This action might not be possible to undo. Are you sure you want to continue?
measured at a distance d
0
from the original source. In
terms of the measured power density, the scattered power
density is
SðinðW=m
2
Þ ¼S
0
d
0
d
1
_ _
2
ðRCSÞ
4pd
2
2
ð38Þ
8. SUMMARY
Since electromagnetic scattering perturbs the incident
ﬁeld, it can create interference, both constructive and
destructive. The three main mechanisms of scattering
are reﬂection, refraction, and diffraction. Reﬂection and
refraction are the most common, and since these scatter
ing mechanisms tend to be specular, they are easiest to
analyze. Diffraction is much more difﬁcult to analyze, and
is typically not as dominant as the other two. Various
approximations can lead to solutions of diffraction
problems. The two main approximations covered were
the GTD, which takes advantage of ray tracing, and
PO, which relies on Huygen secondary sources. Slight
modiﬁcations to the PO solution for the aperture problem
can lead to solutions for backscattering, and forward
scattering from a conductive plate. This simple structure
can serve as a building block for more complicated struc
tures, which can be modeled as composites of conductive
plates.
BIBLIOGRAPHY
1. W. L. Stutzman, Antenna Theory and Design, 2nd ed., Wiley,
New York, 1998.
2. J. D. Kraus, Antennas, 2nd ed., McGrawHill, New York,
1988.
3. C. R. Paul and S. A. Nasar, Introduction to Electromagnetic
Fields, McGrawHill, New York, 1987.
4. C. T. A. Johnk, Engineering Electromagnetic Fields and
Waves, Wiley, New York, 1988.
5. G. G. Skitek and S. V. Marshall, Electromagnetic Concepts
and Applications, PrenticeHall, Englewood Cliffs, NJ, 1982.
6. C. A. Balanis, Advanced Engineering Electromagnetics, Wiley,
New York, 1989.
7. E. C. Jordan and K. G. Balmain, Electromagnetic Waves and
Radiating Systems, PrenticeHall, Englewood Cliffs, NJ,
1968.
8. R. F. Harrington, TimeHarmonic Electromagnetic Fields, Mc
GrawHill, New York, 1961.
9. G. E. Corazza, V. DegliEsposti, M. Frullone, and G. Riva, A
characterization of indoor space and frequency diversity by
raytracing modeling, IEEE J. Select. Areas Commun. 14(3):
411–419 (April 1996).
10. M. Kimpe, V. Bohossian, and H. Leib, Ray tracing for
indoor radio channel estimation, Proc. IEEE, 2nd Int. Conf.
Universal Personal Communication (ICUPC), Oct. 1993, pp.
64–68.
11. J. B. Keller, A geometrical theory of diffraction, in L. M.
Graves, ed., Calculus of Variations and its Applications,
Proc. Symp. Applied Mathematics, Vol III, McGrawHill,
New York, 1958, pp. 27–52.
12. G. E. Athanasiadou, A. R. Nix, and J. P. McGeehan, Indoor 3D
ray tracing predictions and their comparison with high res
olution wideband measurements, Proc. IEEE, 46th Vehicle
Technology Conf., April 1996, Vol. 1, pp. 36–40.
13. O. Landron, M. J. Feuerstein, and T. S. Rappaport, A com
parison of theoretical and empirical reﬂection coefﬁcients for
typical exterior wall surfaces in a mobile radio environment,
IEEE Trans. Anten. Propag. 44(3):341–351 (March 1996).
14. R. H. Clarke and J. Brown\com}, Diffraction Theory and An
tennas, Wiley, New York, 1980.
15. J. D. Kraus, Electromagnetics, 4th ed., McGrawHill, New
York, 1992, pp. 622–627.
16. R. G. Kouyoumjian and P. H. Pathak, A uniform geometrical
theory of diffraction for an edge in a perfectly conducting sur
face, Proc. IEEE 62(11):1448–1461 (Nov. 1974).
17. J. B. Keller, Geometrical theory of diffraction, J. Opt. Soc. Am.
52(2):116–130 (Feb. 1962).
18. J. B. Keller and A. Blank, Diffraction and reﬂection of pulses
by wedges and corners, in The Theory of Electromagnetic
Waves, Interscience Publishers, New York, 1951, pp. 139–158.
19. C. A. Balanis, Antenna Theory, Analysis and Design, Harper
& Row, New York, 1982, pp. 92–94, 448.
20. E. Hecht, Optics, 2nd ed. AddisonWesley, Reading, MA, 1987,
p. 393.
21. M. Spiegel, Advanced Mathematics for Engineers and Scien
tists, Schaum’s Outline Series, McGrawHill, New York, 1971.
22. R. L. Musselman, Analytic scattering model for indoor prop
agation, Proc. IEEE Antennas & Propagation Society Int.
Symp., 2001, Vol. 4, pp. 734–737.
23. H. G. Booker, Slot aerials and their relation to complementa
ry wire aerials, J. Inst. Electric. Eng. (Part IIIA): 620–626
(1946).
24. A. Ishimaru, Wave Propagation and Scattering in Random
Media, Academic Press, San Diego, 1978.
25. H. C. van de Hulst, Light Scattering by Small Particles, Dover
Publications, New York, 1981.
26. M. I. Skolnik, Introduction to Radar Systems, 2nd ed.,
McGrawHill, New York, 1980.
27. S. A. Hovanessian, Radar System Design and Analysis,
Artech House, Norwood, MA, 1984.
ELECTROMAGNETIC WAVES IN IONOSPHERE
ROBERT D. HUNSUCKER
RP Consultants
Klamath Falls, Oregon
Although the emphasis of this encyclopedia is on the high
er frequencies in the radio spectrum, the importance of the
solar–terrestrial environment on radiowaves of all fre
quencies is of considerable importance to engineers de
signing radio systems for various services. An idealized
representation of the solar–terrestrial environment is
shown in Fig. 1.
It is interesting to note that EM waves over a very large
spectral range interact with the terrestrial ionosphere and
with radio receivers on Earth’s surface or in space: (1) so
lar radiation, including the extreme ultraviolet (EUV:
ELECTROMAGNETIC WAVES IN IONOSPHERE 1303
Previous Page
102.7–111.8nm), soft X rays, Lyman Alpha (121.5 nm);
hard X rays (0.2–0.8nm) and cosmic rays produce the
ionospheric layers; (2) solar radio noise of frequencies
B20 MHz up to 1GHz penetrates the terrestrial iono
sphere and can degrade the signaltonoise ratio (SNR)
received on various radio services; (3) radio signals fom
ELF through low VHF (B10 Hz to B35 MHz) propagate
over long terrestrial paths by reﬂection and refraction by
the ionosphere.
The main ionizing agents are listed in the preceeding
paragraph. At high geomagnetic latitudes, energetic elec
trons and protons also can produce areas of high ioniza
tion density. The terrestrial ionosphere is a roughly
spherical shell of weakly ionized plasma that surrounds
Earth. A plasma is a gas that has been ionized by radia
tion or by charged particles, so that it consists of free elec
trons, ions, and neutrals; it is sometimes referred to as the
fourth state of matter [1], the characteristics of various
plasmas are displayed in Fig. 2.
This spherical shell is stratiﬁed into distinct layers; the
lowest region is the D layer, starting at about 50km
height; the E region, starting at about 100km; the F1 lay
er (during the day) near 250km; and the F2 layer at about
350km. At high geomagnetic latitudes, solar energetic
charged particles are also important ionizing agents.
Figure 2 is a plot of log density versus log kinetic temper
ature showing the relative state of ionization of various
plasmas, and Fig. 3 illustrates the various ionospheric
layers and their ionizing agents.
1. THE RADIO SPECTRUM
A considerable portion of the radio spectrum (ELF
through HF) is affected by our ionosphere, and the ITU
nomenclature for the frequency bands, along with the
principal modes and uses, are shown in Table 1. The so
called wireless frequencies lie in the UHFand SHF bands.
Coronal
hole
Flare
Sunspots
Coronal
hole
Mass
ejection
Energetic
particles
Ring
current
Enhanced
solar wind
XRays from flare
Solar wind plasma flow transporting
Interplanetary magnatic field (I M F)
Magnetopause
Ionosphere
Precipitation of
auroral particles
Bow shock
Geomagnetic field
Cusp
Figure 1. Idealized schematic representation of the solar–terrestrial environment.
50% ionization
hydrogen plasma
(4π/3)nλ
D
s
<1
High
pressure
arcs
Shock
tubes
Alkali
metal
plasma
Low
pressure
Flames
Glow
discharge
25
20
15
10
5
Earth
iono
sphere
0
−2 −1 0 1
Solar wind
(1 AU)
log
10
[T(eV)]
2 3 4
Earth
plasma sheet
5
λ
D
, > 1 cm
Solar
corona
Fusion
experiments
Fusion
reactor
Z pinches
Laser
plasma
Focus
ii
,
ee
> 1 cm
l
o
g
1
0
[
n
(
c
m
−
3
)
]
Figure 2. Logarithmic plot of approximate magnitudes of some
typical laboratory and natural plasmas.
1304 ELECTROMAGNETIC WAVES IN IONOSPHERE
2. IONOSPHERIC INTERACTION WITH EM WAVES
The basic interaction mechanism between radio (EM)
waves and the ionosphere involves the oscillation of the
electric component of the wave acting on free electrons,
which are 1800 times less massive than the ionospheric
neutrals or ions. The E wave induces motion of the elec
trons, and at the same time the ionosphere abstracts en
ergy from the electrons, resulting in a bending of the radio
wave and in some energy loss from the wave. Figure 4
illustrates the attenuation of radiowaves in the iono
sphere as a function of frequency from 1 Hz to 30 MHz.
As may be seen in Fig. 4, there is a variation in radio
wave attenuation from day to night. The virtual height
(the height at which radiowaves at vertical incidence are
reﬂected) varies with time of day, with season of the year,
and with geomagnetic activity. The most regular variation
is the local time variation, as shown in Fig. 5.
There are many techniques used to investigate the
characteristics of the ionosphere [2–4]. The propagation
Satellite O
+
H
+
He
+
Ultraviolet
rays
Visible
light
rays
Infrared
rays
IONOSPHERE
600 km
EXOSPHERE
500 km
400 km
300 km
Daytime
200 km
NO
+
O
+
2
100 km
95 km
MESOSPHERE
45 km
10
6
10
5
STRATOSPHERE
10 km
TROPOSPHERE
Electron density
10
4
10
3
10
2
NO
+
Lα
D
E
F
F
2
Gamma rays
X−rays
HF
LF
VHF
MF
n(e)[cm
−3
]
Figure 3. The ionospheric regions and their ionizing agents. (From Ref. 2.)
ELECTROMAGNETIC WAVES IN IONOSPHERE 1305
of radio waves in the ionosphere is described in consider
able detail in Ref. 5, and solar–terrestrial relations and
their effects on radio propagation are covered in Ref. 6.
Radiowave propagation at all frequencies depends to dif
ferent degrees on the geographic and geomagnetic latitu
dinal region of the ionosphere. The most benign
latitudinal region is the midlatitudes; the most disturbed
regions are the auroral, equatorial, and polar regions, as
described in detail in Ref. 4.
In the next section we describe qualitatively the salient
propagation modes and ionospheric effects on radiowaves
as a function of frequency. The section after provides a
mathematical description and the physical principles of
the interaction of radiowaves with the ionosphere. The ﬁ
nal section will introduce the reader to the frontiers
of ionospheric research at the start of the twentyﬁrst
century.
3. EFFECTS ON SPECIFIC RADIO SERVICES
3.1. Extremely Low and Very Low Frequencies (ELF and VLF)
As indicated in Table 1, at the lowest frequencies (ELF–
VLF) the basic propagation mode is a spherical waveguide
mode, with the D and E regions of the ionosphere forming
the upper boundary, and Earth’s surface the lower bound
ary. A simpliﬁed Earth–ionosphere waveguide geometry is
shown in Fig. 6. A schematic diagram of the ﬁrst two
waveguide modes in an ideal earthionosphere mode is
presented in Fig. 7.
In reality, the ELF–VLF waveguide mode is consider
ably more complicated because of its spherical nature and
the electrical characteristics of the upper and lower
boundaries. At ELF frequencies, the wavelength is of the
same order of magnitude as the transverse dimensions of
the waveguide, and the signal propagates deeply into both
land and sea because of the skindepth effect.
At global distances, the signal is very stable, but ex
tremely long antennas and high transmitter powers are
required and the signaling rate is extremely slow. One
unique advantage of ELF is the ability of the signal to
penetrate relatively deeply into seawater (at 100Hz, the
attenuation in seawater is 0.3dB/m, which is %
1
3
of the
Wavelength (m)
10
8
10
7
10
6
10
5
10
4
10
3
10
2
10
1
10
−3
10
−2
10
−1
10
0
10
1
10
2
10
3
10
4
Day
Night
ELF VLF LF
10
−1
10
−2
10
−3
10
−4
Frequency (kHz)
A
t
t
e
n
u
a
t
i
o
n
(
d
B
/
k
m
)
MF HF
Figure 4. Night–day variation of attenuation on radio paths
as a function of frequency from 1 Hz to 30MHz. (Courtesy of
D. Llanwyn Jones.)
Table 1. The Radio Spectrum as Deﬁned by the International Telecommunications Union (ITU); Primary Modes of
Propagation, and Effects of the Terrestrial Ionosphere
ITU Designation Frequency Range Principal Propagation Modes Principal Uses
Extralow frequency (ELF) 30–300Hz Ground wave and Earth–ionosphere
waveguide mode
Submarine communication
Very low frequency (VLF) 3–30kHz Same as above Navigation, standardfrequency and
time dissemination
Low frequency (LF) 30–300kHz Same as above Navigation LORANC
a
Medium frequency (MF) 300–3000kHz Primarily ground wave, but sky
wave
b
at night
AM broadcasting, maritime, aero
nautical communication
High frequency (HF) 3–30MHz Primarily sky wave, some ground
wave
Shortwave broadcasting, amateur,
ﬁxed services
Very high frequency (VHF) 30–300MHz Primarily LoS,
c
some sky wave at
lower VHF
FM broadcasting, television, aero
nautical communication
Ultrahigh frequency (UHF) 300–3000MHz Primarily LoS, some refraction and
scattering by the ionosphere
Television, radar, navigation,
d
aero
nautical communication
Superhigh frequency (SHF) 3–30GHz Same as above Radar, space communication
a
The LORANC system will probably be superseded by the GPS system.
b
Sky wave denotes the Earth–ionosphere–Earth reﬂection mode.
c
Line of sight.
d
Global Positioning System satellite constellation.
Winter
F
F2
F
400
200
400
200
F1
E
0 6 12 18 24
Local time
V
i
r
t
u
a
l
h
e
i
g
h
t
(
k
m
)
0 6 12 18 24
Local time
E
F1
F F
F2
Summer
Figure 5. Average variation of ionospheric layer height as a
function of season and local time. Note the large change in height
of the F2 layer in summer.
1306 ELECTROMAGNETIC WAVES IN IONOSPHERE
attenuation in the Earth–ionosphere waveguide). The
attenuation of ELF signals penetrating normal Earth is
approximately
1
10
th that in seawater, so ELF signals can
also be used to probe or communicate into the solid Earth.
There is also some evidence that highlatitude effects such
as Dregion irregularities can cause some phase anomalies
in ELF and VLF transmissions [7–9].
VLF (3–30 kHz) transmissions also require large an
tennas and highpower transmitters, but are used more
than ELF for time and frequencystandard dissemination
and navigational systems. VLF signals are also inﬂuenced
by anomalies on Earth’s surface [10] and in the high
latitude ionosphere [4].
3.2. Low Frequencies (LF)
Moving up in frequency to the LF band (30–300kHz), the
basic propagation mode below E100kHz is by the ground
(surface) wave, which follows Earth’s curvature, and
above 100kHz is the sky wave and the waveguide mode.
The sky wave is, of course, inﬂuenced by the ionospheric
diurnal, seasonal, and latitudinal variations.
3.3. Medium Frequencies (MF)
Propagation during the daytime in the MF (300–
3000 kHz) band is by ground wave, and for frequencies
above 500kHz, at night by sky wave. At geomagnetic lat
itudes greater than 551, the auroral ionosphere introduces
some anomalous skywave propagation modes [10].
3.4. High Frequencies (HF)
The ionosphere has the most profound effect on signals in
the HF (3–30 MHz) band, making the skywave mode the
dominant means of propagation. The ground wave at HF
is sometimes used in the frequency range of 3–6 MHz, es
pecially over seawater, whose conductivity is much great
er than that of ordinary land. At HF wavelengths
relatively high efﬁciency, gain, and directivity can be
achieved in the antenna systems, so directive communi
cations and broadcasting are realizable. Above 6 MHz, the
sky wave is dominant, so one must really understand
ionospheric behavior and phenomenology in order to pre
dict propagation. Since the ionosphere varies with time of
day, season, solar activity, and sunspot cycle, predicting
HF propagation over a specific path can be somewhat
complicated. Propagation paths up to E10,000 km are
quite common for shortwave (SW) broadcasters, who use
antennas with gains of up to 20 dBi (dBi is gain referenced
to an isotropic source) and transmitter powers of 250kW
and higher. With much less reliability, amateur radio op
erators (hams) sometimes also achieve twoway commu
nications over similar pathlengths using antenna gains of
3–12 dB and transmitter powers of 5–1000 W.
The ionosphere also behaves differently in the equato
rial, midlatitude, auroral, and polar latitudinal regions.
Fortunately, several fairly reliable and easytouse HF
propagation prediction programs are now available to
HF system planners for PCs (ASAPS, VOACAP, PropLab
Pro, etc.). The sources of these programs may be found
in more recent books and articles [4,5,12,13] and in the
amateur radio magazines (QST, CQ, Worldradio, etc.).
Unfortunately, none of the existing prediction pro
grams gives very reliable results at equatorial and high
latitudes. They are based on climatological data, so they
are not intended for forecasting. A following section de
scribes the mathematical essentials of ionospheric propa
gation in considerable detail.
3.5. Very High Frequencies (VHF)
Propagation in the VHF band (30–300 MHz) is primarily
by line of sight (LoS) to the optical horizon, so if the an
tenna patterns direct most of the RF power in the hori
zontal plane, there are essentially no ionospheric effects.
For Earth–space propagation paths, however, the iono
sphere can affect the signal adversely by refraction,
diffraction, scattering, or reﬂection. These effects can be
4h
image
2h
image
Phase front
2 h cos
n
Ray direction
lonosphere z = h
Ground z = 0
−2h
image
Source
Figure 6. Simpliﬁed ray geometry for the ﬁrstorder and second
order VLF–ELF modes. The two conducting planes are represen
tative at the earth surface and the ionosphere.
Perfect reflector
R = +1
R = +1
R = −1
R = +1
Guide wavelength
lonosphere
R = −1
lonosphere
Perfect reflector
p
p
Secondorder (TM
02
) mode
E
H
e
i
g
h
t
H
e
i
g
h
t
E
V
E
V
E
H
E
H
Firstorder (TM
01
) mode
E
Figure 7. The E ﬁeld for ideal Earth–ionosphere waveguide
modes.
ELECTROMAGNETIC WAVES IN IONOSPHERE 1307
especially important when the path traverses the equato
rial, auroral, and polar ionosphere. The amplitude,
phase, and polarization of the signal may change measur
ably. These effects will be quantiﬁed in the following
section.
3.6. Ultra High Frequencies (UHF)
At UHF and above, propagation is primarily LoS, and
because of the higher frequencies (fZ300MHz), these
signals are less affected by the ionosphere than lower
frequencies. On Earth–space paths that traverse the
equatorial and/or highlatitude ionosphere, however, the
signal quality can be significantly degraded at certain
times. At ‘‘wireless’’ frequencies (fE800MHz–20GHz) so
lar radio bursts can sometime seriously degrade certain
wireless systems. Figure 8 shows the variation of peak
ﬂux density of solar radio noise with frequency and time,
and Fig. 9 shows the total number of events per day as a
Freq. range: 1 − 10 GHz
Solar cycle: 20 : 23 (1960−1999)
(a)
10
3
10
4
10
5
10
6
Solar cycle: 20 (1964−1975)
(c)
10
3
10
4
10
5
10
6
Solar cycle: 21 (1976−1985)
(e)
10
3
10
4
10
5
10
6
Solar cycle: 22 (1986−1995)
(g)
100.00
10.00
1.00
0.10
0.01
100.00
10.00
1.00
0.10
0.01
100.00
10.00
1.00
0.10
0.01
100.00
10.00
1.00
0.10
0.01
10
3
10
4
10
5
10
6
Solar cycle: 22 (1986−1995)
(h)
10
3
10
4
10
5
10
6
Solar cycle: 21 (1976−1985)
(f)
10
3
10
4
10
5
10
6
Solar cycle: 20 (1964−1975)
(d)
10
3
10
4
10
5
10
6
Solar cycle: 20 : 23 (1960−1999)
(b)
100.00
10.00
1.00
0.10
0.01
100.00
10.00
1.00
0.10
0.01
100.00
10.00
1.00
0.10
0.01
100.00
10.00
1.00
0.10
0.01
10
3
10
4
10
5
10
6
Freq. range: 10 − 20 GHz
Peak flux density (SFU)
N
u
m
b
e
r
o
f
e
v
e
n
t
s
/
d
a
y
/
b
i
n
*
1
0
0
0
Figure 8. Number of events per day with amplitudes 410
3
SFU (a,b) for 40 years, (c,d) for cycle
20, (e,f) for cycle 21, and (g,h) for cycle 22. The frequency ranges are 1–10GHz (excluding 10GHz)
shown in the left panels and 10–20GHz (excluding 20GHz) in the right panels. Here, the number
of events per day per bin is multiplied by 1000.
1308 ELECTROMAGNETIC WAVES IN IONOSPHERE
function of time in years for two different solar ﬂux
levels [14].
In a 2002 paper [15], the ionospheric effects on satellite
land–mobile services tabulated the estimated maximum
ionospheric effects on these services as shown in Table 2.
The complete report may be seen at http://www.utex
as.edu/mopro/.
To summarize, the radio services most affected by the
ionosphere lie in the frequency range of 1–150MHz [ﬁxed
communication services, AM (amplitude modulation) and
SW (shortwave) broadcasting, amateur radio]. To a lesser
degree, services in the 20 kHz–300MHz region (mainly
some of the navigation services) also suffer some iono
spheric perturbation effects.
There is a plethora of radio instrumentation currently
deployed globally that operates routinely or on a campaign
basis to measure characteristics of the terrestrial iono
sphere. It is beyond the scope of this article to describe
these techniques, but they have been described in consid
erable detail in the literature [2–4].
4. PHYSICAL PRINCIPLES AND MATHEMATICAL
DESCRIPTION OF ELECTROMAGNETIC INTERACTION
WITH THE IONOSPHERE
Because of the complexity of the terrestrial ionosphere (a
weakly ionized plasma with a superimposed magnetic
ﬁeld in which electric currents ﬂow), we must utilize the
magnetoionic theory to quantify the ionosphere physical
parameters. The most successful formulation of the ap
propriate magnetoionic theory was derived by Appleton
and others in the mid1920s [15–18]. We can obtain some
ﬁrstorder properties of the ionosphere by ignoring the
magnetic ﬁeld [19,20]. A simple dispersion equation for
electromagnetic (EM) waves in the ionosphere is
m ¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
1 À
Ne
2
pmf
2
¸
ð1Þ
where
m ¼refractive index of the ionosphere (real part of com
plex refractive index n)
N ¼electron number density of the ionosphere (electrons/
cm
3
or electrons/m
3
)
e ¼electronic charge ¼1.6Â10
À19
C
m¼mass of the electron¼9.1 Â10
À31
kg
f ¼frequency of the radiowave in the ionosphere (Hz)
For reﬂection at vertical incidence, m ¼1 and
N¼
mpf
2
e
2
¼1:24 Â10
4
f
2
ðelectrons=cm
3
and f inMHzÞ
¼1:24 Â10
10
f
2
ðelectrons=cm
3
and f inMHzÞ
ð2Þ
Table 2. Estimated Maximum Ionospheric Effects (Elevation Angle of 301, OneWay Traversal)
Effect Frequency Dependence 0.1 GHz 0.25GHz 0.5GHz 1 GHz 3 GHz 10GHz
Faraday rotation f
À2
30 rot.
a
4.8 rot. 1.2 rot. 1081 121 1.11
Time (group) delay f
À2
25ms 4 ms 1 ms 0.25ms 0.028ms 0.0025ms
Refraction f
À2
o11 o0.161 o2.4
0
o0.6
0
o4
00
o0.36
00
Change in direction of arrival f
À2
20
0
3.2
0
48
00
12
00
1.33
00
0.12
00
Absorption (auroral and/or polar cap) f
À2
5 dB 0.8 dB 0.2 dB 0.05dB 0.006dB 0.0005dB
Absorption (midlatitude) f
À2
o1dB o0.16dB o0.04dB o0.01dB o0.001dB o0.0001dB
Dispersion f
À3
400ps/kHz 26ps/kHz 3.2 ps/kHz 0.4 ps/kHz 45 ts/kHz 0.4 ts/kHz
a
Rotations.
Note: Extrapolated from 1GHz values given by the former CCIR Study Group 6 in Table VII of Report 263, and based on a total electron content (TEC) of 10
18
electrons/m
2
, which is a high value of TEC encountered at low latitudes in daytime with high solar activity. A similar chart may be found in Recommendation
ITUR P.6186 of Study Group III.
Peak flux > 1000.SFU
0.6
0.4
0.2
0.0
4001 : 10000
2001 : 4000
1000 : 2000
1960 1965 1970 1975 1980 1985 1990 1995 2000
(a)
(b)
Peak flux > 10000.SFU
0.10
0.4
0.2
0.0
4001 : 10000
2001 : 4000
1000 : 2000
1960 1965 1970 1975 1980 1985 1990 1995 2000
0.06
0.08
Year
N
u
m
b
e
r
o
f
e
v
e
n
t
s
/
d
a
y
Figure 9. Total number of events/day as a function of time in
years (a) for peak ﬂux 410
3
SFU and (b) for peak ﬂux 410
4
SFU.
The frequency ranges are 1–2, 2–4, and 4–10GHz.
ELECTROMAGNETIC WAVES IN IONOSPHERE 1309
Another useful quantity is the plasma frequency,
F
n
¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃ
Ne
2
pm
_
¼9
ﬃﬃﬃﬃﬃ
N
p
ðf inkHz andNinelectrons=cm
3
Þ
¼9 Â10
À3
ﬃﬃﬃﬃﬃ
N
p
ðf inMHz andNinelectrons=cm
3
Þ
ð3Þ
4.1. The Virtual Height Concept
If we consider an RF pulse traveling vertically upward
into the ionosphere at the speed of light, v ¼c, it will be
reﬂected at the virtual height, h
0
. The time required for
the pulse to be reﬂected from an ionospheric layer and
return to Earth is
t ¼
2
c
_
h
0
dz
m
ð4Þ
then the virtual height can be found from h(f) ¼
1
2
ct, or
h
0
ðf Þ ¼
_
h
0
dz
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
1 Àf
2
n
=f
2
_ ð5Þ
Since the pulse always travels more slowly in the layer
than in free space, the virtual height of a layer is always
greater than the true height. The true height and virtual
height are related by the integral equation
h
0
ðf Þ ¼
_
Zmax
0
dz
mðf ; zÞ
ð6Þ
where z is the true height, Z
max
is the maximum height
reached by the frequency f, and n is the refractive index at
Z
max
for the frequency f. A good discussion of the relation
between true height and virtual height is given in Ref. 20.
4.2. Vertical and Oblique Propagation
Before considering the behavior of a radio signal in a mag
netoionic medium, we will state three theorems that relate
oblique and vertical incidence propagation as depicted in
Fig. 10. The ﬁrst is the secant law, which relates the ver
ticalincidence frequency f
v
reﬂected at B to the oblique
incidence frequency f
ob
reﬂected at the same true height.
A typical derivation of this relation is given in Ref. 5, and
it is usually written as
f
ob
¼f
v
sec j
0
ð7Þ
The secant law, then, relates the two frequencies f
v
and f
ob
reﬂected from the same true height (the distance BD in
Fig. 10).
In order to determine sec j and f
ob
values from vertical
incidence soundings (which measure the virtual height
h
0
), we need two more theorems. Breit and Tuve’s theorem
states that the time taken to traverse the actual curved
path TABCR in Fig. 10 at the group velocity v
g
equals the
time necessary to travel over the straightline path TER
at the freespace velocity c. Referring to the geometry
shown in Fig. 10, we can write the expression
T¼
1
c
_
TER
dx
sin j
0
¼
D
c
sin j
0
¼
TEþER
c
ð8Þ
Martyn’s theorem may be written concisely as
h
0
¼h
0
v
ð9Þ
Smith [21] devised a set of logarithmic transmission
curves, parametric in range, for the curved earth and ion
osphere. They are shown in Fig. 11 and are sufﬁciently
accurate for the distances shown.
4.3. Radio Propagation in a Magnetized Plasma
Before proceeding with a discussion of the Appleton (mag
netoionic) equations, we need to deﬁne two quantities con
tained explicitly in the equations. The ﬁrst is n, the number
of collisions per second (collision frequency) between elec
trons and heavier particles (ions and neutrals). Another
quantity, the gyromagnetic frequency or gyrofrequency, is
the natural frequency (Hz) of gyration of an ion or electron
in a magnetic ﬁeld of strength B
0
(Wb/m
2
) and is given by
f
H
¼
e j j
2pm
B
0
% 2:80Â10
10
B
0
ð10Þ
and the angular gyrofrequency is given by
o
H
¼
e j j
m
B
0
% 1:76Â10
11
B
0
ð11Þ
Since electrons are much less massive than ions, the elec
tron gyrofrequency affects the propagation of HF waves in
the ionosphere more than the ion gyrofrequencies. For ex
ample, since BE0.5Â10
À4
Wb/m
2
, the electron gyrofre
quency is E1.40MHz, which falls at the upper end of the
medium waveband.
E
C
S
A
O
O
F
T R
D
B
D
Figure 10. Plane geometry describing vertical and oblique iono
spheric propagation.
1310 ELECTROMAGNETIC WAVES IN IONOSPHERE
4.4. The Dispersion Relation
Using the recommended URSI (International Union at
Radio Science) notation, the magnetoionic dispersion
equation for a radiowave in a homogeneous, partially ab
sorbing ionized gas on which a constant magnetic ﬁeld is
impressed is given by
n
2
¼1 À
X
ð1 ÀjZÞ À
Y
2
T
2ð1 ÀX ÀjZÞ
_ _
Æ
Y
4
T
4ð1 ÀX ÀjZÞ
1=2
þY
2
L
_ _
1=2
ð12Þ
where
n ¼complex refractive index ¼(m ÀjX)
o ¼angular frequency of the exploring wave (rad/s)
o
N
¼angular plasma frequency
o
H
¼angular gyrofrequency ¼B
o
e/m (rad/s)
o
L
¼longitudinal angular gyrofrequency ¼(B
o
e/m)
cos y
o
T
¼transverse angular gyrofrequency ¼(B
o
e/m) siny
X ¼o
N
2
/o
2
Y ¼o
H
/o
Y
L
¼o
L
/o
Y
T
¼o
T
Z ¼n/o
y ¼angle between the wavenormal and the magnetic
ﬁeld inclination
4.5. The Polarization Relation
We begin by deﬁning the polarization ratio R as
R¼
ÀH
y
H
y
¼
E
x
E
y
ð13Þ
Then we can write the doublevalued polarization equa
tion as
R¼
À
j
Y
L
1
2
Y
2
T
ð1 ÀX ÀjZÞ
Ç
1
4
Y
2
T
1 ÀX ÀjZ ð Þ
2
þY
2
L
_ _
1=2
_ _
ð14Þ
In the upper F region of the ionosphere where the elec
tron–ion collision frequency is very low, we may simplify
the dispersion and polarization equations by dropping the
Z term (since nE0). Equations (12) and (14) then become
(for no absorption)
n
2
¼1 À
2Xð1 ÀXÞ
2ð1 ÀXÞ ÀY
2
T
Æ½Y
4
T
þ4Y
2
L
ð1 ÀXÞ
2
1=2
ð15Þ
and
R¼ À
H
y
H
x
¼ À
j
Y
L
1þ
X
n
2
À1
_ _
ð16Þ
If we further simplify Eq. (12) by dropping the Y terms (no
magnetic ﬁeld), then we obtain n
2
¼1ÀX, which is equiv
alent to Eq. (1).
According to magnetoionic theory, a planepolarized
EM wave traveling in a medium like the terrestrial ion
osphere will be split into two characteristic waves. The
wave that most closely approximates the behavior of a
signal propagating in this medium without an imposed
magnetic ﬁeld, is called the ordinary wave, and the other
is called the extraordinary wave. These terms are taken
from the nomenclature for double refraction in optics, al
though the magnetoionic phenomena are more complicat
ed than the optical ones. The ordinary wave is
represented by the upper sign in the polarization
Eq. (14), except when the wavenormal is exactly along
the direction of the magnetic ﬁeld. Anomalous absorption
800
700
600
500
400
300
200
100
0
12 16 20 24 28 32 36 40 44 48 56 70 80
Angle of departure (deg)
1
0
0
km
2
0
0
km
3
0
0
k
m 4
0
0
k
m
5
0
0
k
m
6
0
0
k
m 8
0
0
k
m
1
0
0
0
k
m
1
2
0
0
k
m
1
4
0
0
k
m
1
6
0
0
k
m
1
8
0
0
k
m
2
0
0
0
k
m
2
5
0
0
k
m
3
0
0
0
k
m
3
5
0
0
k
m
4
0
0
0
k
m
6.0 5.0 4.0 3.0 2.0 1.0
Sec (corrected)
V
i
r
t
u
a
l
h
e
i
g
h
t
(
k
m
)
60 52
Figure 11. Logarithmic transmis
sion curves for curved Earth and ion
osphere, parametric in distance
between transmitter and receiver.
ELECTROMAGNETIC WAVES IN IONOSPHERE 1311
occurs for the extraordinary wave when its frequency
equals the electron gyrofrequency (f
H
¼Be/meE0.8–
1.6 MHz). These frequencies lie in the mediumfrequency
(MF) band; consequently the absorption of the extraordi
nary wave [AE(f Àf
H
)
2
] is large and the polarization of
the transmitted wave is important in the determination of
the fraction of the incident power that goes into the ex
traordinary wave. This is especially true near the mag
netic dip equator, where the magnetic ﬁeld is nearly
horizontal.
In addition to anomalous absorption effects near the
electron gyrofrequency, the wave may also experience
significant lateral deviation. This is illustrated for verti
cal and oblique propagation in Sections 11.2.2–11.2.4 of
Ref. 4.
If Eq. (16) is recast as a function of o and we deﬁne
f ðyÞ ¼
1
2
(sin
2
y)/cos y and o
c
¼(B
0
e/m) f(y), then it will be
seen to describe an ellipse. The quantities f(y) and o
c
play
an important part in the description of the polarization
behavior of waves in magnetoionic theory. The magnitude
of o
c
is independent of frequency, but varies with the angle
between the wavenormal and the magnetic ﬁeld y, where
as the sign of o
c
depends on the sign of the charge e and
the direction of the magnetic ﬁeld. For longitudinal prop
agation o
c
¼0, and for transverse propagation o
c
N. In
the case where X¼1, the quantity o
w
primarily deter
mines the polarization of the wave. A very complete dis
cussion of R as a function of X and of the variation of the
polarization ellipse is given in Ref. 22.
A more complete understanding of the behavior of EM
waves in the terrestrial ionosphere may be obtained by
employing two approximations. The quasilongitudinal
(QL) approximation applies when the wave is propagat
ing nearly parallel to the geomagnetic ﬁeld, and the qua
sitransverse (QT) approximation applies when the wave
propagates in a direction nearly normal to the geomag
netic ﬁeld. References 22 and 23 contain extended discus
sions of the QL and QT approximations:
QT : Y
4
T
b4ð1 ÀXÞ
2
Y
2
L
QL : Y
4
T
54ð1 ÀXÞ
2
Y
2
L
4.6. Absorption of Radiowaves in the Ionosphere
The refractive index n is modiﬁed when one introduces
collisions between the electrons and heavy particles, and
the wave then experiences absorption, which physically is
due to the conversion of ordered momentum into random
motion of the particles after collision. For each collision,
some energy is transferred from the EM wave to the neu
tral molecules and appears as thermal energy. We
will follow the standard treatment of absorption of radio
waves in the ionosphere presented by Davies [23] and
Budden [24].
For the propagation of an EM wave in an unmagne
tized plasma, we can deﬁne the absorption index (or coef
ﬁcient) as
K ¼
o
c
X ð17Þ
where X is the imaginary part of the refractive index n.
For a magnetized plasma without collisions, we can write
K ¼
e
2
2e
0
mcm
.
Nn
o
2
þn
2
ð18Þ
On this basis, we can conveniently divide absorption into
two limiting types, commonly called nondeviative and
deviative absorptions. Nondeviative absorption occurs in
regions where the product Nn is large and mE1, and is
characterized by the absorption of HF waves in the D re
gion. Deviation absorption, on the other hand, occurs near
the top of the ray trajectory or anywhere else on the ray
path where significant bending takes place (for small Nn
and mo1).
When the refractive index E1, there is essentially no
bending of the ray and we can write
K % 4:6 Â10
À2
Nn
mðo
2
þn
2
Þ
dB=km ð19Þ
We can further simplify Eq. (19) for the VHF case, since
o
2
bn
2
; as
K ¼1:15 Â10
À3
Nn
f
2
dB=km ð20Þ
In the MF and HF bands, Eq. (19) may be written as
K ¼4:6 Â10
À2
N
n
dB=km ð21Þ
Unlike nondeviative absorption, deviative absorption oc
curs when the wave experiences significant group retar
dation and consequently spends a relatively long time in
the absorbing layer and there is considerable curvature of
the ray path. The general expression for the absorption
index in a deviating region of a nonmagnetic plasma is
K ¼
n
2cm
ð1 Àm
2
ÀX
2
Þ ð22Þ
In the ionosphere, Eq. (22) reduces to
K ¼
n
2c
m
0
ð23Þ
where m
0
is the group refractive index. For large values of
m
0
, we can write the preceding equation as
K ¼
n
2c
X
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
1 ÀX
p ð24Þ
We should remember that the concepts of deviative and
nondeviative absorption are limiting cases, and that as a
wave approaches the reﬂecting level, ray theory breaks
down, so we must employ full wave theory to obtain a
complete description of the behavior of the wave. Extend
ed discussions of application of the QL and QT approxi
mations to ionospheric absorption may be found in Refs. 5
and 23.
1312 ELECTROMAGNETIC WAVES IN IONOSPHERE
4.7. Scattering of Radiowaves in the Ionosphere
The principles of scattering of radiowaves in general are
discussed in the articles on electromagnetic wave scatter
ing in this encyclopedia. One can qualitatively describe
ionospheric scattering as either strong or weak in terms of
the received signal strength of the scattered signal at the
receiving radar antenna. An example of the former is
VHF–UHF backscatter echoes received from electron den
sity gradients in the auroral E region, and an example of
the latter is incoherent backscatter received by a VHF–
UHF radar from the undisturbed E or F layer.
Another way of classifying scattered echoes is in terms
of their backscatter cross section (using a pulsed radar
system) and their temporal stability. A coherent echo
exhibits a statistical correlation of the amplitude and
phase from one pulse to another and emanates from qua
sideterministic gradients in electron density, which have
correlation times usually greater than 1 ms, correspond
ing to a spectral width of the radar echo of less than
1000 Hz (sometimes less than 100Hz). It also has a back
scatter cross section 10
4
–10
9
times greater than that from
an incoherent echo. Other important considerations in
the case of coherent backscatter are the relation between
the scattering irregularity size relative to the backscatter
sounder freespace wavelength, the mean fractional
deviation in electron density of the scatterer, and the
aspect angle between the radar line of sight and the
major axis of the irregularity. On the other hand, an in
coherent echo arises from random thermal ﬂuctuations in
the ionosphere, which have typical correlation times of
E20 ms, corresponding to a radar echo spectral width
of E50 kHz.
The physical principles governing coherent and
incoherent scattering from the ionosphere are covered
in Refs. 2, 5, and 6, while plasma wave theory is covered
in detail in Ref. 5, and extended descriptions of tech
niques for studying the ionosphere using coherent and
incoherentscattering sounders are given in Refs. 2,3,5,
and 6.
Because of charged particle precipitation of solar ori
gin, ionospheric electric currents and ﬁelds, and plasma
dynamics, there exists a wide spectrum of scale sizes of
ionospheric irregularities, as shown in Fig. 12. Irregular
ities are most prevalent at auroral, polar, and equatorial
latitudes, although they also exist at midlatitudes [23].
The global morphology of ionospheric irregularities is cov
ered in Refs. 4–6.
4.8. Ionospheric Scintillation
Ionospheric scintillations are ﬂuctuations of amplitude,
phase, and angle of arrival of a VHF–UHF signal passing
through irregularities located mainly in the F region.
Ionospheric scintillations can have deleterious effects on
satellitebased communication and navigation systems.
Either extragalactic sources (such as radio stars) or sat
ellite beacon transmitters may be used as the signal sourc
es for Earthobserved studies of ionospheric scintillations,
and both geostationary and orbiting satellite beacons have
been used. There is a voluminous body of literature since
1970 describing the theory, technique, and results of iono
spheric scintillation measurements [2–6].
4.9. Faraday Rotation
One physical principle that makes possible the determi
nation of ionospheric columnar electron content is Fara
day rotation. This effect (for optics) was discovered by
Michael Faraday in 1845, when he subjected a block of
glass to a strong magnetic ﬁeld. He observed that a plane
polarized monochromatic beam of light passing through
the glass in a direction parallel to the imposed magnetic
ﬁeld has its plane of polarization rotated. The amount of
rotation is given by the expression
O¼KHl ð25Þ
where O is the angle of rotation, K is a constant associated
with each substance, l is the pathlength of light through
the substance (m), and H is the magnetic ﬁeld intensity
(A/m).
The Faraday rotation of the electric vector of a radio
wave (see Ref. 45) propagating from a satellite radio bea
con in a direction parallel to Earth’s magnetic ﬁeld (as
seen by an observer looking up, in the Northern Hemi
sphere) is counterclockwise, as shown in Fig. 13.
