PROCEEDINGS OF THE IEEE

Digital Computer Solutions of the Rigorous Equations

for Scattering Problems

Abstract-A survey of recently developed techniques for solving

order of one wavelength in maximum diameter. Large
the rigorous equations thatarise in scattering problems is presented.
scatterers are handled with the aid of physical optics,
These methods generate a system of linear equations fortheun-
geometric optics, and the geometrical theory of diffrac-
known current density by enforcing the boundary conditions at dis-
tion. These optical solutions provide reliable data only
crete points in the scattering body or on its surface. This approach
when the scatterer has a diameter or width which is
shows promise of leading to a systematic solution for a dielectric or
conducting body of arbitrary size and shape. large in comparison with the wavelength. Complications
The relative merits of the linear-equation solution and the varia-
arise when a portion of the surface is concave as, for
tional solutions are discussed andnumerical results, obtained by
example, with thehollow hemisphere. Furthermore, the
these two methods, are presented for straight wires of finite length.
solution for each new scattering shape requires a great
The computation effort required with the linear-equation solution
can be reduced by expanding the current distribution in deal of thought and ingenuity.
a series of
modes of the proper type, by making a change of variables for in- In the past few years, with the widespread availability
tegration, and by employing interpolation formulas. of high-speed digitalcomputers,attentionhasbeen
Solutions are readily obtained for a scattering body placed in an
given t o a distinct approach to the scatteringproblem.
incident plane-wave field or in the near-zone of a source. Examples
First, a system of linear equations is obtained by en-
are included for both cases, using a straight wire of finite length as
the scattering body. forcing the boundary conditions at many points within
The application of these techniques to scattering by a dielectric
the scatterer or on its surface. Next, with the aid of a
body is illustrated with dielectric rods of finite length. digital computer, this system of equations is solved t o
I.INTRODUCTION determine the current distribution on the surface or the
coefficients in the mode expansion forthe scatteredfield.

R IGOROUSSOLUTIONS existforplane-wave Finally, one computes the distant scattering pattern.

scattering by the perfectly conducting plane, cir- This linear-equation technique is valid for scatterers
cularcylinder [I], ellipticcylinder [2], sphere of any convex or concave shape, and the exact solution
[3], and the prolate spheroid [4]. These solutions are can be approached simply by enforcing the boundary
obtained by the method of separation of variables. The conditions a t a sufficiently large number of points. The
wave equation, given by computation time is least for small scatterers (in the
V2# +k2# = 0 Rayleigh region) but i t is reasonable even for bodies of
resonant size or larger, depending on the capacity of the
is separable only in the following eleven coordinate sys- computer.Solutionscanreadilybeobtainedforper-
tems [SI :rectangular, circular cylinder, elliptic cylinder, fectly conducting, imperfectly conducting, and dielectric
parabolic cylinder, spherical, conical, parabolic, prolate bodies. If the body is placed in the near-zone field of a
spheroidal,oblatespheroidal, ellipsoidal, andparabo- source,thesolutionproceedsinthesamestraight-
loidal. Thus, the numberof scattering problems that can forward manner as in the plane-wavecase.
be solved by this classical method is severely limited, Thus, the linear-equation solution shows promise for
A comparable solutionis not possible forthe hemisphere, accurate, systematic calculations for bodies of arbitrary
thecircularcylinder of finitelength,andothersuch material, size, and shape. This paperreviews briefly the
bodies whose surfaces do not coincide with a complete recent progress in thistechnique, discussesseveral
constant-coordinate surface in oneof the systems listed methods for reducing the computation effort, and illus-
previously. In fact,difficulty is experienced in obtaining trates the techniques by considering a wave to be in-
numericaldatawiththe classicalsolution even for cident on a straight wireor a dielectricrod of finite
spheroids [4] and large spheres. length.
Variationalandquasi-staticsolutionshaveshown
considerable success for scatterers of various shapes, but 11. SURVEY OF RECENTPROGRESS
these techniques have been limited t o bodies which are Some of the recent applicationsof the linear-equation
small in comparison with the wavelength or are on the technique follow:
Meiand Van Blade1 [ 6 ] calculated the surface-
Manuscript received March 4, 1965; revised April 30, 1965. current density and the scattering patternsof perfectly
The authoris with the Antenna Lab., Dept. of ElectricalEn-
gineering, The Ohio State University, Columbus, Ohio. conducting rectangular cylinders of infinite length. Re-

