude of the major-
ity-carrier density causing the majority-carrier density to
start to increase, as Eq. (13) implies.
Under reverse bias, the built-in barrier increases in the
space charge region, the repulsive forces on the majority
carriers coming from the quasineutral regions increase,
and injection of minority carriers is not possible. The space
charge region is now totally depleted from both carriers,
and a small leakage current exists as a result of thermal
generation of electronhole pairs in the depletion region.
3.1. Recombination Currents in the Steady State
Here, we will introduce the base and emitter terms as well
as a general expression for the terminal current as the
sum of recombination components. Between the two qua-
sineutral regions, the emitter is the one that is heavily
doped, usually by diffusion or implantation, whereas the
base is more lightly doped and occupies most of the sub-
strate on which the device is made, at least in silicon. The
heavy doping of the emitter excludes the possibility of
moderate- or high-level injection conditions in this region.
At forward bias, majority carriers from the emitter diffuse
as minority carriers to the base where they recombine.
Simultaneously, recombination occurs in the emitter, be-
cause minority carriers are backinjected from the base, as
well as in the space charge region. At steady state, @n/@t
@p/@t 0, and in the dark G0. Now, the continuity equa-
tions [Eqs. (3) and (4)] become after volume integration:
I
D
I
e
I
b
I
SCR
14
where I
D
is the terminal current and I
e
, I
b
, and I
SCR
are
the net recombination currents in the emitter, the base,
and the space charge region, respectively. Equation (14)
expresses the total current as the sum of the recombina-
tion currents in the three regions of the device. Therefore,
excess carrier recombination along with diffusion are the
two basic transport mechanisms that determine the diode
current at a given bias. The carrier recombination occurs
either at the ohmic contacts, at the surface, or in the bulk.
The minority carriers that arrive at the ohmic contact are
supposed to recombine simultaneously there, to preserve
the boundary condition, Eq. (7). The bulk recombination
occurs either through traps or through band-to-band tran-
sitions. In terms of trap-mediated recombination,
the ShockleyReadHall mechanism [4,5] is the most
common:
U
SRH
pn n
2
i
N
t
1
s
p
v
th
nn
i
exp
E
t
E
i
kT
_ _ _ _
1
s
n
v
th
pn
i
exp
E
t
E
i
kT
_ _ _ _
15
In Eq. (15), N
t
is the trap density, s
n
and s
p
are the elec-
tron and hole capture cross sections, respectively, E
t
is the
trap energy level in the gap, and v
th
is the carrier thermal
velocity. The band-to-band recombination is discussed in
Section 6.
4. APPROXIMATE ANALYTICAL EXPRESSIONS IN THE
STEADY STATE
As mentioned previously, the set of Eqs. (1)(6) has no an-
alytical solution in the general case. Approximate closed-
form expressions, though, are possible when low-level in-
jection conditions dominate in the quasineutral base re-
gion. Without loss of generality, we assume that we are
dealing with a p-n diode with a heavily doped p emitter
and an n base. The steady-state situation results when a
terminal bias, say, a terminal voltage V, is steadily applied
on the terminals, and we wait long enough for the initial
transient to disappear. The steady-state version of the
continuity equations [Eqs. (3) and (4)] is simplied be-
cause the time derivatives are set equal to zero. First, we
will derive the expressions for the base current, and then
extensions will be made for the recombination current in
the emitter and the space charge regions. If low-level in-
jection conditions prevail in the base, then, to a good ap-
proximation, the original system of equations [Eqs. (1)(6)]
reduces to the minority-carrier equations [Eqs. (2) and (4)],
which are now decoupled from Eq. (5) (the Poisson equa-
tion). This decoupling results because, as mentioned ear-
lier, at low-level injection the electric eld is practically
bias-independent. Any small eld variations would affect
only the drift current of the majority carriers because of
their high density; the minority carriers would not be in-
uenced. That is why we focus on the minority-carrier
transport to exploit the Poisson equation decoupling. An-
other reason for focusing on the minority carriers is the
fact that the recombination in low-level injection, where
p{n, can always be written as a linear function of their
1038 DIODES
density:
Up p
0
=t 16
In Eq. (16), p
0
is the equilibrium carrier density, whereas
the variable t, the minority-carrier lifetime, is the inverse
of the derivative of the recombination rate with respect to
the minority-carrier density. In the case of Shockley
ReadHall recombination, t 1/s
p
n
th
N
t
. Therefore, from
Eqs. (2b), (4), and (16), we end up with
r
.
Em
p
p
0
D
p
rp
0
p
0
=t G0 17
where p
0
pp
0
is the excess minority-carrier density.
Because of the eld independence on p
0
, Eq. (17) is linear
and becomes homogenous if G0. In the latter case, the
solution is proportional to p
0
(C
j
), the excess minority-car-
rier density at the injecting boundary (Fig. 3).
4.1. Forward Bias and Low-Level Injection
Under forward bias, a basic assumption will be made. This
assumption allows the coupling of the minority-carrier
density to the externally applied terminal voltage; the
Fermi levels are at in the regions where the carriers are a
majority and also in the space charge region. Under this
condition and from Eqs. (1c), (2c), and (7), we obtain
pC
j
nC
j
n
2
i
expeV=kT 18
p
0
C
j

n
2
i
eV=kT 1
N
D
C
j
19
Equation (18) holds under any injection level, provided
that the at Fermi potential assumption holds, whereas
Eq. (19) for the excess minority-carrier density holds only
in low-level injection. The proportionality of the solution
with respect to p
0
(C
j
) forces all carrier densities and cur-
rents to become proportional to the term exp(eV/kT) 1.
Here, we note that the surface recombination is also a lin-
ear function of the excess minority-carrier density when
p5n. More analytically, Eq. (9) becomes
J
n
p
eS
p
p
0
20
where S
p
is the surface recombination velocity. Therefore,
the total base recombination current in Eq. (14) is propor-
tional to the term exp(eV/kT) 1. The same is true for the
quasineutral emitter recombination. Thus, Eq. (14) be-
comes
I
D
I
0e
I
0b
expeV=kT 1 I
SCR
21
where the preexponential factors I
0b
and I
0e
are the base
and emitter saturation currents, respectively. Equation
(19) points out that the saturation currents are propor-
tional to n
2
i
.
The space charge region recombination is a current
component that is hard to express in analytical terms.
This difculty arises because in this region the eld
depends on the bias and there is no such entity as a mi-
nority carrier. Consequently, the linearity conditions that
allowed us to derive Eq. (21) no longer hold. To derive an
approximate expression for the bias dependence of I
SCR
,
certain simplications must be made throughout the
space charge region regarding the integral of Eq. (15).
These simplications result in a bias dependence of the
form exp(eV/nkT) 1, where n, the ideality or slope factor,
takes values from 1 to 2 [2]. The specific value depends on
the trap position in the energy gap, the doping proles,
and the cross section for hole capture relative to the cross
section for electron capture. This range for n holds pro-
vided that the capture coefcients do not depend on the
electric eld. Now the preexponential factor I
0
SCR
is pro-
portional to n
i
. Finally, the expression for the forward cur-
rent of a diode in the base of which low-level injection
conditions prevail becomes
I
D
I
0e
I
0b
expeV=kT 1
I
0
SCR
expeV=nkT 1
22
For voltages higher than 3nkT/e, the unity can be dropped
from Eqs. (21) and (22). Because of a better slope factor,
the emitter and base recombination will dominate the di-
ode current for voltages above a certain level. Below this
level, the space charge region recombination must also be
considered.
Such trends are seen in Fig. 4. Curve 1 is the I V
characteristic of a diode with a base doping of
10
3
10
2
10
2
10
3
10
4
10
5
10
6
10
7
10
8
10
1
1
10
L