Horizontal
scale (km)
Scale ⊥
magnetic field (m)
1000 100 10 1 100 10 1 0.1 0.01
Wandering of normal
to ionosphere
Multiple normal
to ionosphere
Phase
scintillation
Aggregate
of TIDs
(gravitationally
anisotrophic)
Amplitude
scintillation
Plasma turbulence
(magnetically anisotropic)
Blur on
ionograms
Strong back
scattering
and trans
equatorial
propagation
at VHF
Plasma waves
(magnetically
anisotropic)
10
2
10 1 10
−1
10
−2
10
−3
10
−4
10
−5
10
−6
Wave number (m
−1
)
R
a
d
i
u
s
o
f
e
a
r
t
h
A
t
m
o
s
p
h
e
r
i
c
s
c
a
l
e
h
e
i
g
h
t
L
o
g
(
s
p
e
c
t
r
a
l
d
e
n
s
i
t
y
o
f
i
r
r
e
g
u
l
a
r
i
t
i
e
s
) I
o
n
g
y
r
o
r
a
d
i
u
s
E
l
e
c
t
r
o
n
g
y
r
o
r
a
d
i
u
s
Figure 12. Composite spectrum of ionospheric irregularities as a
function of wavenumber over a large spatial scale. (Courtesy of
H. G. Booker.)
ELECTROMAGNETIC WAVES IN IONOSPHERE 1313
Ignoring refraction, the Faraday rotation of the electric
vector is given by
O¼
pf
2c
_
S
R
X
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
Y
4
T
þ4ð1 ÀXÞ
2
Y
2
L
ð1 ÀXÞð1 ÀY
2
L
ÀY
2
T
Þ
¸
ds ð26Þ
where
O ¼Faraday rotation (rad/s)
f ¼wave frequency (Hz)
c ¼2.998 Â10
8
m/s
X ¼kN/f
2
k ¼80.61
N¼electron density (electrons/m
3
)
and Y
L
,Y
T
are as previously deﬁned.
The integration is between the receiver R, and the sat
ellite S. For VHF frequencies the QL approximation holds
and we can express Eq. (26) as
O¼
pK
cf
2
_
S
R
f
L
Nds ð27Þ
where Nds is the ionospheric electron content. Evaluating
the constants yields the relation
O % 8:447Â10
À7
f
À2
_
S
R
f
L
Nds rad ð28Þ
where f
L
¼2.80Â10
10
B
L
, the electron gyrofrequency cor
responding to the longitudinal component of the geomag
netic ﬁeld along the ray path. Details of the application of
Faraday rotation theory and other techniques to deduce
ionospheric columnar electron content may be found in
Refs. 2, 5, and 6.
4.10. Whistlers
Whistlers are bursts of EM radiation at VLF that are ini
tiated by lightning discharges and then travel though the
ionosphere and magnetosphere in ducts approximately
parallel to geomagnetic lines of force. When translated
into sound waves, whistlers are distinguished by tones of
decreasing (or sometimes increasing) frequency, and they
may easily be detected by connecting a suitable antenna to
the input of a very sensitive audio ampliﬁer. In fact, whis
tlers were ﬁrst observed in the last years of the nineteenth
century, and were also heard on the primitive ﬁeld tele
phone systems used in World War I. They have been stud
ied intermittently since 1898, basically as a diagnostic
probe of the ionosphere and magnetosphere [2,26]. A
graphical representation of whistler behavior is shown
in Fig. 14, and the somewhat rarer nose whistler behavior
is illustrated in Fig. 15.
The dispersion relation for whistlers is
T ¼
1
2c
_
S
ðf
N
f
L
dsÞ
f
1=2
ðf
L
Àf Þ
3=2
¼
D
f
1=2
ð29Þ
Impulse Whistler
(a)
f
6000
4000
2000
0
f
(
H
z
)
Time
(b)
0.05
0.04
0.03
0.02
0.01
0
1
/
f
(
H
z

1
/
2
)
√
0 0.5 1.0 1.5 2.0 2.5
Time (s)
(c)
F
r
e
q
u
e
n
c
y
Lightning
(d)
Time
2 hop
4 hop
6 hop
8 hop
Figure 14. Sketch of basic manifestations of a whistler and its
initiating disturbance: (a) the frequency spectrum; (b) frequency–
time curve of a typical whistler; (c) curve of
ﬃﬃﬃﬃﬃﬃﬃﬃ
1=f
_
with time. Ini
tiating disturbance and multiple hops when the source and re
ceiver are at the same end of a magnetic line of force. (After
Helliwell [26].)
F
r
e
q
u
e
n
c
y
t
min
Time
Figure 15. Idealized sketch of the frequency–time characteris
tics of a nose whistler. (After Davies [5].)
Electric vector
Magnetic
field
Satellite
antenna
Horizontal
Ω
Figure 13. Simpliﬁed plane geometry of satellite–Earth propa
gation path in the Northern Hemisphere to explain Faraday ro
tation effects.
1314 ELECTROMAGNETIC WAVES IN IONOSPHERE
where D is dispersion ¼ (1/2c)
_
s
ðf
N
=f
1=2
L
Þ ds, f
N
is plasma
frequency, and f
L
is a longitudinal component of the plas
ma frequency. This is the time T for a signal burst to go
from one hemisphere to its conjugate point in the opposite
hemisphere.
Other natural VLF emissions (called ‘‘dawn chorus,’’
‘‘risers,’’ ‘‘hiss,’’ etc.) that are thought to originate in the
ionosphere can also be heard on whistler detection equip
ment. Since the 1960s, highpower VLF transmitters have
been used to generate whistlers to study properties of the
magnetosphere [27–29].
5. FRONTIERS OF IONOSPHERIC RESEARCH
The use of radiowaves to explore the terrestrial iono
sphere began with the pioneering efforts of Appleton and
of Breit and Tuve in 1926, when they independently used
different techniques to detect the ionospheric layers. Their
work was founded on Marconi’s demonstration of transat
lantic radio transmission and on the hypotheses of Ken
nelly and Heaviside, who independently in 1922
postulated that there must be radioreﬂecting layers in
the upper atmosphere to explain certain experimental re
sults. The foregoing discoveries rested on the bedrock of
the experimental and theoretical work of Heinrich Hertz
(1893) and James Clerk Maxwell (1873), respectively.
There seems to have been several crests in the history
of ionospheric research: (1) in the 1920s, following World
War I; (2), starting shortly after the end of World War II;
and (3) perhaps starting in the mid1970s with the advent
of digital techniques, and (4) more recently with the ad
vent of the National Space Weather Program (see Ref. 30
or http://www.geo.nsf.gov/atm/).
While much of the ionospheric research up until about
1960 was in support of HF communications, the advent of
satellite communications changed the emphasis to iono
spheric research concerning the effects of the ionosphere
on transionospheric propagation and research relating the
ionosphere to the magnetosphere. Most current ionospher
ic research is related to the interrelationship and coupling
between regions over the entire height region of the ter
restrial atmosphere from the troposphere to the magneto
sphere and through interplanetary space to the sun.
There are several areas of ionospheric research that
currently seem to be producing exciting new results, and
these areas will probably continue to be emphasized in the
twentyﬁrst century. These areas include (not necessarily
in order of importance) satellite sensors, ionospheric mod
iﬁcation by using highpower HF transmissions, iono
spheric imaging, coherent radars operating from HF
through VHF, and incoherent scatter radars [2–5]. Most
of these techniques are employed at high geomagnetic lat
itudes as part of the Space Weather Program [31], but
some are also deployed in equatorial regions. We will
briefly describe the essentials of each of these areas of
current ionospheric research emphasis.
5.1. Satellite Sensors
The total number of currently operational solarterrestrial
environment satellite probes is too numerous to list in this
article, but examples of representative satellite sensors
may be found at http://www.nasa.gov, and http://www.e
sa.int. These satellites probe the solar physical properties,
the solar wind, the interplanetary magnetic ﬁeld, and the
terrestrial magnetosphere and ionosphere and provide
realtime data to aid in forecasting and describing solar–
terrestrial conditions.
5.2. Ionospheric Modiﬁcation
In the late 1960s the availability of military surplus equip
ment such as veryhighvoltage and current power sup
plies and HF vacuum tubes capable of many kilowatts of
RF output, together with advances in antenna array the
ory and practice, induced experimenters to design systems
to heat or otherwise modify the ionosphere. As a result of
experiments performed in the early 1970s at the Plattev
ille, Colorado HF highpower heating facility [32], some 10
new ionospheric modiﬁcation facilities were established
and have produced significant information on ionospheric
physics. (see Chapter 14 of Ref. 5). The various modiﬁca
tion facilities are listed in Table 3. Other ionospheric mod
iﬁcation facilities are located in Russia and Ukraine at
Kharkov, Moscow, Zimenki, and Monchegorsk. More in
formation on the HAARP and other heaters may be ob
tained on the Internet at http://www.haarp.alaska.edu.
5.3. Ionospheric Imaging by Radio
For over three decades now (as of 2003), ionospheric sci
entists have investigated using radio methods to image
the ionosphere. Rogers [33] was probably the ﬁrst to sug
gest using the wavefront reconstruction method for this
purpose. Many attempts have been made to produce ho
lographic images of the ionosphere, but it has not proved
to be a very successful technique––probably because of the
difﬁculty in uniformly illuminating a sufﬁciently large
Table 3. Ionospheric Modiﬁcation Facilities (1970–1978)
Facility First Used Latitude Longitude
Geomagnetic
Latitude
Transmit
Power
Frequency Range
(MHz)
Antenna
Gain (dB)
Platteville, CO 1970 40.21N 104.71W 491 1.6MW 2.7–25 18
Arecibo, PR 1980 181N 671W 321 800kW 3–15 25
SURA, Russia 1980 56.11N 46.11E 711 750kW 4.5–9 26
Tromsoe, Norway 1980 69.61N 19.21E 671 1.5MW 2.5–8 28
HIPAS, Alaska 1977 64.91N 146.81W 651 800kW 2.8, 4.5 17
HAARP,
a
Alaska 1997 62.41N 145.21W 621 3.6 MW
a
2.8–10 30
a
a
HAARP is currently under construction. Values given are for the completed facility.
ELECTROMAGNETIC WAVES IN IONOSPHERE 1315
horizontal slab of the ionosphere, not using a sufﬁcient
number of receivers, and the inability to achieve precise
enough measurements of amplitude and phase of the re
ﬂected wave.
On the other hand, another technique (borrowed from
medical technology), computerized ionospheric tomo
graphy (CIT), has produced quite significant results in
imaging the regular (and some irregular) features of the
ionosphere. Basically, this technique utilizes radio bea
cons on satellites in nearpolar orbits and a latitudinal
chain on the Earth subsatellite path of carefully calibrated
TEC receivers, to make many measurements of total elec
tron content (TEC). The basic geometry is illustrated in
Fig. 16.
Currently, VHF–UHF beacons on the TRANSIT,
GLONASS, and GPS satellites are the most used as sig
nal sources to measure TEC to use in CIT ionosphere
reconstructions. One must also use some a priori informa
tion (ionospheric models) and ionosonde data in the algo
rithms in order to achieve realistic results. Some recent
results are summarized in Refs. 34–39 and on the Internet
at http://www.nwraaz.com/iitc, and at http://sideshow.
jpl.nasa.gov:80/gpsiono.
Another ionospheric imaging technique is the IRIS sys
tem (imaging riometer–ionospheric studies) [46], which
uses an antenna array of up to 64 elements to provide
images of enhanced auroral absorption structure in the D
region.
5.4. Coherent Radars
As described in the Section 4.7, HF–UHF coherent back
scatter from ionospheric irregularities can provide very
useful information on the morphology and physics of a
wide range of irregularity scale sizes. At this time there
are about 20 of these backscatter sounders deployed, op
erating on frequencies from 8 to 200MHz, distributed
mainly in the highlatitude and equatorial regions. These
radars are sited so that the main antenna lobe is directed
to intercept irregularities at nearnormal incidence at
E and Fregion heights.
The HF coherent radars are mainly grouped into a
large network, which covers approximately half of the
northern polar cap ionosphere––the SuperDARN network
[40], which is shown on the map in Fig. 17. Much infor
mation has been gained on the Fregion plasma convection
patterns in the polar cap, atmospheric gravity waves, and
other ionospheric phenomena related to ionosphere mag
netosphere interaction; see Refs. 40–43 or http://sdwww.
40
45 50
55
Latitude (deg)
Figure 16. Basic satellite–Earth geometry for computer iono
spheric tomography, illustrating the multiple ray paths on which
the total electron content is measured.
C A
B
T
K
G
E F
Figure 17. Map of Northern Hemisphere showing
coverage of SuperDARN highfrequency radars.
1316 ELECTROMAGNETIC WAVES IN IONOSPHERE
jhuapl.edu/RADAR. The VHF–UHF coherent radars are
documented in Refs. 2–5, and a useful Internet sources is
to be found at http://www.dan.spagency.ca/www/
canopus_home.html.
5.5. Incoherent Scatter Radars
One of the most powerful Earthbased radio methods for
studying the ionosphere is the incoherent scatter radar
(ISR) technique, which has been in use since the early
1960s. ISRs can reveal the electron density, electron and
ion temperature, plasma velocity, and other ionospheric
parameters, even during very disturbed conditions [2–5].
At the present time there are some seven ISRs in opera
tion, located from the north polar cap to the magnetic
equator and spread longitudinally from Scandinavia to
Japan. The newest ISR is located at Svalbard, Norway
[44]. Data from ISRs are essential in studying the relation
between the magnetosphere, ionosphere, and middle at
mosphere at high and equatorial latitudes [47,48].
5.6. Large–Scale Ionospheric Data Assimilation
At this time there is a staggering amount of solar terres
trial and ionospheric data available from many diverse
sources––especially from the ‘‘Space Weather’’ sensors,
and there are several sophisticated ionospheric computer
models [49].
BIBLIOGRAPHY
1. H. Alfve´n, Cosmic Plasma, D. Reidel, Dordrecht, Holland,
1981.
2. R. D. Hunsucker, Radio Techniques for Probing the Terrestrial
Ionosphere, SpringerVerlag, Heidelberg, 1991.
3. R. D. Hunsucker, A review of ionospheric radio techniques:
Present status and recent innovations, in W. R. Stone, ed.,
Review of Radio Science 1990–1992, Oxford Univ. Press, Ox
ford, 1993, Chapter 22.
4. R. D. Hunsucker and J. K. Hargreaves, The HighLatitude
Ionosphere and Its Effects on Radio Propagation, Cambridge
Univ. Press, Cambridge, UK, 2003.
5. K. Davies, Ionospheric Radio, Peter Peregrinus, London,
1990.
6. J. K. Hargreaves, The SolarTerrestrial Environment, Cam
bridge Univ. Press, Cambridge, UK, 1992.
7. J. R. Wait, Electromagnetic Waves in Stratiﬁed Media, Per
gamon Press, Oxford, 1970.
8. J. R. Wait, EM Scattering from a vertical column of ionization
in the earthionosphere waveguide, IEEE Trans. Anten. Prop
ag. 39:1051–1054 (1991).
9. A. D. Watt, VLF Radio Engineering, Pergamon press, Oxford,
1967.
10. R. D. Hunsucker and B. S. Delana, HighLatitude Field
Strength Measurements of Standard Broadcast Band Sky
wave Transmissions Monitored at Fairbanks, Alaska, Final
Report on FCC Contract FCC0375, Geophysical Institute/
Univ. Alaska, 1988.
11. J. M. Goodman, HF Communications––Science and Technol
ogy, Van Nostrand Reinhold, New York, 1992.
12. E. E. Johnson et al., Advanced HighFrequency Radio Com
munications, Artech House, Boston, 1997.
13. R. D. Hunsucker, Auroral and polarcap ionospheric effects on
radio propagation, IEEE Trans. Anten. Propag. 40:818–828
(1992).
14. B. Bala, L. J. Lanzerotti, D. E. Gary, and D. J. Thomson, Noise
in wireless systems produced by solar radio bursts., DOI
10.1029/2001RS002481; Radio Sci. 37(2) (March–April 2002).
15. K. Davies and E. K. Smith, Ionospheric effects on satellite
land mobile systems, IEEE Anten. Propag. Mag. 44(6) (Dec.
2002).
16. E. V. Appleton, Geophysical inﬂuence on the transmission of
wireless waves, Proc. Phys. Soc. Lond. 37(2):16D–22D (1925).
17. H. W. Nichols and J. C. Schelling, The propagation of electric
waves over the earth, Bell Syst. Tech. J. 4:215 (1925).
18. C. S. Gillmor and W. Altar, Edward Appleton and the mag
netoionic theory, Proc. Am. Philos. Soc. 126(5):395–440
(1982).
19. W. H. Eccles, Proc. Roy. Soc. Lond. A, Math. Phys. Sci. 87:79
1912.
20. J. M. Kelso, Radio Ray Propagation in the Ionosphere,
McGrawHill, New York, 1964.
21. N. Smith, The relation of radio skywave transmission to
ionospheric measurements, Proc. IRE 27:332–347 (1939).
22. J. A. Ratcliffe, The Magnetoionic Theory and Its Applications
to the Ionosphere: A Monograph, Cambridge Univ. Press,
London, 1959.
23. K. Davies, Ionospheric Radio Waves, Blaisdell, Waltham, MA,
1969.
24. K. G. Budden, The Propagation of Radio Waves. The Theory of
Radio Waves of Low Power in the Ionosphere and Magneto
sphere, Cambridge Univ. Press, Cambridge, UK, 1985.
25. R. D. Hunsucker, Characteristic signatures of the midlatitude
ionosphere observed with a narrowbeam HF backscatter
sounder, Radio Sci. 6:535–548 (1971).
26. R. A. Helliwell, Whistlers and Related Ionospheric Phenom
ena, Stanford Univ. Press, Stanford, CA, 1956.
27. D. Park and D. Carpenter, Very low frequency radio waves in
the magnetosphere, in L. J. Lanzerotti and C. G. Parrs, eds.,
Upper Atmosphere Research in Antarctica, Res. Ser. Vol. 29,
Monograph 72, American Geophysical Union, Washington,
DC, 1978.
28. D. L. Carpenter, Remote sensing of the magnetospheric plas
ma by means of whistler mode signals, Rev. Geophys. 26:535–
549 (1988).
29. R. A. Helliwell, VLF Whistler Mode Experiments, Cambridge
Univ. Press, Cambridge, MA, in press.
30. J. W. Wright, Jr., ed., National Space Weather Program––
Strategic Plan, FCMP301995, 1995, OFCMSSR; 8455 Cole
sville Road, Suite 1500; Silver Spring, MD 20910.
31. The National Space Weather Program––The Implementation
Plan, FCMP311997, Jan. 1997.
32. W. F. Utlaut, ed., Special issue: Ionospheric modiﬁcation by
high power transmitters, Radio Sci. 9:881–1089 (1974).
33. G. L. Rogers, A new method of analyzing ionospheric move
ment records, Nature 177:613–614 (1956).
34. B. D. Wilson, Subdaily northern hemisphere ionospheric
maps using an extensive network of GPS receivers, Radio
Sci. 30(3):639–648 (1995).
35. H. Na, J. Shen, and H. Lee, A Fourier domain technique
for ionospheric tomography, Radio Sci. 30(3):747–754
(1995).
ELECTROMAGNETIC WAVES IN IONOSPHERE 1317
36. A. J. Mannucci et al., GPS and the ionosphere, in W. Ross
Stone, ed., Reviews of Radio Science, 1996–1999, Oxford Univ.
Press, 1999, Chapter 25.
37. L. Kersley et al., Imaging of electron density troughs by
tomographic techniques, Radio Sci. 32(4):1607–1621 (1997).
38. R. Leitinger, Ionospheric tomography, in W. Ross Stone, ed.,
Reviews of Radio Science 1996–1999, Oxford Univ. Press,
1999, Chapter 24.
39. C. Coker, R. Hunsucker, and G. Lott, Detection of auroral ac
tivity using GPS satellites, Geophys. Res. Lett. 22(23):3259–
3262 (1995).
40. R. A. Greenwald et al., DARN/Superdarn: A global view of the
dynamics of highlatitude convection, Space Sci. Rev. 71:761–
796 (1995).
41. A. S. Rodger, Recent scientific advances in geospace research
using coherent and incoherent scatter radars, in W. Ross
Stone, ed., Review of Radio Science 1996–1999, Oxford Univ.
Press, 1999, Chapter 23.
42. W. A. Bristow and R. A. Greenwald, On the spectrum of ther
mospheric gravity waves observed by the SuperDARN net
work, J. Geophys. Res. 102(A6):11,585–11,595 (1997).
43. A. V. Kustov et al., Dayside ionospheric plasma convection,
electric ﬁelds and ﬁeldaligned currents derived from the Su
perDARN radar observations and predicted by the IZEMEM
model, J. Geophys. Res. 102(A11):24,057–24,067 (1997).
44. G. Wannberg et al., The EISCAT Svalbard radar: A case study
in modern incoherent scatter radar system design, Radio Sci.
32(6):2283–2307 (1997).
45. W. A. S. Murray and J. K. Hargreaves, Lunar radio echoes
and the Faraday effect in the ionosphere, Nature 173:944
(1954).
46. J. K. Hargreaves, D. L. Detrick, and T. J. Rosenberg, Space
time structure of auroral radio absorption events observed
with ImagingRiometer of South Pole, Radio Sci. 26:925–930
(1991).
47. J. V. Evans, Ionospheric movement measured by ISR: A re
view, J. Atmos. Terrestr. Phys. 34:175 (1972).
48. C. LaHoz, ed., Special issue: Selected papers from the 6th Int.
EISCAT Workshop, J. Atmos. Terrestr. Phys. 58:1–507 (1996).
49. R. Schunk et al., Global assimilation of ionospheric measure
ments (GAIM), Radio Sci. (in press).
ELECTROMAGNETICS, TIMEDOMAIN
NATHAN A. BUSHYAGER
MANOS M. TENTZERIS
Georgia Institute of Technology
Atlanta, Georgia
1. INTRODUCTION
1.1. Electromagnetics Overview
The modeling, design, and optimization of RF/microwave
structures is a very challenging and timeconsuming task
because of the difﬁculties in predicting the highfrequency
effects (multimoding, loss mechanisms, parasitics) for spe
cific devices and their interaction with other devices in
close proximity. Even if a single component is well char
acterized in isolation, effects such as surface waves and
parasitics lead to signal corruption and degraded perfor
mance when the element is placed in a circuit with other
components. The study of electromagnetic interaction is
key to properly understanding the operation of any RF or
microwave device. In these devices, it is often better to
think in terms of electromagnetic ﬁelds rather than volt
ages and currents, as is usually done in lowerfrequency
circuits. The characterization of lowfrequency circuits by
voltage and current is possible because Maxwell’s equa
tions, which govern electromagnetic interaction, can be
approximated with circuit equations when a structure is
significantly smaller than the maximum wavelength sup
ported by the device. As the wavelength approaches and
becomes comparable to the size of the device being stud
ied, as in RF devices (and, as another example, in lowfre
quency power transmission lines), these approximations
are no longer valid. When electromagnetic interactions
are not evaluated in these devices, performance degrada
tion often results. For example, when digital circuitry and
RF components are used in close proximity, digital switch
ing noise can corrupt RF signals, rendering the device
useless. Proper consideration of electromagnetics in this
case can account for this interaction and lead to the cre
ation of a successful design.
To analyze these components and their interaction
mechanisms, it is often necessary to use electromagnetic
simulation software. There are a number of commercial
simulation packages, and significant research is ongoing
to ﬁnd new methods and to solve new classes of problems
such as MEMS (microelectromechanical systems), period
ic structures, and metamaterials. These tools allow the
characterization of RF and microwave devices before the
fabrication stage, which can save considerable time and
money in the design process. In addition, they can be used
to optimize devices, allowing hundreds or thousands of
conﬁgurations to be tested. Because of the complexities of
modern microwave structures, it would be impossible to
design many modern RF devices and modules without the
use of electromagnetic simulators.
Electromagnetics is the study of the ﬁelds created from
the motion of charge. If these charges are stationary, a
static, or unchanging, electric ﬁeld results. If these charg
es are moving with constant velocity, a static magnetic
ﬁeld results. If these charges accelerate (change speed or
direction with time), then the result is a coupled electro
magnetic ﬁeld. This is because a changing electric ﬁeld
creates a magnetic ﬁeld, and a changing magnetic ﬁeld
creates an electric ﬁeld. The result is a propagating elec
tromagnetic wave. The equations that govern electromag
netic phenomena are called Maxwell’s equations.
Because of the difﬁculty of applying the necessary
boundary conditions on arbitrary structures, Maxwell’s
equations cannot be solved analytically for most cases of
interest to the RF or microwave designer. Beyond special
cases no analytic expressions can be found that relate the
ﬁelds in the structure to a specific source condition. In
stead, the equations that represent electromagnetic inter
action must be discretized (expressed as simple arithmetic
equations) so that they can be solved numerically. Al
though it is possible to solve these equations by hand, the
1318 ELECTROMAGNETICS, TIMEDOMAIN
number required to accurately characterize a typical RF
device is usually in the millions or billions, necessitating
the use of computers. As these methods use computers to
perform mathematical operations on numbers, they are
referred to as computational or numerical electromagnetic
methods. Numerical methods are a branch of mathematics
in their own right, and are used in all of science and en
gineering, from quantum mechanics to meteorology. These
techniques are chosen and applied largely according to the
type of differential equation modeled. In electromagnetics,
they are often split into two categories: timedomain and
frequencydomain methods.
Maxwell’s equations are usually expressed in terms of
either time or frequency. The time form of the equations
shows the propagation of electromagnetic ﬁelds as they
could be observed by recording the time variation of the
electromagnetic ﬁelds at any point in space. Another
method of representing electromagnetic phenomena is by
considering the response to a harmonic (single sinusoidal
frequency) or ‘‘monochromatic’’ source. Maxwell’s equa
tions can be transformed to give frequencydomain solu
tions in this case. The choice of the domain of the
simulator depends on whether the response of the device
has to be modeled over a wide or narrow frequency band
and whether the ﬁeld variation changes significantly for
different bands (multimode propagation). Results in either
domain can be transformed to the other, but computation
al requirements vary.
In the world of electromagnetics, there are distinct ad
vantages to both time and frequencydomain simulations.
Each frequencydomain simulation is valid for only a sin
gle frequency. If the response of the device over a band is
desired, then several frequency points over the band must
be simulated. Time domain simulations, on the other
hand, can be transformed to give results over a wide fre
quency band. A drawback of timedomain techniques is
that it is more difﬁcult to apply the effects of materials that
have characteristics that change with frequency, such as
dispersive media. These materials are common in RF cir
cuits, and thus complicated methods must be used to allow
their accurate modeling in the time domain. Likewise, it is
difﬁcult to extract transient data from a frequencydomain
simulation.
The two major computational criteria for the selection
of a specific numerical simulator are speed and accuracy.
These are not independent parameters, all simulators can
be made more accurate at the expense of speed. It is not
possible to choose a simulator that is the best for all sit
uations. Many comparisons have been made between sim
ulators [1,2], and depending on the problem type, certain
techniques are often preferred.
The purpose of this article is to give an overview of the
most widely used timedomain simulation techniques. In
addition, specific advantages and weaknesses of each
method are highlighted, along with some references that
can provide more details about each method. With the in
formation presented here, the reader should be able to
understand how the different simulation techniques can
be applied to specific problems and how the features of
that problem place restrictions on the simulation tech
niques. The article begins with a quick overview of elec
tromagnetics and then uses this information to explain
the simulation methods.
2. THEORY
2.1. Maxwell’s Equations
The differential form of Maxwell’s equations in the time
domain can be expressed as
rÂEðtÞ ¼ À
@BðtÞ
@t
ð1Þ
rÂHðtÞ ¼
@DðtÞ
@t
þJðtÞ ð2Þ
r Á DðtÞ ¼rðtÞ ð3Þ
r Á BðtÞ ¼0 ð4Þ
where E is the electric ﬁeld (V/m), H is the magnetic ﬁeld
(A/m), D is the electric ﬂux density (C/m
2
), B is the mag
netic ﬂux density (Wb/m
2
), r is charge density (C/m
3
), and
J is the electric current density (A/m
2
). These equations
relate the electric and magnetic ﬁelds at points in space.
Likewise, these equations can be expressed in integral
form that relates the ﬁelds over contours and areas. In
isotropic, nondispersive linear media the constitutive
equations are
DðrÞ ¼eðrÞEðrÞ ð5Þ
BðrÞ ¼mðrÞHðrÞ ð6Þ
where r is the position vector. In general, the relationship
between the ﬁelds and their ﬂux density is dependent on
the media parameters in tensor form and is a convolution
due to frequency dependence.
An alternative frequencydomain representation of
Maxwell’s equations that is often presented in the litera
ture makes use of the phasor form of the ﬁelds, assuming a
harmonic time dependence (e
jot
). While equations (3) and
(4) remain the same (with the exception that t is trans
formed to o), the time derivatives in (1) and (2) disappear,
giving
rÂ
~
EEðoÞ ¼ Àjo
~
BBðoÞ ð7Þ
rÂH
^
ðoÞ ¼jwD
^
ðoÞ þJ
^
ðoÞ ð8Þ
This representation is often used to derive frequencydo
main methods. The form of the equations given in (1)–(4) is
usually used as the starting point for the derivation of
timedomain computational methods. However, deriva
tions have been given for several techniques using the in
tegral form of Maxwell’s equations as well as the
differential form.
Maxwell’s equations, with the appropriate constitutive
equations and boundary conditions at media interfaces,
can be used to solve for the electric and magnetic ﬁelds
due to any source ﬁelds, charges, and currents in any
ELECTROMAGNETICS, TIMEDOMAIN 1319
given structure. For a general material interface, the
boundary conditions are
^ nn Á ðD
2
ÀD
1
Þ ¼r
s
ð9Þ
^ nn Á B
2
¼ ^ nn Á B
1
ð10Þ
^ nn Á E
2
¼ ^ nn Á E
1
ð11Þ
^ nnÂðH
2
ÀH
1
Þ ¼J
s
ð12Þ
where r
s
and J
s
are the charge and electric current den
sities on the interface and ^ nn is the normal unit vector to
the interface. In the more specific case of a dielectric in
terface, no charge or surface currents exist. As such, (9)–
(12) simply state that the normal components of the elec
tric and magnetic ﬂuxes are constant across the boundary,
while the tangential electric and magnetic ﬁelds are con
stant across the boundary. Likewise, on a perfect electrical
conductor (PEC) interface the tangential electric ﬁeld and
normal magnetic ﬂux are zero, while the normal electric
ﬂux and tangential magnetic ﬁeld are equal to the charge
and current densities, respectively, on the PEC. While (1)–
(12) can be used to create a system of equations that can be
solved for the ﬁelds in any problem, in most cases the re
sulting partial differential equations have no closedform
solution, requiring the use of numerical methods.
The ﬁrst simulation technique presented is the ﬁnite
difference timedomain (FDTD) [3] technique. It is one of
the oldest and most widely used electromagnetic simula
tion techniques. Many methods, such as ﬁnitevolume
timedomain (FVTD) [4], ﬁniteelement timedomain
(FETD) [5], and multiresolution timedomain (MRTD) [6]
techniques, which are also presented in this article, have
advantages over the FDTD technique at the expense of
added complexity. Also presented in this article is the
transmissionline matrix (TLM) [7] technique. Mathemat
ically it can be shown to be similar to FDTD while pos
sessing unique advantages. Another interesting feature of
TLM is that it is directly based on a physical model of
wave propagation, instead of a mathematical discretizat
ion of Maxwell’s equations. The purpose of this article is to
give a basic overview of the operation of these techniques,
a summary of the limitations of each, and an overview of
how they can be used to determine useful device charac
teristics such as S parameters and radiation patterns.
3. TIMEDOMAIN NUMERICAL TECHNIQUES
3.1. FDTD
The ﬁnitedifference timedomain (FDTD) technique was
originally proposed by K. S. Yee in his seminal 1966 paper
[3]. The deﬁning factor of the Yee FDTD scheme is the ar
rangement of the FDTD gridpoints that became known as
the ‘‘Yee cell.’’ This scheme is sometimes referred to as the
‘‘Yee leapfrog scheme,’’ for the manner in which the elec
tric and magnetic gridpoints are interleaved in both space
and time, as shown in Fig. 1 for threedimensional grids.
In this ﬁeld arrangement, the electric ﬁeld Cartesian com
ponents are located half a cell width apart along their axis
from each gridpoint. Similarly, the magnetic ﬁeld compo
nents are located half a cell width in both directions nor
mal to their axis from each gridpoint. This arrangement
places the electric ﬁeld components along the edges of the
cells and the magnetic ﬁelds in the faces. The importance
of this arrangement becomes clear when the FDTD update
scheme is examined.
The FDTD update scheme can be derived by applying
central differences to Maxwell’s curl equations [8]. Specif
ically, Faraday’s and Ampere’s laws, (1) and (2), in differ
ential form are used. The Yee cell ﬁeld arrangement is
convenient in this case because it deﬁnes the ﬁelds at dis
crete points, and the differential form of Maxwell’s equa
tions relates the ﬁelds at a point instead of an area or loop.
Equations (1) and (2) are vector equations, which can
be easily split into three scalar equations. For example, (2)
can be written as
@E
x
@t
¼
1
e
@H
z
@y
À
@H
y
@z
ÀJ
x
_ _
ð13Þ
@E
y
@t
¼
1
e
@H
x
@z
À
@H
z
@x
ÀJ
y
_ _
ð14Þ
@E
z
@t
¼
1
e
@H
y
@x
À
@H
x
@y
ÀJ
z
_ _
ð15Þ
if the constitutive relationship for linear, isotropic, non
dispersive media, (5), is used. Central differences can then
be applied to discretize the time and space derivatives of
these equations. A centraldifference discretization of a
ﬁrst derivative
@F
@x
¼
F
x þDx=2
ÀF
xÀDx=2
Dx
ð16Þ
approximates the derivative of a function F at the point x
by using the ratio of the difference of two surrounding
points, F
x þDx=2
and F
xÀDx=2
, and the distance between
those points, Dx.
Observing the ﬁeld conﬁgurations in Fig. 1 it is clear
that each Eﬁeld component is surrounded by the Hﬁeld
components that appear in its scalar equation, and vice
versa. Furthermore, they are separated from each H com
ponent by half a cell in the direction of differentiation of
the H component. The time offset between the E and H
components can similarly be set to Dt/2. If (13) is expanded
in ﬁnite differences, ignoring the current terms (loss and
source currents), we obtain
E
t þDt
x;i;j;k
ÀE
t
x;i;j;k
Dt
¼
1
e
H
t þDt=2
z;i;j þDy=2;k
ÀH
t þDt=2
z;i;jÀDy=2;k
Dy
_
_
À
H
t þDt=2
y;i;j;k þDz=2
ÀH
t þDt=2
y;i;j;kÀDz=2
Dz
_
_
ð17Þ
1320 ELECTROMAGNETICS, TIMEDOMAIN
where i,j,k specify coordinates in a grid of Yee cells (x,y,z ¼
iDx,jDy,kDz).
In (17) the value of the E
x
component at time t þ1 is
given in terms of prior values of the surrounding Hﬁeld
components and the prior value of the E
x
ﬁeld at the same
point. Solving for E
t þDt
x;i;j;k
E
t þDt
x;i;j;k
¼E
t
x;i;j;k
þ
Dt
e
H
t þDt=2
z;i;j þDy=2;k
ÀH
t þDt=2
z;i;jÀDy=2;k
Dy
_
_
À
H
t þDt=2
y;i;j;k þDz=2
ÀH
t þDt=2
y;i;j;kÀDz=2
Dz
_
_
ð18Þ
gives a discrete update equation for E
x
that requires only
the surrounding ﬁelds at the previous timestep. Similar
equations can be found for the other E and Hﬁeld com
ponents, and a timemarching scheme results. If the cur
rent terms are not neglected a scheme that includes
source currents and Ohmic losses results. It is important
to reinforce that the E and H ﬁelds are not evaluated at
the same time instant; they are separated by half time
step.
The derivation of the FDTD update equations present
ed do not provide a method for choosing the temporal and
spatial steps. It has been shown [8] that the spatial step
required for accurate results with an acceptable numerical
dispersion performance is related to the spectrum of the
excitation waveform and is at maximum onetenth of the
smallest wavelength in the excited waveform. For this
reason, the input waveform must be bandlimited. Gauss
ian pulses, the derivative of Gaussian pulses, and other
easily parameterized waveforms are popular choices.
Using these functions it is possible to simulate the fre
quency band of interest without generating nonphysical
results.
In the FDTD update equations, such as (18), the value
of e and m are speciﬁed by gridpoint. PEC structures are
represented by setting electric ﬁeld components tangen
tial to PEC structures to zero. In order to accurately sim
ulate a structure, several gridpoints must be used across
each feature. Most RF structures have features that are
significantly smaller than the wavelength at their fre
quency of operation, and thus the spatial step is con
strained by the structure.
The timestep used in FDTD must be speciﬁed as a
function of the spatial step [8]. In order for the scheme to
be stable
Dt ¼
1
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
1
me
1
Dx
2
þ
1
Dy
2
þ
1
Dz
2
_ _
¸ ð19Þ
is the maximum value of Dt [known as the CourantFried
richsLevy (CFL) Limit] that can be used in the simula
tion. One other important rule of thumb for the
application of FDTD is that Dx, Dy, and Dz should differ
by a factor of r10.
Because of the discrete nature of the FDTD grid, the
numerical propagation velocity of a wavefront becomes
frequencydependent, an effect that is called numerical
dispersion. This places a limitation on cell size (usually it
must be smaller than
1
10
th of the wavelength) and the
timestep. Often ﬁne grids are required because of the fea
tures of the modeled structure, leading to a large number
of gridpoints and a small timestep. One modiﬁcation to the
FDTD technique that has been created to address this
limitation is the ADI (alternatingdirectionimplicit)
FDTD [9] technique. In this technique the timestep is no
longer ﬁxed by stability requirements and can be made
larger than the CFL limit, although the fact that ADI is an
implicit technique complicates its application. Still the
size of the timestep affects the numerical accuracy of ADI
simulations.
3.2. TransmissionLine Matrix
The transmissionline matrix (TLM) method is a powerful
technique that is unique because of its historical basis and
its derivation from Huygens’ principle. Before modern
computers were available to simulate wave propagation
problems, methods were developed that utilized the rela
tionship between electromagnetic propagation and voltag
es and currents in lumpedelement networks [10,11]. As
modern computers were made available that could repre
sent these circuits, it became possible to translate these
physical simulators to computer models. In this way, the
TLM method is a digital representation of an analog sim
ulator. Both of the models are based on a discretization of
Huygens’ principle.
The TLM method was ﬁrst suggested by Johns and Be
urle [7]. Huygens’ principle states that a propagating
wavefront consists of an inﬁnite number of radiating
sources, and the sum of the radiation from these sources
E
y
E
z
H
z
x
y
H
y
E
x
H
x
z
Figure 1. Yee cell in three dimensions.
ELECTROMAGNETICS, TIMEDOMAIN 1321
forms the new wavefront. This principle is rarely present
ed mathematically, but can be used as the basis of a nu
merical method that describes radiation through a
discrete grid. A twodimensional discretization of the
TLM method is presented here; a number of references
provide a 3D representation as well as a broader overview
of the method [12,13].
In order to discretize Huygens’ principle in two dimen
sions, a Cartesian grid is chosen such that the distance
between elements, Dl corresponds to the distance that the
electromagnetic wave will propagate in one timestep Dt;
therefore
Dt ¼Dl=c ð20Þ
A pulse incident on a node in this grid must be equally
radiated in all four directions. The grid and pulse radia
tion on a node are demonstrated in Fig. 2. From this it can
be seen that this matrix of nodes acts like a network of
transmissionline elements. If these pulses are thought of
as voltages in the transmissionline matrix, then the fol
lowing equation represents the radiated pulses on the
lines surrounding a node at timestep nþ1 based on the
incident pulses at timestep n [14].
nþ1
V
1
V
2
V
3
V
4
_
_
_
_
_
_
_
_
_
_
_
_
_
_
r
¼
À1 1 1 1
1 À1 1 1
1 1 À1 1
1 1 1 À1
_
_
_
_
_
_
_
_
_
_
_
_
_
_
n
V
1
V
2
V
3
V
4
_
_
_
_
_
_
_
_
_
_
_
_
_
_
i
ð21Þ
As the radiated pulses become the incident pulses on the
neighboring nodes, the incident pulses for the next time
step are computed as
k þ1
V
i
1
ðx; zÞ ¼
kþ1
V
i
3
ðx; z À1Þ
k þ1
V
i
2
ðx; zÞ ¼
kþ1
V
i
4
ðx À1; zÞ
k þ1
V
i
3
ðx; zÞ ¼
kþ1
V
i
1
ðx; z þ1Þ
k þ1
V
i
4
ðx; zÞ ¼
kþ1
V
i
2
ðx þ1; zÞ
ð22Þ
if the distance between the gridpoints is normalized.
These equations can be used to in a timestepping fashion
by setting an initial condition on the TLM grid and then
using (21) and (22) to update the values on each node in
the grid.
The grid presented in Fig. 2 and the update equations
given in (21) and (22) demonstrate the propagation of any
wave. The derivation of the TLM technique will be pre
sented for the twodimensional TE simulations, utilizing
the E
y
, H
x
, and H
z
ﬁelds. It was already stated that this
network of grids acts like an electrical network of inter
connected transmission lines, so it is natural to develop a
transmissionline network in which the ﬁeld values can be
determined from the currents and voltages in the net
work. This unit cell is presented in Fig. 3. The equations
that represent signal propagation on these transmission
lines are very similar to Maxwell’s equations. In fact,
there is a simple relationship [14]
E
y
V
y
H
z
I
x
H
x
ÀI
z
m L
e 2C
ð23Þ
that equates the electromagnetic ﬁelds to voltages and
currents in the network.
∆I
Z
X
3
4 2
1
∆
I
Figure 2. Mesh demonstrating incident (dashed) pulse and ra
diated pulses (solid).