796
 Richmond: Computer Solutions of Scattering Problems 797

sultswereobtained for incidentplanewaveshaving perimeter of the conducting body is approximately 20

parallel and perpendicular polarization with respect to wavelengths.
the cylinder axis. Waterman [I31 hasalsodevelopedaprogramfor
Andreasen [7] solved for the surface currents and the solving for the current distribution and the scattered
scattered fields of perfectly conducting cylinders of ellip- field of a perfectly conducting figure of revolution. The
tic and parabolic cross sections, and arrays of parallel scattered field is expressed as an integral of the surface
cylinders. A plane wave was assumed t o be normally currentdensity.Theelectric field intensityis forced
incident on these cylindersof infinite length, and results to vanish throughout a portion of the interior of the
were published for parallel and perpendicular polariza- conductingbodytoobtainasystem of linearequa-
tion.Cylinderswithacircumferenceuptoabout 40 tions. Green’s identity is employed to decouple the pair
wavelengths could be handled with the aid of a CDC of integral equations, to reduce the numberof unknown
1604 computer. currents by a factorof 2, and, after calculating them for
Richmond [SI obtainedsolutionsforarrays of thin one polarization, to obtain by inspection the current ex-
parallel wires of infinitelengthandforconducting pansion coefficients for the orthogonal polarization.
cylinders of semicircular and I-beam cross section. The Schultz,Ruckgaber,Richter,andSchlindler [I41
incident plane wave was assumed t o propagate at any have employed the linear-equation technique to obtain
angle 8 with respect to the cylinder axis, with the inci- the scattering patternsof conducting cones with spheri-
dentmagnetic field intensityperpendiculartothe cal caps. Thefield was expanded in spherical mode func-
cylinder axis. tions of the proper type for each of the two regions of
Richmond [9] calculatedthe field distributionin- space. Relations were obtained for the field expansion
duced in a dielectric cylinder of infinite length, and the coefficients by applying the appropriate boundary con-
distant scattering patterns. The incident plane wave was ditions.Theinfiniteseries for the field wasapproxi-
assumed to be polarized parallel with the cylinder axis. mated with the finite series of spherical modes, and the
Results were published for dielectric cylindrical shells system of linear equations for thecoefficients was solved
of circular and semicircular cross section and for plane with the aid of an IBM 7090 computer.
dielectric slabsof infinite height and finite thickness and Baghdasarian and
Angelakos [15] haveobtained
width. Both lossless and low-loss dielectric cylindrical solutions for the current induced on a circular conduct-
shells were considered, and results are given for homo- ing loop by an incident plane wave, and for the scattered
geneous and inhomogeneous cases. In one case, the in- field of the loop. Excellentresultswereobtainedfor
cident field is that of a parallel line source near the di- normal and oblique incidence.
electric cylinder. Two distinct approaches have been employed to ob-
Mullin, Sandburg, and Velline [lo] calculated the bi- tain a system of linear equations for these scattering
static echo widthof conducting elliptic cylinders for in- problems. One may expand the scatteredfield in a series
cident plane waves having parallel and perpendicular of mode functions (cylindrical modes, spherical modes,
polarization. The scatteredfield external to the cylinder etc.) and obtain a systemof linear equations for the co-
was expanded in a series of outward-traveling modes. In efficients in this series. Alternatively, the surface current
its present form, this technique gives accurate results density can be expanded in a series of mode functions
only for cylinders which do not depart greatlyfrom the and a systemof linear equationsis then obtained for the
circular cross-section shape. coefficients in this series. In the latter case, the scattered
Kennaugh [ l l ] madeone of theearlieststudies of field is expressed as an integral over the surface current
digital-computersolutions of scatteringproblemsin density and one is led to an integral equation. The inte-
which the boundary conditions are enforced a t discrete gral expression for the scattered field is valid a t every
points. Successful calculations were carried out for pro- point in space, whereas a mode expansion for the scat-
late and oblate spheroids illuminated by a plane wave tered field is usually valid only in a particular region.
incident along the symmetry axis.A system of 21 linear Mode-series expansions for the scattered field have been
equations was obtained by setting the tangential electric employed by Sommerfeld [16], JIullin [ l o ] , Kennaugh
field intensity equal to zero a t evenly spaced points on [ l l ] , Schultz [14], and others. Integral expressions for
the surface. Eight coefficients were determined for the the scattered field were used by Mei and Van Blade1 [6],
scattered field expansion in spherical modes by obtain- Andreasen [7], [l2], Richmond [8],[9], Waterman
ing a least-squares solution for the 21 equations. [Is], and Baghdasarian and Angelakos [15].
Andreasen [12] has set up a computer program for
calculating the current distribution and the scattered 111. AN EXAMPLE: THE SLENDER STRAIGHT
field of a perfectly conducting body of revolution. The WIRE OF FINITE LENGTH
incidentplanewavemayhavearbitrarypolarization Consider a harmonicelectromagneticwaveinfree
and angle of incidence. The current distribution is ex- space incident on a slender, finite, perfectly conducting
pressed- asaseries of modecurrents.ThemaximumwireasillustratedinFig. 1. Thetimedependence eiw*
798 PROCEEDINGS OF T H E I E E E A ugust