(
n
A
)

n = 2
n = 1
n = 1
1
3
2
n = 1.23
n = 3.5
1(V)
0.00 0.20 0.40 0.60 0.80 1.00
0.00 10.00 20.00 30.00 40.00
Forward bias (V)
Reverse bias (V)
Figure 4. The experimental I V characteristics of two different
diodes. Diode 1 has a base thickness of 250mm, a base-doping
density of 5.5 10
14
cm
3
, and an area of 2.92.9mm
2
. Diode 2
has a base thickness of 300mm, a base-doping density of 4.5
10
11
cm
3
, and an area of 5 5mm
2
. The base in both devices is of
n type. Plot 3 is the IV characteristic of diode 2 at 78K. The other
plots are at 300K. Curve 1( V) is the reverse bias characteristic,
with reversed sign, of diode 1 (top axis). The straight lines in
curves 1, 2, and 3 are the exponential exp(eV/nkT) 1 ts to the
experimental points. The slope factor n is also shown.
DIODES 1039
5.5 10
14
cm
3
and has an ideality factor of 1 in the bias
range from 0.2 up to 0.4 V. For lower voltages, the space
charge region recombination slightly increases the ide-
ality factor and makes the measured current deviate from
the exp(eV/kT) 1 dependence. The ideality factor also in-
creases for voltages above 0.4 V because of high-injection
effects, which are discussed in the next subsection. The
device of curve 2 has a very light doping density in the
base, 4.5 10
11
cm
3
and is driven in high injection at
even smaller bias. As discussed in the next subsection,
curve 2 exhibits unity slope factor even at very low
voltages.
Curve 3 shows what happens if the temperature is re-
duced to 78 K. The sharp reduction of the intrinsic carrier
density due to its Dexp( E
g
/2kT) dependence requires
much higher voltages to reach the same current as at
300K. In fact, to reach a current density of 10 mA/cm
2
, a
voltage in excess of 1 V is required. The reduction of n
i
re-
duces the recombination in the base and the emitter is
much faster than in the space charge region because the
proportionality constants are n
2
i
and n
i
, respectively.
Therefore, at low temperatures, the bias regions with
higher than 1 ideality factor are expected to be wider.
This is evidenced in curve 3 of Fig. 4, where the ideality
factor is 3.5 for voltages below 950mV. The increase of the
ideality factor above 2 is a result of the PooleFrenkel ef-
fect, which reduces the effective energy separation of the
traps from the bands [6]. The inuence of small values of
n
i
on the ideality factor is evident not only when the tem-
perature drops but also when the bandgap increases. In
several compound semiconductor devices, their large
bandgap, compared to 1.1eV of silicon, results in an in-
trinsic carrier density, which is several orders of magni-
tude smaller than the 10
10
cm
3
value for silicon at 300K
[7,8]. Consequently, their I V characteristics show slope
factors substantially larger than 1 for the entire range of
bias. On the contrary, germanium diodes have slope fac-
tors of 1 even at reduced temperatures because of the
smaller gap, 0.66 eV, of the semiconductor.
4.1.1. The One-Dimensional Case. Equation (22) holds
for any three-dimensional geometry and doping proles
because no assumption, except for low-level injection, was
made so far regarding doping proles and device topology.
If, however, we want to express in closed form the satu-
ration values of the emitter and base recombination cur-
rents, then one-dimensional devices with uniform doping
proles must be considered. In such a case, the one-
dimensional, homogenous, and constant-coefcient ver-
sion of Eq. (17) becomes
d
2
p
0
dx
2

p
0
L
2
p
23
where L
p
D
p
t
_
is the minority-carrier diffusion length.
The rst boundary condition for Eq. (23) is Eq. (19) applied
at the injecting boundary. The other one refers to the
ohmic contact. If it is an ideal ohmic contact deposited di-
rectly on the uniformly doped base, then the second
boundary condition becomes, from Eq. (7), p
0
(l) 0. Here,
l is the base length and the coordinate origin is at the in-
jecting boundary, as shown in Fig. 3. In many cases, be-
tween the ohmic contact and the uniformly doped base, a
thin and heavily doped region intervenes.
This region has a thickness on the order of a micro-
meter and a doping of the same type as the rest of the
base. The purpose of such a layer, called the backsurface
eld, is to provide a better ohmic contact and to isolate
the contact from the lightly doped base so that carrier re-
combinationgeneration is reduced [9]. Such a backsur-
face eld terminates the lightly doped base of diode 2 in
Fig. 4 making it a p-i-n diode, where i stands for intrin-
sic. Therefore, in the presence of this contact layer, the
base ends at a low/high n-n
junction. In terms of mi-

nority-carrier recombination, this interface is character-
ized by an effective recombination velocity S
pe
,
experienced by the minority carriers at the low side of
the junction. The expression for S
pe
is
S
pe
I
0c
N
D
en
2
i
S
24
where I
0c
is the saturation value of the recombination
current in the backsurface eld and S is the device cross
section. Equation (24) can be derived from Eq. (20), by
applying Eq. (19) at the n-n
junction and by equating the

minority current at the low/high junction to the recombi-
nation in the heavily doped region.
Under the previous boundary conditions, the solution
of Eq. (23) yields for the base saturation current:
I
0b
S
en
2
i
N
D
D
p
L
p
1
D
p
S
pe
L
p
tanh=L
p
tanh=L
p

D
p
S
pe
L
p
25
Equation (25) shows that, in terms of the one-dimensional
geometry, the quantity that matters is the ratio l/L
p
. Val-
ues of this ratio much less than one dene the short base,
whereas values above 3 dene the long base. In the long-
base case, Eq. (25) becomes I
0b
Sen
2
i
D
p
=N
D
L
p
. Similar
equations hold for a uniform emitter, too, but now the
heavy-doping effects could modify the value of n
2
i
, as will
be discussed in Section 6. If the base doping is very light,
as in a p-i-n diode, then the increased value of I
0b
will
make the base recombination dominate the current com-
ponents in Eq. (22). Accordingly, the inuence of space
charge region recombination current on the slope factor
will be suppressed even for voltages as low as a few kT/e,
as shown in Fig. 4, curve 2. Also, by extrapolating the
exp(eV/kT) t of curve 2 at zero voltage, a base recombi-
nation current of 0.7nA is obtained. This corresponds to a
300 K saturation current density of 2.8 nA/cm
2
compared
to emitter saturation current densities on the order of
pA/cm
2
. On the other hand, diode 1, with a base doping
density three orders of magnitude higher than that in diode
2, exhibits a saturation current of 24pA/cm
2
. This satura-
tion current comes mainly from the base recombination as a
result of its relatively light doping density and the absence
of a backsurface eld, which gives S
pe
very high values.
1040 DIODES
Equation (25) applies to uniformly doped regions. If the
doping is nonuniform, closed-form expressions are not
possible, in the general case. This is the case because the
one-dimensional version of Eq. (17) is still an ordinary
differential equation with nonconstant coefcients. How-
ever, analytical approximations based on iterative tech-
niques can be derived [10].
4.1.2. Diffusion in Three Dimensions. Equation (23)
holds provided the cross-sectional dimensions of the diode
are much larger than the diffusion length. Otherwise, lat-
eral diffusion of minority carriers in the base becomes im-
portant. In such a case, the three-dimensional version of
Eq. (23) takes the form
r
2
p
0
p
0
L
2
p
26
The last equation can be solved very accurately by semi-
analytical techniques based on the two-dimensional Fou-
rier transform [11]. Simulation results are as shown in
Fig. 5. As illustrated, in the case of a point contact diode
having emitter dimensions of 0.1 L
p
, the base recombina-
tion is expected to increase by a factor of 25 as a result of
the lateral carrier diffusion.
4.2. High-Level Injection
So far, our analytical approaches were based on the low-
level injection assumption. In high-level injection, where
np, an equation similar to Eq. (23) can be derived where
now the hole diffusion length is replaced by the ambipolar
diffusion length [12]. The boundary conditions, however,
are not linear and depend on the electric eld, which, now,
is a function of bias. If the quasi-Fermi potentials are at
in the quasineutral base, then the electronhole plasma
density p is space independent and equals n
i
exp(eV/2kT),
as can be derived from Eq. (18). In such a case, by inte-
grating the recombination current in the emitter, the base,
and the backsurface eld, we obtain
I
D
I
0e
I
0c
expeV=kT
eSl=t
0
expeV=2kTn
i
27
In Eq. (27), t
0
is the high-injection lifetime dened as the
ratio of the recombination rate divided by the plasma den-
sity. The at Fermi level condition can easily be satised
in p-i-n diodes where the light base doping density makes
the high-level injection possible even at a bias of 0.4V. In
Fig. 4, curve 2 shows the exp(eV/2kT) dependence, or slope
factor of 2, for voltages of about 0.4V, which drive the p-i-n
device to high-level injection. For even higher voltages, the
emitter and backsurface eld recombination in Eq. (27)
starts dominating the current, and the slope factor drops
again. For higher base doping densities, as in curve 1 of
diode 1, the required voltage for high-level injection con-
ditions could exceed 0.5Vat 300K. Now, the heavily doped
region recombination in Eq. (27) competes with the bulk
recombination, and the slope factor of 2 does not appear.
The bent of both curves 1 and 2 at voltages near 0.6 V is a
result of series resistance effects, which invalidate the as-
sumption of at Fermi levels across the base. In such a
case, the simulation is possible only by device simulators
that solve the complete system of the transport equations.
4.3. Reverse Bias
Under reverse bias where Vo0, the assumption of at
Fermi levels across the space charge region that led to
Eq. (22) no longer hold. On the other hand, however, the
space charge region can be considered to be fully depleted
from free carriers. In such a case, Eq. (11) holds with V
bi
replaced by V
bi
|V|. Therefore, the depleted space
charge region will expand toward the base according to
the square root of the bias for |V|45 V. In this region, the
ShockleyReadHall Eq. (15) predicts a negative recombi-
nation or generation of electronhole pairs. This genera-
tion current is the basic component of the leakage current
in reverse bias. The contribution of the diffusion compo-
nents from the base and the emitter, I
0b
I
0e
, is usually
negligible unless the base is very lightly doped. The bot-
tomline in Fig. 4 shows the reverse-bias current for diode 1.
The square-root dependence on voltage is not exactly
obeyed because of the PooleFrenkel effect, which
increases the generation rate at higher elds.
5. TRANSIENT RESPONSE OF DIODES
If a diode is subjected to a transient terminal bias, then
in addition to currents due to carrier diffusion and
0.1 1
1
10
10
(1)
(4)
(2)
(3)
L /Ld
N
o
r
m
a
l
i
z
e
d