L∆t/2
C∆t/2
y
4
3
2
1
x
z
L∆t/2
L∆t/2
L∆t/2
Figure 3. Lumpedelement unit cell for TLM [12].
1322 ELECTROMAGNETICS, TIMEDOMAIN
In order to apply the algorithm to a physical problem, it
is important to understand the conditions under which it
can be used. The line matrix does not represent a contin
uous space, but rather a tightly spaced periodic structure.
As such, electromagnetic waves cannot propagate in any
direction, as they can in nature, but only along the lines of
the grid. By reducing the spacing between the gridpoints,
the space can be made to closely represent natural prop
agation. However, because of the nature of the discrete
grid, the speed of wave propagation through the grid will
not match the speed through a physical medium. Further
more, this difference between physical and numerical
propagation speeds is a function of both frequency and
propagation direction, and is thus referred to as grid dis
persion. For diagonal propagation, the grid experiences no
dispersion. The maximum dispersion occurs in the axial
direction and the velocities are related as [12]
v
n
c
¼
o
b
n
c
¼
b
b
n
¼
pDl=l
arcsin
ﬃﬃﬃ
2
p
sinðpDl=lÞ
_ _ ð24Þ
The TLM grid propagation velocity is naturally c/2
1/2
for
square cells in the diagonal direction [12] (as the grid rep
resents a medium with a permittivity twice that of free
space); but for a Dl that is a quarter awavelength, the
network velocity is c/2 in the axial direction. Therefore, for
an accurate simulation, a grid size much smaller than l/4
must be chosen. In practice, however, the grid size is often
constrained by the structure being simulated, resulting in
grids significantly smaller than this limit.
A major advantage of the TLM method is that it sim
ulates actual electromagnetic wave propagation; no ap
proximations are applied to the wave equations. This
makes it more computationally intensive than other meth
ods (although on the same order as others presented here),
an aspect that becomes apparent in the requirement that
is placed on the grid size. However, the advantage of this
method is that it can be used to simulate almost any
structure. The model that has been presented thus far
only allows freespace propagation, but can be easily ex
tended to simulate structures such as perfect electrical
conductors (PEC), perfect magnetic conductors (PMC), di
electric interfaces, and lossy metals.
By modifying the unit cell presented in Fig. 3, the ef
fects mentioned above can be modeled. All of these tech
niques have been presented in the literature [12]. PEC
interfaces can be modeled with short circuits; PMC inter
faces, with an open circuit. In a similar way, the reﬂection
from a lossy boundary can simply be represented by set
ting the reﬂected ﬁeld to be equal to the incident ﬁeld
multiplied by a loss parameter r (which is, of course, less
than 1). Homogeneous materials other than free space can
be modeled by simply setting the values of m and e to match
the desired materials. Of course, it is also possible to mod
el inhomogeneous and dispersive materials, but their
treatment is beyond the scope of this article.
3.3. Conformal Mesh Techniques
The FDTD and TLM techniques that have been presented
are the two most widely used timedomain simulation
techniques. Both techniques have proved to model an ex
tremely wide variety of cases from radar cross sections of
aircraft to circuits in broadband communications devices.
Nevertheless, there are many structures that cannot con
form to a Cartesian grid. When these structures, such as
circular metal and dielectric elements, are encountered,
they are typically treated in FDTD and TLM by staircas
ing a ﬁne grid around the nonconformal shape. When
staircased ﬁnely enough, it is often possible to accurately
simulate the structure. However, this can lead to extreme
ly ﬁne grids and thus excessive execution time. Further
more, some structures cannot be accurately modeled with
a staircase conﬁguration [15]. To more accurately repre
sent these structures, methods have been developed that
use conformal, nonorthogonal meshes with fewer cells, al
though the update of each cell is more computationally
intensive. As an added beneﬁt, a number of codes produce
unstructured meshes that are compatible with these
methods. As all of these methods are quite complex, com
plete derivations will not be presented here. However, the
major points and differences of three of the most popular
methods, the generalized Yee scheme [16,17] and the ﬁ
nitevolume timedomain (FVTD) [4] and ﬁniteelement
timedomain (FETD) [5] techniques will be briefly pre
sented.
The Yee FDTD technique presented earlier in this ar
ticle has been actively researched since the early 1970s.
As such, a number of methods have been developed that
allow FDTD to be used to model structures that do not
conform to the ﬁxed Cartesian grid. One of the simplest of
these methods is the variablegrid technique [18]. A sim
ilar method has been presented for the TLM technique
[19]. This grid is very easy to use with any existing code,
as the orthogonality between the grid directions is main
tained. By simply varying the Dx, Dy, and Dz values as a
function of position along their respective axes, a much
wider variety of structures can be modeled. In addition,
the grid can be dense in the area of ﬁne features while
remaining relatively sparse in large homogeneous areas.
Figure 4a depicts a twodimensional variable grid. In this
grid the two coordinate directions are independently var
ied, thus allowing a much ﬁner grid to be used in a local
ized region.
Another simple extension to FDTD is subgridding [20].
In these methods, the resolution is increased in a subset of
the grid, as shown in Fig. 4b. This allows ﬁne features to
be locally matched while adding no additional computa
tional demands away from the ﬁnely detailed region. The
difﬁculty of applying this method comes from the interface
between the subgrided region and the surrounding mesh.
At these interfaces, cells intersect where there are no
gridpoints. In these regions the ﬁelds must be interpolated
to obtain values where gridpoints do not exist, introducing
nonphysical dispersion and numerical velocity mismatch.
The subgridding method allows the ﬁne grid to be applied
in a more local fashion than variable gridding. However,
as with the variable grid method, the timestep is con
strained by the smallest cell; whereas the simulation uses
fewer gridpoints (and is thus faster), it requires the same
number of timesteps as a grid made entirely of the small
est cell.
ELECTROMAGNETICS, TIMEDOMAIN 1323
A ﬁnal extension that will be mentioned here is a meth
od of local conformal modeling where individual cells are
modiﬁed such that an interface that intersects a cell is
modeled while the surrounding cells are treated as normal
FDTD cells [21]. An example of this grid is presented in
Fig. 4c, where a circle is inscribed into the Cartesian grid.
Cells that are intersected by the circle approximate the
curve with a diagonal edge. These methods can allow the
modeling of curved elements with significantly fewer cells
than staircasing, and they converge more accurately to the
actual shape as the cell size is reduced. For more complex
structures a grid that is conformal everywhere, such as in
Fig. 5, is appropriate. This method can be viewed as a lo
cal, limited version of the generalized Yee scheme.
The generalized Yee scheme [16,17] is so named be
cause it extends the use of the Yee cell, which is based on a
Cartesian grid, into a scheme where the space is ﬁlled
with interlocking polyhedra, the faces and edges on which
the ﬁeld values are constant. As the polyhedra that rep
resent the electric ﬁelds intersect at the centroids of the
magnetic ﬁeld polyhedra, the resulting ﬁeld orientation is
similar to the conventional Yee scheme. However, in the
generalized scheme the points do not have to be regularly
spaced, and the edges that connect them are not orthog
onal. Furthermore, all the polyhedra do not have the same
shape. Removing the structure of the grid adds significant
complexity to the method. However, there are several ben
eﬁts to using the technique: (1) as in traditional FDTD, the
electric and magnetic ﬁelds are offset, allowing a second
order representation while remaining explicit; (2) complex
shapes can be modeled with fewer cells than conventional
FDTD because the need for staircasing has been eliminat
ed; and (3) the update of each ﬁeld point is still dependent
only on the nearest neighbors, allowing the technique to
be efﬁciently parallelized.
Another technique that uses a conformal mesh is the
ﬁnitevolume timedomain (FVTD) method [4]. While it
uses a similar grid as the generalized Yee scheme, it has a
very different derivation and characteristics. The FVTD
method is derived using the integral form of Maxwell’s
equations on contours made in the grid and collocates all
the ﬁelds in both time and space (as opposed to the leap
frog arrangement of the Yee scheme). One difference be
tween this method and FDTD is that it is not dispersive,
but rather dissipative [4].
The ﬁnal conformal method that will be reviewed here
is the ﬁniteelement timedomain (FETD) technique [5].
Finiteelement schemes are very different from the tech
niques presented thus far because the ﬁelds are repre
sented on elements on which a ﬁeld distribution is
assumed and coefﬁcients representing the magnitude of
the ﬁelds on these elements are determined. Like the
FVTD and generalized Yee scheme, the FETD method
can use a conformal mesh and thus represent complex
structures with larger cells than can the FDTD method.
The FETD method can be complicated in a full 3D scheme,
but, as it is an important and widely used timedomain
method, a brief derivation is presented here.
3.4. FiniteElement Time Domain
A multitude of algorithms apply ﬁniteelement techniques
to timedomain electromagnetic modeling. The technique
presented here is the pointmatched timedomain ﬁnite
element method [5], which was developed in the mid
1980s. It is one of the earliest ﬁniteelement methods in
the time domain, and is representative of the techniques
used in ﬁniteelement timedomain (FETD) modeling.
FETD shares many similarities to FDTD. As in FDTD,
Maxwell’s equations in differential form, Eqs. (1)–(4) are
used as a starting point. However, instead of discretizing
the equations using ﬁnite differences, the electric and
magnetic ﬁelds in the equations are replaced with a sum
(a) (b) (c)
Figure 4. Alternate gridding methods for
Yee FDTD: (a) variable gridding; (b) subg
ridding; (c) locally conformal mesh.
Figure 5. Conformal mesh using triangular elements, created
with HFSS (highfrequency structure simulation) [22].
1324 ELECTROMAGNETICS, TIMEDOMAIN
of coefﬁcients multiplied by basis functions chosen by the
user. The electric and magnetic ﬁelds are expressed as
Eðr; tÞ ¼
M
i ¼1
j
i
ðrÞE
i
ðtÞ ð25Þ
Hðr; tÞ ¼
N
j ¼1
c
j
ðrÞH
i
ðtÞ ð26Þ
where M and N are the number of nodes (r
i
and r
j
, re
spectively) used to represent the ﬁelds over the simulated
space. These nodes are not necessarily collocated. It is im
portant to note that, unlike in the ﬁnitedifference tech
nique, all components of each ﬁeld are collocated at their
respective gridpoints. The characteristics of the resulting
simulator are dependent on the basis functions, j and c,
chosen.
The basis functions are restricted such that
j
i
ðrÞ ¼
1 r ¼r
i
0 all other nodes
_
ð27Þ
c
j
ðrÞ ¼
1 r ¼r
j
0 all other nodes
_
ð28Þ
The domain of each basis function ends at the neighboring
grid locations. The effect of the functions is to interpolate
the ﬁelds over the cell. The electric and magnetic grid
points are offset, in much the same way as in the FDTD
method, so that the locations of the magnetic gridpoints
are in the center of the electric ﬁeld cells. With this ar
rangement, update equations very similar to the FDTD
update equations can be developed. The basis functions for
the electric or magnetic gridpoints located at the points r
i
and r
j
, respectively, will only overlap the neighboring
gridpoints of the corresponding magnetic or electric ﬁeld.
If each element has K nodes, and (25) and (26) are inserted
into (1) and (2) (ignoring the current term), the following
results
@H
j
ðtÞ
@t
¼ À
1
mðr
j
Þ
K
l ¼1
rj
l
ðr
j
Þ ÂE
l
ðtÞ ð29Þ
@E
i
ðtÞ
@t
¼
1
mðr
i
Þ
K
l ¼1
rc
l
ðr
i
Þ ÂH
l
ðtÞ ð30Þ
where the points denoted by the index l correspond to the
points where the basis functions overlap.
The coefﬁcients E
i
(t) and H
j
(t) in (25) and (26) are dis
crete functions of time. They can be used to determine the
magnitude of the electric and magnetic ﬁelds at integer
multiples of the timestep Dt. The size of the timestep re
quired for a stable simulation is limited by the grid, but its
calculation will not be covered here. In (29) and (30) the
time derivative is not discretized. If central differences are
used, we obtain
H
nþ 1=2 ð Þ
j
¼H
nÀ 1=2 ð Þ
j
À
1
mðr
j
Þ
K
l ¼1
rj
l
ðr
j
Þ ÂE
l
ðtÞ ð31Þ
E
nþ1
i
¼E
nþ1
i
þ
1
mðr
i
Þ
K
l ¼1
rc
l
ðr
i
Þ ÂH
l
ðtÞ ð32Þ
where n is the timestep index. As in the ﬁnitedifference
technique, the E and H ﬁelds are interleaved in time, and
an explicit timemarching scheme results. The only step
remaining to determine the timemarching scheme is to
calculate the derivatives in (31) and (32), which is an easy
task once the basis functions have been chosen.
While the scheme presented seems relatively straight
forward, several details involved in implementing the
method can be quite daunting. First, the grid used in
this method, unlike that in FDTD and TLM, does not have
to be rectangular. Each element can be shaped differently.
This is an advantage of the method, but the grid genera
tion can be challenging. A common method is to use a
rectangular mesh everywhere except at material bound
aries. This significantly simpliﬁes grid generation. There
are several existing codes and algorithms that can be used
for mesh generation. The creation of an effective meshing
algorithm can be at least as complicated as the FETD
technique itself. Similar grids are used for ﬁniteelement
discretizations in a variety of applications, such as me
chanics, aerodynamics, and frequencydomain electro
magnetics.
The second difﬁculty that must be mentioned is the in
tricacy of representing boundary conditions in the ﬁnite
element method. A system of equations that can be used to
solve for the ﬁelds on the boundary of a PEC are
n
_
ÂE
1
¼ Àn
_
ÂE
2
ð33Þ
r
0
Á ðn
_
ÂH
total
Þ ¼ Àen
_
Á
@E
total
@t
ð34Þ
where r
0
is the surface divergence (derivatives are taken
only along the surface). In the ﬁnitedifference timedo
main method, only (33) is used. This is because the com
ponents of the electric ﬁeld are offset in space, and only the
tangential ﬁelds ever contact a PEC surface. Thus, instead
of simply zeroing components, such as in the FDTD meth
od, a system of equations must be solved at each timestep.
Similar considerations must be taken when implementing
dielectric boundaries.
3.5. Multiresolution Time Domain
The multiresolution timedomain (MRTD) technique [6]
was originally introduced by M. Krumpholz and L. P. B.
Katehi in 1996. The MRTD method is derived through the
application of wavelets to the space discretization of Max
well’s equations (multiresolution analysis). Time differen
tiation is modeled using central differences as in FDTD.
One of the advantages of using MRTD is that the resolu
tion of the discretization can be altered as a function of
ELECTROMAGNETICS, TIMEDOMAIN 1325
time and space. This timespaceadaptive grid can result in
a significant reduction in operations per timestep and sig
nificantly quicker simulations. Another advantage of the
MRTD method is that, depending on the wavelet basis
functions used, the numerical dispersion relationship can
allow the use of significantly fewer basis functions per
wavelength than required in the FDTD method [23]. The
adaptive nature of wavelets and their ability to efﬁciently
represent complex waveforms has led to their use in sev
eral frequency and timedomain techniques.
Many wavelet bases are commonly used in electromag
netics, each with distinct advantages and challenges in its
implementation. Several of these techniques allow not
only adaptive resolution but also the use of fewer points
per wavelength and are comparable to highorder FDTD
techniques. For simplicity, a 2D derivation of MRTD using
the relatively simple Haar wavelets is presented here.
This set of basis functions has the same dispersion prop
erties as FDTD, and thus requires the same number of
basis functions per wavelength [24]. The Haar basis func
tions are chosen over other wavelet basis functions for
several reasons: (1) they are based on pulses and thus are
simpler to apply than other basis functions and (2) when
the basis functions are summed to determine the ﬁeld
values, they form regions of constant ﬁeld value. There
fore, when set to zero, they form a ﬁnite region of zero ﬁeld
value. This enables efﬁcient modeling of PECs. The main
advantage of Haar MRTD over FDTD is the timespace
adaptive gridding capability. The Haar scaling function
and mother wavelet are presented in Fig. 6.
The Haar scaling function is a simple pulse function in
space. If the MRTD update derivation is followed using
only the Haar scaling functions, the FDTD scheme results
[25]. All wavelets used in the Haar MRTD scheme are
based on the mother wavelet. With each increase in reso
lution level, the number of wavelets is doubled, the domain
of each is halved, and the magnitude is increased so that
the norm (integral of the function multiplied by itself over
its entire domain) of each function is 1. The wavelet basis
functions for resolution 1 and 2 are presented in Fig. 7.
MRTD update equations can be determined by expand
ing Maxwell’s curl equations in terms of wavelets and
scaling coefﬁcients in space and pulses in time and then
applying the method of moments. Maxwell’s curl equa
tions, in the 2D TE
z
mode, are given by
@E
x
@t
¼
1
e
@H
z
@y
ð35Þ
@E
y
@t
¼ À
1
e
@H
z
@x
ð36Þ
@H
z
@t
¼
1
m
@E
x
@y
À
@E
y
@x
_ _
ð37Þ
E
x
can be expanded as
E
x
ðxÞ ¼
n;i;j
h
nDt
Â
n
E
x;ff
i;j
j
i
ðxÞj
j
ðyÞ þ
r
max
r ¼0
2
r
À1
p¼0
n
E
x;cf
i;j;r;p
c
r
i;p
ðxÞj
j
ðyÞ
þ
r
max
r ¼0
2
r
À1
p¼0
n
E
x;fc
i;j;r;p
j
i
ðxÞc
r
j;p
ðyÞ
þ
r
max
r ¼0
2
r
À1
p¼0
r
max
s ¼0
2
s
À1
q¼0
n
E
x;cc
i;j;r;p;s;q
c
r
i;p
ðxÞc
s
j;q
ðyÞ
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
ð38Þ
following the notation in Ref. 6. In this case
j
i
ðxÞ ¼jðx=Dx ÀiÞ and c
r
i;p
ðxÞ ¼2
r=2
c
0
ð2
r=2
ðx=Dx ÀiÞ ÀpÞ,
and h(t) is a simple pulse of duration Dt. The maximum
level of wavelet resolution is r
max
. The four groups of
terms in the equations above represent the product of the
x and yoriented wavelet expansions. The terms are the x
scaling/yscaling, xwavelet/yscaling, xscaling/ywavelet,
and xwavelet/ywavelet terms, respectively. In each cell,
represented by (i,j) in 2D, there is a set of 2
Dðr
max
þ1Þ
coef
ﬁcients, where D is the dimension of the simulator
[2
2ðrmax þ1Þ
in the 2D case]. It has been noted [24,25] that
an offset of 1=2
r
max
þ2
between the E and H cells in space
leads to a true doubling of resolution for each increase in
r
max
, and dispersion performance equivalent to an FDTD
grid that has cell spacings of the MRTD equivalent grid
points. For this scheme, there are 2
2ðr
max
þ1Þ
equivalent
gridpoints per cell, distributed evenly in the cell. These
gridpoints exist at each independent ﬁeld value that can
be determined by summing all of the basis functions in a
cell. It should be noted that there are as many of these
points as there are basis functions.
The MRTD update equations can be determined by in
serting the scaling/wavelet function expansions of the
ﬁelds into Maxwell’s equations and applying the method
of moments. The method of moments is applied by multi
plying both sides of the equation by user chosen testing
functions and then taking an integral over their domain.
In MRTD, Galerkin’s method is normally used, where the
testing functions are the same as the basis functions. For
1
1
1
−1
0
0 0.5
0
Figure 6. Haar basis scaling function f and mother wavelet c
0
.
1326 ELECTROMAGNETICS, TIMEDOMAIN
example, time localization (limiting the time coefﬁcients
required for an update to the current and previous time
step) is achieved by multiplying both sides of the equation
by the time pulse, positioned at the point of the ﬁelds in
the space derivative of the equation. As in FDTD, the co
efﬁcients for the ﬁeld values are interleaved in time, using
(38), the electric ﬁeld coefﬁcients are calculated at times
nDt, and in a similar expansion the magnetic ﬁelds are
calculated at times ðnþ
1
2
ÞDt. The method of moments can
be applied to discretize time as
_
h
½nþð1=2ÞDt
@E
x
@t
@t ¼
_
h
½nþð1=2ÞDt
1
e
@H
z
@y
@t ð39Þ
where h
[nþ(1/2)]Dt
represents a pulse of duration Dt cen
tered at time ðnþ
1
2
ÞDt. The integral
_
h
½nþð1=2ÞDt
h
mDt
@t ð40Þ
which appears on the left hand side of (39), evaluates to 1
when n¼m and n¼mÀ1 and 0 elsewhere. Thus, the only
electric ﬁeld components that are needed to update the
coefﬁcients are the current and previous timesteps. On the
right hand side of (39), the integral
_
h
½nþð1=2ÞDt
h
½mþð1=2ÞDt
@t ð41Þ
is nonzero only when n¼m. Thus, the magnetic ﬁeld com
ponents are only required at one location in time. A sim
ilar procedure can be applied for each of the wavelet and
scaling coefﬁcients, so that each the update of each wave
let/scaling coefﬁcient is in terms of its previous value and
the values of the wavelet/scaling values of the surround
ing ﬁelds.
Using a vector notation similar to [25], the following
results are derived
n
E
x
i;j
¼
nÀ1
E
x
i;j
þ
Dt
eDy
U
E
x
nÀ1
H
z
i;j
nÀ1
H
z
i;jÀ1
_ _
ð42Þ
n
E
y
i;j
¼
nÀ1
E
y
i;j
À
Dt
eDx
U
E
y
nÀ1
H
z
i;j
nÀ1
H
z
iÀ1;j
_ _
ð43Þ
n
H
z
i;j
¼
nÀ1
H
z
i;j
þ
Dt
m
1
Dy
U
H
Ex
nÀ1
E
x
i;j
nÀ1
E
x
i;j þ1
_
¸
_
_
¸
_
_
_
_
À
1
Dx
U
H
Ey
nÀ1
E
y
i;j
nÀ1
E
y
i þ1;j
_
¸
_
_
¸
_
_
_
_
ð44Þ
This vector representation appears very similar to its
FDTD analoge. To give an example of the vectors in
(42)–(44), if the maximum resolution r
max
¼0, then
n
E
x
i;j
¼
n
E
x;ff
i;j
n
E
x;cf
i;j
n
E
x;fc
i;j
n
E
x;cc
i;j
_
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
_
ð45Þ
The Umatrices represent the results of the inner products
from the method of moments. For the r
max
¼0 case, we
obtain
U
H
E
^ yy
ðr
max
¼0Þ ¼
À1 À1 0 0 1 1 0 0
1 À3 0 0 À1 À1 0 0
0 0 À1 À1 0 0 1 1
0 0 1 À3 0 0 À1 À1
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
ð46Þ
The MRTD method has a timestep requirement similar to
that of FDTD:
Dt ¼
1
c
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2
r
max
þ1
Dx
_ _
2
þ
2
r
max
þ1
Dy
_ _
2
¸ ð47Þ
In the case of Haar wavelets, the resolution r
max
is
equivalent to having 2
rmax þ1
FDTD gridpoints per cell.
As an alternative to the Haar techniques presented
here, other wavelet bases are commonly used with the
−2
1/2
2
1/2
−2
2
0.5 0 1
1
2
Figure 7. Haar wavelets for resolution levels 1 and 2.
ELECTROMAGNETICS, TIMEDOMAIN 1327
MRTD technique. Two of the most popular are the Battle–
Lemarie [6] and CDF wavelets [26]. These techniques
have the advantage of requiring the use of fewer basis
functions per wavelength than FDTD or Haar MRTD. One
disadvantage of these techniques is that it is difﬁcult to
apply hard boundary conditions, such as PECs and PMCs
using these basis functions. Extending these techniques to
allow the modeling of more structures is currently a busy
research area.
3.6. Parallelization
One common method of decreasing the execution time of
these computationally intensive codes is parallelization.
Code parallelization allows programmers to rewrite code
to distribute the problem over the several processors in a
multiprocessor system. This is becoming especially popu
lar with the advent of lowcost computer clusters. These
clusters, sometimes referred to as Beowulf clusters (as
named by their developers at NASA) or workstation clus
ters, allow the combination of several offtheshelf com
puters into a networked supercomputer. These computers
are significantly cheaper than traditional supercomput
ers, and several rank among the fastest in the world.
While the slower interconnects between the computers
can limit their application, most timedomain methods are
ideal for workstation cluster parallelization.
Most of the timedomain methods presented here are
explicit, meaning that the ﬁelds are updated directly from
other ﬁeld values and no matrices are inverted. Further
more, they are nearestneighbor techniques; the ﬁeld up
date at any given point depends only on the ﬁelds at the
neighboring points. Therefore, these techniques can be
parallelized by dividing the computational space into sub
grids. The only data that needs to be exchanged between
the processes are the data on the edges of the subgrids. As
these data are twodimensional, and the data to be updat
ed are threedimensional, the computation time grows on
the order of the number of gridpoints faster than the time
taken to pass data between grids. As the problem size in
creases, the communication time becomes less of a hin
drance, resulting in excellent performance on workstation
clusters.
4. IMPORTANT FEATURES OF TIMEDOMAIN
SIMULATION TECHNIQUES
4.1. Performing Simulations and Analyzing Output
All the simulators that have been presented in this article
are fullwave simulators. They are so named because they
simulate Maxwell’s equations without approximations,
and thus give the electric and magnetic ﬁelds over the
entire simulation domain as they evolve in time. For most
applications, the quantities of interest from simulation are
not the values of the ﬁelds, but rather macroscopic pa
rameters of the devices being simulated (such as S pa
rameters, input impedance, and bandwidth). In addition,
while the methods that have been presented here allow for
the simulation of ﬁeld propagation through various media,
no mention has been made of how to terminate the com
putational space, or introduce a source condition. This
section contains information about how to use the infor
mation obtained from these simulators.
It is important to remember when using these tools
that they simulate actual physical phenomena. As such,
when an excitation is introduced, it should closely match
the source condition in the physical circuit. This may in
clude introducing a gap voltage in a dipole antenna or a
current on a wire in a waveguide probe feed. In many ways
not only the circuit but also the measurement environ
ment are simulated. A network analyzer, for example, ex
cites a ﬁeld on a feedline and then analyzes the output of
the device. In simulation, ﬁelds can be probed in response
to an excitation. By doing this in as close a fashion to ac
tual measurement as possible, useful information can be
obtained. Of course, simulated ﬁelds can be injected and
measured at any point in space, even where it would be
physically impossible in practice. Used wisely, this allows
for results that are independent of many common sources
of measurement error.
An example structure that can be used to explain many
of these concepts is presented in Fig. 8. The presented
structure is a microstrip coupledline ﬁlter. The physical
structure being simulated consists of metal lines printed
on a dielectric substrate. This is a twoport structure, with
ports 1 (input) and 2 (output) labeled in the ﬁgure. If this
structure were to be fabricated and measured, connectors
would have to be added at the ports to allow connection to
the network analyzer. These could be either coaxial or co
planar waveguide depending on the measurement equip
ment. In either case, the input and output lines would
be made long enough to prevent coupling between the
measurement ports and the structure. In the case of
Absorber
PEC Line
Excitation
Probe location
2
1
Figure 8. Example simulation structure showing location of ab
sorbers, excitations, and probes.
1328 ELECTROMAGNETICS, TIMEDOMAIN
simulation, these connectors are not required. Instead, a
microstrip excitation is impressed on the grid. To prevent
reﬂections of the impressed signal from the input and out
put ports, an absorber is placed around the structure. This
numerical absorber is analogous to the absorbing materi
als used in antenna measurement systems. The voltage
under the microstrip is then measured at the input and
output ports. By simulating the structure over the entire
frequency band of interest and transforming the probed
voltages to the frequency domain, S parameters and other
important parameters such as input impedance and band
width can be determined. Field plots, showing the ﬁeld
distribution in a 2D cross section of the grid, in both the
time and frequency domains, can also be used to optimize
structures by highlighting areas that cause unwanted
crosstalk, radiation, loss, and maximum coupling.
4.2. Source Conditions
When exciting a structure it is necessary to consider both
the spatial distribution of the excitation (how much of the
grid is covered and which mode is excited) and the fre
quency content of the source signal. As most methods are
numerically dispersive, there is a griddependent maxi
mum frequency that can be accurately simulated. For the
simulation of RF and wireless circuits, it is usually desir
able to provide an excitation over a small portion of the
grid to represent a physical incident source condition. The
frequency content of the pulse is chosen to cover the fre
quencies of operation of the device. For many structures,
such as rectangular and circular waveguides, there are
closedform expressions for the modes that can propagate.
These expressions give relative ﬁeld magnitudes as a func
tion of space that are dependent on the structure. To excite
these structures, the mode shape must be discretized, and
then the ﬁeld values can be set to match the mode shape.
Due to discretization error, it is often difﬁcult to excite just
the modes of interest, and mode decomposition must be
performed when the output is analyzed. For other struc
tures, such as microstrip lines, no exact closedform ex
pressions exist. In this case, the mode shape is
approximated [8]. As the signal propagates through the
circuit, it will automatically assume the proper mode
shape. It also possible to build a library of these nonana
lytic excitations by simulating lines and storing the mode
shapes created on the line by a simpler excitation.
To restrict the excitation to a specific frequency band,
its waveform must be shaped in time. The frequency con
tent of the input signal can then be determined by Fourier
transforming the timeshaping function. As the simula
tion time must be ﬁnite, the actual frequency response of
any time waveform is inﬁnite. As such, pulse shapes that
have quickly decreasing magnitude as frequency is in
creased are necessary. In order to accurately control the
frequency content of the pulse, expressions with well
deﬁned Fourier transforms are often chosen. Commonly
chosen waveforms are Gaussian pulses, Gaussian deriva
tive pulses, sinusoidal pulses, and modulated Gaussians
(Gaussians multiplied by sinusoids). It is important to re
mark that while sinusoidal pulses will give the response to
a single frequency, this is not a very efﬁcient use of time
domain methods. One of the chief advantages of time
domain methods is that they simulate a broad band in
comparable time to a single frequency in a frequency do
main code. For this reason it is usually best to use a
broadband excitation.
Alternatives to these excitations are planewave sourc
es. For many applications, such as illuminating targets for
radar cross section evaluation or modeling bulk material
properties [e.g., electronic bandgap (EBG) and metamate
rials], the response to a planewave illumination is de
sired. In FDTD it is often useful to use a totalﬁeld/
scatteredﬁeld formulation [27]. In this formulation the
computational domain is divided into totalﬁeld region
that is surrounded by a scatteredﬁeld region. A connect
ing surface is developed that effectively removes the inci
dent ﬁeld from the scattered region. An alternate method
is a pure scatteredﬁeld formulation [28], in which the in
cident wave is not impressed on the grid, but rather the
response of the surface of the scatterer is applied, and thus
the scattered ﬁeld is generated. As the scatterer responses
must be calculated analytically, this can be a difﬁcult
method to apply, however, there is no error due to the in
cident ﬁeld propagating through the numerical grid.
4.3. Numerical Absorbers
For all of the presented methods, the ﬁelds at any one
point in time and space are dependent on the ﬁelds at
previous timesteps and neighboring spatial points. The
computational domain, however, must be ﬁnite. In order
to update the ﬁeld points, the ﬁelds on the boundaries
must be known. Two conditions on which the ﬁelds are
always known are perfect electrical conductors (PECs)
and perfect magnetic conductors (PMCs). On these sur
faces the tangential electric or magnetic ﬁelds, respective
ly, are zero. In techniques with staggered grids, such as
FDTD and TLM, these conditions can be applied by simply
zeroing the appropriate ﬁeld values. In collocated grid
techniques, such as FVTD and FETD, these conditions
must be applied using all the ﬁeld values. For either type
of simulator, these boundaries will cause total reﬂection of
an incident signal. If the response of a structure is to be
known to an incident signal, then the reﬂection from the
boundaries will combine with the response of the struc
ture, and the results will be useless. To use these hard
boundary conditions, the computational space must be
sufﬁciently large for the time response of the device to end
before reﬂections interfere. This can easily result in im
practically large grids. Therefore, another technique, an
absorbing (sometimes called radiating) boundary condi
tion (ABC), must be applied. The goal of these techniques
is to absorb incoming waves with a minimum of numerical
reﬂection.
There are several techniques for applying absorbing
boundary conditions (ABCs), most of which can be applied
to the methods presented here. A number of these methods,
usually referred to by the surnames of their developers,
such as Bayliss–Turkel [29], Mur [30], and Ramahi [31],
can often reduce reﬂections by 50–80dB. Another tech
nique, which has become the standard for numerical ab
sorbers, is the perfectly matched layer (PML). Introduced
ELECTROMAGNETICS, TIMEDOMAIN 1329
by Berenger in 1994 [32], the PML is manifested in a sim
ulator as a physical material with very specific properties.
It offers excellent performance over a wide frequency band
and range of incident angles. It has been shown to be
equivalent to a similar method, the uniaxial perfectly
matched layer (UPML) [33]. The UPML is the numerical
representation of a uniaxial anisotropic medium with both
electric and magnetic permittivity tensors, that offers nu
merical reﬂection coefﬁcient lower than À100db for most
RF open structures.
4.4. S Parameters
The output of a timedomain simulator is the values of the
electric and magnetic ﬁelds at all points in space over the
time period simulated. To calculate useful characteristics
of the device being simulated, these responses must be
transformed into meaningful parameters. The S parame
ters are frequencydomain characteristics that provide the
response at a port of a structure due to an incident signal.
As such, in timedomain simulators, the input and output
voltages/currents on a structure can be recorded, convert
ed to the frequency domain, and postprocessed to give S
parameters. The structure in Fig. 8 can be used as an
example for determining S parameters.
The nature of the input and output signal depends on
the type of device being simulated. For a quasiTEM struc
ture, such as the microstrip of Fig. 8, the voltage from
ground to signal line and the current in the signal line can
be used to determine S parameters. The voltage can be
determined by summing the electrical ﬁelds normal to the
line from the ground to the signal line, numerically inte
grating
V ¼ À
_
b
a
EÁ ds ð48Þ
Likewise, the current on the line can be determined using
I ¼
_
HÁ dl ð49Þ
which is numerically equivalent to summing the magnetic
ﬁelds tangential to a contour surrounding the conductor.
These parameters are then transformed to the frequen
cy domain using either a fast or discrete Fourier trans
form. In order for the Fourier transform to be accurate, the
simulation must run until the ﬁelds at the measurement
points go to zero (or a very small value relative to their
peaks, usually less than 0.1%). Once the frequency domain
voltages/currents are determined, they can be processed to
determine the S parameters. The general formula for an S
parameter is to divide the output signal at port m by the
incident signal on port n:
S
mn
ðoÞ ¼
V
m;out
ðoÞ
V
n;in
ðwÞ
Z
n
ðoÞ
Z
m
ðoÞ
_ _
1=2
ð50Þ
The expression in (50) is valid for the general case where
the impedance of the input and output ports are not equal.
It should be noted that this is not always the ratio of the
total signal at the input and output probes.
The ﬁelds recorded during the simulation are the total
ﬁelds. For example, the input probe of Fig. 8 will record
the incident ﬁeld from the excitation as well as the reﬂec
tion from the ﬁlter. For the computation of S
21
, however,
voltage used in (50) must be only the incident. In this case
the incident voltage can be determined by performing a
separate simulation with an identical probe on an identi
cal transmission line (without the ﬁlter). To compute S
11
,
the output ﬁeld can be computed by subtracting the inci
dent from the total ﬁeld.
One other parameter that can be easily calculated us
ing voltage and current probes is the characteristic im
pedance of a transmission line. For a signal traveling
through a transmission line, the characteristic impedance
can be determined as the ratio of the voltage to the current
on the line. It is important to note that to measure these
values, only one signal should propagate on the line, with
no reﬂections. For most simulators, exciting a transmis
sion line terminated with PML on both ends will sufﬁce.
For a simulator that has electric and magnetic ﬁeld collo
cated in time and space, the impedance formula is
Z
^
ðoÞ ¼
V
^
ðoÞ
I
^
ðoÞ
ð51Þ
while, for other simulators, the offset must be accounted
for. In FDTD, for example, the halfspacecell offset can be
accounted for by using the geometric average of two cur
rent probes that surround a voltage probe, and an expo
nential multiplier can be used to compensate for the half
timestep offset. For FDTD, the impedance formula is [34]
Z
^
0
ðxÞ ¼
V
^
ðxÞe
ÀjoðDt=2Þ
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
I
^
ðx þ
1
2
Þ I
^
ðx À
1
2
Þ
_ ð52Þ
This discussion has focused on microstrip structures;
however, it can easily be extended to more general lines.
For any given structure, it is necessary to determine what
output voltage can be used to characterize the device. For
a coplanar waveguide, for example, the energy is also car
ried between the signal and ground lines; however, they
are in the same plane and thus the recorded voltage is
tangential to the lines. Similarly, stripline, slotline, and
coaxial structures can be characterized by determining
the voltage between the appropriate conductors. Finally,
using the S parameters, other important parameters such
as input impedance, insertion loss, and bandwidth can be
directly calculated.
4.5. NearField/FarField Transformation
Antennas are often simulated with timedomain simula
tors. The techniques presented in the previous section can
be used to calculate antenna parameters such as input
impedance and bandwidth, but they do not give the po
larization, directivity, gain, and many other antenna pa
rameters. To analyze the antenna in terms of these
1330 ELECTROMAGNETICS, TIMEDOMAIN
parameters, it is necessary to know the farﬁeld radiation
pattern of the antenna. The far ﬁeld is often tens of wave
lengths away from the antenna structure, and an enor
mous grid is required for accurate simulation.
Fortunately, there is another method that can be used to
determine the farﬁeld radiation pattern, the nearto
farﬁeld (NFF) transform.
The farﬁeld radiation pattern is calculated by ﬁrst re
ducing the radiation problem to one set of electric and
magnetic equivalent sources on a closed surface surround
ing the radiating structure. The problem of an antenna
radiating into free space is demonstrated in Fig. 9, in
which the entire antenna and feeding structure are sur
rounded by free space. This space is terminated with an
absorbing boundary condition, such as PML. A transfor
mation surface is placed in between the antenna and the
absorber. This surface has no effect on the simulation; it is
used only to record ﬁelds.
The surface equivalence theorem [35] states that the
problem in Fig. 9 can be represented by electric and mag
netic current sources placed on the NFF transformation
surface. Using these current sources, the ﬁelds outside the
surface are the same as those produced by the radiator,
and the radiation at any point outside the contour can be
determined mathematically. The ﬁrst step in performing
the NFF transformation is to determine the equivalent
currents on the transformation surface.
If the electric and magnetic ﬁelds on the surface are
known, the sources can be calculated as
J
^
s
¼nÂH
^
ð53Þ
M
^
s
¼ ÀnÂE
^
ð54Þ
where n is the unit vector normal to the transformation
surface. Of course, the antenna pattern has meaning only
as a function of frequency, and thus the ﬁelds on the
transformation surface must be transformed to the fre
quency domain. To Fouriertransform the ﬁelds, the ﬁeld
values at all timesteps are required. As the surface com
pletely surrounds the antenna, and thus consists of a very
large number of points, the Fourier transform is per
formed during simulation to avoid the burden of storing
the ﬁeld values for all points on the surface at all time
steps. Once the currents are known, they can be used to
solve for the far (or near)ﬁeld radiation pattern. The
equation that gives the radiation at any point in space
due to these sources can be simpliﬁed to give results for
only the far ﬁeld. For these simpliﬁcations, the distance
from any point on the radiating surface to an observation
point is assumed to be the same, except for the phase
terms. This is possible because the source is very far from
the observation point.
Using these approximations, the electric ﬁeld in the far
ﬁeld is [35]
E
^
r
% 0 ð55Þ
E
^
y
% À
jke
Àjkr
4pr
ðL
^
f
þZ
0
N
^
y
Þ ð56Þ
E
^
f
% À
jke
Àjkr
4pr
ðL
^
y
ÀZ
0
N
^
f
Þ ð57Þ
where
N
^
¼
_
S
_
J
^
s
e
jkr
0
cos c
dS ð58Þ
L
^
¼
_
S
_
M
^
s
e
jkr
0
cos c
dS ð59Þ
where r
0
is the distance between the origin and the source
point on the contour and c is the angle between the vector
to the surface and the vector to the observation point.
This procedure yields the far electric and magnetic
ﬁelds at any position on a sphere surrounding the anten
na. By normalizing these ﬁelds, one can plot the radiation
pattern. The total radiated power can be found by calcu
lating the Poynting vector and integrating over the sur
face. Using the total radiated power and the ﬁeld values,
directivity and gain can be determined. This technique
demonstrates that while the electromagnetic simulation
techniques discussed here are powerful tools, analyzing
the output can also be a difﬁcult, timeconsuming task.
5. SUMMARY
An overview of the purpose and application of time
domain electromagnetic simulation methods has been pre
sented. These methods serve as powerful tools for RF
designers if used correctly. The reader should take from
this section not a complete understanding of timedomain
computational methods, but rather an appreciation of how
they can be applied and, hopefully, where mistakes can be
avoided in simulation. Often the choice of simulator de
pends on software already at hand, and this article should
help the reader gain conﬁdence in determining how dif
ferent packages can be applied. In addition, this article
Antenna
ABC
NFF transformation surface
Figure 9. Diagram demonstrating setup of antenna simulation.
ELECTROMAGNETICS, TIMEDOMAIN 1331
provides a strong background that can be used as a start
ing point for further research in timedomain electromag
netic methods. Through the discussion not only of the
techniques but also how to analyze the results, the use
fulness of timedomain simulation techniques has been
demonstrated. Almost any problem can be simulated us
ing these techniques, and understanding their operation
is the ﬁrst step in using them properly.
BIBLIOGRAPHY
1. M. Schuh and A. Woo, Code scaling, IEEE Anten. Propag.
Mag. 37(4):82–86 (Aug. 1995).
2. S. D. Gedney, A comparison of the performance of ﬁnite dif
ference timedomain, ﬁnite element timedomain, and dis
crete surface integral equation methods on high performance
parallel computers, Proc. IEEE Antennas and Propagation
Symp., June 1994, Vol. 1, pp. 384–387.
3. K. S. Yee, Numerical solution of initial boundary value prob
lems involving Maxwell’s equations in isotropic media, IEEE
Trans. Anten. Propag. AP14:302–307 (March 1966).