is understood. The wire radius a is assumed to be much

smaller than the length L and the wavelength X. The
wire is considered to beahollow metal tube, open a t
both ends. The scattered field may be generated by the If the observation pointis on the axis of the wire, p = O ,
surface current Jon the circular cylinder p = a , radiating and the preceding expression for the scattered field re-
in free space. With good accuracy, the surface current duces to
density on the thin wire can be considered to have only
an axial component. Furthermore, the current density
(representing the sum of the currents on the inner and
outer surfaces of the thin-wall metal tube) can be as-
sumed to be distributed uniformly around the circum-
ference of the wire. T h a t is.
where
J = U(z) (1)
+
r = d a z (2’ - 2)’. (7)
where i represents a unit vector parallel to the wire axis.
Solutions can also be obtained for solid or hollow wires The interior region of the hollow wire (p <a) can be
of large radii, but the purpose here is to illustrate the regarded as a circular waveguide, and the field there can
techniques by considering a relatively simple example. be expanded in a series
of waveguide modes beyond cut-
off. In this way, it can be shown that the total field
essentially vanishes in the interior region if the distance
from the nearest end of the wire exceeds 3 or 4 radii.

P
The electric field intensity is the sum of the incident
and scattered intensities:
E = E’ + ES. (8)

1-A L2.
Thus,

1
E’(p, Z) = - E’(p, Z )
for p < a and for z I <OSL -4a. I t follows from (9) that
(9)

Fig. 1. A plane wave has obliqueincidence on a conductingwire we may set the right-hand side of ( 6 ) equal t o -Ezi(O, z).
of length L and radius a. I t will be noted that we are proposing t o force the
axial electric field intensity to vanish only on the axis of
the wire, although in actuality the vector E must vanish
The z component of the field scattered by the wire is everywhere in the interior region except in the vicinity
given by the following expression: of the open ends. If the wire is slender, it can be shown
that the field will almost vanish everywhere within the
wire if its axial component is forced to vanish on the
axis. Furthermore, the computations are simplified when
this condition is enforced on the axis, since the single
where p and z are the cylindrical coordinates of the ob- integral in ( 6 ) is easier to evaluate than the double inte-
servation point, (a, +’,2’) are the cylindrical coordinates gral in ( 2 ) .
of the source point, r is the distance between these two Even if the current distribution I ( z ) were known, the
points: integral in ( 6 ) could not be evaluated analytically, ex-
+
I = d p z a2 - 2ap cos 4‘+ (z’ - z)? (3) cept in the form of an infinite series. However, numerical
integration is possible andfeasiblewiththeaid of a
+
F(r, z , 2’) = [2r2(1 j k r )
digital computer.
In
manyproblems, the
incident field intensity
e- jkr
+ E,‘(O, z) in ( 6 ) is known. It may,for example, represent
- (3 3jkr - k2r2)(r2- (2’ - z ) ~ ) ]--+- (4) the field of an incident plane wave having oblique inci-
dence asin Fig. 1. The current distributionI ( z ) induced
+
and k = 2a/X. The angular coordinate of the observa- on the wire is then an unknown function which is to be
tion point has been taken equal to zero in(3), since the determined. Techniques for solving this integral equa-
scattered field is independent of + under the assumed tion to any desired degree of accuracy, with the assis-
conditions. Equation ( 2 ) can be derived with the aid of tance of a digital computer, are considered next.
the vector potential or by starting with the expression
for the field of an infinitesimal electric dipole and em- IV. CHANGEOF VARIABLES
ploying the superposition theorem. Integration is often facilitated by making a change of
The current density is related to the current I(z) by variables. This is helpful both in numerical integration
1965 Richmond: Computer Solutions of Scattering Problems 799