3
-
D

c
u
r
r
e
n
t
Figure 5. Three-dimensional diffusion base saturation current of
a planar p-n junction with a square emitter having a sidelength L.
The top surface of the base is supposed to have zero recombination
velocity. The current is normalized with respect to the one-
dimensional diffusion current I
0b
en
2
i
L
2
D
p
=N
D
L
p
. In curve 2,
S
pe
0 and l L
p
/2. In curve 2, l is assumed to be innite. In
curve 3, S
pe
D
p
/L
p
and l L
p
/2, whereas in curve 4, S
pe
is as-
sumed to be innite and l L
p
/2.
DIODES 1041
recombination, we also have the dielectric displacement
current resulting from the time dependence of the electric
eld. If low-level injection is observed in the quasineutral
regions, the displacement current is restricted in the space
charge region. At the same time, low-level injection en-
sures that linearity holds in the base and the emitter, and
Eq. (17) still applies with @p
0
/@t replacing zero in the
right-hand side of the relation. The solution of the time-
dependent edition of Eq. (17) provides the minority-carrier
currents at the injecting boundaries of the base and the
emitter, I
b
(t) and I
e
(t), respectively. These currents have
now two components: the minority-carrier recombination
and the minority-carrier storage current @Q
0
/@t, where Q
0
is the total excess minority-carrier charge. To calculate the
total transient current, reconsider Eq. (14) in its transient
version. Therefore, in addition to I
b
(t) and I
e
(t), the tran-
sient space charge region current is required. Unlike the
base and the emitter, this current in addition to the re-
combination and storage component also includes the dis-
placement current [13]. Insofar as the displacement
current is concerned, the space charge region behaves as
a parallel-plate capacitor with a plate distance WW
A
W
D
, Eq. (11), a dielectric constant e, and a capacitance
C
SCR
eS/W. During transit, the dielectric displacement
current is supplied by the majority carriers from either
side of the junction.
To calculate the transient currents in the base and the
emitter, the boundary conditions must be dened. Bound-
ary condition Eq. (20) holds because of linearity. The other
condition at the injecting boundary depends on the kind of
transient to be considered [14]. Here we will assume that
the device is in equilibrium for to0, whereas at t 0 a
constant voltage Vis applied. We can now assume that Eq.
(19) applies with p
0
(C
j
) replaced by p
0
(C
j
, t) for t40. This
assumption has a validity range depending on how fast
the at quasi-Fermi potential condition can be established
across the space charge region. In fact, even in the absence
of series resistance effects, it takes a short time for this
condition to be established. This short time relates to the
dielectric response time of the majority carriers and the
minority-carrier diffusion time across the space charge re-
gion [14]. For almost all practical cases, the delay in es-
tablishing a xed minority-carrier density at the edge of
the quasineutral region will not exceed the limit of a few
tens of a picosecond [14], in the absence of series resis-
tance effects. Therefore, if the time granularity used in
solving the time-dependent version of Eq. (17) is restricted
to about a nanosecond, then the solutions will be accurate.
In practical cases, however, the very rst part of the tran-
sient current, following the sudden application of a volt-
age, will be determined by charging C
SCR
through the
series resistance of the majority carriers in the base and
the emitter. The respective time constant could be on the
order of a nanosecond. In such a case, the minority-carrier
transport in the base will determine the transient only
after several nanoseconds have elapsed since the applica-
tion of the voltage. The transient base transport can be
expressed in semianalytical forms using Laplace trans-
form techniques [14], especially in the case of uniform and
one-dimensional quasineutral regions. In a long-base
diode, the transition will last for about a minority-carrier
lifetime. In a short-base device with an ohmic contact at
the base end, the transient will last approximately l
2
/2D
p
,
which is the minority-carrier diffusion time through the
base.
5.1. Small-Signal Response
In many cases, the device operates under sinusoidal small-
signal excitation superimposed on a steady-state excita-
tion. In such cases, Eq. (17) still holds, but now 1/t will
have to be replaced by 1/t jo, where j is the imaginary
unit and o is the angular frequency of the excitation. This
is the case because the time derivative of the small-signal
carrier density is the carrier density amplitude times jo.
Having done the complex lifetime replacement, the anal-
ysis that followed Eq. (17) still holds. Now, however, the
small-signal value of the excess minority-carrier density
at the injecting boundary will be the steady-state value in
Eq. (19) times ev/kT. Here, v is the small-signal terminal
voltage, which is supposed to be much less than kT/e. Un-
der low-level injection and in view of the previous tran-
sient response discussion, the small-signal version of
Eq. (22) will refer to a terminal current I
D
having a real
and an imaginary component:
I
D
I
0e
I
0b
ev
kT
exp
eV
kT
_ _
1
_ _
I
SCR
ev
nkT
exp
eV
nkT
1
_ _
joC
SCR
v vGjoC
28
The star exponents denote the complex values of the sat-
uration currents as a result of the complex lifetime. In
Eq. (28), G and C are the diode small-signal parallel con-
ductance and capacitance, respectively. These two compo-
nents are of great importance because their frequency
dependence can reveal minority-carrier properties, such
as diffusivity and lifetime [15], and allow the device circuit
representation when the diode is part of a greater small-
signal circuit. For uniformly doped quasineutral regions,
I
0e
and I
0e
can be obtained from Eq. (25) by replacing the
diffusion length L(Dt)
1/2
with the complex diffusion
length L
L/(1jot)
1/2
. For frequencies sufciently
high, the magnitude of the complex diffusion length will
become much shorter than the base thickness. Then, the
complex version of Eq. (25) predicts that the base current
would change as 1/L
. If the base component in Eq. (28)