4. V. Shankar, A. Mohammadian, and R. W. Ziolkowski, A three
dimensional modiﬁed ﬁnite volume technique for Maxwell’s
equations, Electromagnetics 10:147–161 (1990).
5. A. C. Cangellaris, C. Lin, and K. Mei, Pointmatched time
domain ﬁnite element methods for electromagnetic radiation
and scattering, IEEE Trans. Anten. Propag. 35(10):1160–1173
(Oct. 1987).
6. M. Krumpholz and L. P. B. Katehi, New time domain schemes
based on multiresolution analysis, IEEE Trans. Microwave
Theory Tech. 44:555–561 (April 1996).
7. P. B. Johns and R. L. Beurle, Numerical solution of 2dimen
sional scattering problems using a transmissionline matrix,
Proc. Inst. Electrical Eng. 118(9):1203–1208 (Sept. 1971).
8. A. Taflove and S. C. Hagness, Introduction to Maxwell’s equa
tions and the Yee algorithm, in A. Taflove, ed., Computational
Electrodynamics: The Finite Difference Time Domain Method,
Artech House, Norwood, MA, 1995, Chap. 3.
9. F. Zheng, Z. Chen, and J. Zhang, Towards the development of
a threedimensional unconditionally stable ﬁnitedifference
timedomain method, IEEE Trans. Microwave Theory Tech.
48(9):1550–1558 (Sept. 2000).
10. J. R. Whinnery and S. Ramo, A new approach to the solution
of highfrequency ﬁeld problems, Proc. IRE 32:284–288 (May
1944).
11. J. R. Whinnery, C. Concordia, W. Ridgway, and G. Kron, Net
work analyzer studies of electromagnetic cavity resonators,
Proc. IRE 32:360–367 (June 1944).
12. W. J. R. Hoefer, The transmission line matrix (TLM) method,
in T. Itoh, ed., Numerical Techniques for Microwave and Mil
limeterWave Passive Structures, Wiley, New York, 1989,
Chap. 8.
13. C. Christopoulos, The TransmissionLine Modeling Method:
TLM, Oxford Univ. Press, Oxford, UK, 1995.
14. P. B. Johns, The solution of inhomogeneous waveguide prob
lems using a transmissionline matrix, IEEE Trans. Micro
wave Theory Tech. MIT22:209–215 (March 1974).
15. A. C. Cangellaris and D. B. Wright, Analysis of the numerical
error caused by the stairstepped approximation of a conduct
ing boundary in FDTD simulations of electromagnetic phe
nomena, IEEE Trans. Anten. Propag. 39(10):1518–1525 (Oct.
1991).
16. N. Madsen, Divergence Preserving Discrete Surface Integral
Methods for Maxwell’s Equations Using Nonorthogonal
Unstructured Grids, Technical Report UCRLJC109787,
Lawrence Livermore Natl. Lab., Berkeley, CA, Feb. 1992.
17. S. Gedney, F. Lansing, and D. Rascoe, A generalized Yee
algorithm for the analysis of MMIC devices, IEEE Trans.
Microwave Theory Tech. 44:1393–1499 (Aug. 1996).
18. P. Monk and E. Suli, Error estimates for Yee’s method on non
uniform grids, IEEE Trans. Magn. 30(5):3200–3203 (Sept.
1994).
19. D. A. AlMukhtar and J. E. Sitch, Transmissionline matrix
method with irregularly graded space, IEE Proc. H: Micro
waves, Opt. Anten. 128:299–305 (Dec. 1981).
20. J. Ritter and F. Arndt, Fast subgrid FDTD matrix pencil
technique for the rigorous analysis of resonant 3D microwave
structures, Proc. IEEE Microwave Symp., June 1998, Vol. 3,
pp. 1271–1274.
21. S. Dey and R. Mittra, A locally conformal ﬁnitedifference
timedomain algorithm for modeling threedimensional per
fectly conducting objects, IEEE Microwave Guided Wave Lett.
7:273–275 (Sept. 1997).
22. HFSS, Ansoft Corporation, Pittsburgh, PA.
23. E. M. Tentzeris, R. L. Robertson, J. F. Harvey, and L. P. B.
Katehi, Stability and dispersion analysis of BattleLemarie
based MRTD schemes, IEEE Trans. Microwave Theory Tech.
47(7):1004–1013 (July 1999).
24. C. Sarris and L. P. B. Katehi, Fundamental griddingrelated
dispersion effects in multiresolution timedomain schemes,
IEEE Trans. Microwave Theory Tech. 49(12):2248–2257 (Dec.
2001).
25. T. Dogaru and L. Carin, Application of Haarwaveletbased
multiresolution timedomain schemes to electromagnetic
scattering problems, IEEE Trans. Anten. Propag. 50(6):774–
784 (June 2002).
26. T. Dogaru and L. Carin, Efﬁcient timedomain electromagnet
ic analysis using CDF biorthogonal wavelet expansion, Proc.
IEEE Int. Microwave Symp., May 2001, Vol. 3, pp. 2019–2022.
27. K. R. Umashankar and A. Taflove, A novel method to analyze
electromagnetic scattering of complex objects, IEEE Trans.
Electromagn. Compat. 24:397–405 (1982).
28. R. Holland, Threde: A ﬁeld free EMP coupling and scattering
code, IEEE Trans. Nucl. Sci. 24:2416–2421 (1977).
29. A. Bayliss and E. Turkel, Radiation boundary conditions for
wavelike equations, Commun. Pure Appl. Math. 23:707–725
(1980).
30. G. Mur, Absorbing boundary conditions for the ﬁnitedifference
approximation of the timedomain electromagnetic ﬁeld equa
tions, IEEE Trans. Electromagn. Compat. 23:377–382 (1981).
31. O. M. Ramahi, The complementary operators method in
FDTD simulations, IEEE Anten. Propag. Mag. 39(6):33–45
(Dec. 1997).
32. J. P. Berenger, A perfectly matched layer for the absorption of
electromagnetic waves, J. Comput. Phys. 114:185–200 (1994).
33. S. Gedney, An Anisotropic Perfectly Matched Layer Absorbing
Media for the Truncation of FDTD Lattices, Technical Report
EMG95006, Univ. Kentucky, Lexington, KY, May 1997.
34. J. Fang and D. Xeu, Numerical errors in the computation of
impedances by FDTD method and ways to eliminate them,
IEEE Microwave Guided Wave Lett. 5(1):6–8 (Jan. 1995).
35. A. Taflove and S. C. Hagness, Neartofarﬁeld transforma
tion, in A. Taflove, ed., Computational Electrodynamics:
The Finite Difference Time Domain Method, Artech House,
Norwood, MA, 1995.
1332 ELECTROMAGNETICS, TIMEDOMAIN
ELECTRONIC WARFARE
ANTHONY E. SPEZIO
ALAN N. DUCKWORTH
FRANCIS J. KLEMM
STANLEY A. MOROZ
JAMES M. TALLEY
Naval Research Laboratory
Electronic warfare (EW) is the systems discipline that ex
ploits an adversary’s use of the electromagnetic spectrum
to overcome threats that use communications, navigation,
and radar systems. It is an important tool in pursuing
military objectives and advancing national policy and sov
ereignty. EW provides the means to counter, in all battle
phases, hostile actions that use the electromagnetic spec
trum—from the beginning, when enemy forces are mobi
lized for an attack, through to the ﬁnal engagement. EW
exploits the electromagnetic spectrum through electro
magnetic sensing, analysis, and countermeasures to
establish operational advantage in a hostile encounter.
The use of electronic warfare accelerated rapidly dur
ing World War II, and it has been used in most military
conﬂicts since. The aircraft used by Nazi Germany to
bomb the fogshrouded British Isles were guided by radio
beacons from the European mainland. By using false guid
ance signals, the British were able to redirect the German
bombing attacks from densely populated urban areas to
less populated rural areas. In this same conﬂict, U.S.
bombers used chaff (packets of tinfoil cut into thin strips)
jettisoned from the attacking U.S. aircraft to reﬂect anti
aircraft radar signals, thereby reducing the effectiveness
of the German antiaircraft batteries and bomber force at
trition. In the Pacific theater of operations during World
War II, U.S. Navy submariners detected and determined
the bearing and location of Japanese ship radio transmis
sions for weapons targeting. In the Korean conﬂict, detec
tion and location of North Korean antiaircraft radar
signals provided targeting data for subsequent air strikes.
In Vietnam, the exploitation of antiaircraft and missile
radars was reﬁned with the use of U.S. Air Force Wild
Weasel weapons—suppression aircraft that used sensors
to detect and locate the weaponsassociated threat signals
to provide targeting information for ordnance delivery.
Electronic warfare applications are described extensively
in military accounts of the past half century.
Military operations use EW as one means to gather
tactical intelligence from noncooperative forces and to
counter their electromagnetic, radio, and radarcon
trolled weapons. Land, sea, and air forces use the electro
magnetic spectrum for command and control, weapons
targeting, and weapons control. Figure 1 shows multiple
land, sea, and air platforms in a typical tactical environ
ment. Also indicated are links for sensing, communica
tions, and navigation in support of the military mission.
Electronic warfare provides use of the electromagnetic
(EM) spectrum by the host force and denial or limitation of
its use by an adversary. Realization of this goal occurs
when host force systems use the EM spectrum while ad
versary systems are denied its use. Countermeasures
(CM) to threat systems that use the EM spectrum can be
selectively applied on a time and/or frequencymulti
plexed basis so that host force use of the EM spectrum is
uninhibited.
Electronic warfare includes the operational functions of
electronic support (ES), electronic selfprotection (EP),
and electronic attack (EA). ES provides surveillance and
warning information for EW system use. CM to threat
systems, including jamming, falsetarget generation, and
decoying, are performed for EP (protection of the host
platform against an electronically controlled threat). EA
performs these same CM functions to protect a battle force
Figure 1. Tactical operational concept indi
cating systems that use the EM spectrum.
ELECTRONIC WARFARE 1333
composed of several platforms or battle units. The ES, EA,
and EP functions are interrelated because EA and EP can
be queued using ES information, and EA and EP can use
some of the same sensing and CM equipment for distinct
operational objectives.
This article includes a description of the EW timeline
and the various phases of conﬂict. Also provided is a sum
mary description of the signal environment in which EW
systems operate. Those interested in more detailed de
scriptions of the EM communications, radar, and naviga
tion technology against whose signals EW systems
operate are referred to the appropriate sections of this en
cyclopedia. A discussion of EW functional areas ES, EP,
and EA provides a functional framework for supporting
EW technologies.
1. ELECTRONIC WARFARE TIMELINE
Electronic warfare is used in a layered operational interac
tion with electronically controlled threat systems. The elec
tronic warfare system provides its own force with data for
selfprotection and threat weapons suppression. Figure 2
graphically illustrates the EW functional timeline.
Electronic support provides operational intelligence re
lating to electronically controlled threat systems and com
munications systems in the battle group or theater
environment. Electronic threatwarning information de
rives from ES surveillance data, recognizing that hostile
force deployments or weaponsrelated transmissions con
stitute a threat. Air defense combines electronic and radar
surveillance with tactics and countermeasures to control
the air battle. EA and active EP, using countertargeting
(CTAR) jamming, false target generation, and/or decoying,
attempt to deny target acquisition by adversary sensors.
CTAR endeavors to deny weapon’s sensors use of the spec
trum, and decoys dispersed into the environment provide
preferred target signatures to the threat weapon’s sensor.
The EW battle timeline provides the general context in
which the discipline of EW is used in the tactical environ
ment. The EW timeline stage in a specific engagement
depends on the deployment of forces and the perceived
imminence of hostile engagement. Note that the technol
ogies used in the various stages of the engagement are
dynamic, and EW systems and weapon systems technol
ogies evolve to overcome susceptibilities. The boundaries
and definitions of EW timeline stages are redeﬁned with
each new advance in weapon and EW technology.
1.1. Electronic Support
Electronic support provides operational intelligence that
is related to radiated signals in the battle group or theater
environment. Surveillance includes monitoring of both
combatants and commercial transports. Control of contra
band and critical materials is an EW surveillance mission
that provides critical intelligence data to the area com
mander. Surveillance of noncooperative combatant forces
provides deployment intelligence in the area of observa
tion. Early threatwarning information extracted from
surveillance data occurs by recognizing hostile force weap
onsrelated transmissions.
Within the lethal range of hostile force weapons, battle
space surveillance updates are required rapidly. Deploy
ment and operational modes of hostile forces are moni
tored closely to determine imminence of hostile activity. In
some environments, potentially hostile forces remain
within weapons’ lethal range and a high level of vigilance
is necessary to maintain security.
1.2. Air Defense
Air defense is used to maintain control of the battle group
airspace and defend against threat aircraft and missiles.
Battle group surveillance, implemented by the combina
tion of EW, infrared/electrooptic (IR/EO), and radar sen
sors, provides environmental data required for air
defense. Electronic combat techniques and weapons are
used to counter an airborne threat.
Air defense is an extensive, complex, electronic combat
interaction between hostile forces. EW assets are a key
tool of the battle force commander and of the individual
elements within the command. These assets provide in
formation for developing tactical intelligence in all phases
of the engagement. The outcome of the air battle is by no
means established by the quantity of EWassets possessed
by each of the opposing forces, but depends greatly on how
the EW assets are used in conjunction with other sensor
systems, weapons, and air defense tactics.
Aircraft ships and/or battleﬁeld installations partici
pate in air defense. Own force aircraft operating at alti
tude can engage a threat force at long lineofsight ranges.
Aircraft, together with ship and battleﬁeld installations,
provide coordinated air defense as the hostile force ap
proaches own force locations. The EWobjective in the ear
ly air defense or outer air battle is to prevent threat force
detection and location of own force. Electronic combat ac
tions that prevent or delay own force detection provide a
distinct advantage by allowing additional time to develop
tactics to counter the threat force. In addition, the threat
force battle timeline and interplatform coordination are
perturbed. Fragmentation or dissolution of the hostile
Surveillance and warning
Air defense
Countertargeting
Missile
defense
Own force
Hostile force
Figure 2. Electronic warfare battle situation showing various
phases of the engagement timeline.
1334 ELECTRONIC WARFARE
force attack can occur if own force electronic combat is ef
fective in the outer battle.
As the hostile force overcomes the outer battle elec
tronic attack and approaches the own force within weap
ons range, air defense assumes the role of denying
targeting information to the hostile sensors. The EW ob
jective at this stage of the engagement is to prevent hostile
force weapons launch by denying targeting data to their
sensors. Electronic combat surveillance, warning, and
countermeasure assets are used for countertargeting. Sur
veillance sensors assess hostile force deployment and pro
vide information about the adversarial tactics being used.
Warning sensors indicate the status of threat sensors as
they attempt to acquire targeting data for weapons sys
tems handoff. Countermeasure assets, including jamming,
spoofing, and decoying, continue to provide a virtual en
vironment to threat platform sensors to prevent own force
target acquisition by the hostile force.
The terminal phases of an air defense engagement are
characterized by heightened activity. The combatants,
both hostile and own force, are conﬁned to a smaller por
tion of the battle space. Weapons and decoys in ﬂight add
to the physical and EM signal density. Electronically, both
own force and hostile forces struggle to exploit the EM
environment to achieve their respective operational objec
tives. Countermeasures jamming and spoofing are used
with full appreciation that coordinated jamming produces
degradation of hostile force sensors, but that weapons
with homeonjam (HoJ) capability can exploit this action
to the destructive detriment of the radiating platform.
1.3. Countertargeting
Countertargeting (CTAR) is a subset of radar electronic
countermeasures (ECM) used in electronic attack. CTAR
provides specially modulated radiofrequency (RF) signal
transmissions to counter hostile force longrange surveil
lance or targeting radar. The transmission modulation can
be amplitudemodulated (AM) or frequencymodulated
(FM) noise, or combinations of these, and they can be
pulsed or continuouswave. CTAR transmission is used
both to disrupt and interfere with the threat radar oper
ation, thereby preventing it from correctly locating and
identifying own force target(s).
Countertargeting success criteria includes mission
completion prior to threat force interdiction or weapon
launch. Realistically, the results of a CTAR electronic at
tack against a hostile force are probabilistic, in that some
opposing forces at some time during the battle timeline
succeed in launching missiles. CTAR can delay and reduce
the coordination of hostile missile ﬁrings and, consequent
ly, reduce the number of missiles ﬁred and the attrition of
personnel, ships, and aircraft.
1.4. Terminal Defense
Terminal defense against electronically controlled mis
siles and guns is the ﬁnal phase of the EW battle time
line. Weapons are launched in the terminal phase of
hostile force engagement, and EP and EA capability is
brought to bear on the weapons and their electromagnetic
(EM) guidance and control signals. Onboard jamming and
falsetarget radiation that is effectively used for counter
targeting is less effective for terminal defense. Jamming or
falsetarget radiation makes the target platform vulnera
ble to missiles with homeonjam capability. Home on jam
is an electronic countercountermeasure that exploits the
target countermeasure’s radiation to steer the missile to
the target. Consequently, offboard countermeasures, or
decoys, are used to lure the missile away from the high
value target.
2. THE ELECTRONIC WARFARE ENVIRONMENT
2.1. Threat Systems
Electronic warfare interacts with an adversary’s EM sys
tems for signal exploitation and potentially for electronic
attack. Threat systems of EW interest include radar, com
munications, and weapons control. Some of the threat
systems exploited by EW are briefly described in the fol
lowing.
2.1.1. Radar. Radar uses radiofrequency transmissions
ranging from high frequency (HF) to millimeter waves
(from 30 MHz to 40 GHz) in pulsed and continuouswave
(CW) modes to illuminate targets and collect reﬂected ech
oes. Radartransmissionreﬂected echoes are used to mea
sure target characteristics and determine target location.
Military forces use radar for both offensive and defensive
weapon systems. Radar functions include target detection
and identiﬁcation, target acquisition, target tracking, and
navigation. Weapons systems using radar may be land
based, airborne, shipboard, or in space. A typical radar
system contains a transmitter that produces a highpow
ered RF signal, tunable over a band of frequencies; an an
tenna system that radiates energy and collects reﬂected
echoes; a receiver that detects signal return; and signal
processing electronics that extract target measurements,
such as range, bearing, and speed. Target location infor
mation is provided to a weapon system to control and
direct the weapon onto the target.
Landbased radars function as groundcontrolled inter
cept (GCI) systems, surfacetoair missile (SAM), antiair
craft artillery (AAA) batteries, and space tracking
systems. GCI is used to direct interceptor aircraft against
attacking aircraft and to coordinate the air battle. SAM
sites use early warning/surveillance radar, target acquisi
tion radar, target tracking (TT) radar and/or illuminators
for missile guidance, beamriding systems. AAA radars
have operating frequencies and data rates similar to those
of SAM tracking radars and usually receive targeting in
formation from SAM surveillance and target acquisition
(TA) facilities. Advanced SAM systems handle ballistic
missile defense with higher data rates against highspeed
targets. Airborne intercept and control (IC) radars provide
early warning and information for the command and con
trol of forces operating in the tactical environment. Space
surveillance and tracking radars usually use large, ﬁxed,
phased arrays operating in the HF (3–30 MHz) to 1 GHz
frequency range. Table 1 gives parameters of typical ra
dars categorized by radar function. The reader is referred
ELECTRONIC WARFARE 1335
to radar and electromagnetic wave propagation articles
within this encyclopedia.
Radar advancements can be expected in the areas of
phasedarray antennas, complex modulations on the ra
dar pulse, improved signal processing to extract enhanced
data from the radar return, and frequency diversity to
cover the less used regions of the spectrum. Advanced de
signs from the U.S., European, and Russian inventories
can be expected because of operational needs for enhanced
sensor performance and the availability of affordable tech
nologies to provide additional capability.
2.1.2. Communications. Communications systems pro
vide information exchange for command and control to
coordinate between surveillance sites and between combat
units. Communications networks range from basic ﬁeld
radio networks to longdistance, widearea systems and
pointtopoint, highdatarate installations. Communica
tions systems cover the spectrum from very low frequency
(5Hz) to the frequencies of visible light, and they can be
either freespace transmissions or conﬁned to a transmis
sion line. Freespace transmission links may be line of
sight or cover longer distances by reﬂecting from the ion
osphere, atmospheric layers, or troposcatter, or by relay
ing via satellite.
Command and control communication links, using HF
direct microwave and satellite relay, disseminate voice and
digital data transmissions to land forces, air forces, and
ships. Land combat units use ultrahigh frequency (UHF)
(300MHz–3GHz), very high frequency (VHF) (30–
300MHz), landlines, and cellular phones over shorter dis
tances mainly for voice transmissions. Surveillance activ
ities and weapons sites may exchange data via voice or
digital data link over a transmission path appropriate for
the link span. Such links are used to transmit surveillance
radar reports to an operations center or directly to a SAM
battery. Communication link data rates depend on link
bandwidth, modulation technique, and signaltonoise ra
tio. Individual transmission link throughput rates are in
the range of hundreds of megabytes per second. Computer
technology has enabled increased communication link ca
pacity for handling and processing data. The high data
rates attainable permit transmission fromairborne observ
ers and between precision weapons and launch platforms.
Communications in hostile environments are trans
mitted via protected cable between ﬁxed sites, thus
providing protection from physical damage, security
from intercept, and immunity from jamming. Mobile com
munications require freespace transmissions that are
susceptible to intercept and jamming. Communications
countercountermeasures, complex modulation, encryp
tion, and spatial radiation constraints are used to miti
gate the effects of EA. The use of modulation techniques
increases privacy, reduces interference, improves recep
tion, and reduces the probability of detection. Spread
spectrum communication systems that use four categories
of signal modulation [direct sequencemodulated, frequen
cy hopping, intrapulse FM (chirp), and time hopping] pro
vide some level of signal protection from detection,
demodulation, and interference. However, this is at the
expense of increased bandwidth.
2.1.3. Passive Weapons Sensors. Electrooptical and in
frared (EO/IR) systems sense spectral energy that is ra
diated by an object or reﬂected from an object from a
source such as the sun, moon, or stars. The electrooptical
spectral regions are categorized according to atmospheric
propagative characteristics or spectral transmittance. The
EO/IR spectrum used for passive weapons sensors spans
the 0.2–15 mm wavelength range.
Electrooptical/infrared guidance provides angle target
tracking information only. EO/IR weapons system guid
ance sensors fall into three classes: nonimaging, pseudo
imaging, and imaging. Generally, countermeasure
techniques exhibit preferential effectiveness against guid
ance approach. Some countermeasures techniques may be
effective against pseudoimaging sensors and less effective
against nonimaging and imaging sensors. Other counter
measures techniques may be preferentially effective
against nonimaging and imaging sensors. Figure 3 illus
trates the most common seeker design approaches. These
approaches are quadrant, spin scan, conical scan (con
scan), transverseline scan, and rosette scan. In the quad
rant approach, an intentionally defocused spot images on
a fourelement square array. Tracking is achieved by bal
ancing the signal on all four detectors. In spin scan, a
spinning reticle provides phase and amplitude informa
tion with respect to a ﬁxed reference. With conscan, the
target image is nutated by using a scanning mirror or op
tical wedge imaged onto a ﬁxed reticule or pattern of de
tectors. The nutated target image generates a modulated
frequency proportional to the angular and radial offset
from the center. In the transverseline scan approach, a
rotating or reciprocating mirror at a depressed elevation
angle generates a scanline transverse to the missile axis,
and the forward motion of the missile creates the orthog
onal axis of the search pattern. With the rosette scan, a
petal pattern is scanned over a small instantaneous ﬁeld
of view (IFoV) by two counterrotating optical elements.
Rosettescan tracking is accomplished by balancing the
signal output from all petals with the target present in the
central apex of the rosette. The small IFoV of the trans
verseline scan and rosette scan provide high spatial
resolution and the ability to resolve multiple sources
Table 1. Parameter Ranges Associated with Radar Functions
Radar Function
Radar Parameter GCI IC Surveillance TA TT, AA Space Surveillance
Frequency range 30MHz–3.0GHz 3.0–10.0GHz 30MHz–3.0GHz 3.0–8.0GHz 6.0–10.0GHz 30MHz–1.0GHz
PRF range 100–500pps 1000–3000pps 100–500pps 1000–2000pps 2000–4000pps —
1336 ELECTRONIC WARFARE
within the scanned ﬁeld of view. Focal plane arrays, scan
ninglinear arrays, or twodimensional arrays of detectors
in the image plane provide highresolution ‘‘pictures’’ of
the target space. Many imageprocessing algorithms are
available to classify targets and establish track points.
Figure 3 illustrates the basic features of common seekers.
Passive electrooptic sensors are desirable targeting
and weapons guidance systems because they radiate no
energy to warn the target of an impending attack. These
sensor systems are vulnerable to decoys, with thermal
signatures similar to true targets and to highintensity
sources that can saturate the electrooptic sensor detector
or cause physical damage.
3. ELECTRONIC WARFARE FUNCTIONAL AREAS
Threat systems use the EM spectrum extensively.
This section discusses functional aspects of EW. The
relationships that govern their systems’ application are
described in the following section. These functional areas
are electronic support (ES), electronic protection (EP), and
electronic attack (EA). Electronic attack uses countertar
geting (CTAR), jamming, falsetarget generation, and de
coys to defeat the threat sensors. Electronic protection
uses electronic support and electronic attack for ownplat
form selfprotection.
3.1. Electronic Support
Electronic support provides surveillance and warning in
formation to the EW system. ES is a passive, nonradiat
ing, EW system function that provides a fast, accurate
assessment of the EM radiating environment. ES is the
aspect of EW that involves techniques to search for, inter
cept, locate, record, and analyze radiated energy for ex
ploitation in support of military operations. Electronic
support provides EW information for use in EA and EP
Image
Error
Image
decenter
Nonimaging reticles
Pseudoimaging
Spinscan
reticle
Time
Time
Time
V
V
Time
Time
Time
V
V
Time
V
V
V
V
Target
Freq.
Rosette
IFOV
Petal
Target
0
1
Full imaging
Linear
detector
array
Target
Scanned
array
Target
image
2 D array
Search
FOV
1
Search
FOV
Constant
Transverse line scan
Video
Detectors (4)
No error
Error
Nutation
Quadrantfixed reticle
Conscan
Figure 3. Common IR/EO sensor types in
cluding nonimaging reticules, line scanning
detectors, and area array imagers.
ELECTRONIC WARFARE 1337
and in tactical planning. ES directly provides threat iden
tiﬁcation/detection and early warning. It also provides
data for electronic countermeasures (ECM), electronic
countercountermeasures (ECCM), threat avoidance, tar
get acquisition, and homing.
Electronic support provides timely EM environment
information for the EW system. The spatial and spectral
environment over which ES operates may span a hemi
spherical spatial segment and a spectrum of tens of giga
hertz. In tactical EW systems, signals in the environment
are analyzed and reports of environment activity are
provided on the order of a second after threat signal
reception.
3.2. Electronic Attack
As an EW function, EA provides an overt active response
capability against enemy combat systems with the intent
of degrading, deceiving, neutralizing, or otherwise render
ing them ineffective or inoperative. EA responds to threat
systems to protect multiple platform or battle group units.
EA includes measures and countermeasures directed
against electronic and electrooptical systems by using
the electromagnetic spectrum (radio, microwave, infrared,
visual, and ultraviolet frequencies). EA technical func
tions include radio and radar signal jamming, false target
generation, and the use of decoys for threat system con
fusion and distraction.
Electronic attack is reactive to environment threats. To
function effectively, therefore, the EA system requires
threat information from the environment, including
threat classiﬁcation, bearing and, if possible, range. These
functions are performed by the ES system or by other sur
veillance systems such as radar or infrared search and
track (IRST). Effective EA response selection requires
knowledge of the threat class and operating mode. Threat
signal data are derived from measuring signal parameters
(frequency, scan type, scan rates, pulse repetition frequen
cy, or continuouswave radiation characteristics). Absence
of radiation may indicate that the threat uses a passive RF
or an electrooptical sensor. The detected threat electronic
parameters are compared to an extensive emitter
database. The EW database, derived from intelligence
sources, is used to identify the threat and correlate the
threat and operating mode with effective EA techniques.
Operational threat exploitation is often impeded by intel
ligence gaps and/or threat use of parameters reserved
for wartime.
3.2.1. Nondestructive Electronic Attack. Nondestructive
EA produces electromagnetic signals at a predetermined
radio, infrared, visual, or ultraviolet frequency with char
acteristics that temporarily interfere with the threat’s re
ceiving system, that is, power level, frequency, and
polarization. EA degrades or overcomes threat system op
eration by overpowering the target signal at the threat
sensor. ‘‘Dazzling’’ is laser or highpower lamp EO/IR jam
ming. Dazzling saturates the detectors or focal plane ar
rays of electrooptical (infrared, visual, ultraviolet) guided
missiles and target tracking systems. Deceptive EA pre
sents a confusing signal to the threat sensor that degrades
its performance to the point where it is no longer effective.
Power levels used for deception are less than those re
quired for jamming because deception does not require
threat sensor saturation.
3.2.2. Destructive Electronic Attack. Destructive EA
physically damages or destroys the threat electronic sys
tem. Specially designed missiles such as the HARM (high
speed antiradiation missile) missile, shown being released
from an A6 aircraft in Fig. 4, are equipped with radar
homing seekers that attack the threat radar antenna and
nearby electronic equipment within the blast radius of the
missile warhead. More recently, similar seekers have been
ﬁtted to loitering remotely piloted vehicles for a similar
purpose. Advances in highpower microwave and laser
technology have made directed energy more practical. At
very high power levels, microwave energy destroys the
components in a missile seeker or threat radar, rendering
them inoperative, Highpower lasers also physically dam
age both RF and electrooptical threat systems.
3.3. Electronic Protection
Electronic protection provides EW protection for the host
platform. Key environment surveillance and threatwarn
ing information is provided by the ES system function (as
it is for EA). EP responds to threats in the environment
with information for evasive action and with the counter
measure responses described previously. EP is primarily
directed against the terminal threat targeted on the host
platform, and preferred EP techniques use decoys that are
less susceptible to the homeonjam weapon mode.
4. ELECTRONIC WARFARE TECHNICAL AREAS
Technical areas that support the ES, EA, and EP func
tional EW systems areas are discussed in this section. All
aspects of EW are addressed by modeling and simulation
because this is the most practical means for functional
Figure 4. HARM missile (shown after separation from an EA6B
aircraft) is an EW weapon for physically destroying the source of
hostile radiation.
1338 ELECTRONIC WARFARE
evaluation. System architectural analyses address the for
mulation of efﬁcient EW system conﬁgurations to provide
the operational functions required within the constraints
of available equipment, techniques, and technology. Tech
nical areas that address ES primarily are signal detection,
measurement, and processing issues that deal with
environment surveillance and warning. Technical areas
associated with EA and EP include CTAR jamming and
falsetarget generation, EO/IR CM, and decoys. Also
included in these technical area discussions are technol
ogy challenges to EW technologies for future capability.
4.1. Modeling and Simulation for Electronic Warfare
Electronic warfare uses modeling and simulation exten
sively in three areas of investigation: research into new
hardware, threat domination/exploitation, and tactics de
velopment. The effectiveness of an EW architecture or
equipment suite is assessed by using a computer model
and parametric studies run against the model. Estimates
of a threat system’s capabilities are incorporated into the
model as environment sources because acquiring foreign
hardware and measuring its performance is difﬁcult. En
vironment signal models stimulate the EW system model.
The EA effectiveness modeled against the threat is mea
sured, and tactics are developed to further reduce threat
system efﬁciency.
Modeling and simulation (M&S) combine detailed an
tiship missile models with ship models, antiair missile
models with aircraft models, electromagnetic propagation
models, and chaff RF decoy models. (Chaff RF decoys are
described later.) Chaff effectiveness evaluation considers
the spatial relationship between the missile seeker and
the ship while accounting for radar clutter and multipath
returns. Signals at the missile are processed through the
seeker receiver and missile guidance and tracking logic. A
chaff cloud(s) injected into the simulation provides a false
radar target signal at the missile seeker. By varying the
amount of chaff and/or the chaff round spatial relationship
with respect to both the defended ship and the threat
missile, chaff effectiveness and tactics can be evaluated.
However, the accuracy of the M&S results depends on the
accuracy of the models used. An accurate missile sensor
and control model is necessary to determine the effects of
the complex signal returns from the target ship and the
chaff on the missile controls and resultant ﬂight path. In a
simulated engagement, detailed missile functions are re
quired to provide an accurate assessment of chaff effec
tiveness. These functions include monopulse antenna
processing, range and angle tracking, missile guidance,
and aerodynamics. Multiple threat seeker modes, such as
acquisition, reacquisition, track, homeonjam (HoJ), and
simulated coherent combinations of signal segments are
also required in the model.
Target ship, aircraft, and chaff radar cross section
(RCS) must be accurately modeled. Typically, a multire
ﬂector target simulation is used to represent the RCS sig
nature. Ideally, a model of thousands of scatterers would
provide greater accuracy. However, careful selection of
several hundred scatterers is adequate.
The accuracy of the missile and target interaction de
pends on the propagative environment model including
multipath. Typically, a raytracing algorithm models the
propagation of RF energy. Useful models rely on a sto
chastic representation of clutter as a function of wind
speed, grazing angle, frequency, polarization, and ducting.
Modeling of an ocean environment can be extended to in
clude reﬂection from wave segments. Models are veriﬁed
by using measured ﬁeld test data.
4.1.1. Electronic Warfare System Architectures. The EW
system architecture ties system functional elements into
an efﬁcient conﬁguration optimized to the operational mis
sion. Figure 5 shows a typical EWsystemarchitecture. The
system performs signal acquisition and parameter mea
surement, direction ﬁnding, countermeasure generation,
and decoy deployment. The system central processing unit
(CPU) provides sensor and countermeasure coordination
and EW system interface with other onboard systems.
Fusing the measurements of EW sensors and proces
sors is a complex technological challenge. This informa
tion includes radar, communications, EO/IR, direction
ﬁnding, and signal analysis. Data fusion within the EW
system requires algorithmic development and significant
enhancement in computational throughput. The EW sys
tem includes antenna(s), receiver(s), and processor(s) ele
ments that provide data on signals in the environment.
System sensors detect and measure threat signal charac
teristics. Multiple sensor subsystems measure the char
acteristics of the signal. For example, a signal acquisition
detects the presence of a signal and measures the envelope
characteristics (frequency, time of arrival, and signal du
ration). Another sensor that may include multiple anten
nas and receivers provides signal bearingangle data.
Separate subsystem sensors measure intrapulse signal
modulation and/or received polarization.
A countermeasures receiver may use an independent
electromagnetic environment interface. The countermea
sures receiver accepts signals from the environment and
provides them to the techniques generator. Target signals
designated by CPU algorithms are selected for counter
measure generation as are the countermeasure modulation
WB
converter
Channelizer
Synthesizer
Encoder Tuner
Tuners
Decoys
Phase
quantizer
CPU
Receiver
Techniques
generator
Transmitter
DF
antenna
CM RCVR ANT CM XMIT ANT
Comm
Nav
Display
Figure 5. Electronic warfare system architecture indicating sys
tem functional elements required to provide ES, EA, and EP
functions to the host platform and operational battle group.
ELECTRONIC WARFARE 1339
techniques to be applied. The resulting jamming signals
are ampliﬁed to the desired power levels and radiated into
the environment.
Decoys are part of the EW system architecture. This
subsystem is controlled by the CPU based on sensor in
puts. Decoys provide the important function of separating
the countermeasure signal source from the host platform.
In this operational mode, decoys provide alternative high
ly visible targets to divert a weapon from its intended tar
get. Also required are the means, such as the coordination
of jamming with the use of decoys, to neutralize the HoJ
weapons threat.
4.2. Surveillance and Warning
Electronic support surveillance and warning perform the
functions of noncooperative intercept and exploitation of
radiated energy in the EM environment. Surveillance and
warning detection relationships are those associated with
communications systems. Additional signal detection con
straints result because the signal’s spatial location and its
characteristics may not be known. Signal unknowns re
quire tradeoffs of detection sensitivity and environment
search. Once detected and measured, environment signals
require sophisticated signal processing for signal sorting,
formation, and characterization before they can be corre
lated with signal intelligence libraries for classiﬁcation.
Some fundamental tradeoff relationships for detection and
warning are discussed below.
4.2.1. Threat Signal Detection. Threat signal detection
occurs as the electronic support system is illuminated
above the system sensitivity level with signals that satis
fy the singlepulse detection criteria. Detection is per
formed as the ES system scans the environment.
Detection metrics include incident radiation sensitivity,
detection probability, false detection probability, corrup
tion probability, simultaneous detection, and throughput
rate.
Aircraft are often used to carry electronic warfare bat
tleﬁeld surveillance equipment. The operating altitude of
surveillance aircraft provides a long lineofsight range to
the horizon. The range to the electromagnetic horizon ac
counting for nominal atmospheric refractions is given by
R¼
3
2
_ _
h
_ _
1=2
ð1Þ
where h is the aircraft sensor altitude in feet and R is the
observertohorizon range in statute miles.
The time required to survey the environment depends
on the surveillance alert status, system sensitivity, instan
taneous observation segment, and rate of environment
search. The number of instantaneous environment seg
ments in frequency and bearing establish the number of
environment dwell periods required for an environment
scan. The larger the environment segments, the more rap
idly the system performs the scan. The dwell at given en
vironment segment is scheduled to span the signal event
period. Time to intercept is modeled by
T
I
¼
ðT
D
NMÞ
P
T
ð2Þ
where T
I
, is the time required to survey the environment,
T
D
is the EW support system dwell period, Nis the number
of frequency segments in the environment, Mis the number
of spatial segments in the environment, and P
T
is the prob
ability that the signal occurs above the sensitivity level.
In Eq. (2), spatial environment segmentation, spectral
environment segmentation, and detection probability
combine multiplicatively to deﬁne the time required to
survey the environment. Wide instantaneous bandwidths
and a large instantaneous ﬁeld of view reduce environ
ment survey time unless equipment choices reduce system
sensitivity and the corresponding probability of signal de
tection. Equations (3) and (4) describe receiver sensitivity
and aperture gain functional relationships:
S¼ðNFÞðSNRÞðkTBÞ ð3Þ
where S is receiver sensitivity, NF is receiver noise factor,
SNR is the required sensitivity for detection and false
alarm criteria, k is Boltzmann’s constant, T is tempera
ture in degrees kelvin, and B is bandwidth in hertz
G¼2Kp=y ð4Þ
where G is antenna gain, K is antenna efﬁciency (less than
unity), and y is antenna beamwidth in steradians.
A tradeoff between sensitivity and time to intercept is
implied in Eqs. (3) and (4). By using multichannel pro
cessing, the tradeoff can be resolved in either domain. A
wideband channelizer provides instantaneous spectral
coverage equal to the span of the channelizer frequency
coverage, and receiver sensitivity is established by the
bandwidth of an individual channel. Multichannel spatial
processing provides the instantaneous spatial coverage of
the sum of the channels being processed. System antenna
gain is based on channel beamwidth.
Detection sensitivity requires consideration of the de
sired detection range. Equation (5) deﬁnes the electronic
support detection range in terms of the threat signal pa
rameters and the electronic support antenna, receiver,
and processor system parameters:
R
MAX
¼
P
t
G
t
G
r
l
2
ð4pÞ
3
S
N
_ _
MIN
kTB
n
L
_
¸
¸
_
_
¸
¸
_
1=2
ð5Þ
where R
MAX
is the maximum detection range, P
t
is the
threat signal transmit power, G
t
is the threat signal an
tenna gain, G
r
is the antenna gain of the electronic support
subsystem, l is the wavelength of the threat signal trans
mission, (S/N)
MIN
is the minimum signaltonoise ratio re
quired by the electronic support subsystem for detection, k
is Boltzmann’s constant, T is absolute temperature, B
n
is
the effective noise bandwidth of the electronic support
1340 ELECTRONIC WARFARE
receiver, and L represents the combined feed losses of the
threat transmitter and the electronic support receiver.
The probabilistic characteristic of signal detection is il
lustrated by considering the intercept of a threat at a sig
nal level considerably above the receiver threshold level.
Detection probability arises primarily from the indepen
dent probabilities that the ES system observes the envi
ronment in the spatial and spectral location of the threat
emitter and that the threat emitter illuminates the re
ceiver with the required power for detection.
Also of importance is the probability of signal detection
once the ES system is steered to the signal spatial and spec
tral location. Then detection probability P
D
is based on the
signal characteristics, that is, the probability that the threat
signal illuminates the EW system during the observation
period. The time required to perform a detection T
I
is de
rived from the scan internal T
S
and is given by T
I
¼T
S
/P
D
.
False reports from the electronic support receiver are
highly undesirable. Limited computational resources are
needed to process each pulse received in an attempt to
form an association with other pulse reports. The rate of
false reports is established by the proximity of the detector
threshold level to the noise level. Figure 6 shows the re
lationship between the singleevent probability of detec
tion, the probability of false signal report generation, and
the signaltonoise ratio. This ﬁgure shows that both the
probability of detection and the probability of false gener
ation are strong functions of the signaltonoise ratio.
The probability of pulse interference P
OL
depends on
the duration T
D
of the signal and the rate R at which sig
nals are expected. A reduction in P
OL
results from adding
parallel measurement channels. The functional relation
ship approximating P
OL
is
P
OL
¼
ðT
D
RÞ
N
N!
1 þ
N
N¼1
T
D
R
N
_ _ ð6Þ
where T
D
is the event duration, R is the event repetition
rate, N is the number of parallel measurement functions
provided, and P
OL
is less than 0.9.
4.2.2. Electronic Support Signal Processing. The ES sig
nal processor derives signal information from the multi
tude of environment event measurements. Signal
processing is the focal point of the ES subsystem where
operationally relevant sense is made of large data inputs.
ES processing includes sorting event data and correlating
sorted event data with emitter libraries to establish the
class or family of signals to which the emitter belongs.
Beyond sorting, intensive processing is applied to identify
intercepted emitters specifically and to locate them pre
cisely within the battle space.
Sorting, a key electronic support signal processing
function, correlates event descriptors from the same emit
ter. Correlation is performed on the basis of both instan
taneous and temporal signal parameters. Instantaneous
parameter sorts are less computationally demanding than
temporal deinterleaving.