and inanalyticintegration.Althoughtheoptimum where

change of variable may be difficult to determine, one G(#) = e-jka/cos 8'
that has been found very helpful in the problem of the
slender finite wire is the following: .[(cosO' +jka)(2 - 3 cos2e') + k2a2cose']. (17)
z' - z = a atn@' (10)
The limits of integration are again given by (13) and
dz' = a sec2e' de'. (11) (14).
If (16) is enforced at N points z =z1, z2, , z . along
~ -
With this change of variables, the integral equation (6) the wire axis, a systemof N linear equations is obtained
for the slender wire becomes: which have the following form:
.v

n=l
(CmJn + CmntInt) = -~z'(0, (18)

[(jka + cos e') (2 - 3 cos2e') + k2a2cos e ' ] d8' C,, represents the scattered field a t a point (0, z,) gen-
(12) erated by the current distributionI = cos (2%- l)az/L,
= - E,"O, z). and C,,' is the scattered field a t zm generated by the
The limitsof integration are current I=sin 2nazlL. The complex mode amplitudes
Inand I,' are obtained by solving this system of linear
e: = - tan-'
0.5L +z equations. The method of Crout [18] has been found t o
a be convenient and efficient forsolving the system of
0.5L - z linear equations on a digital computer.
8; = t a r 1 TheFourier seriesfor thecurrent I(z) converges
a rapidly a t first,andthenmore slowly, as shownin
If theintegral is evaluatednumericallywiththe Table I. In calculating the data shown in Table I, the
Newton-Cotes formulas [17], the integrand is sampled integrals in (16) were evaluatedwiththefifth-order
at equally spaced points of the variable of integration. Newton-Cotes formula (which is exact for integrals of
Theintegrandvariesmostrapidlywhenthesource fifth-degree polynomials) using 1000 terms for the inte-
point is in the vicinity of the observation point. The gration along thewire.
number of terms required in numerical integration is
fixed by the requirement that the integrand be ade- TABLE I
quatelysampledinthe regionwhere itvariesmost FOCRIER
COEFFICIEKTS
FOR THE CURRENT I ( z ) FOR A N
INCIDENT PLANE LvA4VE: L=O.Sx; U=0.005X.
rapidly. If theintegration is on z' as in(6),alarge
number of termsare required toobtainadequate
sampling. If onemakesthechange of variablesde-
I . ei = 300 ei = 90"

scribedpreviously,equalintervalsaretakenin 0' -1-

(instead of in z'), providingadensesamplingin the
vicinity of z' = z where theintegrandvariesmost 1 1.4272 -35.30.1408
~~

179.1 -I- 3 ,4762 -36.1

rapidly, and tapering to a lighter sampling as one moves 2 0.0800 135.8 0.0230 -0.9 0.1700 146.9
away from this region. Thus, fewer terms are required 3 0.0373 -41.10.0101 179.1 0.0849
-34.2
4 0.0235 140.0 0.0064 -0.9 0 .OS47 145.4
in the numerical integration. 5 0.0169
-39.3 0.0045 179.1 0.0399
-34.8
6 0.0132 141.0 0.0035 -0.9 0.0312 145.1
V. MODEEXPANSION OF THE CKRRENT DISTRIBUTION 7 0.0108
-38.8
0.0029 179.1 0.0256 -35 .O
8 0.0092 141.3 0.0024 -0.9 0.0218 145.0
The current on a hollowwirewith thin walls must 9 0.0080
-38.6
0.0021 179.1 0.0190
-35.1
10 0.0072 141.5
0.0019 -0.9 0.0169 144.9
vanish at the ends.A suitable expansion for the current 11 0.0065 -38.4 0.0017 179.1 0.0152 -35.1
is the Fourierseries given by 12 0.0060 141.6 0.0016 -0.9 0.0140 144.9
13 0.0056 -38.3 0.0015 179.1 0.0130 -35.1
14 0.0053 141.7 0.0014
2nrz] .
-0.9 0.0121 144.9
7rz (15)
n=l
I, cos (2n - 1) -
L
+ I,,' sin -
L
15 0.0050 -38.3 0.0007 179.1 0.0115 -35.1