were to dominate, then beyond a certain frequency, C
would change as o
1/2
while G would change as o
1/2
.
This frequency dependence is conrmed in Fig. 6,
which shows the frequency response of diode 1, from Fig.
4, at two bias points. The theoretical t to the experimen-
tal results was obtained on the basis of the diode equiva-
lent circuit shown in Fig. 7. This circuit includes all the
components relating to carrier injection and storage in the
devices three regions in accordance with Eq. (28). It also
includes the base resistance R
N
, which has been ignored in
Eq. (28). In Fig. 6, the square root law is better obeyed at
the higher bias point and for frequencies less than 1 MHz,
1042 DIODES
especially for the capacitance. This is a combined result of
the space charge region capacitance, the relative contri-
bution of which increases at lower bias, and the series re-
sistance, the inuence of which is stronger at high
frequencies. The corner frequencies of the conductance
and the capacitance depend on the base thickness and the
lifetime. The t shown in Fig. 6 gave a minority-hole life-
time in the n-type base of about 30 ms. Such a lifetime and
Eq. (25) imply that the saturation current density of
24 pA/cm
2
at 300K, as shown in Fig. 4, is 90%, due to
base recombination. The emitter contribution of 10% is
discussed in Section 6.
The series resistance R
N
becomes the bulk majority-
carrier resistance under reverse bias or even under for-
ward bias, provided that the frequency is high (|L
|5l).
The capacitance C
N
QNR
in parallel with R
N
, as shown in
Fig. 7, is the geometric capacitance of the quasineutral
base [16,17]. For ordinary resistivity devices, it can be ig-
nored unless the frequency is in the gigahertz range. How-
ever, for diodes made on high resistivity substrates, this
capacitance must be considered especially at reverse bias
and high frequencies [17]. From Fig. 7 and in the limit of
very high frequencies under forward bias, the parallel
conductance saturates at 1/R
N
whereas the parallel ca-
pacitance does so at C
N
QNR
. This is because of the combi-
nation of the increasing injection conductancies and the
space charge region capacitance. Then, the product R
N
C
N
QNR
becomes the dielectric response time of the majority
carriers in a uniform base. At high injection, the parallel
conductance will saturate at the sum of the two carrier
conductances [18]. Under reverse bias, the circuit of Fig. 7
reduces to the space charge region capacitance in series
with the parallel combination of C
N
QNR
and R
N
. Unlike the
forward-bias case, where the circuit parameters depend
roughly exponentially on the terminal voltage V, in re-
verse bias the voltage dependence would be restricted to
V
1/2
. In the sense of the voltage dependence, the circuit
of Fig. 7 is the circuit of a varactor.
6. HEAVY-DOPING EFFECTS IN THE EMITTER
In the previous subsection, the emitter saturation current
density was estimated to be about 2 pA/cm
2
. From Eq. (25)
and by assuming microsecond lifetimes, we would expect
saturation currents on the order of a fA/cm
2
from an emit-
ter doped in the range 10
19
10
20
cm
3
. Such a discrepancy
by three orders of magnitude is due to the heavy-doping
effects, namely, the short lifetime resulting from Auger
recombination and the effective increase of n
i
due to band-
gap narrowing. In the Auger recombination process, a
minority carrier recombines directly with a majority one,
and the energy is transferred to another majority carrier.
Because of such kinetics, the Auger minority-carrier life-
time is inversely proportional to the square of the major-
ity-carrier density. The proportionality constant is
B10
31
cm
6
/s for minority electrons in p
emitters and
3 10
31
cm
6
/s for minority holes in n
emitters [19]. In
heavily doped regions, the Auger recombination rate is by
far higher than the ShockleyReadHall rate and deter-
mines the lifetime. Therefore, nanosecond lifetimes are
expected, especially for holes, in emitters doped in the vi-
cinity of 10
20
cm
3
.
In a heavily doped region, every minority carrier inter-
acts strongly with the majority carriers because of their
high density. The minority-majority carrier attraction
along with the carrierdopant interaction and the semi-
conductor lattice random disruption by the dopant atoms
reduces the bandgap and changes the density of states in
both bands [20,21]. The net result is an effective shrinkage
of the gap depending on the doping type and density
[2224]. This shrinkage changes the intrinsic carrier den-
sity n
i
to a much higher effective n
ie
. The result of the band
distortion is that the original system of transport equa-
tions [Eqs. (1)(6)] no longer holds. More specifically,
1
10
3
10
4
10
5
Frequency (Hz)
10
6
10
10
2
10
1
C
a
p
a
c
i
t
a
n
c
e

(
n
F
)
C
o
n
d
u
c
t
a
n
c
e

(
m
S
)
Figure 6. Experimentally measured capacitance (dots) and con-
ductance (squares) at 300K for diode 1. The bottom and the top
curves correspond to two different bias points: 420mV and
450mV, respectively. The solid curves are the theoretical ts
from the equivalent circuit of Fig. 7.
Emitter
contact
Base
contact
G
SCR
C
SRC
G
p
C
p
R
N
C
N
QNR
G
+
N
Figure 7. The equivalent-circuit model of a diode. The base in-
jection currents in Eq. (28) correspond to C
p
, imaginary part, and
G
p
, real part. The injection in the emitter is represented by G
N
.
The space charge region recombination is represented by G
SCR
,
whereas C
SRC
is the space charge region capacitance. The rest of
the components are accounted for in the text.
DIODES 1043
Eqs. (1b,c) and (2b,c) are not valid for the majority carriers
even if n
ie
substitutes n
i
because Boltzmann statistics
must be replaced by FermiDirac statistics. Also, Eq. (6)
no longer holds in a nonuniform region because the band
edges are not parallel any more and each carrier experi-
ences a different eld. However, the minority carriers still
follow the Boltzmann statistics, and Eq. (17) holds for the
minority carriers. Now E is the minority-carrier eld
(1/erE
c
for electrons), and the boundary condition Eq.
(19) is valid with n
i
replaced by n
ie
. Therefore, Eq. (25) still
applies for the minority-carrier recombination in a uni-
formly doped emitter. For an emitter doped at about
10
20
cm
3
, a gap narrowing of about 100 meV is expected
[2224], which makes n
ie
several tens higher than n
i
. If
such an n
ie
as well as nanosecond lifetimes replace n
i
and
microsecond lifetimes in Eq. (25), an emitter saturation
current on the order of pA/cm
2
is predicted, in accordance
with the experimental results of the previous section.
7. DIODES OF NONCONVENTIONAL TRANSPORT
So far in this article, devices based on the drift and diffu-
sion model of Eqs. (1) and (2) were studied. Charge carri-
ers can be transported from one region to another by
tunneling. Also, they can be temporarily trapped in ener-
gy gap states, atom clusters, or crystallites imbedded in
insulating lms, thereby affecting the tunneling or the
conventional transport of the free carriers.
In this respect, the rst device to be examined is p-i-n
diode 2 of Fig. 4, operating at cryogenic temperatures.
Around 4.2K, the equilibrium Fermi level in the lightly
doped n
region is pinned at the donor level. These levels,