The initial signal sorting is histogramming based on
instantaneous signal parameters. The signal parameters
used for histogrambased sorting are those available from
a single event or pulse measurement. They include exter
nal signal parameters, such as signal frequency, start
time, duration, power level, and angle of arrival. Other
instantaneous parameters used are measurements of sig
nal modulation. Signals measurements with like param
eters are binned together, and it is postulated that each
bin contains event descriptor data from the same emitter.
After sorting, event descriptors are placed in individual
emitterassociated groups. The monopulse and interpulse
characteristics of the event group measurements are
quantiﬁed into a signal descriptor. The signal descriptors
are classiﬁed into an emitter class by correlation with a
library database.
In some instances, highresolution signal measure
ments identify specific emitters. As might be expected,
identiﬁcation parameter sets and the processing required
to establish them are significantly in excess of that re
quired for classiﬁcation. Here, as in the case of classiﬁca
tion, detailed signal descriptors are correlated with a
library to deﬁne a specific emitter.
The spatial distribution of threat signals in the envi
ronment is operationally important. Determining the
threat signal bearing angle with respect to own platform
is a key step toward establishing threat signal position
information. Conventional techniques used for direction
ﬁnding measurements include the use of differential am
plitude processing of squinted antennas (antennas aimed
in different directions), differential phase measurements
from a phasedarray antenna, and differential time of
arrival measurements from spatially separated receivers.
Both hostile and benign operational scenarios require
information about the location of both noncooperative
ﬁxed and mobile emitter installations. Electronic warfare
target location exploits directionﬁnding data and naviga
tional data to provide a signal location solution. Single or
multiple platforms are used to generate location data.
20
15
10
5
0
−5
−10
−15
10
−14
P
f0
= 10
−16
10
−2
10
−1
10
−12
10
−10
10
−8
10
−6
10
−5
10
−4
10
−3
0.001 0.01 0.1 0.5 0.9 0.99 0.999
Probability of detection
V
i
s
i
b
i
l
i
t
y
f
a
c
t
o
r
(
s
i
g
n
a
l

t
o

n
o
i
s
e
r
a
t
i
o
)
(
d
B
)
Figure 6. Detection probability and false detection probability
for various signaltonoise ratio conditions.
ELECTRONIC WARFARE 1341
The accuracy of target location depends on the preci
sion of the directionﬁnding data and the navigation mea
surement and on the length of the baseline between
measurements and the range to the target. Figure 7 shows
target location geometry. The major error location axis A is
modeled by
A¼Rj csc
c
2
_ _
ð7Þ
where R is the range from observer to the target emitter, j
is the directionﬁnding measurement error, and c is the
angle subtended by the maximum difference in observa
tion bearings with respect to the target, which provides
location measurement error for the condition cop/2. The
range R from the observer to the target is given by
R¼L
sinðp Ày ÀgÞ
sin y
ð8Þ
where L is the separation between observations, y is the
angle between the baseline and the opposite bearing
angle, and g is the angle between the baseline and the
adjacent bearing angle.
4.2.3. Electronic Support Digital Signal Processing
Technology. Electronic warfare system processing, both
dedicated and programmable, assimilates environment
data from the receivers and wideband processors. It uses
these data to sort, classify, and identify the sources of
emissions to represent the environment relevantly. The
digital signal processor provides the means for applying
an array of algorithms to both predetection and detection
signal data to extract threat information for EW system
use. Digital signal processing metrics include highrate
signal throughput processing in a compact module.
Digital signal processing is the heart of the ES func
tion. It provides the ﬂexibility of applying an extensive
array of algorithms to system data. Critical digital signal
processing technology challenges include processing
throughput and developing efﬁcient processing algo
rithms. Although signal data can be reﬁned by applying
sequential algorithms, the ES response is timecritical; it
must provide the most accurate assessment of available
data within the required response time. Great potential
exists for advancing digital signal processing technology,
but optimum ES performance can be expected from judi
cious allocation of processing tasks between wideband
processors and the digital signal processor.
An example of digital signal processing technology is L
MISPE (little monopulse information signal processing el
ement), a specialpurpose signal processor designed to op
erate with highquality superheterodyne RF receiver
systems. LMISPE provides extremely accurate pulse
analysis and parameter extraction for signal classiﬁcation
and specific emitter identiﬁcation (SEI). It is contained in
a single rackmounted enclosure.
4.2.4. Surveillance and Warning Technology. Surveil
lance and warning are the sensor and environment pro
cessing functions for the EW system. Speed and accuracy
of measurements and processing functions are the prima
ry metrics for ES. Accurate throughput is important in
providing sufﬁcient time for effective threat response to
the EA or platform commander. In addition, precision
threat assessment provided to the EA subsystem facili
tates optimum technique selection and conservation of EA
power resource for engaging multiple threats. The ES per
formance challenge is further constrained by space limi
tations aboard platforms, particularly aircraft. Receiver
technology performs environment sensing for the EW
application.
4.2.5. Receiver Technology. Electronic support through
put and physical displacement metrics are addressed in
developing wideband, smallsize monolithic microwave in
tegrated circuit (MMIC) technology. MMIC monolithic in
tegrated analog processing at multigigahertz operating
frequencies provides a capability suited to ES receiver ap
plications. Advantages sought in the exploitation of this
technology base include economies of size, weight, power,
and cost. Increased receiver dynamic range for continuous
environment intercept during active countermeasures
transmission remains a receiver technology challenge.
The MMIC receiver shown in Fig. 8 is an example of this
technology.
4.2.6. Wideband Processing. Wideband receivers pro
vide high probability of signal intercept. Wide spectral
segment processing is necessary to increase signal detec
tion sensitivity and to provide copulse reception of multi
ple simultaneous signals and rejection of interference
signals. Requirements for wide instantaneous bandwidth,
rapid throughput, and small modules are wideband pro
cessing metrics.
Acoustooptic channelization technology is being devel
oped for wideband processing as a compact, economical
means for performing highresolution environment seg
mentation. Wideband signal frequency demultiplexing is
performed using Bragg regime acoustooptic diffraction
and electronic signal detection and encoding. Functions
performed by these acoustooptic processors include chan
nelized correlation, convolution, and spectral processing.
Observer
Baseline
L
Target
R
Range
R /2 ϕ
/2 ϕ
Figure 7. Emitter location geometry supporting Eq. (7), with
observer track and signal measurement angles indicated.
1342 ELECTRONIC WARFARE
Acoustooptic channelizers are based on Bragg diffrac
tion of light (Fig. 9). The Bragg cell serves as the optical
deﬂection or optical modulator element within the proces
sor. The Bragg cell is an optically transparent medium,
such as a crystal, that is driven at the applied RF fre
quency by using a piezoelectric RFtoacoustic transducer.
The Bragg cell transduces the RF signal into acoustic
waves that are collimated into the Bragg cell crystal. The
propagating acoustic wave creates sequential regions of
crystal compression and extension that correspond to the
period of the acoustic wave. The acoustically induced dif
fraction grating in the Bragg cell interacts with a coherent
optical source to perform RF input frequency demultiplex
ing. The deﬂected lightbeams output from the Bragg cell
are focused onto a detector array where light is detected to
indicate energy in segments of the applied RF spectrum.
4.2.7. Wideband Interconnections. Electronic warfare
sensors require broad access to the electromagnetic
environment to provide quick response to hostile electro
magnetic activity. For convenience and efﬁciency, central
stowage of signal processing functional elements is impor
tant. To assure signal visibility, environment apertures,
antennas, and EO/IR sensors must occupy locations on the
periphery of the aircraft, ship, or land vehicle. Wideband
interconnects transmit electromagnetic environment data
from the EW system apertures to processing subsystems.
With the current RF bandwidth of the electronic war
fare environment expanding through tens of gigahertz,
just ﬁnding a medium that supports that level of frequen
cy coverage is a challenge. At light frequencies, however, a
100GHz spectrum spans less than a third of 1% of light
frequency. In addition, lowlosstransmission optical ﬁbers
provide a nearly lossless means to transfer wide spectra
across a platform. Indeed, wideband interconnect technol
ogy is developing the use of ﬁberoptics.
Usable optical ﬁber bandwidth is limited by dispersion.
Conventional ﬁber exhibits dispersion of 20 ps/km/nm of
bandwidth. A typical signal operating within a 10 MHz
bandwidth would exhibit dispersion of less than 0.11.
Clearly, bandwidth limitations are elsewhere in the link.
Detectors have also been developed to provide
bandwidths on the order of tens of gigahertz. High RF
operating frequency detection is performed by using
smallgeometry detectors that exhibit maximum power
limitations. Limitation in maximum power levels applied
to the detector restricts the output signal intensity range.
More recent developments in distributed detector ele
ments are extending detector powerhandling capabilities.
Dynamic range is a significant ﬁberoptic link metric
because the EW sensor system must process lowpower
signals on the horizon in an environment with highpower
local transmissions. Modulator and detector attenuation
reductions are technological issues being addressed to
enhance the dynamic range performance of ﬁberoptic
links.
4.3. Countertargeting
Countertargeting (CTAR) is the technical area that pro
vides the means for protecting the host platform or force
from weapons targeting by a hostile force. CTAR functions
include obscuration, falsetarget generation, and confu
sion. Associated techniques include jamming and onboard
and offboard falsetarget generation.
Countertargeting operates against radars that feature
a target, locating or surveillance mode, as shown in the
functional sequence of Fig. 10. Airborne surveillance ra
dar is generally used against ship and ground forces be
cause the aircraft altitude provides extended surface
target detection range. Conversely, when defending
against aircraft with CTAR, the radar could be ground
based. Some radars are designed with the sole purpose of
surveillance, whereas others are multimode and can track
targets. By using imaging processing, modern surveillance
radars that include synthetic aperture, inverse synthetic
aperture, high range resolution, and moving target indi
cation processing can accurately determine target location
and identify the type of target.
Figure 8. The MMIC receiver, a combination of monolithic
microwave, analog, and digital circuits, performs signal selection
and conversion to a convenient intermediate frequency.
Laser
Bragg cell
D
e
fle
c
te
d
lig
h
t
Undeflected light
Acoustically induced
diffraction grating
Transducer
Signal input
D
e
fle
c
tio
n
a
n
g
le
f
1
f
2
O
p
t
i
c
a
l
s
e
n
s
o
r
Bragg cell principle
f
1
f
2
Figure 9. The acoustooptic Bragg regime signal transform pro
cessing principle used for signal frequency analysis, sensitivity
enhancement, and directionﬁnding functions.
ELECTRONIC WARFARE 1343
4.3.1. Countertargeting Techniques. Figure 10 shows
the CTAR functional sequence. CTAR EA techniques are
categorized as environment obscuration and jamming and
falsetarget signal generation. CTAR provides either con
fusing or ambiguous data to adversary surveillance and
targeting radar displays to confuse the human operators
who interpret these presentations. Radar displays include
plan position indicators (PPIs), A or B scopes, or combi
nations of these. Obscuration screens targets over selected
portions of the display with a jamming signal power above
that of the target signal in environment segments span
ning both range and azimuth (see radar articles for de
scriptions of radar displays). The amplitude of the
obscuration CTAR signal exceeds that of any targetre
ﬂected signal in the screened sector.
Experienced operators can recognize obscuration and
radar jamming and initiate procedures to mitigate its
effects. The falsetarget CTAR technique, however, is a
more subtle form of EA that is less apparent to the oper
ator. Here, the CTAR signal creates false indications on
the radar display that appear as real targets to the oper
ator. When the display is cluttered with false targets, ra
dar operator time is consumed sorting through them.
Selecting a false target for missile engagement dissipates
an expensive weapon.
CTAR EA systems can be used to protect an entire mil
itary force. CTAR force protection systems are generally
large and use human operators for system control. An ex
ample is the AN/ALQ99 system installed on the EA6B
(Fig. 11), and EF111 EWaircraft. Some EA systems, such
as the AN/SLQ32 installed on surface ships (Fig. 12), are
for selfprotection and support EA functions.
The EA system selects a specific technique from a large
EA technique library. Selection is based on knowledge of
the threat location, class, electronic parameters, and op
erating mode. The EA system, using an embedded receiver
subsystem, rapidly adapts to threat signal operating mode
changes. The threat changes operating mode as either a
Figure 11. EA6B aircraft equipped with the AN/ALQ99 EA
system for airborne CTAR.
Figure 12. Shipboard installation of the AN/SLQ32 EW equip
ment used for CTAR.
Program start
Program finish
Initialized serial port
and generators
Set generator 1 at center frequency
and sens pwr +4dBm
Set generator 2 at first sensitivity
frequency/power reading
Set generator 2 for next
frequency/sens power.
Set PRI/pulse width
(100/0.1 50/0.5 10/1.0 20/10.0)
Send dwell
check for data
Record data in file
Five repetitions
completed?
Data
available?
Are all
PRI/pulse widths
done?
Freq = 3.5GHz?
No
No
No
No
Yes
Yes
Yes
Ye
Figure 10. CTAR functional diagram showing sequence used in
engaging a surveillance or targeting radar signal.
1344 ELECTRONIC WARFARE
countercountermeasures technique to circumvent EA or
as part of the hostile targeting and homing sequence.
Adaptive EA provides rapid changes in techniques as
the threat sequences through operating modes.
Both jamming and deception CTAR techniques may be
used. RF jamming techniques are either ‘‘barrage’’ or
‘‘spot.’’ Barrage jamming covers a wider frequency band
at lower power density levels and is used to jam either
several radars at once or spreadspectrum systems where
the precise frequency of the threat is uncertain. Spot jam
ming concentrates the entire jamming power within the
bandwidth of a single threat radar receiver with corre
spondingly better results. In both cases, a radial jamming
strobe will appear on the threat radar PPI scope, as shown
in Fig. 13. If the ratio of jamming signal power to the re
ﬂected radar signal power (J/S) is insufﬁcient, the real
target will ‘‘burn through’’ the jamming signal and become
visible within the jamming strobe. For greater jamming
effectiveness, it is desirable to have sufﬁciently large J/S to
prevent burn through in the mainbeam and the principal
sidelobes (see jamtosignal calculations later).
Deception techniques are more varied and are gener
ally threatspecific. Many deception techniques are direct
ed against threattracking radars or missileseeker
radars. These techniques attack the threat radar target
tracking loops in range, angle, or Doppler. Deception tech
niques are often used in combinations and can be
sequenced as the threat modes vary, or they can sequence
according to a programmed pattern. Falsetarget decep
tion techniques are generated to emulate true target re
turns. The threatradar operator, in response to deception,
may conclude that all detected targets are genuine and
simply select false targets for weapons engagement, or, if
deception is suspected, time and computational resources
must be used to identify the true target prior to engage
ment. In automated weapons systems, the EA subsystem
may create so many false targets that the radar computer
becomes overloaded. Because Doppler radar and missile
seekers process large numbers of ambiguous radar re
turns to ﬁx the true target, they are particularly vulner
able to coherent falsetarget techniques. An effective
CTAR approach combines jamming and deception. Jam
ming creates a radial strobe that obscures the true target,
whereas the deceptive CTAR provides false targets that
project through the jamming strobe.
4.3.2. Countertargeting Effectiveness. Countertargeting
effectiveness is assessed by comparing threat system per
formance in benign and CM environments. The ability of
the threat system to detect, acquire, and target true tar
gets, including parameters, such as target acquisition
time and weapon release range, is assessed by evaluating
threat performance against live targets on test ranges.
Evaluating missileseeker countermeasure effectiveness
presents a more difﬁcult problem. Computer simulations
model the missile ﬂyout from an actual or surrogate threat
system against a live target. A measure of CTAR effec
tiveness (MoE) is the ratio of the number of missiles that
approach their target outside of the missile lethal range to
those missiles that approach the target within lethal
range. Software simulates multiunit engagements. U.S.
Navy ship EA is evaluated by ﬂying test aircraft carrying
captive instrumented seekers against the ships and re
cording the threat system performance.
A statistical technique to assess CTAR effectiveness
compares the number of missiles required to defeat an EA
equipped aircraft versus the number required to defeat a
nonEAequipped aircraft. Similar statistics assess the
number of antiradiation missiles ﬁred versus the number
of radar systems defeated. Additional effectiveness infor
mation can also be gleaned from intelligence sources.
4.3.2.1. Obscuration Burnthrough. A measure of CTAR
obscuration effectiveness is the range at which the radar
displays the target in the presence of jamming. This is
called the burnthrough range. At this range, the radar is
sufﬁciently close to the target that the processed target
reﬂected radar power exceeds the jamming signal display
masking. The real target becomes visible superimposed on
the jamming signal. Burnthrough is modeled in Eq. (9) by
using the radar range equation and freespace propaga
tion. The radar range equation provides the signal power
S that is received at the radar after being transmitted to
and reﬂected from the target. The freespace signal prop
agation equation models the jammer power J that is re
ceived at the radar from the jammer. The quotient of
jammer to signal power constitutes a ﬁgure of merit
known as the jamtosignal (J/S) ratio. This ratio is unique
for each radar and depends on radar processing gain and
on the display format and screen phosphor. Operator pro
ﬁciency also plays a significant role. Rearranging the
terms of this equation to solve for range yields the burn
through equation
R
b
¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
J
S
P
R
sB
J
P
J
4pB
R
_ _
¸
ð9Þ
where R
b
is the burn through range, J/S is the ratio of
jammertosignal power required to jam the victim radar,
P
R
is the effective radiated power of the radar, P
J
is the
effective radiated power of the jammer, s is the radar cross
section of the target, B
J
is the jamming signal bandwidth,
and B
R
is the processing bandwidth of the radar receiver.
This equation models the case with the jammer located on
the radar target platform.
Figure 13. PPI radar scope without and with jamming, showing
the effects of CTAR jamming on the threat radar display.
ELECTRONIC WARFARE 1345
4.3.2.2. JammertoSignalPower Relationships. The J/S
power ratio at the threat radar is a concept central to pre
dicting EA effectiveness. To degrade the threat radar, an
interfering jammer power J of sufﬁcient strength is re
quired to overcome the targetreﬂected signal at the radar
S. For effective EM noise jamming, the J/S required is 0–
6 dB minimum, depending on the noise modulations used
and the detailed characteristics of the threat. The mini
mum J/S ratio required for effective CTAR deception tech
niques varies from 0dB for false targets, to 0–6 dB for
range deception, to 10–25 dB for angletracking deception,
and to 20–40 dB for monopulse deception. Equations (10)–
(12) are based on two typical EA tactical situations. Self
protection CTAR [Eq. (10)] addresses the case with the
target in the threat radar mainbeam. Support CTAR [Eq.
(11)] addresses the case of the target in the threat main
radar beam but with the EA jamming emanating from a
separate platform and radiating into an arbitrary bearing
of the threat radar antenna pattern. In both cases, the
radar is assumed monostatic (i.e., the radar receiver and
transmitter are collocated):
J/S for selfprotection EP CTAR:
J=S¼
4pP
j
G
j
B
r
R
2
P
r
G
r
sg
2
B
j
ð10Þ
where P
j
is jammer power output; G
j
is gain of jammer
antenna in direction of radar; B
r
is radar receiver noise
bandwidth; R is radartojammer range; P
r
is radar power
output; G
r
is gain of radar antenna in target direction; s is
target radar cross section; g
2
is propagation oneway pow
er gain (square of the ratio of ﬁeld strength to freespace
ﬁeld strength due to direct and reﬂected ray combination),
0og
2
o4 (interferometer lobing); and B
j
is the jammer
noise bandwidth.
J/S for support EA:
J=S¼
4pP
J
G
jr
G
rj
B
r
R
4
t
g
2
j
P
r
G
2
r
sB
j
R
2
j
g
4
t
ð11Þ
where G
jr
is the gain of the jammer antenna in the direc
tion of the radar, G
rj
is the gain of the radar antenna in the
direction of the jammer, R
t
is the radartotarget range, g
j
is the jammertoradar propagation factor, R
j
is the radar
tojammer range, and g
t
is the radartotarget propagation
factor. The remaining terms are as deﬁned previously.
Effect of target radar crosssectional reduction:
S¼
P
r
G
r
sl
2
g
4
ð4pÞ
3
R
4
ð12Þ
where l is the wavelength of the radar operating frequen
cy. All the remaining terms are as deﬁned previously.
Equation (12) deﬁnes the signal at the receiver of a
monostatic radar. Note that the power received at the ra
dar is directly proportional to the target radar cross sec
tion s and inversely proportional to the fourth power of the
range R (R is the separation between the target and ra
dar). Therefore, as the radar cross section is reduced, the
signal at the radar is correspondingly reduced. If the cross
section is sufﬁciently reduced, the target becomes indis
tinguishable from the radar noise and background clutter.
Low observable platforms, such as the B2 and F117 air
craft, provide sufﬁciently low radar cross section to make
radar detection difﬁcult. The implication of radar cross
sectional reduction technology to CTAR is twofold: ﬁrst,
with sufﬁciently low radar cross section, EP may not be
necessary, and secondly, if the cross section merely lowers
the signal power at the radar, then a lower power, lowcost
CTAR transmitter becomes sufﬁcient to provide the J/S
necessary to achieve the desired level of survivability.
4.3.3. Countermeasure Technology. Countermeasure
technology addresses the evolving threat in addition to
the need for economic force protection. Significant ad
vances in radar, communications, EO/IR weapons’ sen
sors, and weapons control present heightened
challenges to maintaining effective EA capability.
4.3.3.1. Radar Countermeasures Technology. Counter
targeting equipment for use against advanced synthetic
aperture radar (SAR) or inverse synthetic aperture (ISAR)
surveillance and targeting radar requires wide instanta
neous bandwidths and high processing speeds. Further
more, because these radars use coherent processing,
CTAR effectiveness consequently requires coherent radar
signal storage and reproduction to enhance effectiveness.
Digital RF memory (DRFM) technology is being developed
to convert the analog radar RF signals into a digital for
mat for convenient storage. As required, the radar signal
is retrieved from storage and converted to RF for use in
countermeasure waveform generation. Technology limita
tions and costs constrain currently available DRFM de
signs, each optimized for a specific application.
Radiofrequencytapped delay lines provide precise tim
ing between portions of the CTAR waveform. Analog RF
tapped delay lines use surface acoustic wave (SAW) and
acoustic charge transport technology. Research is under
way to create digital tapped delay lines. Noise modulation
is commonly applied to CTAR signals, and highquality
tunable noise sources are required. The output EA stage is
the transmitter/antenna combination that generates and
radiates the CTAR signal. Antennas for EA applications,
once considered a dedicated asset, are currently envi
sioned as multifunction phasedarray antennas with ele
ments fed by solidstate ampliﬁers.
Radiofrequency isolation between the countermea
sures transmitter and the receiver is a common problem
of countermeasuresequipped platforms. The countermea
sure signal appears at the receiver antenna. When the
transmitter and receiver are insufﬁciently isolated, the
countermeasure signal interferes with lower level threat
signal reception from the environment. Interference de
mands careful attention to antenna design, isolation, and
platform siting.
4.3.3.2. Radar Countermeasure Signal Source Technology.
Electronic attack transmitters require signal sources that
can be rapidly switched in azimuth, elevation, frequency,
and polarization to generate multiple highpower beams
1346 ELECTRONIC WARFARE
with low sidelobes over large multioctave bandwidths.
CTAR requirements for economical compact transmitters
are challenged by the lack of appropriate lowcost EM
power sources. Furthermore, few commercial applications
exist for wideband EM power source technology. Research
and development in this area is limited primarily to EA
applications. Original EW power sources, tunable mag
netrons, and crossﬁeld ampliﬁers provided only narrow
operating bandwidths. Travelingwave tubes (TWTs)
evolved to ﬁll the need for wide, instantaneous bandwidth.
Over time, TWT bandwidths grew from a singleoctave 2–
4 GHz band to multiple octaves at frequencies beyond
40 GHz. However, TWTs are expensive and unreliable. Al
though new miniTWTs and microwave power modules
have become available, their basic design remains vacu
umenvelopebased. MMIC technology is steadily advanc
ing, and it now provides solidstate chips with multioctave
signal generation capability, wide instantaneous band
width, and signal power levels approaching 5 W. With
MMIC technology, solidstate active aperture arrays be
come achievable, and such arrays for EA applications are
now being developed. Although MMIC active aperture
array signal source promises good performance and
reliability, the system remains expensive.
4.4. Passive Electrooptic/Infrared Electronic Warfare
Electronic warfare in a passive EO/IR target acquisition
and weapons sensors environment applies to a growing
threat capability. The openocean bluewater scenario re
quires EO/IR EA and EP ship protection, typically 200
nautical miles or more from shore, against massive and
coordinated attack. EO/IR EA applications have more re
cently focused on littoral scenarios involving amphibious
operations in support of peacekeeping operations for re
gional conﬂicts, providing humanitarian assistance in po
litically and militarily unstable regions; evacuating
civilians from regions of conﬂict, and ensuring safe pas
sage of commerce through disputed littoral waters and
choke points.
The traditional EO/IR threat, the longrange antiship
missile, has been intensiﬁed in the littoral areas by a large
variety of airtosurface, airtoair, and surfacetoair EO/
IR missile weapons. These missiles can inﬂict severe dam
age to the smaller craft used for littoral warfare.
Electrooptic system target detection range depends on
detector sensitivity and resolution. A target image is de
ﬁned by contrast with the background. Sensitivity deter
mines whether the contrast is discernible. Resolution
depends on the spatial environment angle illuminating
the detector, which is a function of detector surface area
and focusing optics. The distance at which target features
are resolvable determines the maximum operating range
of the system.
The target signature detectability is determined not by
the absolute temperature of the object but rather by the
contrast between the target and background within a giv
en spectral band. Environment backgrounds range from
the cold, uniform background of space to thermally clut
tered land areas. Solar interaction with the target and
background reﬂection and heating further degrade the
background contrast with the target. Typical target con
trasts range from about 1 kW/sr (kilowatt per steradian)
in the 2–3 mm atmospheric window for an aircraft engine
to tens of kilowatts per steradian for ships in the 8–12 mm
window. Target aspect, especially the location of hotspots,
greatly inﬂuences the signature.
4.4.1. Electrooptic/Infrared Countermeasures. Electro
optic/infrared countermeasures are constrained by specu
lar atmospheric propagative characteristics, as is the
threat (Fig. 14). The contrast of the target to the back
ground within the weapon sensor’s specular passband,
the type of seeker spatial localization processing, and
Figure 14. EO/IR atmospheric transmission spectral segments and laser and laser harmonics
countermeasures source spectral regions.
ELECTRONIC WARFARE 1347
available practical radiation sources are also prime con
siderations.
The missile ﬂyout and CM sequence of events occurs in
several seconds. As part of an integrated electronic war
fare suite, the EO/IR EA system is designed to engage a
large number of missiles launched in a coordinated attack.
Figure 15 shows a typical timeline of the CM response to
an attack by a subsonic antiship missile. The time line
indicates the interaction of EO/IR EA with other ship de
fense systems.
To preclude detection by a threat EO/IR sensor, target
signature can be reduced through a combination of con
vective, conductive, and radiative mechanisms. Exterior
surfaces of ship stacks are cooled by convective air ﬂow
between the engine exhaust ports and the outer stacks.
Engine plume and exhaust gases from all types of engines
can be cooled by dilution with air. Radiation from hotspots
can be reduced by spectral emissivity modiﬁcations or by
obscuring the hot areas from view. On new platforms, low
observability design criteria have led to lowsignature air
craft and ships.
Onboard aircraft CM sources initially generated false
target location and/or guidance degradation through
weapon automatic gain control (AGC) manipulation.
This technique remains highly effective against many
threats. The onboard jammer sources can be chemically
fueled IR sources or electrically powered incandescent and
metal vapor lamps. As the wavelength passbands of anti
air and antiship seekers gradually migrate to longer wave
lengths, out to the 8–14 mm window, noncoherent sources
will no longer be practical.
Basic spin scan and conical scan (conscan) ‘‘hotspot’’
seekers are vulnerable to ﬂare decoys. Almost universally,
these ﬂares are composed of magnesium and
polytetraﬂuoroethylene and are designed with a radiant
intensity several times that of the target. In the distrac
tion mode, the decoy is an excellent target; in the seduc
tion mode, the weapon’s seeker control signal is biased by
the decoy or transferred to it. Because pseudoimaging
seekers exhibit spatial and temporal processing capabili
ties, simple ﬂares are relatively ineffective, and simple
ﬂares perform even more poorly against imaging sensors.
Newer decoys overcome advanced seekerdiscriminating
processing with improved spectral characteristics that
more closely match the target platform spectral emissions.
Improved decoy spatial distribution in the form of clouds
and multiple hotspots, temporal risetimes, and persis
tence match target signature increase rates and lifetimes,
thus preventing time–history discrimination. Kinematics
model realistic target movement.
The small beam divergence of lasers can result in high
radiance, lowpower sources that provide the J/S power
ratios needed for effective EA. Two laser sources, primary
lasers and nonlinearly shifted lasers, are available for CM
applications. Lasers shifted by nonlinear conversion in
clude harmonic generation and tunable optical paramet
ric oscillators (OPOs). Primary lasers do not produce
spectral lines in all the potential threat passbands of in
terest and are susceptible to notchﬁlter countercounter
measure techniques. Although harmonic generating EA
techniques provide additional wavelengths, they are also
subject to counterCM. Promising sources for IR/EO CM
are tunable OPOs pumped by diodepumped, solidstate
lasers. Two nonlinear materials currently demonstrating
the highest potential are periodically poled lithium nio
bate (PPLN) and zinc germanium phosphide (ZnGeP
2
).
Figure 14 shows the primary lasers of interest and the
wavelength coverage possible with PPLN and ZnGeP
2
OPOs.
Although noncoherent sources provide wide angular
protection, highresolution detection is necessary to point
and track the threat system and effectively use laser pow
er. Timely threat detection and warning ES is essential to
the success of all nonpreemptive EA.
60
50
40
30
20
10
0
0 5 10 15 20
ASM Range from the ship (km)
Handoff
to
EO/IR EA
Handoff
from
EO/IR EA
Range by which
EO/IR EA
Inform must be
passed
Minimum range for intercept
Decoy
RF decoy
IR decoy
CIWS
CIWS
Antimissile
missiles
Antimissile
missile launch
T
i
m
e
f
r
o
m
A
S
M
i
m
p
a
c
t
(
s
)
Figure 15. Missile attack timeline showing launch, acquisition, and homing phases of the missile
as well as the CM attack on missile sensors and control circuits.
1348 ELECTRONIC WARFARE
4.4.2. Electrooptic/Infrared Countermeasure Technology.
Key EO/IR EA technologies required to counter threat
performance improvements include higher throughput
data processing using more capable algorithms, laser
beamsteering, and decoy launcher design. Needed pro
cessing improvements include faster signal processing,
more efﬁcient image processing, and false alarm reduc
tion. Highperformance, highspeed beamsteering, prefer
ably nonmechanical, is required to reduce response time in
multiple threat environments. Improved decoy launchers
to position decoys quickly and accurately within the sce
nario are also needed.. Low observability technologies are
being developed to decrease or mask the IR/EO signatures
of targets. Target signature reduction increases the effec
tiveness of conventional countermeasure responses by re
ducing the jamming power required to counter the missile
systemeffectively. Lowobservability enables application of
new technologies to IR/EO countermeasures by reducing
the size, weight, and power requirements of decoy and la
ser CM sources. For example, diode laser and diode
pumped nonlinear optical sources can be integrated with
unmanned aerial vehicles to produce new classes of CM
devices and tactics. Largearea spectrally selective sources
and obscurants provide advanced capability against spa
tially and spectrally discriminating threats. Primary laser
and laserpumped nonlinear sources are important evolv
ing technologies. Launchers and vehicles that provide rap
id and precise CM placement with realistic kinematic
performance are areas of increasing importance.
4.5. Decoy Countermeasures
Decoys are EW devices, usually expendable, deployed
from the platforms to be protected. Decoys generate a
jamming response to the threat or false targets. In either
case, the decoy lures the threat away from the intended
target toward the decoy. A jamming decoy generates a
cover signal that masks the target signal. Thereby the
threat sensor signal ﬁdelity is degraded, making detection
and tracking of the intended target more difﬁcult. A jam
ming signal may also activate the antijam homeonjam
mode of the weapon system. As false targets, the decoys
generate credible target signatures to provide weapon sys
tem seduction or distraction. Decoys create confusion that
causes weapons to attack false targets.
Decoys may be either passive or active. A passive decoy
generates a countermeasure response without the direct,
active ampliﬁcation of the threat signal. Principal exam
ples of passive decoys are chaff and corner reﬂectors in the
RF spectrum and ﬂares in the EO/IR spectrum.
4.5.1. Decoy Operational Employment. Decoys provide
EA capability across the entire EW battle timeline. Decoys
are used primarily for EP missile defense and selfprotec
tion missile defense but also for countersurveillance and
countertargeting applications.
Jamming is used in conjunction with decoys to obscure
the target signal at the threat radar during decoy deploy
ment. As decoys are deployed, jamming ceases and the
threat radar acquires the decoy as a target or transfers
radar tracking from the target to the decoy. Threat radar
acquisition of the decoy as a target is probable because
decoys present prominent signatures.
Decoys used for missile defense perform either seduc
tion, distraction, or preferential acquisition functions. A
singledecoy type may perform multiple functions, de
pending on deployment geometry with respect to the
launch aircraft or ship and the stage of electronic combat.
Decoys are used in a seduction role as a terminal de
fense countermeasure against missile weapons systems. A
seduction decoy transfers the lock of the missile guidance
radar or EO/IR sensor from the defending platform onto
itself. The decoy that generates a falsetarget signature is
initially placed in the same threat tracking gate, missile
sensor range, and/or angle segment as the defending tar
get and is subsequently separated from the launching plat
form. The decoy signature captures the missile guidance
sensor, and the target lock is transferred from the ship or
aircraft to the decoy. Typically, the decoy is separated in
both range and angle from the defending target to assure
targettomissile physical separation greater than the mis
sile warhead’s blast range. The seduction decoy missile in
teraction is typically initiated within 10s of deployment.
Distraction decoys are deployed prior to missileseeker ac
quisition and provide multiple false targets from which the
seeker may select. Deployed distraction decoys provide a
confusing environment to the missile seeker, causing it to
attack a decoy rather than the intended target.
The ALE129 chaff decoy (Fig. 16) is representative of
RF seduction decoys for aircraft defense. The NATO Sea
Gnat MK214 cartridge shown ﬁred from a shipboard
launcher in Fig. 17 provides surface defense against
radarguided weapons. Figure 18 shows a TORCH decoy
deployed at sea for IR defense.
Distraction decoys are observed for extended periods in
the engagement scenario. Consequently, the distraction
decoy must generate a credible signature that is sufﬁcient
to preclude shortterm and extended missile decoy dis
crimination.
The AN/SLQ49 inﬂatable corner reﬂector (Fig. 19) and
the rocketlaunched NATO Sea Gnat MK216 chaff
cartridge (Fig. 20) are representative of distraction decoys
for surface ship defense. The TALD decoy (Fig. 21) is an
example of a distraction decoy used for aircraft defense.
Figure 16. ALE129 RF chaff round with the bundle of reﬂector
elements partially deployed from the canister.
ELECTRONIC WARFARE 1349
Frequently, persistent seduction decoys perform a dis
traction function after separating sufﬁciently from the de
fended platform. This ‘‘residual distraction’’ further
minimizes the number of distraction decoys required in
an engagement.
An EA preferential acquisition decoy provides a signa
ture to the missile seeker such that during acquisition the
missile seeker senses the real target only in combination
with the decoy signature. In the endgame, the decoy sig
nature in the missile ﬁeld of view biases the aim point of
the missile tracker away from the intended target.
The preferential acquisition concept requires decoys
positioned close to the defending platform. Decoys can be
towed behind the target aircraft or tethered to the defend
ing ship. The AN/ALE50 (Fig. 22) is a towed decoy used
for air defense preferential acquisition, and the EAGER
decoy (Fig. 23) is being developed for ship defense prefer
ential acquisition.
4.5.2. Chaff Decoys. A chaff decoy is composed of mul
tiple—tens of thousands to millions—of electrically con
ductive dipole ﬁlament elements deployed in the air to
reﬂect and scatter radar signal radiation and create a
falsetarget radar response. Figure 24 shows a typical de
ployed chaff decoy. The chaff decoy frequency response is
determined by the length of the dipole elements, and the
chaff radar crosssectional (RCS) magnitude results from
the number of dipoles deployed. Figure 25 shows a radar
PPI display of an environment containing numerous chaff
clouds.
The RCS of a chaff cloud is tuned for a given frequency
(with the dipole length onehalf the wavelength of the in
cident radar signal), and its RCS can be approximated by
RCSðm
2
Þ ¼
0:018c
2
N
f
2
ð13Þ
where c is the speed of light (3 Â 10
8
m/s), f is the fre
quency in hertz, and N is the number of dipoles in the
cloud.
4.5.3. Corner Reﬂector Decoys. Corner reﬂectors are
conductive geometric structures that are typically shaped
in the form of a perpendicular triangular corner. The
shape maximizes the reﬂection of incident radar signals
and provides a large apparent target signature. Figure 26
shows a multifaceted triangular corner reﬂector that pro
vides wide angular coverage.
The apparent RCS normal to a triangular corner re
ﬂector is given by
RCSðm
2
Þ ¼
4pL
4
f
2
3c
2
ð14Þ
Figure 18. TORCH EO/IR decoy deployed at sea.
Figure 19. AN/SLQ49 inﬂatable corner reﬂector decoy deployed
at sea.
Figure 20. NATO Sea Gnat MK216 distraction decoy deployed
from a rocket launcher.
Figure 17. NATO Sea Gnat MK214 seduction RF decoy de
ployed from a shipboard rocket launcher.
1350 ELECTRONIC WARFARE
where L is the length from the outside corner to the apex
of the reﬂector, f is the frequency in hertz, and c is the
speed of light (3 Â10
8
m/s). The 3dB beamwidth of this
type of corner reﬂector is 401.
4.5.4. Flare Decoys. Flares are typically incendiary de
vices that produce EO/IR radiation to generate a false
target. Figure 27 is an IR image of a magnesium–Teflon
ﬂare deployed from an aircraft.
4.5.5. Active Decoys. An active decoy uses direct threat
signal ampliﬁcation to generate the countermeasure re
sponse. In the case of RF systems, it is generally an RF
ampliﬁer (transistor or tube). In the EO/IR spectrum, a
laser or ﬂash tube ampliﬁes the threat signal. Jammer
and repeater decoys are active decoys.
Repeater decoys receive, amplify, and retransmit the
received signal to generate a false target. Multiple signals
may be retransmitted to generate multiple target returns.
Modulation techniques (amplitude and frequency) may
also be applied to the signal before retransmission to en
hance effectiveness. The apparent radar cross section of an
active RF decoy is given by
RCSðm
2
Þ ¼
ðP
d
G
d
4pR
2
Þ
P
r
G
r
ð15Þ
Figure 21. TALD decoy distraction decoy.
Figure 22. AN/ALE50 towed decoy deployed from a tactical air
craft in ﬂight.
Figure 23. EAGER shipboardtethered decoy in ﬁeld trials.
Figure 24. Deployed chaff round shown as a burst of reﬂector
elements against a sky background.
Figure 25. Radar PPI display showing target reﬂections from
multiple chaff decoys.
ELECTRONIC WARFARE 1351
where P
d
G
d
is the effective radiated power (ERP) of the
decoy, R is the range between the decoy and the radar in
meters, and P
r
G
r
is the effective radiated power (ERP) of
the radar.
For a decoy operating with linear gain, that is, a decoy
whose transmission signal power is directly proportional
to the input signal level (up to the signal compression
level), the RCS relationship simpliﬁes to the relationship
given by
RCSðm
2
Þ ¼
ðG
t
c
2
Þ
4pf
2
ð16Þ
where G
t
is the combined electronic and antenna gains
(receive and transmit) of the decoy, c is the speed of light
(3Â10
8
m/s), and f is the frequency in hertz.
4.5.6. Decoy Effectiveness. A distraction decoy is de
ployed at an extended range from the defending platform
and provides an alternate target for seeker lockon. Dis
traction decoys require deployment before seeker lockon to
engage the radar in its acquisition process. Usually more
than one distraction decoy is used to defend a platform. An
estimate of the effectiveness of the distraction decoy is
given by
P
s
¼1 À
1
Nþ1
ð17Þ
where P
s
is the probability that the missile will be dis
tracted to the decoy and N is the number of distraction
decoys deployed.
Equation (17) assumes that all of the distraction decoys
exhibit viable target signatures and are equally likely to
be acquired by the missile sensor. The number of decoys
deployed can be reduced with the same probability of suc
cess with knowledge of the seeker acquisition logic, for
example, a neartofar/righttoleft acquisition search.
Seduction decoy effectiveness is primarily determined
by the intensity of the decoy signature compared with the
target being defended. However, the radar track bias, for
example, leadingedge tracker and discrimination algo
rithms, can significantly impact decoy effectiveness. In
some cases, the radar track bias can be exploited to in
crease decoy seduction effectiveness.
4.5.7. Decoy Countermeasure Technology. Diverse tech
nologies are required to support decoy launch and station
keeping and countermeasure generation. Because most
decoys are singleevent, shortterm items, cost plays a
major role in selecting and developing technology for de
coy use. Furthermore, because the defending platform
must generally deploy a number of decoys throughout an
engagement, decoy size and weight criteria also are crit
ical. Attendant decoy platform technologies include aero
dynamics, aircraft/projectile design, propulsion systems,
avionics, and mechanical structures. Decoy payload tech
nologies that will have significant importance in future
systems include broadbandwidth microwave and milli
meterwave components (e.g., antennas and ampliﬁers).
Microwave and millimeterwave output power sources
are required with high power, efﬁciency, and duty cycle to
support the projected threat environments. The future RF
threat environment is expected to be densely populated
with longpulse radar. Higher decoy radiated power at
higher duty cycles will be needed to prevent decoy satu
ration as the number of simultaneous threat signals in the
environment increases.
Ultrahighspeed countermeasure frequency set on cir
cuitry is necessary to queue jammer frequency rapidly.
Signals with rapid frequency hopping and frequency
chirping require rapid activation for effective countermea
sures. Spatially large and efﬁcient spectrally matched IR
materials and radiating structures are needed to counter
multispectral, imaging IR seekers. Safe, nontoxic, highly
opaque, broadspectrum IR and electrooptical obscuration
materials are required to mask targets and confuse image
processing seekers. Efﬁcient, primary power sources ca
pable of high peak power and dense energy storage are
needed to provide the increasing demand for electrical
power used in decoy systems.