Inserting this expression in (12) we obtain: In order t o reduce the number of lin ear equations
which must be solved, it would be advantageous to ex-
Adz
-
8h2a2
[. G(0') cos (2n -
rd
l) I de
'
pand the current I ( z ) in a series of functions which con-
vergesmorerapidlythantheFourier seriesgivenin
Table I. Table I1 lists the coefficients for the current
expansion in several types ofseries. The Chebyshev and
Legendre series appear most promising,but this matter
= - E,'(O, z) (16) requires
further
investigation.
 800 PROCEEDINGS O F T H E I E E E A ugust
TABLE I1 Figure 2 shows several calculated scattering patterns
COEFFICIENTS
FOR THE CURRENT I ( z ) FOR AN INCIDENTP L A N E based on the Fourier series expansion for the current.
; = O . @ X X ; ei=900
WAVE: L = O S X ~
The results shown in this figure correspond to the cur-
"",' No' Fourier
Maclaurin
Chebyshev
Hermite Legendre
rent distribution listed in Table I. Fifteen cosine modes
and 15 sine modes were used, and 1000 terms were em-
12.2762
8.2929
1.7589
3.374
3.476 ployed in each integration along the wire.
2 0.170 i.5581
4.037 14.3644 Z.ioOs A comparison of the measured and calculated back-
3 0.085 0.0655
4.4135
0.0319
3.128
4 0.055 4.101 0.0112 0.3453 0.0421 scatter echo areaof straight wires is illustrated in Fig. 3.
5 0.040 1.871 0.0146 0.0073 0.0372 The. measured data are those published by KOUYOUM-
Table I1 gives the magnitude of In in mA for the following series: jian [19]. In the calculations,15 cosine terms were used
in the Fourierseries for the current and1000 terms were
employed in the numerical integrations. The variational
solution of Tai [20] is also shown. It may be noted that
thelinear-equationsolutionshowsbetteragreement
with the experimental measurements.
In principle, a high degree of accuracy may be ob-
tained with the variational solution by including a suf-
where ficient number of terms in the trial functionfor the cur-
x = 2z/L. rentdistribution.Inpractice,however,thecomputa-
tional effort increases rapidly as the numberof terms is
VI. THEDISTANTSCATTERED
FIELD increased. Inthescattering problemfor theslender
The distant scattered field of a slender wire of finite wire, only two terms are usually included in the trial
length has only a 8 component which is given by function for normal incidence and four termsfor oblique
incidence. On the other hand,it is practical t o include a
jw e-jkro sin e,
Ea(@,)= -
4aYo S -LIP
L'2
I(z')eikz' 01 &'
(19)
much greater numberof terms in the current expansion
with the linear-equation technique.
An interesting property of the zero-order variational
where ro is the distancefrom the origin (at the centerof solution [20] is that it yields for normal incidence an
the wire) and 8, is the scattering angle measured from echo area which is independent of the wire radius when
the axis of the wire. the wire lengthis 0.5 X, 1.5 X, etc. Experimental measure-
If thecurrent I ( z ) isrepresentedbya piecewise ments [21], however, show that the echo area is defi-
uniform function,' the distant scatteredfield is given by nitely a functionof the radius even when the length is an
oddnumber of half wavelengths.Higher-ordervaria-
tional calculations by Hu[22] show quite accurately the
dependence on the radius forwirelengths up to two
where I , is the currenton segment n,z , is the coordinate wavelengths. The linear-equation solution also shows a
of the centerof segment n, and N is the total numberof dependenceontheradius whichagrees with experi-
segments. mental data and the calculations of Hu. This depen-
If the current distribution is expressed as a Fourier dence is slight for the half-wave wire,but notfor the 3/2
series as in (15), the distant scattered field is given by wavelength case.
Van Vleck,Bloch, and Hamermesh [23], King [24],
and Dike and King [25] have obtained an approximate
solution for the thin wire by solving the integral equa-
tion to obtain the leading terms in the seriesfor the
currentdistribution.AlthoughLindroth [26] hasob-
tained useful results by this methodfor wire lengths up
sin (aL' cos e,)
+ 2jnI,'
4n2 - (2L' cos eJ2
t o 2.5 wavelengths, it does not appear to show much
promise for longerwires. The rigorous solution of HallCn
[27] has apparently not yet been exploited to obtain ac-
where L' = L/X. curate solutionsfor scattering bylong wires.
If the incident field is that of a plane wave of ampli- Solutions were readilyobtainedwiththelinear-
tude EO,the echo areaof the wire is given by equationtechniquefor wires of length L=2.865 X
(kL= 18) and radiia = 0.00415 X and 0.0105 X (ka= 0.026
and 0.066). For normal incidence, the backscatter echo
areas were foundto be1.65 and 2.63 square wavelengths,
See Baghdasarian and Angelakos 1151 for an example employing respectively,
showing excellent
agreement with the
a piecewise uniform function for the current. measurementsby As and Schmitt [21]. For a length
1965 Richmond: Computer Solutions
Problems
of Scattering 801