now, are not ionized except for a fraction to compensate
the charge of the unintentionally introduced acceptor ions.
At such low temperatures, there are no free carriers in the
base, and no measurable conduction is possible unless the
voltage is raised enough to achieve the atband condition
[25,26]. For silicon, this voltage V
0
would be about 1.1V.
For even higher voltages, conduction is possible only if
electrons and holes can be injected in the frozen substrate
from the n and p regions, respectively. In this sense, Eq.
(24) based on the assumption of at majority-carrier Fermi
levels no longer holds. For To10K, injection is possible by
carrier tunneling through the small potential barrier ex-
isting at each of the p-i and i-n interfaces [26]. These bar-
riers exist because of the band distortion in the heavily
doped regions and the smaller gap there, as outlined in the
previous section. For V4V
0
, electrons tunnel in the i layer,
and the higher the forward bias, the higher the current
due to a eld-induced effective lowering of the barriers.
As shown in Fig. 8, for temperatures below10 K it takes
at least several volts to establish a current of few nano-
amperes. The injected electrons in the i layer are trapped
by the ionized donors and built a space charge and a sub-
sequent potential barrier. For even higher voltages ap-
proaching 10 V, the barrier at the i-p interface lowers,
holes now enter the i layer in large numbers. Their charge
neutralizes the trapped electron charge and causes the
voltage breakdown and the negative differential resis-
tance that appears in Fig. 8 for To10 K. The negative
resistance persists and beyond breakdown as a result of
new carrier generation by the impact ionization of occu-
pied shallow donors by the injected carriers. The interac-
tion of free and trapped carriers through impact ionization
gives rise to a negative dynamic conductance and capac-
itance that for frequencies high enough change as o
2
[27]. For T410 K the injection mechanism changes to
thermion emission over the interface potential barriers,
whereas the space charge effects are now less pronounced.
Another example of tunneling injection mechanism is
the breakdown effect in zener diodes. Here, the base is
quite heavily doped (E10
18
cm
3
), and the strong electric
eld in the space charge region increases even further by
applying a reverse bias. For elds approaching 10
6
V/cm, a
valence band electron can tunnel to a conduction band
state of the same energy. This way, electronhole pairs are
created, and the reverse current sharply increases. An-
other diode structure based on tunneling is a new metal
insulatorsemiconductor device having silicon nanocrys-
tals imbedded in the thin insulating lm [28]. One way to
realize such diodes is by depositing an aluminum elec-
trode on a thin (on the order of 10 nm) SiO
2
layer contain-
ing silicon nanocrystals. The substrate is n-type
crystalline silicon. The silicon nanocrystals can be creat-
ed either by oxidizing deposited amorphous silicon layers
[28] or by low-energy silicon-ion implantation in the SiO
2
lm [29]. In the absence of the nanocrystals, by applying a
negative voltage of a few volts on the aluminum electrode
relative to the n-type silicon substrate, only a small tun-
neling current would be present.
1
10
10
2
10
3
10
4
10
5
10
6
10
7
1
V (V)
10
13 K
17 K
9 K
7 K
I

(
n
A
)
Figure 8. Measured I V characteristics of diode 2 at cryogenic
temperatures. The square points correspond to 4.2K. The
To10K plots exhibit a distinct voltage breakdown. [Reprinted
from K. Misiakos, D. Tsamakis, and E. Tsoi, Measurement and
modeling of the anomalous dynamic response of high resistivity
diodes at cryogenic temperatures, Solid State Electron. 41:1099
1103 (1997), with kind permission from Elsevier Science Ltd, The
Boulevard, Langford Lane, Kidlington 0X5 1GB, UK.]
1044 DIODES
When the nanocrystals are introduced, much higher
currents are observed while the conductance curve exhib-
its characteristic peaks. Such peaks are shown in Fig. 9
showing the reverse current and conductance of a quan-
tum dot diode formed by low-energy implantation of sili-
con in a 10-nm SiO
2
layer [29]. The conductance peaks
appear when the metal Fermi level is swept across the
discrete energy states of the nanocrystals, thus enabling
resonant tunneling from the metal to the semiconductor
[28]. The three-dimensional connement of electrons in
the quantum box crystallites creates a large separation
between energy states, which along with the Coulomb
blockade effect of the occupied states explains the large
voltage separation of the three rst conductance peaks in
Fig. 9 [28,29]. Such quantum dot devices hold the promise
of single-electron transistors [30] and silicon-based light
emitting diodes [31].
Ending this article, we would like to mention the basic
uses of the diode as a device. The most frequent use of the
diode is the protection of complementary metal oxide
semiconductor (CMOS) integrated circuits from electro-
static discharges by clamping the output pads to the pow-
er-supply voltages through reverse-biased p-n junctions.
In analog integrated circuits, forward-biased diodes are
used for voltage shifting. Such diodes usually come from
properly wired bipolar transistors (e.g., emitterbase di-
odes with basecollector short-circuited). Diodes, as dis-
crete devices, nd applications mainly as rectifying
elements in power circuits. The breakdown effect of zen-
er diodes makes these devices useful as voltage reference
sources in power supplies. Photodiodes are widely used for
detecting photons or charge particles. Finally, large area
diodes with exposed front surface and proper design and
engineering can efciently convert solar light into elec-
tricity and are used as solar cells [32].
BIBLIOGRAPHY
1. W. Shockley. Electrons and Holes in Semiconductors, Van
Nostrand, Princeton, NJ, 1950.
2. C. T. Sah, R. N. Noyce, and W. Shockley, Carrier generation
and recombination in p-n junction and p-n junction charac-
teristics, Proc. IRE 45:12281243 (1957).
3. J. A. del Alamo, Charge neutrality in heavily doped emitters,
Appl. Phys. Lett. 39:435436 (1981).
4. W. Shockley and W. T. Read, Statistics of the recombination of
holes and electrons, Phys. Rev. 87:835842 (1952).
5. R. N. Hall, Electron-hole recombination in germanium, Phys.
Rev. 87:387 (1952).
6. J. C. S. Woo, J. D. Plummer, and J. M. C. Stork, Non-ideal base
current in bipolar transistors at low temperatures, IEEE
Trans. Electron. Devices 34:131137 (1987).
7. A. B. Sproul and M. A. Green, Intrinsic carrier concentration
and minority carrier mobility from 77 K to 300K, J. Appl.
Phys. 74:12141225 (1993).
8. K. Misiakos and D. Tsamakis, Accurate measurements of the
intrinsic carrier density from 78 to 340K, J. Appl. Phys.
74:32933297 (1993).
9. J. G. Fossum, Physical operation of back surface eld solar
cells, IEEE Trans. Electron. Devices 24:322325 (1977).
10. J. S. Park, A. Neugroschel, and F. A. Lindholm, Systematic
analytical solution for minority-carrier transport in semicon-
ductors with position dependent composition with application
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
0.00
0.05
0.10
0.15
0.20
0.25
16 14 12 10
Gate voltage (V)
8 6 4 2 0
C
u
r
r
e
n
t