FURTHER READING
J. S. Accetta and D. L. Shumaker, eds., The Infrared and Electro
Optical Systems Handbook; D. H. Pollock, ed., Vol. 7, Counter
measure Systems, Infrared Information Analysis Center, Ann
Arbor, MI; SPIE Optical Engineering Press, Washington DC,
1993.
Figure 26. Multifaceted corner reﬂector deployed on a ship bow
to provide a high crosssectional reﬂection at several frequencies.
Figure 27. Flare IR decoy deployed from a tactical aircraft in
ﬂight.
1352 ELECTRONIC WARFARE
B. Blake, Jane’s Radar and Ele+9ctronic Warfare Systems, Jane’s
Information Group, Surrey, UK, 1993.
J. A. Boyd et al., Electronic Countermeasures, Peninsula Publish
ing, Los Altos, CA, 1978.
E. J. Chrzanowski, Active Radar Electronic Countermeasures,
Artech House, Norwood, MA, 1990.
N. C. Currie, Techniques of Radar Reﬂectivity Measurement,
Artech House, Dedham, MA, 1984.
R. D. Hudson, Jr., Infrared Systems Engineering, WileyIntersci
ence, New York, 1969.
W. L. McPherson, Reference Data for Radio Engineers, Howard W.
Sams, Indianapolis, 1977.
R. J. Schlesinger, Principles of Electronic Warfare, Peninsula
Publishing, Los Altos, CA, 1961.
M. I. Skolnik, Radar Handbook, McGrawHill, New York, l970.
L. B. Van Brunt, Applied ECM, Vol. 1, EW Engineering, Dunn
Loring, VA, 1978.
W. Z. Wolfe and G. J. Zississ, eds., The Infrared Handbook, rev.
ed., Environmental Res. Inst. Michigan, Ann Arbor, MI, 1985.
ELLIPTIC FILTERS
KENNETH V. NOREN
University of Idaho
Moscow, Idaho
JERRY MICHAEL GRIMM
JMG Consulting
Irving, Texas
1. INTRODUCTION
Real electrical signals are usually composed of an inﬁnite
number of components of various frequencies. Each single
component occurs at a single frequency has a distinct
magnitude and phase. An electrical ﬁlter will be deﬁned as
an electrical network that processes electrical signals on
the basis of the individual frequencies of the signals com
posing that signal. The signal processing can affect both
the magnitude and phase of each component. For example,
the output signal of an antenna may represent an electri
cal signal that requires processing by a receiver. The sig
nal may have a fairly wide spectrum, is composed of
signals at a wide range of frequencies, but it is often de
sired that the receiver process a small range of these fre
quencies, or channel, where important information is
contained. One solution is to use a bandpass ﬁlter in one
of the stages following the antenna. The ﬁlter would pro
cess the signal so that the range or band of frequencies
containing the desired information would pass through to
the output, and the signals outside the range would be
rejected. From here, signal processing of the desired in
formation may take place. Although this example is great
ly simpliﬁed in comparison to what actually happens in a
receiver, it nonetheless illustrates the general idea of
ﬁltering.
Over time, several ‘‘standard’’ types of ﬁlters, or signal
processing schemes for ﬁlters, have evolved. These are
lowpass, highpass, bandpass, bandreject, and allpass ﬁl
ters. Lowpass ﬁlters strive to allow frequencies below a
specific ‘‘passband frequency’’ to pass, while rejecting
those frequencies above the passband frequency. High
pass ﬁlters strive to allow frequencies above a specific
passband frequency to pass, while rejecting those frequen
cies below the passband frequency. Bandpass ﬁlters allow
a range or band of frequencies to pass, while rejecting fre
quencies outside of that band. Bandreject ﬁlters reject a
band of frequencies, allowing frequencies outside that
band to pass. The main objective of these four types of ﬁl
ter is to process the magnitude of the signal as a function
of frequency in a particular way. The allpass ﬁlter lets all
signals pass through, but selects a band of frequencies for
phase angle processing, while not affecting the magni
tudes. Time delay of the input to the output is an impor
tant consideration for allpass ﬁlters. The overall choice of
ﬁlter type depends on the application.
Filter design can be broken down into two broad stages.
The ﬁrst stage is the selection of a transfer function pos
sessing the mathematical properties of ﬁltering that will
meet a set of speciﬁcations that are determined from the
particular application. A transfer function is a mathemat
ical description that describes the relationship between
the input signal and the output signal of a network. We
will use it in the sense that for a given input signal, we will
have a specific output signal that is determined by the
transfer function. Since ﬁlters process electrical signals
according to the frequency content, the transfer function
for a ﬁlter is a function of s ¼jo¼j2pf, where o is the fre
quency in rad/s (radians per second) and f is the frequency
in hertz.
The second stage of ﬁlter synthesis is the realization of
a circuit that possesses the same transfer function as the
mathematical function selected to do the ﬁltering. In gen
eral, the circuit may be an analog, digital, or a mixed an
alog–digital circuit depending on the application.
However, to date, most ﬁlters used at RF and microwave
frequencies are of the analog type.
2. THE APPROXIMATION PROBLEM
When ﬁltering, engineers usually desire ideal ﬁlters. For
example, when deciding to use a lowpass ﬁlter, the engi
neer typically desires that all frequencies above a speciﬁed
passband or cutoff frequency be rejected by the ﬁlter. An
ideal lowpass transfer function magnitude with a pass
band frequency, o
p
¼1 rad/s, is shown in Fig. 1, and the
ideal lowpass transfer function phase characteristics are
shown in Fig. 2. For the magnitude–frequency plot, all
frequencies below o
p
are passed, with a gain of 1 V/V, and
all frequencies above o
p
are rejected. It is a ‘‘brick wall’’
function. It is intuitively obvious that this is an ideal mag
nitude characteristic for a lowpass ﬁlter. The ideal phase
characteristics are not so intuitive. The important feature
of the ideal phase characteristics are not the values of the
phase, but that the phase response is linear. A transfer
function that has linear phase characteristics means that
a signal composed of two different frequencies applied at
the same instant of time at the input of the ﬁlter will
ELLIPTIC FILTERS 1353
arrive, after processing, at the output of the ﬁlter at the
same time. If the two input signals add together to create a
distinct waveform in the time domain, it may be important
that they reconstruct together at the output to maintain
the ‘‘shape’’ of the input signal. Sometimes this is impor
tant, and sometimes it is not. A deviation from the linear
phase response results in phase distortion.
The characteristics depicted in Figs. 1 and 2 are for a
normalized ﬁlter; that is, they have passband frequencies
o
p
equal to 1 rad/s and the magnitude–frequency charac
teristics have a maximum gain of 0 dB, or 1 V/V, in the
passband. It is conventional to begin a lowpass ﬁlter de
sign with a normalized ﬁlter. This allows for a common
starting point for all lowpass ﬁlters, for example, and is
also a convenient way of comparing the characteristics
of different types of lowpass ﬁlter functions with each
other. Moreover, numerous tables exist that provide the
coefﬁcients or poles and zeros of ﬁlter transfer functions.
These tables provide information for normalized ﬁlters.
Since there is an inﬁnite number of possibilities for pass
band frequencies, and similar parameters, it would be im
practical, if not impossible, to generate tables for all
possible cases. Thus, the tables deal only with the normal
ized case. It is trivial to scale a ﬁlter for a desired passband
frequency from the normalized frequency of 1rad/s while
retaining all other characteristics of the ﬁlter.
The ﬁrst step in ﬁlter design is to ﬁnd a transfer func
tion having magnitude characteristics similar to those of
the magnitude–frequency plot depicted in Fig. 1. Unless a
transfer function has an inﬁnite number of terms, it is
impossible to devise a transfer function with these char
acteristics. Although an engineer may desire and ideal re
sponse, he/she must live with a nonideal response. Hence,
from this simple example arises the approximation prob
lem. In other words, may we ﬁnd a transfer function whose
magnitude response comes close enough to the ideal char
acteristics of Fig. 1 to meet or slightly exceed the ﬁlter
speciﬁcations for a particular application. In general, the
higher the order of the ﬁlter, the closer the transfer func
tion will approach the ideal case. However, the higher the
order of a ﬁlter, the more complex the design and the more
components that are needed to realize the transfer func
tion. Thus, the concept of tradeoffs and compromises aris
es. In general, a set of ﬁlter speciﬁcations must be
determined before selecting the transfer function. The
speciﬁcations may be viewed as to how far the actual ﬁl
ter application may deviate from the ideal case. The ideal
ﬁlter response provides a metric with which to compare
practical transfer functions.
Since it is impossible to come up with an ideal transfer
function that is practically realizable, several terms have
been deﬁned and have been accepted as conventions that
allow the description of the deviation of a practical ﬁlter
function from the ideal ﬁlter function. These terms are de
picted in Fig. 3 and may be referred to as ﬁlter speciﬁca
tions. The speciﬁcations are: the passband, the stopband,
the passband ripple (PBR), the stopband ripple (SBR), and
the stopband attenuation A. PBR, SBR, and A are usually
speciﬁed in decibels. There are three distinct regions. The
passband is located from 0rad/s to o
p
rad/s. The second
region is the stopband region located from o
s
to inﬁnity.
Finally, there is the transition region, composed of the
range between o
p
and o
s
. Figure 3 should be interpreted as
follows: Signals at or below o
p
will have a gain of at least of
G dB and at the most H dB, and signals above o
s
will be
attenuated by at least A dB or have a maximum gain of
SBR dB. Note that (GÀH) dB¼PBR in decibels. Filter
types other than lowpass ﬁlters have similar speciﬁcations
and the reader is encouraged to investigate these [1,2].
Previous research in network theory has resulted in
several classic mathematical approximations to the ideal
ﬁlter magnitude function. Each of these was designed to
optimize a property of the ﬁlter function. The lowpass
ﬁlter approximations are usually of the form
jHðjoÞj
2
¼
1
1þe
2
P
2
ðoÞ
ð1Þ
0
−50
−100
−150
−200
−250
−300
−350
−400
−450
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Frequency (radians)
Figure 2. The ideal lowpass ﬁlter transfer function phase char
acteristics, which may be summarized as a linear phase response.
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.8 1.6 2 1.4
Frequency (radians)
Figure 1. Magnitude–frequency plot of the ideal lowpass ﬁlter
transfer function, showing that all frequencies of a signal below
1 rad/s are passed while those above 1rad/s are rejected.
1354 ELLIPTIC FILTERS
When P(o) is replaced with different types of functions,
different approximations to the ideal lowpass ﬁlter re
sponse arise. The standard approximations to the ideal
lowpass magnitude response are the Butterworth approx
imation, the Chebyshev approximation, the inverse
Chebyshev approximation, and the elliptic approximation.
Each of these has strong points and weak points. A ﬁfth
classic approximation worth mentioning is the Bessel
function. This approximation will not be discussed be
cause it strives to approximate the ideal phase response.
This article focuses on the elliptic approximation.
3. THE ELLIPTIC APPROXIMATION
Before beginning a mathematical discussion on the elliptic
approximation, it is useful to examine a plot of an elliptic
ﬁlter function. A plot of the magnitude versus frequency of a
ﬁfthorder lowpass elliptic ﬁlter transfer function is depicted
in Fig. 4. The frequency range of the passband is oro
p
. The
frequency range of the stopband is oZo
s
. The passband and
stopband may be characterized as equiripple; that is, the
amplitude oscillates between a maximum and minimum
throughout each band. If the order of the ﬁlter function is
even, there are n/2 peaks in the passband and n/2mini
mums or zeros in the stopband. If the order of the ﬁlter is
odd, there are (nÀ1)/2 peaks, plus one at o¼0 in the pass
band. Also for the oddorder case, there are (nÀ1)/2
minimums or zeros, plus one at o¼N, in the stopband.
In discussing the properties of the elliptic ﬁlter, it is
beneﬁcial to compare its characteristics with those of the
other classic ﬁlter types. The magnitude responses for
normalized ﬁfth order low pass Butterworth, Chebyshev,
and elliptic ﬁlter magnitude functions are compared in
Fig. 5 and the phases are compared in Fig. 6. Table 1
summarizes the comparisons and the main attributes of
the elliptic ﬁlter may be stated. For a given ﬁlter order, the
elliptic ﬁlter provides the sharpest cutoff characteristics,
and thus out of all three ﬁlter types, it best approximates
the ideal lowpass magnitude function in terms of a sharp
transition region. This is very important if ﬁltering is re
quired for frequencies close to each other, if it is desired to
pass one of these frequencies and reject the other. The
compromise in using the elliptic ﬁlter is its very poor
phase characteristics.
The theory underlying the mathematics of the elliptic
ﬁlter is complicated, and beyond the scope of this article.
Interested readers may consult Refs. 2 and 3. A summary
of the mathematics is discussed in this article. The deri
vation assumes that o
p
¼1rad/s and that o
norm
s
¼o
s
=o
p
:
This results in normalized passband and stopband
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
0 wp ws
Figure 4. The magnitude characteristics of a ﬁfthorder elliptic
ﬁlter showing an equiripple passband and stopband, and a zero at
o¼N.
G
H
PBR
0
0
Passband
w
s
w
p
Frequency (radians)
Stopband SBR
A
Figure 3. Filter speciﬁcations. PBR, SBR, A, passband, and stop
band are ways of characterizing an actual ﬁlter function’s mag
nitude versus frequency response to that of an ideal response. For
an ideal ﬁlter function, PBR¼SBR¼0 V/V and o
s
and o
p
are
equal.
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Elliptic
Chebychev
Butterworth
Figure 5. Amplitude comparison. The ﬁfthorder elliptic func
tion magnitude characterisitics show a much sharper transition
for the passband to the stopband than do the Butterworth and
Chebyshev function characteristics. (This ﬁgure is available in
full color at http://www.mrw.interscience.wiley.com/erfme.)
ELLIPTIC FILTERS 1355
frequencies. The normalized stopband frequency o
norm
s
is
determined by divided the o
s
by o
p
. In a latter stage in the
design process, the circuit is then denormalized, so the
passband and stopband frequencies are those speciﬁed by
the application.
The general form of the elliptic ﬁlter magnitude
squared transfer function is given by Eq. (1). For the low
pass elliptic ﬁlter function, P(jo) is replaced with R
n
(jo).
R
n
(jo) has two different forms, one for an evenorder func
tion and one for an oddorder function. R
n
(jo) will be de
scribed for a normalized lowpass ﬁlter. For the evenorder
case
R
n
ðoÞ ¼M
n=2
i ¼1
o
2
Àðo
norm
s
=o
i
Þ
2
o
2
Ào
2
i
ð2Þ
and for the oddorder case
R
n
ðoÞ ¼No
ðnÀ1Þ=2
i ¼1
o
2
Àðo
norm
s
=o
i
Þ
2
o
2
Ào
2
i
ð3Þ
where M and N are normalization constants and are cho
sen so that R
n
(1) ¼1. The o
i
are calculated for the even or
oddorder case. For the former case, we have
o
i
¼
o
norm
s
sn
ð2i À1ÞK
1
o
norm
s
_ _
n
_
¸
¸
_
_
¸
¸
_
ð4Þ
and for the latter case
o
i
¼
o
norm
s
sn
2iK
1
o
norm
s
_ _
n
_
¸
¸
_
_
¸
¸
_
ð5Þ
where sn is the Jacobian elliptic sine function and K(k) is
the complete elliptic integral of the ﬁrst kind, and is de
ﬁned as
KðkÞ ¼
_
p=2
0
ð1 Àk
2
sin
2
xÞ
À1=2
dx ð6Þ
The order of the ﬁlter function may be determined
by rounding up n to the nearest integer in the
expression
n¼
K
1
o
norm
s
_ _
K
0
1
L
_ _
K
0
1
o
norm
s
_ _
K
1
L
_ _ ð7Þ
where L is deﬁned as
L¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
10
0:1PBR
À1
10
0:1A
À1
_
ð8Þ
and PBR and A are in decibels. Finally
K
0
ðkÞ ¼K
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
1 Àk
2
p
ð9Þ
When R
n
(o) is found, the substitution s ¼o/j is made, and
R
n
(o/j) may be inserted into Eq. (1). The poles of H(s)H(s
Ã
)
¼H(jo)
2
are then found. This is a standard synthesis
technique [4]. The left halfplane poles and half of the
zeros are selected and combined to give the ﬁnal form of
the elliptic ﬁlter transfer function. For n even, we obtain
HðsÞ ¼H
n=2
i ¼1
ðs
2
þo
2
i
Þ
a
0
þa
1
s þ Á Á Á þa
nÀ1
s
nÀ1
þa
n
s
n
ð10Þ
For the case of n odd, we obtain
HðsÞ ¼H
ðnÀ1Þ=2
i ¼1
ðs
2
þo
2
i
Þ
a
0
þa
1
s þ Á Á Á þa
nÀ1
s
nÀ1
þa
n
s
n
ð11Þ
0
−50
−100
−150
−200
−250
−300
−350
−400
−450
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Chebychev
Elliptic
Butterworth
Frequency (radians)
Figure 6. Phase comparison. The ﬁfthorder elliptic function
phase characteristics deviate much farther from the desired ide
al linear phase characteristics than do the Butterworth and
Chebyshev function characteristics. (This ﬁgure is available in
full color at http://www.mrw.interscience.wiley.com/erfme.)
Table 1. Filter Comparisons
Filter Type Transition Region Linear Phase Properties
Butterworth Poor Good
Chebyshev Good Poor
Elliptic Best Very poor
1356 ELLIPTIC FILTERS
Note that the evenorder transfer function given by
Eq. (10) has n poles and zeros and thus no zeros at inﬁn
ity while the oddorder transfer function of Eq. (11) has n
poles and nÀ1 zeros and thus has a single zero at inﬁnity.
It may be convenient to express the denominator in the
form of coefﬁcients or in terms of products of poles and
zeros, depending on the type and method of realization.
Because of the complexity of the calculations required
to ﬁnd the transfer function, the usual method of ﬁnding
H(s) is usually either with a computer program, or using
one of the numerous tables that have been generated and
published [2,5]. Today’s simulators allow the ﬁlter design
er to readily vary zeros to modify the elliptic ﬁlter function
to take on more specialized tasks.
4. FREQUENCY AND IMPEDANCE SCALING
Frequently, a normalized design is the ﬁrst step in ﬁlter
realization. A frequencynormalized ﬁlter is designed for a
passband frequency of 1 rad/s. A typical normalized real
ization has component values on the order of ohms, farads,
and henries. The design is then frequencyscaled so that
the normalized response is shifted into place; that is, the
passband and stopband frequencies are transformed from
normalized values to the design values. The procedure is
performed by ﬁnding the scaling constant o
p
and replac
ing s with s/o
p
in the circuit. This results in new values for
the capacitances and inductances, while the values for re
sistances remain unchanged. In certain circumstances it
may be desirable to frequencyscale the normalized trans
fer function ﬁrst and then do the circuit synthesis.
Frequency scaling usually results in values for capac
itors and inductors that are close to practical, but still not
practical. Moreover, the impedances of the resistors re
main unchanged. The next step in denormalizing a nor
malized realization is to impedancescale. Impedance
scaling amounts to multiplying each impedance by a spec
iﬁed constant. The constant is picked so that after scaling,
the components have reasonable values. If all impedances
are scaled by the same factor, the voltage transfer function
of the circuit remains the same. With good selection of the
constant, practical values may be achieved.
5. ELLIPTIC FILTERS
5.1. Highpass Filters
In the preceding sections we discussed the properties and
applications of lowpass elliptic ﬁlters; there is little differ
ence when discussing those of the highpass elliptic ﬁlter.
The ﬁrst step in highpass ﬁlter design is to normalize
the highpass parameters to the parameters that describe
the normalized lowpass ﬁlter. The parameters that de
scribe the highpass are identical to those of the lowpass
ﬁlter. One difference is that o
s
oo
p
. In general, a lowpass
ﬁlter transfer function may be transformed into a high
pass transfer function by a sto1/s transform. This simply
means that wherever s appears in the transfer function,
1/s is substituted.
Once the normalized lowpass elliptic transfer function
has been determined and a normalized circuit has been
synthesized, a lowpasstohighpass transform is applied.
This means that everywhere in the circuit, s is replaced
with 1/s. This results in capacitors becoming inductors,
and inductors becoming capacitors. If, in an active RC cir
cuit, for example, inductors are not desired, the circuit
may be magnitudescaled by 1/s. This results in the in
ductors becoming resistors and the resistors becoming ca
pacitors.
Alternatively, if a normalized lowpass elliptic function
has been determined, it is possible to apply the sto1/s
transform on the transfer function, resulting in a normal
ized highpass elliptic transfer function. It is now possible to
synthesis a circuit fromdirectly fromthis transfer function.
5.2. Bandpass Filters
Bandpass ﬁlters may be classiﬁed as wideband or narrow
band. A wideband bandpass ﬁlter allows a wide range of
R
s
V
s
+
−
+
−
2 1 3
(a)
+
−
R
load
V
out
n1 n
R
s
V
s
2 1 3
(b)
+
−
R
load
V
out
n2 n1 n
Figure 7. Typical nthorder lowpass pas
sive elliptic ﬁlters are realized with induc
tors and capacitors, and include the source
and load resistances in the overall design.
The circuit in (a) is for the oddorder case
and has n capacitors and (nÀ1)/2 induc
tors. The circuit in (b) is for the evenorder
case and has (nÀ1) capacitors and n/2 in
ductors. Both have a total of n stages, and
n is the order of the ﬁlter transfer function.
ELLIPTIC FILTERS 1357
frequencies to pass with equal magnitude scaling, ideally.
A narrowband ﬁlter seeks to allow only one frequency or a
very small band of frequencies to pass. One definition of
narrowband versus wideband ﬁlters is given in Ref. 1.
This particular definition states that if the ratio of the
upper cutoff frequency to the lower cutoff frequency is
greater than 1 octave, the ﬁlter is considered a wideband
ﬁlter.
Synthesis of wideband bandpass ﬁlters may be per
formed by a cascade of a highpass ﬁlter and a lowpass ﬁl
ter where the cutoff frequency of the lowpass ﬁlter is
greater than that of the highpass ﬁlter. The lower bound
of the definition of wideband results in a separation of the
highpass and lowpass ﬁlters such that there is minimal
interaction between the ﬁlters. If the ratio of the lowpass
cutoff frequency to the highpass cutoff frequency is smaller
than 1 octave, the cutoff frequencies are too close together
and the ﬁlters interact and must be treated as a single
ﬁlter.
Narrowband ﬁlters require alternative synthesis tech
niques, such as a transformation. Like the highpass ﬁlter
functions, bandpass ﬁlter functions may be synthesized
from lowpass ﬁlter functions. This is done by performing
the transformation
s !
o
0
BW
s
o
0
þ
o
0
s
_ _
ð12Þ
on a normalized lowpass ﬁlter function, where o
0
is the
center frequency and BW is the bandwidth of the ﬁlter.
Equation (12) should be interpreted as s is replaced by the
expression on the right. The effect of performing this op
eration on the lowpass circuit is that inductors are re
placed by a series combination of an inductor and a
capacitor and capacitors are replaced by a parallel combi
nation of a capacitor and an inductor. This transform may
also be used in the design of wideband bandpass ﬁlters.
5.3. BandReject Filters
Like bandpass ﬁlters, bandreject ﬁlters may also be clas
siﬁed as wideband or narrowband. Awideband bandreject
ﬁlter seeks to block a wide range of frequencies while al
lowing frequencies outside that band to pass with ideally
equal magnitude scaling. A narrowband bandreject ﬁlter
seeks to block only one frequency or a very small band of
frequencies. In much the same way that it deﬁnes nar
rowband versus wideband bandpass ﬁlters, Ref. 1 deﬁnes
narrowband versus wideband bandreject ﬁlters. The def
inition is identical to that for the bandpass ﬁlter.
Synthesis of wideband bandreject ﬁlters may be per
formed by a cascade of a highpass ﬁlter and a lowpass ﬁl
ter where the cutoff frequency of the highpass ﬁlter is
greater than that of the lowpass ﬁlter. The lower bound of
the definition of wideband results in a separation of the
highpass and lowpass ﬁlters such that there is minimal
interaction between the ﬁlters. If the ratio of the cutoff
frequency of the highpass ﬁlter to that of the lowpass ﬁlter
is less than 1 octave, the cutoff frequencies are too close
together and the ﬁlters interact and must be treated as a
single ﬁlter.
Narrowband ﬁlters require alternative synthesis tech
niques, such as a transformation. Bandreject ﬁlters may
also be synthesized from normalized lowpass ﬁlter func
tions by performing a transform of
s !
BW
o
0
s
o
0
þ
o
0
s
_ _ ð13Þ
on a normalized lowpass ﬁlter, where o
0
is the center
frequency and BWis the bandwidth of the ﬁlter. The affect
of performing this operation on the lowpass circuit is that
inductors are replaced by a parallel combination of an in
ductor and capacitor and capacitors are replaced by a se
ries combination of a capacitor and an inductor. This
transform may also be used in the design of wideband
bandpass ﬁlters.
5.4. Realizations of Elliptic Filters
There are an inﬁnite number of possible synthesis algo
rithms that may be used. In this section we describe one.
The ﬁrst step in elliptic ﬁlter design is to ﬁnd a transfer
function that will meet a set of speciﬁcations that have
been determined from the application. The design usually
begins with specifying the passband frequency o
p
, PBR,
os, and A. If ﬁlter tables are to be used, the frequencies of
the ﬁlter speciﬁcations must ﬁrst be normalized. This is
achieved by dividing o
s
and o
p
by o
p
. Other normaliza
tions are possible. This results in a normalized passband
frequency of 1rad/s, and a normalized stopband frequency
of o
s
/o
p
rad/s. If a highpass ﬁlter is desired, the speciﬁca
tions must be transformed further, by inverting o
p
to 1/o
p
.
Once the desired transfer function is determined, a meth
od of realization is selected. Presently, most RF and mi
crowave realizations of elliptic ﬁlters are analog circuits.
Moreover, the greater majority of the analog realizations
are passive, although active realizations are achievable at
the lower end of the RF frequency spectrum.
5.4.1. Passive Realizations. One may ask if the passive
realizations may utilize only discrete elements. There is
considerable ongoing research in realizating passive ﬁlters
in integrated circuit (IC) technologies, and there are pub
lished research results of such ﬁlters, although this is not
yet commonplace. A principal problem with passive LC
realizations in an IC technology are low quality factors (Qs)
of the inductors presently available in standard IC tech
nologies. The lower the inductor Q, the more losses in the
inductor, and the more difﬁcult it becomes to design high
performance ﬁlters. At the time of this writing, there is
considerable interest in fabricating highQ onchip or on
package inductors in integrated circuit (IC) technologies. If
progress in these inductortechnologies continues at today’s
present rate, it is not inconceivable that passive synthesis
could become commonplace in integrated ﬁlter design.
Passive realizations utilize capacitors, inductors, and
resistors. The source and load resistances are considered
part of the realization. Systematic procedures exist for
the synthesis of passive ﬁlters to realize elliptic transfer
1358 ELLIPTIC FILTERS
functions. Moreover, normalized passive ﬁlters are avail
able as part of table lookup approaches [5]. An example of
odd and even passive ﬁlter topologies is depicted in Fig. 7.
Evenorder passive realizations synthesized from Eq.
(10) will have negative elements because they do not have
at least a zero at inﬁnity. This problem can be solved by
shifting the highestfrequency zero to inﬁnity. The result
ing elliptic ﬁlter function will have a double zero at inﬁn
ity, and the passive ﬁlter realization will now have positive
elements but unequal terminating resistances. This even
order elliptic transfer function is known as case B, while
the original evenorder transfer function given by Eq. (10)
is called case A. Equal terminating resistances can be ob
tained by shifting the ﬁrst maximum to the origin. The
resulting evenorder elliptic transfer function is known as
case C [2,5]. The new ﬁlter functions will be of the forms
given by Eq. (14) for case B and Eq. (15) for case C:
H
B
ðsÞ ¼H
n=2
i ¼2
ðs
2
þo
2
B;i
Þ
b
0
þb
1
s þ Á Á Á þb
nÀ1
s
nÀ1
þb
n
s
n
ð14Þ
H
C
ðsÞ ¼H
n=2
i ¼2
ðs
2
þo
2
C;i
Þ
c
0
þc
1
s þ Á Á Á þc
nÀ1
s
nÀ1
þc
n
s
n
ð15Þ
The magnitude response in for the case B and C ﬁlter func
tions are now slightly modiﬁed from the original type A.
5.4.2. Active Realizations. Active realizations of elliptic
ﬁlters for highfrequency applications tend to be limited to
a few types. These types include GmC, syntheticinductor
based, and negativeresistancebased. GmC ﬁlters utilize
transconductors and capacitors to realize elliptic ﬁlter
functions. Syntheticinductorbased ﬁlters utilize trans
conductors and capacitors to synthesize a impedance that
is inductive. These are often classiﬁed as GmC ﬁlters as
well. Negative resistance based utilize active elements to
synthesis an impedance that looks like a negative resistor.
The ‘‘negative resistor’’ used to cancel out the resistance of
an inductor, thereby providing the inductor with a higher
‘‘effective Q.’’ All include realizations in both IC and dis
crete technologies. However, because of limited bandwidth
of active components, the upper end of the frequency range
tends to be around 200–400MHz for practical applications,
although there have been reports in the literature demon
strating research results at greater frequency, particularly
in monolithic microwave integrated circuit (MMIC) tech
nologies. But the majority of elliptic ﬁlters are passive in
nature, fabricated using discrete techniques. Pushing
active ﬁlter synthesis, and in particular higherorder ﬁl
ters such as elliptic ﬁlters, into the high MHz and even low
GHz ranges appears to be a rich research area.
5.5. Microwave Filters
Elliptic ﬁlters are common for microwave ﬁlters; typical
applications occur after the ﬁnal power ampliﬁer or before
the ﬁrst lownoise ampliﬁer in a system where insertion
losses have direct impact on system performance. Typical
applications include transmit–receive duplexers, multi
band combiners, interference rejection, spurious emission
control, and channelized power combining.
The circuits depicted in Fig. 7 are often used as
the starting point for a passive elliptic ﬁlter regardless
of the technology being used to realize the circuit. How
ever, the zeros of elliptic ﬁlters are frequently modiﬁed
from the ideal values to provide design freedoms that can
sharpen rejection (at the cost of phase linearity), or can
better phase response (at the cost of rejection). The result
is a quasielliptic ﬁlter. In conjunction with appropriate
technology and physical layout, elliptic microwave ﬁlters
realize high performance in compact geometries.
At microwave frequencies, lumped inductors and ca
pacitors are physically large when compared to the wave
lengths of the application and seldom perform well
(especially inductors) because of parasitic effects. More
over, the lumped elements must be considered as distrib
uted elements. Microwave ﬁlters use stages that consist of
distributed elements (e.g., microstrip lines or resonant
cavities) and transitions between elements (e.g., coupling
slots or impedance steps) that appear inductive or capac
itive. Performance depends on the electrical size of ele
ments. There are many ways to realize microwave ﬁlters.
Choosing the technology to use depends on a variety of
factors, some obvious, some subtle, but all with impact on
realizing the needed performance.
5.5.1. Tradeoffs. A primary consideration is the place
ment of the ﬁlter used in the system. Filters integrated
with MMICs require planar layouts such as with micro
strip or coplanar waveguide (CPW). Filters integrated on
RF circuit boards may be stripline, surface acoustic wave
(SAW), ceramic, or helical, depending on the frequency
and space available. If the ﬁlter is a separate unit, handles
high power, or interfaces with an antenna system, wave
guides and coupled cavities are the best choices. Bandpass
and bandreject ﬁlters are easily realized using resonant
elements; lowpass and highpass ﬁlters use propagating
elements.
Frequency also plays a role in technology selection.
Waveguides and waveguide cavities are natural choices
above 5 GHz. Waveguides become too large for the wire
less bands from 400 MHz to 2.5GHz; here, combline cavity
ﬁlters and coaxial line ﬁlters are useful, as are planar ﬁl
ters. Helical and SAW ﬁlters are useful in the lowerMHz
RF bands.
Another consideration is the natural tradeoff that oc
curs between insertion loss versus quality factor of the
components. Each stage in a multistage microwave ﬁlter
loses power to the ﬁnite conductivities of metal and losses
in dielectrics (even lumped inductors and capacitors pos
sess parasitic resistances that absorb power). In general,
ﬁlters composed of a larger number of stages with identi
cal Q values will have higher insertion loss; using higher
Q stages can reduce insertion loss. The Q factor of a ﬁlter
stage can be increased by increasing unit volume, silver
plating the RF path, or loading with lowloss ceramics.
Superconducting ﬁlters have extremely high Q values. Di
electric material allows ﬁlter size to be physically reduced
and Q factors to generally improve; both effects can reduce
ELLIPTIC FILTERS 1359
insertion loss. Using multimode stages, where one phys
ical element supports multiple electrical stages, also can
reduce insertion loss.
There are less obvious considerations to choosing a
technology as well. Power handling may be critical, which
tends toward using waveguides and cavities and places
limits on how small gaps may be. Microwave elements
have dispersive propagation effects and support multiple
modes; these in turn lead to bandwidth limits and higher
frequency spurious responses in which the stopband
begins to resemble the passband. Coupling and tuning
methods also have bandwidth limits. Manufacturability is
also important; it is possible for unwanted coupling mech
anisms to occur and change the ﬁlter response, or for pro
duction tolerances to change passbands. Last but not least,
cost plays a role, especially for commercial applications.
5.5.2. Elliptic Filter Design Concerns Unique to Micro
wave Frequencies. Microwave ﬁlters start with elliptic pro
totype ﬁlters, designed as mentioned previously. The
challenge is mapping lumped inductors and capacitors
into the correct distributed equivalents for the microwave
technology chosen. Fundamental to all microwave ﬁlters
realization is inclusion of transmissionline effects. A very
useful property that follows from this is the ability to in
vert impedances to more easily realize certain ﬁlter top
ologies; impedance inverters convert impedances to
admittances (for e.g., convert a series inductor into a
shunt capacitor). A common way of realizing an inverter
is to employ the quarterwave transformer; for cavity ﬁl
ters, an impedance inverter is realized as a shunt induc
tance in cascade between equallength negative line
lengths. Admittance inverters convert admittances into
impedances; they realize as quarterwave transformers or
as series capacitance in cascade between equallength
negative line lengths. The negative line lengths serve to
reduce the lengths of the loads on either side, usually re
sulting in more compact designs. The most limiting trans
missionline effect is that the effective L and C values
change as frequency deviates from the design center and
the response changes.
Transmission lines can be implemented into prototype
ﬁlters converting lumped impedances, for example, Z
1
¼
joL, to their transmissionline form of Z
1
¼jOL using
Richard’s transformation, O¼tan b1. For a lowpass pro
totype, the passband frequency o
c
of unity results in an
electrical length of 451 (or l/8). Inductors become a short
circuited stub with characteristic impedance (Z
0
) of L;
capacitors become opencircuited stubs with Z
0
¼1/C.
However, the stub impedance is not the same as the
lumped element for frequencies away from passband,
and the ﬁlter response differs.
The physical elements used in microwave ﬁlters have
mode structures with cutoff frequencies determined by
their geometries; at frequencies below cutoff, the mode re
jects energy. As frequencies increase, more modes ‘‘turn
on’’ and again, spurious responses can occur. Cutoff fre
quencies can be changed by small perturbations to the ge
ometries; this allows ﬁlter tuning by elements such as
screws. Square and circular geometries support degener
ate modes that can be coupled to each other; this easily
supports multimode ﬁlters but can add limitations to
exploiting the design freedom of elliptic zeros.
Discontinuities in transmission lines and cavities have
responses that behave inductive or capacitive, depending
on whether primarily magnetic or electric energy is being
stored at the discontinuity. Microwave resonators have
equivalent circuits that look like lumped LC resonators.
Intercavity couplings via slots or probes also appear in
ductive or capacitive, depending on how the modes couple
between cavities. All these discontinuities have effects on
the electrical lengths of ﬁlter sections and adjustments
must be made. Slots in cavity walls seem inductive; ex
ploiting this idea and reducing cavity electrical length on
either side of the slot will implement an impedance in
verter like that mentioned previously. Marcuvitz [7] pro
vides indepth discussions on ﬁnding equivalent circuits
for all types of discontinuities, perturbations, and cou
plings. Another good reference overall for traditional mi
crowave ﬁlter synthesis is Matthaei et al. [8].
Today, numerical simulation tools are employed to de
termine the actual broadband equivalent circuits and their
impact on the basic mode behavior of waveguides and cav
ities. Their use has improved the ability to design ﬁlters,
reduced design cycle time, and provided insight into build
ing ﬁlters. Using these tools effectively still requires an
understanding of the underlying basic principles.
6. TECHNOLOGY OVERVIEW
Striplinetype ﬁlters are planar in design. This makes
them appropriate for use with MMICs and for use on RF
and microwave board layouts. While the microstrip ﬁlter
is most common, ﬁltering structures using coplanar wave
guide (CPW), slotguide, and suspended stripline are used
as well. These ﬁlters generally are small because of the
shorter wavelengths of the planar substrates. They also
are lowQ ﬁlters (200–300), due to etch thickness and sub
strate losses. Planar ﬁlters have limited powerhandling
capability as well. For microstrip ﬁlters, shunt open stubs
are commonly used, since shortcircuiting elements in mi
crostrips involves extra fabrication steps. Alternating
highimpedance/lowimpedance stages and coupledline
segments help realize lowpass or highpass ﬁlters; micro
strip resonators and coupledline ﬁlters allow for realiza
tion of bandpass and bandreject ﬁlters. Coupledline
ﬁlters are generally edgecoupled for microstrip; coupled
lines gain an extra design dimension when considering
suspended stripline ﬁlters, with both vertical and planar
separation contributing to the performance. Planar ﬁlters
also take up board or die space.
Ceramic ﬁlters also ﬁnd use on RF boards and for small
devices from 800MHz and up; a typical application is for
duplexers in mobile phones and wireless LAN cards. Ce
ramic, highdielectric materials dramatically reduce the
physical size of a ﬁlter. They have lowQ(200–400) and low
power handling primarily because of their small size.
Helical ﬁlters are used on RF and IF boards. Helical
ﬁlters use wound wire coils as bandpass resonators; this
allows mediumQ performance in compact packaging in
the 75–800 MHz range.
1360 ELLIPTIC FILTERS
Surface acoustic wave (SAW) ﬁlters are typically used
more at IF frequencies (50–400 MHz). These ﬁlters use pi
ezoelectric materials that support acoustictype leaky
waves; typical wavelengths are 10
À5
those of light. This
means that ﬁlter dimensions can also be much smaller;
SAW ﬁlters also achieve brickwalltype responses. SAW
ﬁlters have lowQ (50–100) and low powerhandling capa
bilities; more recent developments are pushing SAW per
formance up into the 1–2 GHz range.
Waveguide ﬁlters have highQ (typically 10,000) and
high powerhandling capabilities. They are common above
2 GHz, or in lowfrequency, highpower applications.
Waveguides types used include circular, rectangular,
square, ridged, and corrugated, depending on the appli
cation. Walls are placed approximately l/4 apart to gen
erate resonant cavities within the waveguide, the
resonant modes are endwallcoupled to each other. Wave
guide ﬁlters are large in size and can be quite heavy. As
frequencies increase above 50 GHz, mechanical tolerances
become extremely critical in the production process.
Coaxial resonator ﬁlters (also called combline ﬁlters or
TEM ﬁlters) are common from 400MHz to 6GHz. They
consist of sidewallcoupled cavities whose resonance is
controlled by an inductive resonator and a capacitive air
gap. They can be realized as interdigital ﬁlters, with l/4
resonator lengths and airgaps between top and bottom
walls. More commonly, they are realized as combline ﬁl
ters where all resonators are on the same wall. This allows
for ease of manufacture. The resonator length can be less
than l/4; this allows freedom to control when cavity har
monic frequencies occur. These are mediumQ (3000–
6000) and mediumpower devices. Combline resonators
may be dielectrically loaded to increase cavity Q (typically
15,000 or higher). The resonators can also be of dielectric
material; this increases Q (typically above 10,000) and
adds the ability to generate multiple modes.
The highestQ resonators exist in superconducting
ﬁlters. These ﬁlters use either thinﬁlm or bulk highT
c
(critical temperature) materials, and operate at 77K. The
thinﬁlm resonators are used in planar circuits; bulk mate
rials show up in cavity resonators. Because of the zero re
sistance of the superconductor, almost no power is lost per
stage (some loss occurs because cavity walls are still metal
and substrates are still lossy). Consequently, true brickwall
ﬁlters, comprising 30, 40, or more stages, are practical with
small insertion loss. Superconducting ﬁlters require exter
nal cooling support to maintain 77K temperatures; conse
quently, as a system they are more complicated than other
choices. Superconducting ﬁlters have found use in situa
tions with severe interference; here, the performance they
provide outweighs the added complexity of the system.
The advent of MEMS technology is having an impact on
microwave ﬁlters as well. MEMS techniques allow realiza
tion of lumped inductors with reasonable Q values at mi
crowave frequencies. MEMS switches also have potential to
work with tunable ﬁlters and switched ﬁlterbanks. MEMS
processes are generally compatible with IC production and
stand to improve stripline ﬁltering. For standalone MEMS
parts, packaging is a concern—it is usually packaging that
provides most of the parasitics that ruin the performance of
lumped elements at microwave frequencies.
7. SUMMARY
Elliptic ﬁlters are a class of ﬁlters used to shape the mag
nitude of an electric signal. They may be used in applica
tions for any of the standard magnitude processing ﬁlters.
In comparisons to other available ﬁlters, they provide the
sharpest transition from the passband to the stopband for
a given order. The magnitude response is equiripple in the
passband and the stopband. The drawback of elliptic ﬁl
ters is very poor phase characteristics in comparison to
other ﬁlter types. Furthermore, evaluation of elliptic ﬁlter
parameters is considerably more difﬁcult than other ﬁlter
approximation functions, due to the use of elliptic sine
functions and elliptic integrals.
BIBLIOGRAPHY
1. A. B. Williams, Electronic Filter Design Handbook, McGraw
Hill, New York, 1981.
2. L. P. Huelsman and P. E. Allen, Introduction to the Theory and
Design of Active Filters, McGrawHill, New York, 1980.
3. R. W. Daniels, Approximation Methods for Electronic Filter
Design, McGrawHill, New York, 1974.
4. A. Budak, Passive and Active Network Analysis and Synthesis
(reissue), Waveland Press, Prospect Heights, IL, 1991.
5. A. I. Zverev, Handbook of Filter Synthesis, Wiley, New York,
1967.
6. E. P. Cunningham, Digital Filtering: An Introduction,
HoughtonMifflin, Boston, 1992.
7. N. Marcuvitz, Waveguide Handbook, IEE Electromagnetic
Wave Series 21, IEE Publishing, 1986.
8. G. Matthaei, L. Young, and E. M. Jones, Microwave Filters,
Impedance Matching Circuits and Coupling Structures,
Artech House, Norwood, MA, 1980.
FURTHER READING
C. S. Lindquist, Active Network Design with Signal Filtering
Applications, Long Beach, CA: Steward & Sons, 1977.
W.K. Chen, ed., The Circuits and Filters Handbook, CRC Press,
Boca Raton, FL, and IEEE Press, New York, 1995.
ENHANCEMENTS OF THE FINITE DIFFERENCE
TIME DOMAIN METHOD
RAJ MITTRA
WENHUA YU
YINCHAO CHEN
Pennsylvania State University
University Park, Pennsylvania
1. INTRODUCTION
The fundamentals of the conventional ﬁnitedifference
timedomain (FDTD) algorithm on a Cartesian grid have
been adequately covered by the article FINITEDIFFERENCE
TIMEDOMAIN ANALYSIS, and it has been demonstrated that
the algorithm can be used to simulate a wide variety of
electromagnetic problems because of its ﬂexibility and
ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD 1361
versatility. In this article we discuss some variations and
extensions of the method that have been designated to
enhance the performance of the algorithm. The ﬁrst of
these deals with the problems of handling conductors or
dielectrics with curved surfaces and edges by using a
‘‘conformal’’ FDTD scheme, which mitigates the errors
introduced by the staircasing employed in the conven
tional FDTD. The second, referred to in the literature as
the ‘‘multiresolution timedomain (MRTD) method,’’ has
been introduced for the purpose of reducing the require
ments of both the CPU time and memory.
The conformal FDTD technique has had a long history
with many contributors [1–6]. It is beyond the scope of this
article to discuss the various proposed approaches in
detail. Instead we brieﬂy describe one conformal FDTD
technique [7–12], which not only circumvents the need for
complex mesh generation but also does not suffer from
latetime instability problems that plague many of the
proposed schemes.
The method we detail below handles both the PEC and
dielectric objects of arbitrary shape, and does not require a
signiﬁcant modiﬁcation of the Cartesiantype FDTD;
hence, it preserves the computational efﬁciency of the
conventional FDTD technique—which is desirable.
2. CONFORMAL FINITEDIFFERENCE TIMEDOMAIN
TECHNIQUE
In order to describe the conformal approach, we ﬁrst
present below two typical difference equations derived
from the timedependent Maxwell’s curl equations (for
details, see FINITEDIFFERENCE TIMEDOMAIN ANALYSIS). As we
have seen in the above chapter, the electric ﬁelds are
located along the edges of cells, while the magnetic ﬁelds
are positioned at the centers of these cells. Maxwell’s
difference equations in this scheme are written as
H
nþð1=2Þ
x
ði; j; kÞ ¼H
nÀð1=2Þ
x
ði; j; kÞ
þ
Dt
m
x
ði; j; kÞ
E
n
y
ði; j; k þ1Þ ÀE
n
y
ði; j; kÞ
DzðkÞ
À
E
n
z
ði; j þ1; kÞ ÀE
n
z
ði; j; kÞ
DyðjÞ
_
_
_
_
_
_
_
_
_
_
_
_
_
_
ð1Þ
E
nþ1
y
ði; j; kÞ ¼
e
y
ði; j; kÞ À0:5 Á Dt Á s
y
ði; j; kÞ
e
y
ði; j; kÞ þ0:5 Á Dt Á s
y
ði; j; kÞ
E
n
y
ði; j; kÞ
þ
Dt
e
y
ði; j; kÞ þ0:5 Á Dt Á s
y
ði; j; kÞ
Â
H
nþð1=2Þ
x
ði; j; k þ1Þ ÀH
nþð1=2Þ
x
ði; j; kÞ
DzðkÞ
À
H
nþð1=2Þ
z
ði þ1; j; kÞ ÀH
nþð1=2Þ
z
ði; j; kÞ
DxðiÞ
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
ð2Þ
where the superscripts represent the time index and the
arguments correspond to spatial sampling locations. The
equations for the other ﬁeld components can also be
written similarly. As pointed out by Taﬂove in FINITEDIF
FERENCE TIMEDOMAIN ANALYSIS, to avoid numerical instabil
ities, the time increment Dt must satisfy the Courant
condition
Dt
1
c
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
1
D x
_ _
2
þ
1
D y
_ _
2
þ
1
D z
_ _
2
¸ ð3Þ
where D x; D y; D z are the lattice dimensions of the cells.
2.1. Conformal FDTD Technique for a Perfect Electric
Conductor (PEC)
If the equations above are used for conductors with curved
surfaces and/or edges (see Figs. 1 and 2), the staircasing
version of the conventional FDTD may introduce signiﬁ
cant errors, unless a very ﬁne discretization is employed
at a considerable cost of CPU time and memory. It has
been shown [7–9] that this difﬁculty can be overcome by
employing a modiﬁcation of the conventional FDTD up
date equations, called the conformal FDTD algorithm. A
key step in this approach is to assume that both the
electric and magnetic ﬁelds inside the distorted cell are
located at the same positions as those in the conventional
FDTD scheme, and that Faraday’s law is applied over the
entire FDTD cell, rather than only in the distorted part—
thus implying that the contour path follows the edges of
the FDTD in its entirety. Since parts of the contour path
are located inside the PEC region, we discard their con
tributions and write the
_
~
EE
.
d
~
ll contour integral as
∆z
0
(k)
∆x
0
(i )
∆z(i +1,j,k)
∆x(i,j,k+1)
∆E
x
(i,j,k+1)
E
x
(i,j,k)
H
y
(i,j,k)
E
z
(
i
,
j
,
k
)
E
z
(
i
+
1
,
j
,
k
)
z
x
PEC
Free space
(Dielectric)
×
Figure 1. Intersection between the FDTD mesh and a PEC
surface in the x–z plane.
1362 ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD
follows:
_
~
EE
.
d
~
ll ¼
_
Dx
0
ðiÞ
E
x
ði; j; kÞ
.
dx þ
_
Dz
0
ðkÞ
E
z
ði; j; kÞ
.
dz
þ
_
Dxði;j;k þ1Þ
E
x
ði; j; k þ1Þ
.
dx
þ
_
Dzði þ1;j;kÞ
E
z
ði þ1; j; kÞ
.
dz
_
¸
¸
¸
_
_
¸
¸
¸
_
ð4Þ
The deﬁnitions of the variables used in this equation are
shown in Fig. 1. We note that the Eﬁeld representation
remains unchanged from that in the conventional FDTD
scheme. However, the magnetic ﬁelds inside the partially
ﬁlled cells are updated slightly differently to account for
the deformation of the cell, as shown below:
H
nþð1=2Þ
y
ði; j; kÞ ¼
Â
Dzði; j; kÞE
n
z
ði; j; kÞ ÀDzði À1; j; kÞE
n
z
ði À1; j; kÞ
Dx
0
ðiÞ ÂDz
0
ðkÞ
þ
Dxði; j; k À1ÞE
n
x
ði; j; k À1Þ ÀDxði; j; kÞE
n
x
ði; j; kÞ
Dx
0
ðiÞ ÂDz
0
ðkÞ
_
¸
¸
¸
¸
_
_
¸
¸
¸
¸
_
ð5Þ
Equations (4) and (5) enable us to handle PEC objects
whose geometries do not conform to the Cartesian coordi
nates.
2.2. Conformal FDTD Technique for Dielectric Objects
For the partially ﬁlled cells, the conformal technique for
dielectric objects utilizes a modiﬁed version of the concept
of effective dielectric constant (see Figs. 3 and 4). Rather
than using a weighted volume average to deﬁne the
effective dielectric constant, the conformal dielectric algo
rithm [12] utilizes a weightedline average concept in
stead. The calculation of the weightedline average
dielectric constant is carried out on the bold lines in Fig.
5a. The effective dielectric constants e
eff
x
ði; j; kÞ and
e
eff
y
ði; j; kÞ are deﬁned below:
e
eff
x
ði; j; kÞ ¼
Dx
2
ði; j; kÞ
Dx
e
2
þ
Dx ÀDx
2
ði; j; kÞ
Dx
e
1
ð6Þ
e
eff
y
ði; j; kÞ ¼
Dy
2
ði; j; kÞ
Dy
e
2
þ
Dy ÀDy
2
ði; j; kÞ
Dy
e
1
ð7Þ
We note from the Figs. 5b and 5c that E
x
ði; j þ1; kÞ and
E
y
ði þ1; j; kÞ do not penetrate the dielectric; hence, the
dielectric constant e
1
is used in place of the effective
dielectric constant in the conformal dielectric FDTD up
date equations for the electric and magnetic ﬁelds. Typical
update equations take the form
E
nþ1
x
ði; j; kÞ ¼E
n
x
ði; j; kÞ þ
Dt
e
eff
x
ði; j; kÞDy
Â½H
nþð1=2Þ
z
ði; j þ1; kÞ ÀH
nþð1=2Þ
z
ði; j; kÞ
À
Dt
e
eff
x
ði; j; kÞDz
Â½H
nþð1=2Þ
y
ði; j; k þ1Þ ÀH
nþð1=2Þ
y
ði; j; kÞ
ð8Þ
E
nþ1
x
ði; j þ1; kÞ ¼E
n
x
ði; j þ1; kÞ þ
Dt
e
1
ði; j þ1; kÞDy
Â½H
nþð1=2Þ
z
ði; j þ2; kÞ ÀH
nþð1=2Þ
z
ði; j þ1; kÞ
À
Dt
e
1
ði; j þ1; kÞDz
Â½H
nþð1=2Þ
y
ði; j þ1; k þ1Þ
ÀH
nþð1=2Þ
y
ði; j þ1; kÞ
ð9Þ
where Dy and Dz are the mesh dimensions along the y and
z directions, respectively. Note that the effective dielectric
constant is used in (8) because the edge on which E
x
ði; j; kÞ
lies does penetrate the dielectric. Furthermore, the update
equation in (9) has a form that is identical to that used in
the conventional FDTD formula, because E
x
ði; j þ1; kÞ is
located entirely within a single dielectric region, whose
Deformed cell
and
Integral path
Perfectly electric conductor
Border of the PEC
x
x
E
x
y
E
z
z
Free space or dielectric
Figure 2. Contour path of the conformal FDTD in a deformed
cell.
x
z
y
(b)
s
0
= s
1
+ s
2
eff
1
, s
1
2
, s
2
Weighted area
or volume averaging
(a)
Figure 3. Conformal dielectric techniques: (a) ori
ginal problem; (b) after averaging.
ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD 1363
permittivity is e
1
. Similar update equations can also be
written for other electric ﬁeld components.
3. ILLUSTRATIVE EXAMPLES
We now present some examples to illustrate the applica
tion of the conformal FDTD method. We use a Gaussian
pulse modulated by a sine function for the excitation, and
a sixlayer unsplit PML [13] to truncate the FDTD domain
on the open boundary. On the basis of the Courant
condition, the timestep is chosen to be:
Dt ¼
0:995
c
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
1
Dx
_ _
2
þ
1
Dy
_ _
2
þ
1
Dz
_ _
2
¸ ð10Þ
which is very close to that dictated by the Courant
condition for the undistorted cell.
3.1. Cylindrical Cavity
For the ﬁrst example, we consider the problem of comput
ing the resonant frequencies of the dominant TE and TM
modes of a cylindrical cavity of circular cross section. The
results for the resonant frequencies are shown in Table 1,
together with the dimensions of the cavities, with R and
H representing the radius and height of the cylindrical
cavities, respectively. The spatial discretization is chosen
to be 0.005 m for each of the four cases studied. We observe
good agreement between the conformal FDTD results and
those obtained analytically. Furthermore, the CFDTD
algorithm, described herein, has been found to be totally
stable.
1
, s
1
2
, s
2
(a)
s
0
= s
1
+ s
2
eff
(b)
Weighted area or
volume averaging
2
, s
2
1
, s
1
1
, s
1
1
, s
1
2
, s
2
2
, s
2
Figure 4. Illustrative examples showing that dif
ferent material distributions lead to the same
effective dielectric constant: (a) original problem;
(b) after averaging.
Weighted area or
volume averaging
∆x
∆y
1
2
∆y
2
∆x
2
(a)
(b) (c)
y
e
f
f
(
i
,
j
,
k
)
E
x
(i, j, k)
H
z
(i, j, k)
E
x
(i, j +1, k)
1
(i, j +1, k)
E
y
(
i
,
j
,
k
)
E
y
(
i
+
1
,
j
,
k
)
1
(
i
+
1
,
j
,
k
)
x
(i, j, k)
eff
Figure 5. Linear weighted dielectric constant aver
aging procedure used in the present conformal di
electric FDTD scheme: (a) Intersection between
FDTD mesh and curved dielectric surface; (b) origi
nal problem; (c) equivalent problem.
1364 ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD
3.2. Radial Stub
For the next example of the CFDTD simulation, we con
sider a radial stub whose geometry is shown in Fig. 6. The
dimensions of the computational domain are 8.12 Â 8.12
Â 0.889mm
3
, and the domain is discretized into 40 Â 40
Â 7 cells. In order for the edges of the microstrip lines to
coincide with those of the FDTD grid, we choose a nonuni
form mesh whose dimensions are Dx(i) ¼0.20175mm,
i ¼1,2,3,4; Dx(i) ¼0.203357mm, i ¼5; . . . ; 18; Dx(i) ¼
0.203mm, i ¼19,20,21; Dx(i) ¼0.20333mm, i ¼26; . . . ; 36;
and Dx(i) ¼0.20175mm, i ¼37,38,39,40. A uniform mesh
is used in the z direction with a cell dimension of
Dz(k) ¼0.127mm. A voltage source between the ground
plane and the left microstrip line is placed at a distance of 5
cells away from the domain boundary (y ¼0), while the
observation point is located at a distance of 10 cells
away from the boundary (y ¼y
max
) in the y direction. The
3dB cutoff frequency of the excitation source is 25GHz,
and the timestep is Dt ¼0.3146ps. The scattering para
meter S
21
versus frequency is shown in Fig. 7. Good
agreement with the experimental data reported in Ref.
14 is observed.
3.3. Microstrip Patch Antenna of Circular Shape
Next, we compute the scattering parameter of a microstrip
patch antenna of circular shape, as shown in Fig. 8. The
dimensions of the computational domain are 0.11 Â 0.088
Â 0.00795m
3
and are subdivided into 55 Â 44 Â 10 cells.
The cell dimensions are chosen to be 0.002, 0.002, and
0.000795 m, in the x, y, and z directions, respectively. The
simulation was run with a timestep of Dt ¼2.29847 ps, and
no stability problems were encountered. The reﬂection
plot of the circular patch antenna as a function of fre
quency is shown in Fig. 9. The results reported in Ref. 2
via the application of the symmetric and asymmetric
planar generalized Yee (PGY) approaches, with a timestep
of 0.725 ps [2], and those obtained using the MoM (method
of moments) [15] are also plotted in the same ﬁgure for
comparison. The CFDTD results are seen to compare quite
favorably with those obtained via the MoM.
3.4. DielectricLoaded Cavity
Finally, we turn to an example pertaining to a curved
dielectric structure, namely, rectangular cavity loaded
with a cylindrical dielectric rod (see Fig. 10) [16].
The relative dielectric constant of the rod is 38, while it
is 1 for the pedestal below. The computational domain is
discretized in to 48 Â44 Â47 cells, and the Dx and Dy are
chosen to be 0.519112 mm inside, and 0.549275 and
0.563033 mm outside the dielectric rod. Also Dz is
0.537307, 0.5537, and 0.53658 mm below, inside, and
above the dielectric resonator, respectively. For the other
structures listed in Table 2, the discretizations are chosen
in a similar manner.
The simulation results, shown in Table 1, demonstrate
that the results agree well with both the modematching
values and the measured data, thus validating the accu
racy of the CFDTD algorithm, even for these narrowband
structures.
4. SUMMARY OF CFDTD
In this section, we have described a conformal FDTD
approach, designated for both PEC and dielectric objects
without the use of staircasing and the need for complex
mesh generation techniques. We have shown, by compar
ison with analytical and experimental results, that the
CFDTD algorithm is both accurate and computationally
efﬁcient. For further information, the reader may refer to
Refs. 17 and 18.
Table 1. Comparison of Resonant Frequencies of a Circularly Cylindrical Cavity Derived Using Analytical and CFDTD
Techniques
Geometry I Geometry II Geometry III Geometry IV
Modes Methods R Â H¼(0.1Â0.1 m) R Â H¼(0.1 Â0.08m) R Â H¼(0.1 Â0.06m) R Â H¼(0.06Â0.06m)
TE
111
CFDTD 1.734 2.062 2.639 3.133
Analytical 1.738 2.071 2.650 3.148
Difference 0.23 0.43 0.42 0.48
TM
011
CFDTD 1.88 2.194 2.742 2.872
Analytical 1.889 2.199 2.751 2.897
Difference (%) 0.48 0.23 0.33 0.86
x
y
z
0.61 mm
0.61 mm
3
.2
5
3
m
m
0.635 mm
r
= 9.8
PEC ground plane
Figure 6. Microstrip line with a radial stub.
ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD 1365
5. INTRODUCTION TO MRTD
The multiresolution timedomain (MRTD) scheme for
solving electromagnetic ﬁeld problems, introduced by
Krumpholz and Katehi [19,20], is based on expansions of
unknown ﬁelds in terms of scaling or wavelet functions. It
has been shown that the use of the MRTD improves the
computational capacity and efﬁciency over those of the
conventional ﬁnitedifference timedomain (FDTD)
method [19–23].
The MRTD approach, described in this section, retains
the philosophy of the leapfrog algorithm employed in the
conventional FDTD, and places the MRTD scheme in the
context of the FDTD, although it differs considerably in
detail, as we will see below.
To illustrate the application of MRTD, we analyze a
twolayer, dielectricloaded cavity, and develop a systema
tic formulation for constructing the constitutive relations;
then update equations in the MRTD transform domain by
utilizing only the ﬁeld quantities deﬁned in the real
structures. We introduce the concepts of the multiple
image technique (MIT) and present a criterion for deter
mining the number of images needed along each side of
the structure that is based on the localization properties of
the scaling basis functions.
In this section, we also present a twodimensional (2D)
version of the MRTD, in conjunction with the unsplit
anisotropic perfectly matched layer (APML), for the ana
lysis of printed transmission lines. We apply the MIT
APML concept for the analysis of microwave structures
with inhomogeneous materials, including different dielec
trics and conductors with inﬁnite conductivity. Such ima
ging is needed to handle the boundaries in the MRTD
formulation.
6. MIT MRTD METHOD FOR PECSHIELDED STRUCTURES
6.1. MRTD Update Equations
By using the generalized differential matrix operators
(GDMOs) [24], the two curl relations of the Maxwell’s
equations can be expressed as
0 À@
z
@
y
@
z
0 À@
x
À@
y
@
x
0
_
¸
¸
_
_
¸
¸
_
H
x
H
y
H
z
_
¸
¸
_
_
¸
¸
_
¼
@
@t
D
x
D
y
D
z
_
¸
¸
_
_
¸
¸
_
ð11Þ
0 À@
z
@
y
@
z
0 À@
x
À@
y
@
x
0
_
¸
¸
_
_
¸
¸
_
E
x
E
y
E
z
_
¸
¸
_
_
¸
¸
_
¼ Àm
0
@
@t
H
x
H
y
H
z
_
¸
¸
_
_
¸
¸
_
ð12Þ
where, for generality, we assume that the relative permit
tivity is a biaxial tensor. Then the associated constitutive
0 5 10 15 20 25
1
CFDTD
Experiment [20]
Staircase FDTD
0
0.1
0.2
0.3
CFDTD
Experiment [20]
Staircase FDTD
Frequency (GHz)
0.4
0.5
0.6
0.7
0.8
0.9
I
S
2
1
I
(
d
B
)
Figure 7. Transmission coefﬁcient of the microstrip line with a
radial stub comparison of CFDTD and experimental results.
Top view
3.18 mm
1.59 mm 4 mm
56 mm
1
7
m
m
Side view
PEC
Figure 8. Microstripcoupled circular patch antenna. The microstrip line is located on a 1.59mm
substrate ðe
r
¼2:62Þ backed by a ground plane, and the patch antenna is printed on the same
dielectric at a height of 1.59mm, above the microstrip line.
0.5 1 3.5 4 4.5 5
0
Frequency (GHz)
I
s
1
1
I
(
d
B
)
1.5 2 2.5 3
−15
−10
−5
0
CFDTD
PGY [4]
MoM [24]
Figure 9. Comparison of reﬂection loss of the microstripcoupled
circular patch antenna computed by using the CFDTD, the MoM,
and the PGY techniques.
1366 ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD
relation in the space domain becomes
D
x
D
y
D
z
_
¸
¸
_
_
¸
¸
_
¼e
0
e
xx
ð~rrÞ 0 0
0 e
yy
ð~rrÞ 0
0 0 e
zz
ð~rrÞ
_
¸
¸
_
_
¸
¸
_
E
x
E
y
E
z
_
¸
¸
_
_
¸
¸
_
ð13Þ
In the MRTD domain, the local relationship between
the D and E ﬁelds shown above is no longer valid; hence,
we must work simultaneously with both the D and E
ﬁelds.
As a ﬁrst step in the MRTD formulation, we begin
by representing all the ﬁeld quantities in terms of the
scaling function in space and pulse function in time as
follows
D
x
ð~rr; tÞ
¼
1
i;j;k;n¼À1
fx
D
n
i þð1=2Þ;j;k
f
i þð1=2Þ
ðxÞf
j
ðyÞf
k
ðzÞh
n
ðtÞ
ð14Þ
H
x
ð~rr; tÞ
¼
1
i;j;k;n¼À1
fx
H
n
i;j þð1=2Þ;k þð1=2Þ
f
i
ðxÞf
j þð1=2Þ
ðyÞ
Âf
kþð1=2Þ
ðzÞh
nþð1=2Þ
ðtÞ
ð15Þ
where fðxÞ denotes the cubic spline Battle–Lemarie scal
ing function [25,26] and h
n
(t) is a rectangular pulse
function. Substitution of these ﬁeld expansions into the
Maxwell equations, followed by the application of the
Galerkin method [27], leads us to the following set of
update equations
fx
D
nþ1
i þð1=2Þ;j;k
¼
fx
D
n
i þð1=2Þ;j;k
þ
n
aðnÞ
Â
fz
H
nþð1=2Þ
i þð1=2Þ;j þn þð1=2Þ;k
Dt
Dy
_
À
fy
H
nþð1=2Þ
i þð1=2Þ;j;k þn þð1=2Þ
Dt
Dz
_
ð16Þ
fx
H
nþð1=2Þ
i;j þð1=2Þ;kþð1=2Þ
¼
fx
H
nÀð1=2Þ
i;j þð1=2Þ;kþð1=2Þ
þ
1
m
0
n
aðnÞ
Â
fy
E
n
i;j þð1=2Þ;k þn þ1
Dt
Dz
_
À
fz
E
n
i;j þn þ1;k þð1=2Þ
Dt
Dy
_
ð17Þ
where the quantities being updated are the expansion
coefﬁcients in the ﬁeld expansions, and their correspond
ing superscripts and subscripts represent the discretized
time and space positions, respectively. The coefﬁcient aðnÞ,
appearing in (16) and (17), may be derived by utilizing a
nonorthogonal relationship [19,22]. Although the summa
tion index n in (16) and (17) spans from positive to negative
inﬁnity, it is usually sufﬁcient to truncate it at 9, by taking
advantage of the localized nature of the Battle–Lemarie
scaling function.
6.2. MIT for Derivation of the Constitutive Relationship
The constitutive relationship in the transform domain of
the MRTD is considerably more involved than its counter
part in the spatial domain presented in (13). This is
because the MRTDtransformed D ﬁeld at a particular
point is determined not only by the E ﬁeld at the same
location but also by the distribution of the E ﬁeld in
its neighborhood. To solve for the constitutive relation in
the MRTD domain, we once again represent the E ﬁelds
as expansions in terms of the same basis functions as
(14), and substitute the above expansion into (13). Next,
an application of the Galerkin’s method leads us to
the constitutive relationships in the MRTD domain that
Table 2. Comparison of Different Methods for a Dielectric Rod in a Rectangular Cavity
a
Size (in.)
Resonant Frequencies (GHz)
2R t Present Kaneda’s [11] Mode Matching [17] Measured [17]
0.654 0.218 4.388 4.40 4.3880 4.382
0.689 0.230 4.163 4.17 4.1605 4.153
0.757 0.253 3.725 3.78 3.721 3.777
a
a¼1.0 in., b¼1.0 in., e ¼0.92 in., h¼0.275 in., r ¼0 in.; e
r
¼38.
b
a
2R
2r
e
h
t
Figure 10. Dielectric resonator comprising a cylindrical rod in a
rectangular cavity. The pedestal below is assumed to be free space
in this work.
ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD 1367
read
fx
D
n
i þð1=2Þ;j;k
¼e
0
þ1
i
0
;j
0
;k
0
¼À1
e
xx
ð Þ
ii
0
;jj
0
;kk
0
fx
E
n
i
0
þð1=2Þ;j
0
;k
0 ð18Þ
where
ðe
xx
Þ
ii
0
;jj
0
;kk
0 ¼
_
þ1
À1
_
þ1
À1
_
þ1
À1
f
i þð1=2Þ
ðxÞf
j
ðyÞf
k
ðzÞe
xx
ð~rrÞ
Âf
i
0
þð1=2Þ
ðxÞf
j
0 ðyÞf
k
ðzÞ
dx dy dz
Dx Dy Dz
¼d
i;i
0 d
j;j
0 d
k;k
0 þ
P
p¼1
e
xx
ðpÞ À1 ½ x
ii
0
;jj
0
;kk
0 ðpÞ
ð19aÞ
x
ii
0
jj
0
kk
0 ¼a
i;i
0 ðpÞb
j;j
0 ðpÞg
k;k
0 ðpÞ
¼
_
x
2
ðpÞ
x
1
ðpÞ
f
i þð1=2Þ
ðxÞf
i
0
þð1=2Þ
ðxÞ
dx
Dx
Â
_
y
2
ðpÞ
y
1
ðpÞ
f
j
ðyÞf
j
0 ðyÞ
dy
Dy
_
z
2
ðpÞ
z
1
ðpÞ
f
k
ðzÞf
k
ðzÞ
dz
Dz
ð19bÞ
where we assume that all dielectric objects are
rectangular and P is the total number of dielectric
objects.
We now present an example of deriving the constitutive
relationship for a twolayer, dielectricloaded cavity shown
in Fig. 11. By expressing
e
aa
¼e
aa
ðyÞ
¼
e
r
ðpÞ; y
1
ðpÞ y y
2
ðpÞ
1; otherwise
_
_
_
ðp¼1; 2Þ
ð20Þ
we can write the x component of the constitutive relation
in the MRTD domain
fx
D
n
i þð1=2Þ;j;k
¼e
0
j þM
j
0
¼jÀM
ðe
xx
Þ
j;j
0
fx
E
n
i þð1=2Þ;j
0
;k
ð21Þ
with
ðe
xx
Þ
j;j
0 ¼d
j;j
0 þ
P
p¼1
e
r
ðpÞ À1 ½
Â
_
y
2
ðpÞ
y1ðpÞ
f
j
ðyÞf
j
0 ðyÞ
dy
Dy
ð22Þ
where M is the number of the sampled E ﬁelds in the
MRTD domain. It is usually set equal to a number slightly
greater than 9, by taking advantage of the localization of
the scaling function.
It is very important to note that the summations in
(21) cover not only the original computation domain but
also the entire image regions, as shown in Fig. 12. In fact,
for a PECshielded structure, all the ﬁeld quantities in
the image regions can be expressed in terms of the ones
inside the original cavity structure. Toward this end, we
employ the principle that the tangential components of
the image E ﬁelds (parallel to the PEC mirror wall) or the
normal components of the image H ﬁelds (normal to
the PEC mirror wall) are always odd symmetric about
the original ﬁelds, whereas the normal components of the
image E ﬁelds and the tangential components of
the image H ﬁelds are even symmetric. Using these above
guidelines we can express all the image ﬁeld quantities
in terms of the ﬁelds in the original region, and this
leads to a degeneration of the summation index in (21)
to N
y
, the grid number along the y direction, deﬁned
only inside the original cavity structure. Consequently,
N
y2
∆y
N
y
∆y
N
x
∆x
N
z
∆z
N
y1
∆y
x
z
y
r1
r2
Figure 11. Geometry of a twolayer, dielectricloaded cavity.
Image structures
Original structure
Image structures
Figure 12. Original structure and multiple images in the y–z
plane.
1368 ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD
(21) can be simpliﬁed to read
fx
D
n
i þð1=2Þ;j;k
¼e
0
N
y
j
0
¼0
ð^ ee
xx
Þ
j;j
0
fx
E
n
i þð1=2Þ;j
0
;k
ð23Þ
where we deﬁne the equivalent permittivity matrix in the
MRTD domain as follows
Ny
j
0
¼0
ð^ ee
xx
Þ
j;j
¼
Ny
j
0
¼0
ðe
xx
Þ
j;j
0 À
ÀNy
j
0
¼À1
ðe
xx
Þ
j;j
0
À
2Ny
j
0
¼Ny þ1
ðe
xx
Þ
j;j
0 ÀÁ Á Á
þðÀ1Þ
m
jÀM
j
0
¼ÀðmÀ1ÞNyÀ1
ðe
xx
Þ
j;j
0
þðÀ1Þ
m
j þM
j
0
¼mN
y
þ1
ðe
xx
Þ
j;j
0
ð24aÞ
or in a matrix form as
½^ ee
xx
¼½e
xx
orig
À½e
xx
ImgðÀ1Þ
À½e
xx
ImgðÀ1Þ
ÀÁ Á Á þðÀ1Þ
m
½e
xx
ImgðÀmÞ
þðÀ1Þ
m
½e
xx
Imgð þmÞ
ð24bÞ
where the m is the number of images. The ﬁrst term in
(24b) corresponds to the original structure, and the rest
are associated with the images.
Next, we present a criterion for determining the
number of image m along the y direction (positive or
negative). The criterion is derived from knowledge of the
effective range of the basis functions at the PEC boundary
locations
m¼
INT
M
N
y
_ _
þ1; if
M
N
y
is not aninteger
M
N
y
; if
M
N
y
is aninteger
_
¸
¸
¸
_
¸
¸
¸
_
ð25Þ
where INT is the integerconverting function that trun
cates all the decimal parts of a number. For example, if
M¼10, N
y
¼5, then m¼INT(2.5) þ1 ¼3, this implies that
we need three images in both sides of the þy and –y
directions.
We will now illustrate the procedure for deriving the
equivalent permittivity matrix in the MRTD domain.
We consider a cavity with the discretization grid number
N
y
¼6 in the y direction. To simplify the notation, we omit
the subscript x, and write ðe
xx
Þ
j;j
simply as e
j;j
0 . In this
example we choose M¼9 and m¼2. We then have
fx
D
n
i þð1=2Þ;0;k
fx
D
n
i þð1=2Þ;1;k
fx
D
n
i þð1=2Þ;2;k
fx
D
n
i þð1=2Þ;3;k
fx
D
n
i þð1=2Þ;4;k
fx
D
n
i þð1=2Þ;5;k
fx
D
n
i þð1=2Þ;6;k
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¼e
0
^ ee
0;0
^ ee
0;1
^ ee
0;2
^ ee
0;3
^ ee
0;4
^ ee
0;5
^ ee
0;6
^ ee
1;0
^ ee
1;1
^ ee
1;2
^ ee
1;3
^ ee
1;4
^ ee
1;5
^ ee
1;6
^ ee
2;0
^ ee
2;1
^ ee
2;2
^ ee
2;3
^ ee
2;4
^ ee
2;5
^ ee
2;6
^ ee
3;0
^ ee
3;1
^ ee
3;2
^ ee
3;3
^ ee
3;4
^ ee
3;5
^ ee
3;6
^ ee
4;0
^ ee
4;1
^ ee
4;2
^ ee
4;3
^ ee
4;4
^ ee
4;5
^ ee
4;6
^ ee
5;0
^ ee
5;1
^ ee
5;2
^ ee
5;3
^ ee
5;4
^ ee
5;5
^ ee
5;6
^ ee
6;0
^ ee
6;1
^ ee
6;2
^ ee
6;3
^ ee
6;4
^ ee
6;5
^ ee
6;6
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
Â
fx
E
n
i þð1=2Þ;0;k
fx
E
n
i þð1=2Þ;1;k
fx
E
n
i þð1=2Þ;2;k
fx
E
n
i þð1=2Þ;3;k
fx
E
n
i þð1=2Þ;4;k
fx
E
n
i þð1=2Þ;5;k
fx
E
n
i þð1=2Þ;6;k
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
ð26Þ
with
e
xx
½
orig
¼
e
0;0
e
0;1
e
0;2
e
0;3
e
0;4
e
0;5
e
0;6
e
1;0
e
1;1
e
1;2
e
1;3
e
1;4
e
1;5
e
1;6
e
2;0
e
2;1
e
2;2
e
2;3
e
2;4
e
2;5
e
2;6
e
3;0
e
3;1
e
3;2
e
3;3
e
3;4
e
3;5
e
3;6
e
4;0
e
4;1
e
4;2
e
4;3
e
4;4
e
4;5
e
4;6
e
5;0
e
5;1
e
5;2
e
5;3
e
5;4
e
5;5
e
5;6
e
6;0
e
6;1
e
6;2
e
6;3
e
6;4
e
6;5
e
6;6
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
ð27aÞ
e
xx
½
ImgðÀ1Þ
¼
0 e
0;À1
e
0;À2
e
0;À3
e
0;À4
e
0;À5
e
0;À6
0 e
1;À1
e
1;À2
e
1;À3
e
1;À4
e
1;À5
e
1;À6
0 e
2;À1
e
2;À2
e
2;À3
e
2;À4
e
2;À5
e
2;À6
0 e
3;À1
e
3;À2
e
3;À3
e
3;À4
e
3;À5
e
3;À6
0 e
4;À1
e
4;À2
e
4;À3
e
4;À4
e
4;À5
0
0 e
5;À1
e
5;À2
e
5;À3
e
5;À4
0 0
0 e
6;À1
e
6;À2
e
6;À3
0 0 0
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
ð27bÞ
ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD 1369
e
xx
½
Imgð þ1Þ
¼
0 0 0 e
0;9
e
0;8
e
0;7
0
0 0 e
1;10
e
1;9
e
1;8
e
1;7
0
0 e
2;11
e
2;10
e
2;9
e
2;8
e
2;7
0
e
3;12
e
3;11
e
3;10
e
3;9
e
3;8
e
3;7
0
e
4;12
e
4;11
e
4;10
e
4;9
e
4;8
e
4;7
0
e
5;12
e
5;11
e
5;10
e
5;9
e
5;8
e
5;7
0
e
6;12
e
6;11
e
6;10
e
6;9
e
6;8
e
6;7
0
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
ð27cÞ
e
xx
½
ImgðÀ2Þ
¼
0 0 0 e
0;À9
e
0;À8
e
0;À7
0
0 0 0 0 e
1;À8
e
1;À7
0
0 0 0 0 0 e
2;À7
0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
ð27dÞ
e
x
½
Imgð þ2Þ
¼
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 0 0 0 0 0 0
0 e
4;13
0 0 0 0 0
0 e
5;13
e
5;14
0 0 0 0
0 e
6;13
e
6;14
e
6;15
0 0 0
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
ð27eÞ
where (27a) is associated with the original structure,
whereas (27b)–(27e) are contributed by the image regions.
As seen from the above, these latter contributions to the
equivalent permittivity matrix decrease dramatically with
the increase in the number of image regions. Since the
elements of the equivalent permittivity matrix depend
only on the basis functions and the material properties
of the structure, both the ^ ee
xx
½ and its inverse ^ ee
xx
½
À1
can be
computed in advance and saved in the MRTD Maxwell
solver for the update equations. Consequently, the E ﬁeld
can be updated as follows
fx
E
n
i þð1=2Þ;j;k
¼
1
e
0
Ny
j
0
¼0
ð½^ ee
xx
À1
Þ
j;j
0
fx
D
n
i þð1=2Þ;j
0
;k
ð28Þ
where, apparently, we need to compute the product of only
two small matrices, whose dimensions are typically 6 or
less in this work, to update the E ﬁelds at each timestep.
6.3. MIT in Update Equations
Although, in principle, (16) and (17) can be employed to
update the ﬁelds, they are numerically inefﬁcient and
somewhat impractical for coding and computation, be
cause the summations in these update expressions include
both the original and image regions. In this section, we
truncate the summations in these update equations by
excluding some of the image ﬁeld quantities. This
is accomplished by carrying out the following two steps:
(1) expressing all the ﬁeld quantities in terms of those
deﬁned in the original structure by the MIT technique
described in the previous section and, (2) maintaining the
values and positions of the coefﬁcients aðnÞ unchanged for
all of the terms, but expressing them only by using the
grid indices inside the structure. Following this procedure,
we can rewrite the update equations for the x components
using the MIT as follows
fx
D
n
i þð1=2Þ;j;k
¼
fx
D
n
i þð1=2Þ;j;k
þ
N
y
À1
n ¼0
aðn ÀjÞ
fz
H
nþð1=2Þ
i þð1=2Þ;n þð1=2Þ;k
_
þ
Img ¼even
NyÀ1
n ¼0
aðÀ Img À2 ð Þ
.
N
y
Àn Àj À1Þ
_ _
þ
N
y
À1
n ¼0
aðImg
.
N
y
þn ÀjÞ
_
fz
H
nþð1=2Þ
i þð1=2Þ;n þð1=2Þ;k
_
þ
Img ¼odd
NyÀ1
n ¼0
aðÀImg
.
N
y
Àn Àj À1Þ
_ _
þ
N
y
À1
m¼0
aðImg
.
N
y
þn ÀjÞ
_
Â
fz
H
nþð1=2Þ
i þð1=2Þ;N
y
ÀnÀ1 þð1=2Þ;k
__
Dt
Dy
þ
NzÀ1
n ¼0
aðn ÀkÞ
fy
H
nþð1=2Þ
i þð1=2Þ;j;n þð1=2Þ
_
þ
Img ¼even
N
z
À1
n ¼0
aðÀ Img À2 ð Þ
.
N
z
Àn Àk À1Þ
_ _
þ
NzÀ1
n ¼0
aðImg
.
N
z
þn ÀkÞ
_
Â
fy
H
nþð1=2Þ
i þð1=2Þ;j;n þð1=2Þ
_
þ
Img ¼odd
N
z
À1
n ¼0
aðÀImg
.
N
z
Àn Àk À1Þ
_ _
þ
NzÀ1
n ¼0
aðImg
.
N
z
þn ÀkÞ
_
Â
fy
H
nþð1=2Þ
i þð1=2Þ;j;N
z
ÀnÀ1 þð1=2Þ
__
Dt
Dz
ð29aÞ
1370 ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD
H
nþð1=2Þ
i;j þð1=2Þ;k þð1=2Þ
¼
fx
H
nÀð1=2Þ
i;j þð1=2Þ;k þð1=2Þ
þ
1
m
0
N
z
n ¼0
aðn Àk À1Þ
fy
E
n
i;j þð1=2Þ;n
_
þ
Img ¼even
À
N
z
À1
n ¼0
aðÀ Img À2 ð Þ
.
N
z
Àn Àk À2Þ
_ _
þ
NzÀ1
m¼0
aðImg
.
N
z
þn ÀkÞ
_
fy
E
n
i;j þð1=2Þ;n þ1
_
þ
Img ¼odd
N
z
À1
n ¼0
aðÀImg
.
N
z
Àn Àk À2Þ
_ _
À
N
z
À1
n ¼0
aðImg
.
N
z
þn ÀkÞ
_
Â
fz
E
nþð1=2Þ
i;j þð1=2Þ;NzÀnÀ1
__
Dt
Dz
À
1
m
0
N
y
n ¼0
aðn Àj À1Þ
fz
E
n
i;n;k þð1=2Þ
_
þ
Img ¼even
À
N
y
À1
n ¼0
aðÀ Img À2 ð Þ
.
N
y
Àn Àj À2Þ
_ _
þ
NyÀ1
n ¼0
aðImg
.
N
y
þn ÀjÞ
_
Â
fz
E
n
i;n þ1;kþð1=2Þ
_
þ
Img ¼odd
N
y
À1
n ¼0
aðÀImg
.
N
y
Àn Àj À2Þ
_ _
À
N
y
À1
n ¼0
aðImg
.
N
y
þn ÀjÞ
_
Â
fz
E
n
i;N
y
ÀnÀ1;kþð1=2Þ
__
Dt
Dy
ð29bÞ
where N
y
and N
z
are the numbers of cells along the
y and z directions inside the structure, respectively, and
the summation indexes, ‘‘odd’’ (1,3, y) and ‘‘even’’ (2,4,
y), denote the contributions to the updated ﬁeld quan
tities from the speciﬁed odd or even image regions,
respectively.
Before closing this section, it would be worthwhile to
point out that although the (29a) and (29b) are quite
lengthy, they are actually quite efﬁcient in terms of coding
and computation. Because all the ﬁeld quantities usually
have small dimensions, the coefﬁcients aðnÞ can be
computed in advance and saved in the MRTD solver for
later use.