(
m
A
)
C
o
n
d
u
c
t
a
n
c
e

(
m
S
)
Figure 9. Current and conductance plots of a
reverse-biased quantum dot diode. The con-
ductance peaks correspond to steps in current
curve. [Reprinted from P. Normand et al.,
Silicon nanocrystal formation in thin ther-
mal-oxide lms by very low energy Si
ion
implantation, Microelectron. Eng. 36(14):
7982 (1997) with kind permission of Else-
vier Science-NL, Sara Burgerharstraat 25,
1055KVAmsterdam, The Netherlands.]
DIODES 1045
to heavily doped silicon, IEEE Trans. Electron. Devices
33:240249 (1986).
11. S. Kavadias and K. Misiakos, Three-Dimensional simulation
of planar semiconductor diodes, IEEE Trans. Electron. Devic-
es 40:18751878 (1993).
12. S. M. Sze, Physics of Semiconductor Devices, 2nd ed., Wiley,
New York, 1981, p. 87.
13. F. A. Lindholm, Simple phenomenological model of transition
region capacitance of forward biased p-n junction diodes or
transistor diodes, J. Appl. Phys. 53:76067608 (1983).
14. T. Jung, F. A. Lindholm, and A. Neugroschel, Unifying view
of transient responses for determining lifetime and surface
recombination velocity in silicon diodes and back-surface
eld solar cells with application to experimental short circuit
current decay, IEEE Trans. Electron. Devices 31:588595
(1984).
15. A. Neugroschel et al., Diffusion length and lifetime determi-
nation in p-n junction solar cells and diodes by forward biased
capacitance measurements, IEEE Trans. Electron. Devices
25:485490 (1978).
16. B. M. Vul and E. I. Zavatitskaya, The capacitance of p/n junc-
tions at low temperatures, Sov. Phys.JETP (Engl. transl.)
11:611 (1960).
17. S. Kavadias et al., On the equivalent circuit model of reverse
biased diodes made on high resistivity substrates, Nucl. In-
strum. Methods Phys. Res. A322:562565 (1992).
18. K. Misiakos and D. Tsamakis, Electron and hole mobilities
in lightly doped silicon, Appl. Phys. Lett. 64:20072009
(1994).
19. J. Dziewior and W. Schmid, Auger coefcients for lightly
doped and highly excited silicon, Appl. Phys. Lett. 31:
346348 (1977).
20. G. D. Mahan, Energy gap in Si and Ge: Impurity dependence,
J. Appl. Phys. 51:26342646 (1980).
21. P. T. Landsberg et al., A model for band-p shrinkage in semi-
conductors with application to silicon, Phys. Status Solidi B
130:255266 (1985).
22. J. W. Slotboom and H. C. de Graaff, Measurements of band
gap narrowing in Si bipolar transistors, Solid-State Electron
19:857862 (1976).
23. A. W. Wieder, Emitter effects in shallow bipolar devices: mea-
surements and consequences, IEEE Trans. Electron. Devices
27:14021408 (1980).
24. J. A. del Alamo and R. M. Swanson, Measurement of steady-
state minority-carrier recombination in heavily doped n-type
silicon, IEEE Trans. Electron. Devices 34:15801589 (1987).
25. A. K. Jonscher, p-n junctions at very low temperatures, Br. J.
Appl. Phys. 12:363371 (1961).
26. Y. N. Yang, D. D. Coon, and P. F. Shepard, Thermionic emis-
sion in silicon at temperatures below 30 K, Appl. Phys. Lett.
45:752754 (1984).
27. K. Misiakos, D. Tsamakis, and E. Tsoi, Measurement and
modeling of the anomalous dynamic response of high resist-
ivity diodes at cryogenic temperatures, Solid-State Electron.
41:10991103 (1997).
28. E. H. Nicollian and R. Tsu, Electrical properties of a silicon
quantum dot diode, J. Appl. Phys. 74:40204025 (1993).
29. P. Normand et al., Silicon nanocrystal formation in thin ther-
mal-oxide lms by very-low energy Si
ion implantation,
Microelectron. Eng. 36(14):7982 (1997).
30. K. Yano et al., Romm-temperature single-electron memory,
IEEE Trans. Electron. Devices 41:16281638 (1994).
31. D. J. Dimaria et al., Electroluminescence studies in silicon
dioxide lms containing tiny silicon islands, J. Appl. Phys.
56:410 (1984).
32. F. Zhang, S. Wenham, and M. A. Green, Large area, concen-
trator buried contact solar cells, IEEE Trans. Electron.
Devices 42:145149 (1995).
DIPOLE ANTENNAS AND ARRAYS*
CYNTHIA M. FURSE
OM P. GANDHI
University of Utah
Salt Lake City, Utah
GIANLUCA LAZZI
North Carolina State University
1. INTRODUCTION
A dipole antenna is most commonly a linear metallic wire
or rod with a feedpoint at the center as shown in Fig. 1.
Most often, this type of antenna has two symmetric radi-
ating arms. Because of the symmetry of the antenna rel-
ative to the xy plane containing the feedpoint, the
resultant radiation is independent of f (rotationally sym-
metric about the z axis). Dipole antennas and arrays of
dipoles are commonly used for HF and UHF broadcasting,
TV, and FM communications, and as electric eld probes.
This article describes the basic nature and applications of
dipole antennas and some of their variations such as
biconical and bowtie antennas, slot dipoles, folded dipoles,
sleeve dipoles, and shunt-fed dipoles. The commonly used
broadband log periodic and YagiUda dipole arrays are
also discussed.
2. DIPOLE ANTENNA TYPES
2.1. Innitesimal Dipole (Hertzian Dipole)
An innitesimal dipole (L5l) is a small element of a lin-
ear dipole that is assumed to be short enough that the
current (I) can be assumed to be constant along its length
L. This is also called a Hertzian dipole. The electric and
magnetic eld components of this dipole are [6]
H
1
4p
IL sin y e
jb
0
r
jb
0
r

1
r
2
_ _
a
f
1
E
jZ
0
IL
2pb
0
cos y
jb
0
r
2

1
r
3
_ _
e
jb
0
r
a
r
jZ
0
IL
4pb
0
sin y
b
2
0
r

jb
0
r
2

1
r
3
_ _
e
jb
0
r
a
y
2
where Z
0
(m
0
/e
0
)
1/2
is the intrinsic impedance ( 377O)
for free space and b
0
o(m
0
e
0
)
1/2
is the propagation con-
stant ( o/c, where c is the velocity of light). The elds are
*See also Section on Monopole Antennas
1046 DIPOLE ANTENNAS AND ARRAYS
seen to decay rapidly (1/r
3
and 1/r
2
variation) very near
the antenna, and less rapidly (1/r variation) farther away.
The elds with terms 1/r
2
and 1/r
3
(the induction terms)
provide energy that is stored near the antenna. The elds
with 1/r variation (the radiation terms ) provide actual
energy propagation away from the antenna. The distance
away from the antenna where the induction and radiation
terms are equal is dl/2p. When dol/2p, this is the near
eld of the antenna, and the induction terms dominate.
When d4l/2p, this is the far eld, and the radiation terms
dominate. In the far eld, the wave propagation is in the
transverse electromagnetic (TEM) mode, which is charac-
teristic of far-eld radiation from nite structures.
The far-zone radiated elds of the Hertzian dipole fol-
low from (1) and (2) by retaining the 1/r varying terms:
H
j
4pr
IL sin y e
jb
0
r
a
f
3
E
jZ
0
4pr
IL sin y e
jb
0
r
a
y
4
As expected for TEM wave propagation, the E and Helds
are perpendicular to each other and to the outward prop-
agation in the r direction. Also the ratio of E/HZ
0
(m
0
/e
0
)
1/2
, which is the intrinsic impedance of free space.
The radiation pattern of this short dipole is shown in
Fig. 2, and exhibits the classical symmetry expected of di-
pole antennas, which is both independent of f and sym-
metric about the xy plane through the center (feedpoint)
of the dipole. The magnitude of the total radiated power is
P
rad
40 p
2
I
0
2
(L/l)
2
. From Eqs. (3) and (4) it is interesting
to note that even for this constant-current innitesimal
dipole, the radiated power density is proportional to sin
2
y.
Hence, it is maximum for y 901 (i.e., in the xy plane
normal to the orientation of the dipole) and zero for the
directions along the length of the dipole (y 01 and 1801).
The latter property for zero radiation along the length of
the dipole will be seen for all linear dipoles regardless of
length. It follows from the fact that a linear antenna may
be considered to be composed of innitesimal dipoles that
do not create E and H elds or radiated power density for
the y 01 and 1801 directions.
2.2. Linear Dipole Antennas
The geometry of a linear dipole antenna of length L2h is
shown in Fig. 1. The current distribution is sinusoidal,
and is given by
Iz
0