7. 2D MRTD ANALYSIS FOR PRINTED
TRANSMISSION LINES
7.1. Field Construction
Following the procedure described in Refs. 28 and 29, we
develop a twodimensional (2D) MRTD analysis by deriv
ing a set of governing equations for the problem at hand,
speciﬁcally, a lossless guidedwave structure shown in
Fig. 13. We begin by expanding the ﬁeld quantities as
E
x
ðx; y; z; tÞ; E
y
ðx; y; z; tÞ; E
z
ðx; y; z; tÞ
H
x
ðx; y; z; tÞ; H
y
ðx; y; z; tÞ; H
z
ðx; y; z; tÞ
_
_
_
_
_
_
t
¼
jE
x
ðx; y; tÞ; jE
y
ðx; y; tÞ; E
z
ðx; y; tÞ
H
x
ðx; y; tÞ; H
y
ðx; y; tÞ; jH
z
ðx; y; tÞ
_
_
_
_
_
_
t
e
Àjbz
ð30Þ
where b is the propagation constant for the structure.
Next, we use the GDMOs to obtain the following govern
ing equations from the Maxwell’s curl equations
0 jb @
y
Àjb 0 À@
x
À@
y
@
x
0
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
H
x
H
y
0
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
¼ e
0
@
@t
e
xx
0 0
0 e
yy
0
0 0 e
zz
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
þ
s
e
x
0 0
0 s
e
y
0
0 0 s
e
z
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
jE
x
jE
y
E
z
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
ð31aÞ
0 jb @
y
Àjb 0 À@
x
À@
y
@
x
0
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
jE
x
jE
y
E
z
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
¼ Àm
0
@
@t
m
xx
0 0
0 m
yy
0
0 0 m
zz
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
þ
s
m
x
0 0
0 s
m
y
0
0 0 s
m
z
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
Â
H
x
H
y
jH
z
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
ð31bÞ
In practice, we set s
m
a
¼0. Note that a printed planar
transmission line is usually inhomogeneous in the vertical
ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD 1371
direction, but homogenous in the horizontal direction,
which leads to e
aa
¼e
aa
ðyÞ; ða ¼x; y; zÞ.
7.2. 2D MRTD Algorithm
Next, we represent all the ﬁeld quantities by using
scaling functions in space and pulse functions in
time. For instance, for the x component of the ﬁelds we
write
E
x
ðx; y; tÞ ¼
þ1
n;i;j ¼À1
fx
E
n
i þð1=2Þ;j
f
i þð1=2Þ
ðxÞf
j
ðyÞh
n
ðtÞ ð32Þ
H
x
ðx; y; tÞ ¼
þ1
n;i;j ¼À1
fx
H
nþð1=2Þ
i;j þð1=2Þ
f
i
ðxÞf
j þð1=2Þ
ðyÞh
nþð1=2Þ
ðtÞ
ð33Þ
Substitution of the ﬁeld expansions into Maxwell’s
equations, followed by application of Galerkin’s method,
leads us to obtain a set of ﬁeld update equations. For
instance, the x components of the update equations are
given by
fx
H
nþð1=2Þ
i;j þð1=2Þ
¼
fx
H
nÀð1=2Þ
i;j þð1=2Þ
þ
Dt
m
0
m
x
Â b
fy
E
n
i;j þð1=2Þ
À
n
aðnÞ
fz
E
n
i;j þn þ1
1
Dy
_ _
ð34Þ
þ1
i
0
;j
0
¼À1
ðs
e
x
Þ
ii
0
;jj
0
Dt
2
fx
E
nþ1
i
0
þð1=2Þ;j
0
þ
fx
E
n
i
0
þð1=2Þ;j
0
_ _
þe
0
þ1
j
0
¼À1
ðe
xx
Þ
j;j
0
fx
E
nþ1
i þð1=2Þ;j
0
À
fx
E
n
i þð1=2Þ;j
0
_ _
¼Dt b
fy
H
nþð1=2Þ
i þð1=2Þ;j
þ
m
aðmÞ
fz
H
nþð1=2Þ
i þð1=2Þ;j þmþ1
1
Dy
_ _
ð35Þ
where the coefﬁcient a(n) is as given in Ref. 19. Assuming
that the conducting strip is inﬁnitely thin, we can express
its conductivity as
s
e
a
¼
s; i
1
Dx x i
2
Dx; j
0
Dy À
t
2
_ _
oyo j
0
Dy þ
t
2
_ _
0; otherwise
_
¸
_
¸
_
ðD¼x; yÞ
ð36Þ
Thus, we can express ðe
xx
Þ
j;j
0 and ðs
e
x
Þ
ii
0
;jj
0 as
ðe
xx
Þ
j;j
0 ¼
_
þ1
À1
f
j
ðyÞe
xx
ðyÞf
j
0 ðyÞ
dy
Dy
ð37aÞ
ðs
e
x
Þ
ii
0
;jj
0 ¼
_
þ1
À1
_
þ1
À1
f
i þð1=2Þ
ðxÞf
j
ðyÞsf
i
0
þð1=2Þ
ÂðxÞf
j
0 ðyÞ
dx dy
Dx Dy
ð37bÞ
By including the image contributions and using the ﬁrst
order approximation, we can derive
ðs
e
x
Þ
ii
0
;jj
0 ¼s
i;i
0 d
i;i
0 B
j;j
0 d
j;j
0 ðd
j;j
0
þd
j;Àj
0
þd
j;2j
0
þ Á Á ÁÞ ð38Þ
where d
ii
0 is the Kro¨necker symbol, which is 1 for i ¼i
0
,
otherwise 0. Then, we further rewrite the discretized
timedomain equation as
ð
fx
E
nþ1
i þð1=2Þ;j
þ
fx
E
n
i þð1=2Þ;j
Þ
s
e
ðd
j;j
0
þd
j;Àj
0
þd
j;2j
0
þ Á ÁÁÞDt
2
þe
0
þ1
j
0
¼À1
ðe
xx
Þ
j;j
0 ð
fx
E
nþ1
i þð1=2Þ;j
0
À
fx
E
n
i þð1=2Þ;j
0 Þ
¼Dt b
fy
H
nþð1=2Þ
i þð1=2Þ;j
þ
n
aðnÞ
fz
H
nþð1=2Þ
i þð1=2Þ;j þn þ1
1
Dy
_ _
ð39Þ
The summation index n in (39) includes not only the
regions in the interior of the original structure but also
the image regions. Two types of boundary are frequently
used for truncating the grids: the absorbing boundary
condition for open structures and the perfect electric
conductor (PEC) for shielded ones. Usually, the PEC
boundary is handled by using the MIT described earlier.
Figure 13 shows the original structures: shielded and open
microstrip lines, as well as their images.
ð
fx
E
nþ1
i þð1=2Þ;j
þ
fx
E
n
i þð1=2Þ;j
Þ
s
e
d
j;j
0
Dt
2
þe
0
Ny
j
0
¼0
ð^ ee
xx
Þ
j;j
0 ð
fx
E
nþ1
i þð1=2Þ;j
0
À
fx
E
n
i þð1=2Þ;j
0 Þ
¼Dt b
fy
H
nþð1=2Þ
i þð1=2Þ;j
þ
n
aðnÞ
fz
H
nþð1=2Þ
i þð1=2Þ;j þn þ1
1
Dy
_ _
ð40Þ
By taking an inverse of (30), we can obtain the
update equation for the x component of the E ﬁeld as
1372 ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD
follows
fx
E
nþ1
i þð1=2Þ;j
¼
Ny
j
0
¼0
ð½AÞ
j;j
0
fx
E
n
i þð1=2Þ;j
0
þ
1
e
0
Ny
j
0
¼0
ð½BÞ
j;j
0
_
b
fy
H
nþð1=2Þ
i þð1=2Þ;j
0
Dt
þ
m
aðmÞ
fz
H
nþð1=2Þ
i þð1=2Þ;j
0
þmþ1
Dt
Dy
_
ð41Þ
with
½A ¼ ½^ ee
xx
j;j
0 þ½I
s
e
d
j;j
0
Dt
2e
0
_ _
À1
.
½^ ee
xx
j;j
0 À½I
s
e
d
j;j
0
Dt
2e
0
_ _
ð42aÞ
½B ¼ ½^ ee
xx
j;j
0 þ½I
s
e
d
j;j
0
Dt
2e
0
_ _
À1
ð42bÞ
For a shielded structure, we need to employ the MIT for
constructing the ½^ ee
xx
matrix. Since the matrix [A] and [B]
depend only on the basis functions and the material
properties of the structure, they can be computed in
advance and saved in the MRTD Maxwell solver for the
update equations.
7.3. Application of APML
In this section we address the problem of mesh truncation
for open structures. We ﬁnd that the 2D version of the
lossy, uniaxial, anisotropic PML (APML) [30] medium is
well suited for this purpose. Since the APML is applied
only in the x and y directions, the frequencydomain
Maxwell curl equation in the APML medium can be
written as
j
@H
z
@y
þjbH
y
ÀjbH
x
Àj
@H
z
@x
@H
y
@x
À
@H
x
@y
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¼joe
0
e
xx
s
y
s
x
0 0
0
e
yy
s
x
s
y
0
0 0 e
zz
s
x
s
y
_
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
_
jE
x
jE
y
E
z
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
ð43aÞ
@E
z
@y
þbE
y
bE
x
À
@E
z
@x
j
@E
y
@x
Àj
@E
x
@y
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
¼ Àjom
0
m
xx
s
y
s
x
0 0
0
m
yy
s
x
s
y
0
0 0 m
zz
s
x
s
y
_
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
_
H
x
H
y
jH
z
_
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
_
ð43bÞ
with
s
a
¼
1 þ
s
a
joe
0
; for absorption in a direction
1; elsewhere
_
_
_
ða ¼x; yÞ
ð44Þ
h
h
y
x
r
(b) (a)
Imag
Imag
Origin Imag
Imag
b
w
APML
y
x
r
a
APML
W
Figure 13. Cross section of printed transmis
sion lines and their images: (a) shielded micro
strip line; (b) open microstrip line.
ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD 1373
Depending on combination of the values of s
x
and s
y
, the
equations above are applicable to all of the APML regions,
which include two side edges, the top wall, as well as the
two corners. By expanding these equations for the case of
the corners, we immediately obtain the x components of
the governing equations given below:
@H
z
@y
À
@H
y
@z
¼joD
x
þ
s
y
e
0
D
x
ð45Þ
@E
z
@y
À
@E
y
@z
¼ ÀjoB
x
À
s
y
e
0
B
x
ð46Þ
We use a twostep update approach to timestepping
ﬁelds inside APML regions. Following the MRTD discre
tization procedure, described previously, we obtain the
desired Eﬁeld update equations:
fx
D
nþ1
i þð1=2Þ;j
¼
1 À
s
y
Dt
2e
0
1 þ
s
y
Dt
2e
0
fx
D
n
i þð1=2Þ;j
þ
1
1þ
s
y
Dt
2e
0
Â b
fy
H
nþð1=2Þ
i þð1=2Þ;j
Dt
_
þ
m
aðmÞ
fz
H
nþð1=2Þ
i þð1=2Þ;j þmþ1
Dt
Dy
_
ð47Þ
fx
E
nþ1
i þð1=2Þ;j
¼
fx
E
n
i þð1=2Þ;j
þ
1
e
0
j
0
ð½e
x
À1
Þ
j;j
0
Â 1 þ
s
x
Dt
2e
0
_ _
fx
D
nþ1
i þð1=2Þ;j
0
_
À 1 À
s
x
Dt
2e
0
_ _
fx
D
n
i þð1=2Þ;j
0
_
ð48Þ
We can employ the same procedure to obtain the remain
ing E and Hﬁeld update equations.
In applying the MRTD method, we chose the time step
Dt to satisfy the stability and limit condition [19] in all
simulations. Starting from the convergence condition of
the update equations of D ﬁelds given above, it can be
shown that, to obtain good absorption, the APML para
meters may be chosen as
s
a
¼s
max
a Àa
0
N
p
Da
¸
¸
¸
¸
¸
¸
¸
¸
p
¼g
e
0
Dt
i Ài
0
N
p
¸
¸
¸
¸
¸
¸
¸
¸
p
; ða ¼x; yÞ ð49Þ
where a
0
is the starting position in the a direction, d is the
thickness of the APML, and N
p
is the number of cells
within the APML region. In principle, it can be shown that
this relationship is consistent with the standard form,
which can be derived from a plane wave, normally in
cident on a PML region. We found that we can minimize
the reﬂection error, and have good absorption, if we choose
p¼2 and g ¼0.8–1.6.
7.4. Computation of the Propagation Characteristics
Typically, for the printed transmission lines, the two
parameters of interest are the effective dielectric constant
e
eff
, and the characteristic impedance Z
0
, which can, in
general, be derived from the MRTD calculated ﬁeld quan
tities. For example, the timedomain voltage V deﬁned as
an integral from the PEC ground to the stripline can be
calculated as follows
V ¼ À
_
h
0
E
y
ðx
0
; y; t
0
Þdy
¼ À
Ns
l ¼0
_
ðl þ1ÞDy
lDy
E
y
ðx
0
; y; t
0
Þdy
¼ À
N
s
l ¼0
_
ðl þ1ÞDy
lDy
__
E
y
ðx; y; tÞdðx Àx
0
Þdðt Àt
0
Þdx dt
_ _
dy
¼ À
N
s
l ¼0
1
i;jÀl ¼À1
fy
E
n
i;jÀl þð1=2Þ
f
i
ðx
0
ÞbðjÞDy
_ _
ð50Þ
where N
s
is the number of the cells from the bottom
ground to the PEC strip and the indices i and j of the
summation span the original microstrip line and its image
regions. The integral coefﬁcient b(j) forms a localized
distribution with even symmetric b(j) ¼b( Àj) listed in
Table 3. We can apply a similar procedure to solve for all
required ﬁeld quantities.
7.5. BHW Function Truncation and Modulation
To eliminate the Gibb phenomenon that results from an
abrupt truncation of the timedomain response, the Black
man–Harris (BH) window function is used to truncate and
modulate the time domain [29]. We write
½E
W
k
ðt
n
Þ; H
W
k
ðt
n
Þ ¼ E
k
ðt
n
Þ; H
k
ðt
n
Þ ½
W
BH
ðt
n
Þ ðk ¼x; y; zÞ
ð51Þ
where ½E
k
ðt
n
Þ; H
k
ðt
n
Þ represent the original timedomain
electromagnetic ﬁelds deﬁned in the MRTD algorithm,
and ½E
W
k
ðt
n
Þ; H
W
k
ðt
n
Þ are the windowed versions. Since the
frequencydomain response is the convolution of the fre
quencydomain signature and its window function, the
Table 3. Integral Coefﬁcient b(j)
j b(j)
0 0.9143952
1 0.0385998
2 0.0095740
3 À0.0086579
4 0.0050629
5 À0.0027034
6 0.0014053
7 À0.0007203
8 0.0000695
1374 ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD
choice of the window function can signiﬁcantly affect the
accuracy of the propagation characteristics. Extremely
low sidelobe levels ( À92 dB) of the BH window ensures
the accuracy of the extracted frequency domain signals,
whereas the use of the conventional rectangular window
function, which has a much higher sidelobe level
( À13 dB), can seriously corrupt the frequency signatures,
especially for a resonant structure.
8. NUMERICAL RESULTS
8.1. MIT MRTD Application
We have applied the MRTD scheme in conjunction with
the MIT to analyze a number of dielectricloaded cavities,
viz., an empty cavity, two partially ﬁlled (25% and 50%)
cavities containing a single dielectric medium, and two
different twolayer, dielectricﬁlled cavities. The dimen
sions of the cavity were set to be 1 Â 2 Â 1.5 m
3
, in the x, y,
and z directions, respectively, in all these cases.
We begin with an empty cavity that is discretized with
2 Â 4 Â 3 cells. The frequency spectrum of E
x
sampled at
the gridpoint [1,3,1] is displayed in Fig. 14, and the
extracted resonant frequencies for the dominant and
higherorder modes are summarized in Table 4, along
with the discretization parameters. It is evident that the
results obtained with the MRTD scheme show good agree
ment with both the analytical results and those derived
using the FDTD. However, the MRTD scheme demands
only 0.8% of the computational resources needed in the
conventional FDTD technique. In addition, we have in
vestigated the required CPU time for conventional MRTD
and the MIT MRTD schemes. For a total timestep of
N
t
¼10
4
, it was found that the CPU time for the conven
tional MRTD was about 19.27 s, whereas of the MIT
MRTD required about 27.08 s on a 500MHz Alpha digital
workstation. Obviously, this increase in the CPU time is
due to the use of additional images in the process of
implementing the MIT MRTD technique. Specially, we
use ﬁve images along the positive and negative x direc
tions, and three in the remaining directions. By compar
ison, the conventional MRTD uses only a single image
irrespective of the direction. We also note that the intro
duction of the MIT in the MRTD scheme enhances the
accuracy of the results.
The MRTD analysis of the half and quarterﬁlled
dielectric cavities, with e
r1
¼e
r2
¼64 and e
r1
¼64, e
r2
¼1,
respectively, also yields excellent results, as is evident
from Figs. 15 and 16 and Tables 5 and 6. Once again, the
MRTD scheme is found to be highly efﬁcient when com
pared to the conventional FDTD method, since it requires
only 0.8% and 1.2% of the CPU memory for the two cases
described above. In general, the accuracy of the MRTD
technique is again improved when it is combined with the
multipleimage technique.
Next, we consider a twolayer, dielectricﬁlled cavity,
using only 2 Â 6 Â 3 cells to discretize the computational
domain. We investigate the following two cases: (1)
e
r1
¼64, e
r2
¼8 and (2) e
r1
¼8, e
r2
¼64, with N
y1
¼N
y2
¼
1:5 cells. The structure dimensions still remain
0 50 100 150 200 250 300 350
0
100
200
300
400
500
600
700
800
Empty cavity
Frequency (MHz)
M
a
g
n
i
t
u
d
e
o
f
E
x
a
t
[
1
,
3
,
1
]
Figure 14. Frequency spectrum of E
x
at the location [1,3,1] for
an empty cavity.
Table 4. Resonant Frequencies (in MHz) for an Empty
Cavity
a
Analytic
FDTD [20] SMRTD [20]
This
Method
Percentage
Difference
(10Â20Â15) (2 Â4Â3) (2Â4 Â3) (%)
125.00 124.85 125.10 124.95 À0.040
180.27 179.75 180.50 180.39 0.067
213.60 212.40 214.60 214.30 0.328
246.22 244.50 248.70 248.55 0.946
250.00 248.70 251.00 250.75 0.300
301.04 298.95 303.90 303.65 0.580
336.34 334.35 339.20 338.91 0.764
a
V ¼1 Â2Â1:5m
3
; Dx ¼Dy ¼Dz ¼
1
2
m.
0 10 20 30
40 50 60 70
0
100
200
300
400
500
600
700
800
900
Half−filled dielectric cavity
M
a
g
n
i
t
u
d
e
o
f
E
x
a
t
[
1
,
2
,
1
]
Frequency (MHz)
Figure 15. Frequency spectrum of E
x
at the location [1,2,1] for a
dielectric cavity with 50% ﬁlling.
ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD 1375
V ¼1Â2 Â1:5m
3
. We also analyze the same problem
using the traditional FDTD mesh with a discretization of
Dx ¼Dy ¼2:5 cm and Dz ¼3 cm, speciﬁcally, 40 Â40 Â50
cells. As seen from Figs. 17 and 18 and from Table 7, the
two sets of computed results agree very well with each
other and the percentage differences are 0.038% and
0.343%, for cases 1 and 2, respectively. However, the
MIT MRTD scheme uses only 0.045% of the computational
resources relative to the conventional FDTD algorithm.
8.2. Application of 2D MRTD
We consider a shielded microstrip line shown in Fig. 13a,
whose trace is assumed very thin and perfectly conduct
ing. Figure 19 shows that the propagation characteristics
derived from an application of the present 2D MRTD
scheme agree quite well with those derived from the
spectraldomain approach (SDA) [31].
Next, we study the propagation characteristics of a
shielded microstrip line, shown in Fig. 13a, whose dimen
sions and substrate material are given in the caption of
Fig. 20. We observe, from Fig. 20, that the MRTDcom
puted results are in good agreement with those published
in the literature, derived by using the FDTD [29]. For this
case, the distance from the air–dielectric interface of the
0 10 20 30 40 50 60 70
0
0.5
1
1.5
2
2.5
3
Quart−filled dielectric cavity
Frequency (MHz)
M
a
g
n
i
t
u
d
e
o
f
E
x
a
t
[
1
,
3
,
1
]
Figure 16. Frequency spectrum of E
x
at [1,3,1] for a dielectric
cavity with 25% ﬁlling.
Table 5. Resonant Frequencies (in MHz) for a Partially
Filled Dielectric Cavity with 50% Filling
a
FDTD [20] SMRTD [20]
This
Method Percentage
Analytic (10Â20 Â15) (2Â4 Â3) (2Â4 Â3) Difference (%)
18.627 18.615 18.715 18.692 0.349
27.172 27.140 27.350 27.313 0.519
29.375 29.215 29.580 29.526 0.514
35.069 34.970 35.280 35.247 0.507
a
e
r1
¼e
r2
¼64; N
y1
¼N
y2
¼1:5; V ¼1 Â2 Â1:5 m
3
; Dx ¼Dy ¼Dz ¼
1
2
m.
Table 6. Resonant Frequencies (in MHz) for a Partially
Filled Dielectric Cavity with 25% Filling
a
Analytic
FDTD [21] SMRTD [21] This Method
Percentage
Difference
(10Â20Â15) (2 Â6 Â3) (2Â6 Â3) (%)
27.290 27.250 27.370 27.313 0.084
37.136 37.000 37.370 37.323 0.503
42.343 42.200 42.420 42.602 0.611
a
e
r1
¼64; e
r2
¼1; N
y1
¼N
y2
¼1:5; V ¼1 Â2 Â1:5 m
3
; Dx ¼Dz ¼
1
2
m; Dy ¼
1
3
m.
0 5 10 15 20 25 30 35
0
20
40
60
80
MRTD Method
0 5 10 15 20 25 30 35
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
FDTD Method
Frequency (MHz)
Frequency (MHz)
M
a
g
n
i
t
u
d
e
o
f
E
x
a
t
[
1
,
3
,
1
]
M
a
g
n
i
t
u
d
e
o
f
E
x
a
t
[
2
0
,
1
2
,
3
0
]
r1
= 64,
r2
= 8
f
0
= 26.398 MHz
r1
= 64,
r2
= 8
f
0
= 26.388 MHz
Figure 17. Frequency spectrum of E
x
of a twolayer, dielectric
loaded cavity.
0 5 10 15 20 25 30
0
200
400
600
800
1000
1200
1400
0 5 10 15 20 25 30
0
0.5
1
1.5
2
2.5
3
Frequency (MHz)%
Frequency (MHz)
M
a
g
n
i
t
u
d
e
o
f
E
x
a
t
[
1
,
3
,
1
]
M
a
g
n
i
t
u
d
e
o
f
E
x
a
t
[
2
0
,
1
2
,
3
0
]
MRTD Method
FDTD Method
r1
= 8,
r2
= 64
f
0
= 20.752 MHz
r1
= 8,
r2
= 64
f
0
= 20.681 MHz
Figure 18. Frequency spectrum of E
x
for a twolayer, dielectric
loaded cavity.
1376 ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD
structure to the inner surface of the APML is seven cells.
The corresponding discretization parameters employed in
the FDTD and the MRTD methods are summarized in
Table 8, and a comparison of the total number of cells
needed in the MRTD and FDTD technique is presented in
Table 8, and the relative advantage of the former is
evident from the table.
Finally, we investigate an open microstrip line, whose
geometry is speciﬁed in Fig. 13b. Once again, from Fig. 21
we observe good agreement between the computed results
and those derived by using the FDTD [29,32], as well as
the empirical approach [33,34]. The corresponding discre
tization parameters employed in the FDTD and MRTD
methods are summarized Table 9. We note that the
memory required in the 2D MRTD method is only about
6.25% or 7.64% of the 2D FDTD scheme for the same
shielded and open microstrip lines, respectively. We
further notice that, to ensure computational accuracy,
we must employ at least two cells along the width of the
narrow PEC strip.
9. SUMMARY OF MRTD
In this section we have presented an adjustable multiple
image technique, which can be incorporated in the multi
resolution timedomain (MRTD) scheme for the boundary
truncation of PECshielded. We have also developed a 2D
version of the MRTD, and have applied it in conjunction
with an APML mesh truncation technique, for the analy
sis of both shielded and open microstrip lines. We have
developed a systematic technique for constructing the
multiple images, extracting the constitutive relations,
Table 7. Resonant Frequencies (in MHz) for a TwoLayer,
DielectricFilled Cavity
a
Case
10.1 FDTD This Method
Percentage
Difference
(40Â40Â50) (2 Â6 Â3) (%)
1 26.388 26.398 0.038
2 20.681 20.752 0.343
a
N
y1
¼N
y2
¼1:5; V ¼1Â2 Â1:5 m
3
; Dx ¼Dz ¼
1
2
m; Dy ¼
1
3
m.
0 2 4 6 8 10 12 14 16 18 20
1
2
3
4
5
6
7
8
Frequency (GHz)
SDA, Itoh and Mittra
2DMRTD
E
f
f
e
c
t
i
v
e
d
i
e
l
e
c
t
r
i
c
c
o
n
s
t
a
n
t
(
e
f
f
)
r
= 8.875
r
= 2.65
Figure 19. Effective dielectric constant e
eff
versus frequency of a
shielded microstrip line with w¼h¼1.27mm and a¼b ¼
12.7mm. (SDA, Itoh and Mittra [31].)
Table 8. Discretization Dimensions of a Shielded
Microstrip Line
a
Method Dx (mm) Dy (mm) N
x
N
y
2D MRTD 0.5 0.5 13 7
2D FDTD 0.125 0.125 52 28
3D FDTD 0.125 0.125 52 28
a
e
xx
¼e
zz
¼9:4; e
yy
¼11:6.
Table 9. Discretization Dimensions of an Open Microstrip
Line
a
Method Dx (mm) Dy (mm) N
x
N
y
2D MRTD 0.05 0.0333 21 12
2D FDTD 0.0125 0.0125 110 30
3D FDTD 0.0125 0.0125 55 30
0 5 10 15 20 25
5.5
6
6.5
7
7.5
8
8.5
9
9.5
10
0 5 10 15 20 25
35
40
45
50
55
60
65
70
M
a
g
n
i
t
u
d
e
o
f
Z
0
(
Ω
)
(b)
Frequency (GHz)
E
f
f
e
c
t
i
v
e
d
e
l
e
c
t
r
i
c
c
o
n
s
t
a
n
t
(
e
f
f
)
(a)
Frequency (GHz)
2DMRTD
2DFDTD, Tong and Chen
3DFDTD
2DMRTD
2DFDTD, Tong and Chen
3DFDTD
Figure 20. Frequency dependence of propa
gation characteristics of a shielded microstrip
line with w¼h¼1.5mm, a¼6.5mm, b ¼
3.5 mm; e
xx
¼e
zz
¼9:4; e
yy
¼11:6: (a) e
eff
;
(b) magnitude of Z
0
in ohms. (Tong and Chan
[29].)
ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD 1377
deriving the update equations, and determining the num
ber of images in the transform domain of the MRTD. The
propagation characteristics computed via the MRTD have
been shown to be in good agreement with those obtained
by using other numerical techniques. The computational
resources needed in the MRTD scheme are only a small
fraction of those of the FDTD algorithm, albeit the com
putational complexity is higher in the MRTD. Although
we have presented only a few representative examples of
the application of the MRTD, its scope is wider, as is
evident from a variety of problem to which it has been
applied.
10. CONCLUSION
In this article we have brieﬂy discussed two techniques for
enhancing the FDTD algorithm: the conformal method
and the MRTD algorithm. Space does not permit us to
include two other approaches that are also quite impor
tant: the pseudospectral timedomain (PSTD) method and
the multiscale analysis for dealing with geometries re
quiring a ﬁne discretization in a subregion of the compu
tational domain. We have simply chosen, as a compromise,
to include some selected set of references on these two
topics. The references on the multiscale analysis have
been limited to those related to the FDTD/ADI algorithm
and do not include a vast array of literature dealing with
the problem of subgridding in FDTD.
BIBLIOGRAPHY
1. A. R. Bretones, R. Mittra, and R. G. Martin, A hybrid
technique combining the method of moments in the time
domain and FDTD, IEEE Microwave Guided Wave Lett.
8(8):281–283 (Aug. 1998).
2. S. Gedney, F. Lansing, and D. Rescoe, A full wave analysis of
passive monolithic integrated circuit devices using a general
ized Yee algorithm, IEEE Trans. Microwave Theory Tech.
44:1393–1400 (1996).
3. M. Fusco, FDTD algorithm in curvilinear coordinates, IEEE
Trans. Anten. Propag. 38(1):76–89 (Jan. 1990).
4. T. G. Jurgens, A. Taﬂove, K. Umashankar, and T. G. Moore,
Finitedifference timedomain modeling of curved surfaces,
IEEE Trans. Anten. Propag. AP40(4):357–366 (April 1992).
5. S. Dey and R. Mittra, A locally conformal ﬁnite difference time
domain (FDTD) algorithm for modeling threedimensional
perfectly conducting objects, IEEE Microwave Guided Wave
Lett. 7(9):273–275 (Sept. 1997).
6. K. S. Yee and J. S. Chen, The ﬁnitedifference timedomain
(FDTD) and the ﬁnitevolume timedomain (FVTD) methods
in solving maxwell’s equations, IEEE Trans. Anten. Propag.
AP45(3):355–363 (March 1997).
7. W. Yu, R. Mittra, D. Arakaki, and D. H. Werner, A conformal
ﬁnite difference time domain (CFDTD) algorithm for model
ing perfectly conducting objects, 16th Annual Review of
Progress in Applied Computational Electromagnetics, Mon
terey, CA, 2000.
8. W. Yu and R. Mittra, Accurate modeling of planar microwave
circuit using conformal FDTD algorithm, IEE Electron. Lett.
36(4) (2000).
9. W. Yu, and R. Mittra, A conformal FDTD software package for
modeling of antennas and microstrip circuit components,
IEEE Anten. Propag. Mag. 42(5):28–39 (1Oct. 2000).
10. N. Kaneda, B. Houshmand, and T. Itoh, FDTD analysis of
dielectric resonators with curved surfaces, IEEE Trans.
Microwave Theory Tech. 45(9):1645–1649 (Sept. 1997).
11. S. Dey and R. Mittra, A conformal ﬁnitedifference time
domain technique for modeling cylindrical dielectric resona
tors, IEEE Trans. Microwave Theory Tech. 47(9):1737–1739
(Sept. 1999).
12. W. Yu and R. Mittra, A conformal ﬁnite difference time
domain technique for modeling curved dielectric surfaces,
IEEE Microwave Guided Wave Lett. 11(1):25–27 (Jan. 2001).
E
f
f
e
c
t
i
v
e
d
i
e
l
e
c
t
r
i
c
c
o
n
s
t
a
n
t
(
e
f
f
)
(a) (b)
Frequency (GHz) Frequency (GHz)
20 40 60 80 100 120 140 160 180 200
9
9.5
10
10.5
11
11.5
2DMRTD
2DFDTD, Tong and Chen
3DFDTD, Becker
Empirical, Pramamick and Bharti
2DMRTD
2DFDTD, Tong and Chen
3DFDTD, Becker
Empirical, Gupta et al
20 40 60 80 100 120 140 160 180 200
32
34
36
38
40
42
44
M
a
g
n
i
t
u
d
e
o
f
Z
0
(
Ω
)
Figure 21. Frequency dependence of propagation characteristics of an open microstrip line with
w=h¼1:5 mm; h¼0:10mm; e
r
¼13:0; (a) e
eff
; (b) magnitude of Z
0
in ohms. (Tong and Chan [29],
Becker [32], Gupta, Grag, and Chadha [34].)
1378 ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD
13. S. D. Gedney, An anisotropic PML absorbing media for FDTD
solution of ﬁelds in lossy dispersive media, Electromagnetic
16:399–415 (1996).
14. I. Wolff, Applications of ﬁnitedifference timedomain techni
que to planar microwave circuit design, in T. Itoh, ed., Time
Domain Methods for Microwave Structures, IEEE Press,
1998.
15. IE3D, Zeland Software, 39120 Argonaut Way, Suite 499,
Freemont, CA 94538.
16. X.P. Liang and K. A. Zakim, Modeling of cylindrical dielectric
resonators in rectangular waveguides and cavity, IEEE
Trans. Microwave Theory Tech. 41(12):2174–2181 (Dec.
1993).
17. W. Yu and R. Mittra, CFDTD: Conformal FiniteDifference
TimeDomain Maxwell’s Equations Solver: Software and
User’s Guide, Artech House, Norwood, MA, 2003.
18. W. Yu and R. Mittra, A conformal FDTD software package for
modeling of antennas and microstrip circuit components,
IEEE Anten. Propag. Mag. 42(5):28–39 (Oct. 2000).
19. M. Krumpholz and L. P. B. Katehi, MRTD: New time domain
schemes based on multiresolution analysis, IEEE Trans.
Microwave Theory Tech. 44:555–571 (April 1996).
20. R. Robertson, E. Tentzeris, M. Krumpholz, and L. P. B.
Katehi, MRTD analysis of dielectric cavity structures, IEEE
MTTS Digest, 1996, pp. 1861–1864.
21. E. Tentzeris, M. Krumpholz, and L. P. B. Katehi, Application
of MRTD to printed transmission lines, IEEE MTTS Digest,
1996, pp. 573–576.
22. Q. Cao and Y. Chen, MRTD analysis of a transient electro
magnetic pulse propagating through a dielectric layer, Int. J.
Electron. 86:459–474 (April 1999).
23. R. L. Roberson, E. M. Tentzeris, M. Krumpholz, and L. P. B.
Katehi, Modeling of dielectric cavity structures using multi
resolution time domain analysis, Int. J. Num. Model. Elec
tron. Networks Devices Fields 11:55–68 (April 1998).
24. Y. Chen, K. Sun, B. Berker, and R. Mittra, Unﬁeld matrix
presentation of Maxwell’s and wave equations using general
ized different differential matrix operators, IEEE Trans.
Educ. 41:61–69 (Feb. 1998).
25. G. Battle, A block spin construction of ondelettes, Commun.
Math. Phys. 67:601–615 (1987).
26. P. G. Lemarie, Ondelettes a localization exponentielle, J.
Math. Pure Appl. 67:277–236 (1988).
27. R. F. Harrington, Field Computation by Moment Methods,
Krieger, Malabar, FL, 1982.
28. S. Xiao and R. Vahldieck, An efﬁcient 2D FDTD algorithm
using real variables, IEEE Microwave Guided Wave Lett.
3:127–129 (May 1993).
29. M. Tong and Y. Chen, Analysis of propagation characteristics
and ﬁeld images for printed transmission lines printed
on anisotropic substrates using a 2DFDTD method, IEEE
Trans. Microwave Theory Tech. MTT46:1507–1510 (Oct.
1998).
30. S. D. Gedney, An anisotropic perfectly matched layered
absorbing medium for the truncation of FDTD lattices,
IEEE Trans. Anten. Propag. 44:1630–1939 (1996).
31. T. Itoh and R. Mittra, A technique for computing dispersion
characteristics of shielded microstrip line, IEEE Trans. Mi
crowave Theory Tech. MTT22:896–898 (Oct. 1974).
32. W. D. Becker, The Application of TimeDomain Electromag
netic Field Solvers to Computer Package Analysis and Design,
Ph.D. dissertation, Dept. Electrical and Computer Engineer
ing, Univ. Illinois at Urbana–Champaign, 1993.
33. P. Pramanick and P. Bharti, An accurate description of
dispersion in microstrip, Microwave J. 89–96 (Dec. 1981).
34. K. C. Gupta, R. Grag, and R. Chadha, Computeraided
Design of Microwave Circuits, Artech House, Dedham, MA,
1981.
REFERENCES ON PSTD
Q. H. Liu, A spectraldomain method with perfectly matc
hed layers for timedomain solutions of Maxwell’s equations,
paper presented at 1996 URSI Meeting, Baltimore, MD, July
1996.
Q. H. Liu, The PSTD algorithm: A timedomain method requiring
only two cells per wavelength, Microwave Opt. Technol. Lett.
15(3):158–165 (1997).
Q. H. Liu, Largescale simulations of electromagnetic and acous
tic measurements using the pseudospectral timedomain
(PSTD) algorithm, IEEE Trans. Geosci. Remote Sens.
37(2):917–926 (1999).
Q. H. Liu, PML and PSTD algorithm for arbitrary lossy aniso
tropic media, IEEE Microwave Guided Wave Lett. 9(2):48–50
(1999).
Q. H. Liu and G. X. Fan, A frequencydependent PSTD algorithm
for general dispersive media, IEEE Microwave Guided Wave
Lett. 9(2):51–53 (1999).
Q. H. Liu and G. X. Fan, Simulations of GPR in dispersive media
using the PSTD algorithm, IEEE Trans. Geosci. Remote Sens.
37(5):2317–2324 (1999).
Q. H. Liu and J. Q. He, An efﬁcient PSTD algorithmfor cylindrical
coordinates, IEEE Trans. Anten. Propag. 49(9):1349–1351
(2001).
Q. H. Liu, X. M. Millard, B. Tian, and Z. Q. Zhang, Applications of
nonuniform fast transform algorithms in numerical solutions
of differential and integral equations, IEEE Trans. Geosci.
Remote Sens. 38(4):1551–1560 (2000).
H. O. Kreiss and J. Oliger, Comparison of accurate methods for
the integration of hyperbolic equations, Tellus 24:199–215
(1972).
S. A. Orszag, Comparison of pseudospectral and spectral approx
imation, Stud. Appl. Math. 51:253–259 (1972).
D. Gottlieb and S. A. Orszag, Numerical Analysis of Spectral
Methods, SIAM, Philadelphia, 1997.
C. Canuto, M. Y. Hussani, A. Quarteroni, and T. Zang, Spectral
Methods in Fluid Dynamics, SpringerVerlag, New York, 1988.
B. Fornberg, On a Fourier method for the integration of hyperbolic
equations, SIAM J. Num. Anal. 12:509–528 (1975).
B. Fornberg, A Practical Guide to Pseudospectral Methods, Cam
bridge Univ. Press, New York, 1996.
B. Yang, D. Gottlieb, and J. S. Hesthaven, Spectral simulation of
electromagnetic wave scattering, J. Comput. Phys. 134(2):
216–230 (1997).
J. S. Hesthaven, Spectral collocation timedomain modeling of
diffractive optical elements, J. Comput. Phys. 155(2):287–306
(1999).
B. Yang and J. S. Hesthaven, A pseudospectral method for
timedomain computation of electromagnetic scattering by
bodies of revolution, IEEE Trans. Anten. Propag. 47(1):
132–141 (1999).
B. Yang and J. S. Hesthaven, Multidomain pseudospectral com
putation of Maxwell’s equations in 3D general curvilinear
coordinates, Appl. Num. Math. 33(1–4):281–289 (2000).
ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD 1379
B. Tian and Q. H. Liu, Nonuniform fast cosine transform and
Chebyshev PSTD algorithm, Progr. Electromagn. Res. 28:259–
279 (2000).
G. X. Fan, Q. H. Liu, and S. A. Hutchinson, FDTD and PSTD
simulations for plasma applications, IEEE Trans. Plasma Sci.
29(2):341–348 (2001).
G. X. Fan, Q. H. Liu, and J. S. Hesthaven, Multidomain pseudos
pectral timedomain method for simulation of scattering from
buried objects, IEEE Trans. Geosci. Remote Sens. 40(6):1366–
1373 (2002).
G. X. Fan and Q. H. Liu, Pseudospectral timedomain algorithm
applied to electromagnetic scattering from electrically
large objects, Microwave Opt. Technol. Lett. 29(2):123–125
(2001).
G. Zhao and Q. H. Liu, The 2.5D multidomain pseudospectral
timedomain algorithm, IEEE Trans. Anten. Propag.
51(3):619–627 (2003).
G. Zhao, Y. Q. Zeng, and Q. H. Liu, The 3D multidomain
pseudospectral timedomain method for wideband simula
tion, IEEE Microwave Wireless Compon. Lett. 13(5):184–186
(2003).
J. S. Hesthaven, Spectral penalty methods, Appl. Num. Math.
33(1–4):23–41 (2000).
D. Funaro and D. Gottlieb, A new method of imposing boundary
conditions in pseudospectral approximations of hyperbolic
equations, Math. Comput. 51(184):599–613 (1988).
REFERENCES ON MULTISCALE ANALYSIS
M. W. Chevalier, R. J. Luebbers, and V. P. Cable, FDTD local grid
with material traverse, IEEE Trans. Anten. Propag. 45:411–
421 (March 1997).
M. Okoniewski, E. Okoniewska, and M. A. Stuchly, Three dimen
sional subgridding algorithm for FDTD, IEEE Trans. Anten.
Propag. 45:422–429 (March 1997).
M. Feliziani and F. Maradei, Mixed ﬁnitedifference/Whitney
elements time domain (FD/WETD) method, IEEE Trans.
Magn. 34:3222–3227 (March 1998).
B. Z. Wang, Y. Wang, W. Yu, and R. Mittra, A hybrid 2D ADI
FDTD subgridding scheme for modeling onchip interconnects,
IEEE Trans. Adv. Pack. 24:528–533 (Sept. 2001).
Z. Chen and J. Zhang, An unconditionally stable 3D ADIMRTD
method free of the CFL stability condition, IEEE Microwave
Wireless Compon. Lett. 11:349–351 (Aug. 2001).
O. Podebrand, M. Clemens, and T. Weiland, New ﬂexible subgrid
ding scheme for the ﬁnite integration technique, IEEE Trans.
Magn. 39:1662–1665 (May 2003).
M. Marrone, R. Mittra, and W. Yu, A novel approach to deriving a
stable hybridized FDTD algorithm using the cell method, Proc.
IEEE APS URSI, Columbus, OH, June 22–27, 2003.
M. Marrone and R. Mittra, A theoretical study of the stability
criteria for hybridized FDTD algorithms for multiscale analy
sis, IEEE Anten. Propag. Mag. (in press).
1380 ENHANCEMENTS OF THE FINITEDIFFERENCE TIMEDOMAIN METHOD