I0
sin kh
sin kh jz
0
j for hoz
0
oh 5
where I(0) is the current at the feedpoint of the antenna, h
is the half-length of the antenna, and k o(me)
1/2
is the
propagation constant in the material surrounding the
dipole. The current distributions for several lengths of
dipole antennas are shown in Fig. 3.
r
h =
L
2
Figure 1. Dipole antenna.
Figure 2. Radiation pattern for an innitesimal (or Hertzian)
dipole.
Figure 3. Current distributions and associated radiation pat-
terns for several different lengths of dipole antennas.
DIPOLE ANTENNAS AND ARRAYS 1047
The electric and magnetic elds around the dipole are
calculated by modeling the antenna as a series of Hertzian
or elemental dipoles and integrating the elds from each
of these elements. The resultant elds far from the anten-
na at a distance r are
E
jZI0
2pr sinkh
Fy e
jotkr
y 6
and
H
jI0
2pr sinkh
Fy e
jotkr
f 7
where Z (m/e)
1/2
, and where the y dependence of the ra-
diated elds F(y) is called the pattern factor and is given
by the following:
Fy
coskh cosy cos kh
sin y
8
The radiation power density (radiation pattern) is given
by
Py
E
!
E
!
2Z
r
0
!
ZI
2
0
8p
2
r
2
0
sin
2
kh
F
2
y 9
Using Z Z
0
120p, this can also be expressed in terms of
the total radiated power W ( I
2
(0)R
a
/2) and the feedpoint
resistance R
a
as follows:
Py
30
pr
2
W
R
a
F
2
y
sin
2
kh
10
The normalized radiation patterns are shown in Fig. 3b for
several different lengths of dipoles.
The directivity of a dipole antenna related to the max-
imum power density that an antenna can create at a dis-
tance r is given by
D
P
max
P
0
F
2
y
max
1
2
_
p
0
F
2
y sin y dy
120
R
a
F
2
yj
max
sin
2
kh
11
where P
0
W=4pr
2
is the isotropic power density that
would have been created at the eld point if the antenna
had a directivity of one and radiated isotropically for all
angles (clearly a mathematical possibility but not physi-
cally realizable).
The input resistance R
a
of a center-fed dipole antenna of
length 2h is twice that of an end-fed monopole of length h.
This may therefore be obtained by using the graphs given
in the related encyclopedia article, MONOPOLE ANTENNAS.
The ohmic losses of a dipole antenna [given by I
2
(0)
R
ohmic
/2] are quite small, particularly for h/l40.1.
The resultant antenna radiation efciencies (given by
R
a
/(R
a
R
ohmic
)) are on the order of 9099%.
Two effects cause the behavior of physical dipoles to
slightly differ from that of ideal dipoles: (1) realistic an-
tennas have some nite thickness, and (2) the ends of the
dipole capacitively couple to air, effectively making the di-
pole electrically longer by 29% than its physical length.
For a half-wave dipole (length2hl/2), for instance, the
physical length must be slightly shortened in order to cre-
ate a resonant length antenna (X
a
0). Table 1 shows the
wire lengths required to produce a resonant half-wave di-
pole. This shortening varies from 2% to 9%, depending on
the thickness of the dipole.
Since a dipole antenna is a physically resonant struc-
ture, its feedpoint impedance (particularly the reactance
X
a
) varies greatly with frequency. Thus, these antennas
have a fairly narrow bandwidth. The VSWR of a dipole
antenna as a function of frequency and wire thickness is
shown in Fig. 4 for an antenna that would be half-wave
resonant at 300MHz. Using a measure of usable band-
width that the measured VSWR should be less than 21,
this antenna has bandwidths of 31026248 MHz for
the thicker wire and 30428024 MHz for the thinner
wire. As fractions of the design frequency (300 MHz), the
bandwidths are 16 and 8%, respectively.
2.3. Slot Dipole
A slot dipole antenna is a dual to the linear dipole anten-
na. The radiation pattern of a slot antenna is identical to
that of the linear dipole of the same length (see Fig. 3)
except that orientations of the E and H are interchanged.
Also the feedpoint impedance Z
s
of a slot antenna is re-
lated to that of the dual linear antenna by the following
equation
Z
s
Z
2
4Z
a
12
where Z
s
is the impedance of the slot and Z
a
is the
impedance of the dual linear antenna.
2.4. Biconical Dipoles
A biconical dipole such as that shown in Fig. 5a is com-
monly used for broadband applications when the are
angle y is between 301 and 601. The exact are angle is not
Table 1. Wire Lengths Required to Produce a Resonant Half-Wave Dipole for a Wire Diameter of 2a and a Length L
Length: diameter ratio L/(2a) Percent Shortening Required Resonant Length L Dipole Thickness Class L
5000 2 0.49l Very thin
50 5 0.475l Thin
10 9 0.455l Thick
critical, so it is generally chosen so that the impedance of
the dipole nearly matches the impedance of the feedline to
which it is connected. The impedance of the biconical di-
pole varies as a function of wavelength and are angle,
with a relatively at impedance response for wide are
angles. Hence, this antenna is of broader bandwidth than
a simple linear dipole.
Some variations of this method of using aring to in-
crease bandwidth are the at bowtie antenna (which may
be built on a printed circuit board) and the wire version of
the biconical antenna shown in Figs. 5b and 5c, respec-
tively.
2.5. Folded Dipole Antennas
A folded dipole antenna is shown in Fig. 6. The dipole is
created by joining two cylindrical dipoles at the ends and
driving the entire structure by a transmission line (often a
two-wire transmission line) at the center of one arm as
shown. The feedpoint impedance of a folded dipole of two
identical-diameter arms is 4 times larger than that for an
unfolded dipole of the same length. This can actually be
advantageous, since the feedpoint resistance may now be
comparable to the characteristic impedance Z
0
of the
transmission line or feedline. The reactance of the anten-
na may easily be compensated by using a lumped element
with a reactance that is negative of the reactance at the
terminals of the folded dipole antenna or else by using a
foreshortened antenna length to resonant length arms so
that X
a
0 (see Table 1).
2.6. Shunt-Fed Dipoles
Matching networks of reactive elements are generally re-
quired to match the feedpoint impedance (R
a
jX
a
) of cen-
ter-fed dipoles to transmission lines. Typically these lines
have characteristic impedance on the order of 300600 O.
To alleviate the need for matching networks, the antennas
are at times shunt-fed at symmetric locations off the cen-
terpoint as shown in Fig. 7. This procedure using either
the delta match (Fig. 7a) or the T-match (Fig. 7b) is often
used for half-wave dipoles (2hl/2) with A and B dimen-
sions that are typically on the order of 0.100.15l.
2.7. Sleeve Dipole
The sleeve dipole antenna and its equivalent electrical
model are shown in Fig. 8. In practice, this antenna is
built from a coaxial line with the outside conductor and
insulation stripped away from the center conductor, which
is left protruding. The outer conductor is connected to the
ground plane, and the image produced by the ground
plane creates an equivalent sleeve dipole antenna. These
dipoles are useful, because they have a broadband VSWR
over nearly an octave of bandwidth.
3. APPLICATIONS
Dipole antennas and arrays of dipole antennas are used
for shortwave (330 MHz) and for VHF and UHF (30
900 MHz) radio and TV broadcasting. If directional com-
munication is desired such as for shortwave radio trans-
mission via the ionosphere, a phased array of horizontal
dipoles may be used mounted above a ground plane. The
spacing is chosen to send the major lobe of radiation to-
ward the sky at a suitable angle to reect off the iono-
sphere and provide broadcast coverage over the desired
service area.
Figure 4. VSWR of a dipole antenna as a function of frequency
and wire thickness (from Ref. 3).
Figure 5. Biconical dipole antenna and variations: (a) biconical
dipole antenna; (b) at bowtie antenna; (c) wire version of bicon-
ical dipole antenna.
Figure 6. Folded dipole antenna.
(a) (b)
A
B
Figure 7. Shunt-fed dipoles: (a) delta match; (b) T-match.
For VHF and UHF radio and TV broadcasting over a
3601 azimuthal angle, collinearly mounted vertical dipoles
that are excited in phase with each other are often used.
Two examples of this are shown in Figs. 9a and 9b. An
example variation of this is a three- to eight-bay turnstile
antenna used for TV broadcasting shown in Fig. 10a. Each
turnstile is made of two perpendicular slot antennas as
shown in Fig. 10b.
3.1. Log periodic Antennas
For broadband applications log periodic antennas are com-
monly used as both transmitting and receiving antennas.
The bandwidth is easily controlled by adjusting the rela-
tive lengths of the longest and shortest elements in the
array. The geometry of a log periodic array is shown in
Fig. 11a, which shows how the phase reversal feed sys-
tem for this antenna is constructed. The equivalent an-
tenna model of this array is shown in Fig. 11b. The
elements of the array are dipole antennas that increase
in both length and spacing according to the formula
t
R
n1
R
n
d
n1
d
n
13
where t f
n
/f
n1
is the ratio of the resonant frequencies f
n
and f
n1
of the adjacent dipole elements. Since lengths
and spacings are interrelated, the choice of one initial
value controls the design of the remaining elements.
The spacing between one half-wave dipole and its adja-
cent shorter neighbor is given by
s
d
n
2L
n
1 t
4
cot a 14
L
L
2S
(a) (b)
L
S
Ground
plane
Coaxial
line
Figure 8. Sleeve dipole antenna: (a) physical model; (b) equiva-
lent electrical model.
(a) (b)
Figure 9. Collinearly mounted vertical dipoles for VHFand UHF
radio and TV broadcasting: (a) pole-mounted array of collinear
dipoles; (b) vertical dipoles spaced around a pole.
(a) (b)
Metal
sheet
Slot
Figure 10. Variation on collinearly mounted vertical dipoles:
(a) turnstile antenna used for TV broadcasting; (b) two perpen-
dicular slot antennas constituting each turnstile.
Figure 11. Log periodic dipole array: (a) geometry of a log
periodic array showing how the phase reversal feed system for
this antenna is constructed (from Ref. 6); (b) equivalent antenna
model of the log periodic array.
Log periodic arrays are generally constructed with
small values of a (101rar451 [5]) and large values of t
(0.95rtr0.7 [5]), which essentially gives a traveling wave
propagating to the left in the backre direction, away from
the antenna array. The nature of this array is that only
the elements that are approximately half-wavelength long
radiate, and since they are radiating to the left, the small-
er elements do not interfere with them. This is accom-
plished by the phase reversal of the feeds. An array that is
built without the phase reversal radiates in the endre
direction. The interference of the longer elements to the
right of radiating elements results in spurious reections
and erratic impedance behavior, known as end effect.
An effective way to further increase the bandwidth of a
log periodic antenna is to change from dipole elements to
elements with individual broader bandwidths, similar to
changing from a dipole antenna to a biconical antenna.
This is accomplished for log periodic arrays by using a
conguration of wires such as shown in Fig. 12, where
each element is a sawtooth element and therefore has
broader bandwidth than the individual dipole elements.
3.2. Broadband Dipole Curtain Arrays
A broadband dipole curtain such as shown in Fig. 13 is
commonly used for high-power (100500-kW) HF iono-
spheric broadcasting and shortwave broadcasting sta-
tions. The curtain is composed of several dipoles, usually
a half-wavelength long, mounted horizontally or vertically
in a rectangular or square array, often backed by a re-
ecting plane or wire mesh. This array has several de-
sireble features, including high gain, broad bandwidth,
independent control of horizontal and vertical radiation
patterns, ease of matching (low VSWR), and the ability to
efciently broadcast efciently. Using a phased-feeds
system, this array allows beamsteering of the radiation
pattern in both the azimuthal and elevation planes, pro-
viding a very high degree of exibility.
3.3. VHF/UHF Communication Applications:
3.3.1. YagiUda Dipole Array. YagiUda arrays are
commonly used as general-purpose antennas from 3 to
3000 MHz, in particular as home TV antennas. They are
inexpensive, have reasonable bandwidth, and have gains
up to 17 dBi or more if multiple arrays are used. They have
unidirectional beams with moderate sidelobes [6].
A typical YagiUda array is shown in Fig. 14. This ar-
ray is a simple endre array of dipole antennas where only
one of the elements is driven and the rest are parasitic.
The parasitic elements operate as either reectors or di-
rectors. In general [6], the longest antenna, which is about
l/2 in length, is the main reector, and is generally spaced
l/4 in back of the driven dipole. The feed element is com-
monly a folded dipole antenna 0.450.49l long. Adding
directors, which are generally 0.40.45l long, to the front
of the driven element increases the gain of the array. The
directors are not always of the same length, diameter, or
spacing. Common arrays have 612 directors and at most
two reectors. Additional improvements in gain by adding
more elements are limited; however, arrays have been de-
signed with 3040 elements [5]. A gain (relative to isotro-
pic) of 59 per wavelength of array length is typical
for YagiUda arrays, for an overall gain of 5054
(14.817.3 dB).
The YagiUda array is characterized by a main lobe of
radiation in the direction of the director elements and
small sidelobes. The beamwidth is small, generally 30601
[5]. Typical E- and H-plane patterns of a YagiUda array
are shown in Fig. 15. Typically, the performance of a Yagi
Uda array is computed using numerical techniques
[5, Ref. 17]. For the simple case where all the elements
are approximately the same size, the electric eld pattern
can be computed from the array factors of the various el-
ements.
The input impedance of a YagiUda array is often
small. For example, for a 15-element array with reector
length0.5l, director spacing0.34l, and director length
0.406l, the input impedance is 12, 22, 32, 50, or 62O
for reector spacings of 0.10, 0.13, 0.15, 0.18, and 0.25l, re-
spectively. This can make matching to typical transmis-
sion lines (50, 75, or 300O) difcult. Folded dipoles used
for the driven element are therefore used to boost the input
impedance by a factor of Z4.
Extensive studies of the design of YagiUda arrays
have been made [5, Ref. 20], and tables are provided to
optimize the YagiUda array for a desired gain.
Figure 12. LPAwith sawtooth wire elements for increased band-
width. Dots indicate feedpoint locations; heavy wires indicate di-
pole elements; light wires indicate wires for structural support
only.
Figure 13. A broadband dipole curtain. Dots indicates feedpoint
locations. Figure 14. YagiUda array.
3.3.2. Dipoles for Circular Polarization. For applications
that require a circularly polarized antenna such as TVand
FM broadcasts and space communications, at least two
dipoles, each of which has a linear polarization, must be
combined in an array, often referred to as crossed dipoles.
In a crossed dipole conguration, dipoles are mounted
perpendicular to each other for circular polarization or at
other angles for elliptical polarization. Currents are fed
901 out of phase between the two dipoles. These can also be
used as probes for sensing vector elds to isolate individ-
ual components of the electric eld. Adaptations of the
crossed dipole are shown in Figs. 16a and 16b. Dipole ar-
rays such as the YagiUda can also be combined to provide
circular polarization, as shown in Fig. 16c.
BIBLIOGRAPHY
1. J. D. Kraus, Antennas, 2nd ed., McGraw-Hill, New York, 1988.
2. M. F. Iskander, Electromagnetic Fields and Waves, Prentice-
Hall, Englewood Cliffs, NJ, 1992.
3. W. L. Stutzman and G. A. Thiele, Antenna Theory and Design,
J Wiley, New York, 1991.
4. O. P. Gandhi, Microwave Engineering and Applications, Per-
gamon Press, New York, 1985.
5. C. A. Balanis, Antenna TheoryAnalysis and Design, J Wiley,
New York, 1982.
6. R. C. Johnson, Antenna Engineering Handbook, 3rd ed.,
7. G. A. Theile, Analysis of YagiUda antennas, IEEE Trans.
Anten. Propag. 17 (1969).
DIRECT SATELLITE TELEVISION
BROADCASTING
JOHN P. GODWIN
DirecTV
Direct-to-home (DTH) satellite television broadcasting
has no strict technical or legal definition. Since the late
1970s the term has been used to delineate commercial
systems that deliver television directly to consumer homes
using communication satellites in geosynchronous orbit.
Systems originally intended for DTH applications have
operated at downlink frequencies above 11GHz and with
antennas of 1 m or less. Certain systems operated at 4 GHz
were planned for cable television distribution and became,
secondarily, DTH systems with customer parabolic anten-
nas in the 2.53.0 m range. Most systems have been
supported primarily from subscription and pay-per-
view revenues rather than advertising revenues. In the
(a) (b)
Figure 15. Typical E- and H-plane patterns of
a YagiUda array; total number of elements
27, number of directors 25, number of reec-
tors 1, number of driven elements 1, total
length of reector 0.5l, total length of feeder
0.47l, total length of each director 0.406l,
spacing between reector and feeder 0.125l,
spacing between adjacent directors 0.34l,
radius of wires 0.003l (from Ref. 7).
(a)
(c)
(b)
Figure 16. Cross-dipole applications for circular or elliptical po-
larization (from Ref. 6): (a) two shunt-fed slanted V dipoles; (b)
series-fed slanted dipoles; (c) circularly polarized YagiUda array.
1052 DIRECT SATELLITE TELEVISION BROADCASTING
Next Page

04b

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

04b

Transféré par

Droits d'auteur :

Formats disponibles

polarization and increased bandwidth for cylindrical dielectric

resonator antenna, Electron. Lett. 37:406408 (March 2001).

Figure 11. Graphical oscillator solution showing one-port oscil-

0, d/D2.0, L/D0.4, ML, e

and the other from the RF port

which is the RF signal to be detected, that is, measured by

junction. In terms of mi-

junction and by equating the

. If the base component in Eq. (28)

region is pinned at the donor level. These levels,

Vous aimerez peut-être aussi