This provides a useful check on the accuracyof numeri-

cal scatteringcalculations,except in caseswhere the
reciprocity theorem has been incorporated into the solu-
tion.Furthermore,thereciprocitytheoremcanbe
utilized t o speed up thecalculations. Of course, the reci-
procity theorem is automatically satisfied by any rigor-
ous solution.
In the linear-equation technique, the currents on the
wire satisfy a systemof equations of the following type:
N
CmnIn = - E,'(&, Zm) (24)
n=l
ScatteringAngle os (Degrees)
Fig. 2. Calculated bistatic echo area of strai ht wire vs. scattering
where Ezi(0i,z,) represents the tangential electric field
angle for incident plane wave with angle of incidence of 30°,60°, intensity of the incident plane wave. TheI , may repre-
and 90". sent the mode-current amplitudes or the piecewise uni-
form currents induced on the wire. From (24) it is obvi-
ous that the currents will be a function of the angle of
incidence B i . The distant scattered field in any angular
direction 0, is given by (19)-(21), which can be written
in the following general form:
N
- w e i , 0,) = C In(eJsn(e8>. (25)
n=l

The scattering functions S,(e,) aregiven by (20) and

(21) when the piecewise uniform or the Fourier series
representation is employed for the current.
From (23) and (25), it is required that
N N
In(SJSn(eJ= In(e,)Sn(e;>. (26)
n=l n-1

- Experimental Data From

I t does not necessarily follow that In(eJ =A.S,,(Bi), al-
o.il+- - -- Variational Solution By T a i though this is the most obvious solution of (26). I t is
8 8 8 Linear-Equation
Solution possible to determine by means of (26) the currents I ,
o p - 1 I I I I I
forallangles of incidenceonce they have been calcu-
0.30 0.35 0.40
0.45
0.50 0.55 0.60 0.65 0.70
WireLength L/A
lated for one angleof incidence. However, this approach
will not be pursued here since the technique described
Fig. 3. Measured and calculated backscatter echo area of straight in Section VI11 accomplishes the same end result with
wire for normal incidence.
equal efficiency.
I t is not difficult to show that thereciprocity theorem
L = 3.82 X (kL= 24) and radius a = 0.0035 X the calcu- and (26) are satisfied precisely by the solution using the
lated result is 2.8 square wavelengths and the measured piecewise uniform representation, even when the solu-
result by Sevick [21] is 2.01 square wavelengths. The tion is inaccurate as a result of dividing the wire into a
difference is believed t o arise from phase errors along the small number of segments. Thus the reciprocity theo-
wire in the experimental measurements. rem does not in this case provide a useful check on the
accuracy. On the other hand, the reciprocity theoremis
VII. THERECIPROCITY THEOREM not preciselysatisfied bythesolutioninSection V,
which employs a Fourierseries expansionfor the current.
When a plane wave is incident on a straight wire, the For the case illustrated in Table I and Fig. 2, the reci-
distant scattered field is a function of the angle of inci- procity theorem is satisfied t o a high degree of accuracy,
dence Qiand the angle of scattering 8,. The reciprocity but it is not satisfied so accurately when the same prob-
theorem assures us that the phase and strength of the lem is solved with an inadequate numberof terms in the
scattered field is unchanged when the angle of incidence Fourier series and in the numerical integration. This is
and the angle of scattering are interchanged. Thus, illustrated in Table111.
802 PROCEEDINGS O F THE I E E E A ugust

TABLE I11 When a large systemof linear equations is involved, the

ECHOAREA OF STRAIGHT
WIRE WITH PLANE-WAVE IXCIDEXT efficiency is greatly increased by repeating only the cal-
L = 0.5x a = 0.005x
culations like those in ( 3 3 ) through ( 3 6 ) for each new
Number of Terms Number of Terms
Calculated
Echo incident field or angle of incidence, instead of treating
in Integration in Fourier Series " Area/X* each one like an entirelynew problem.
300 4 cos and 4 sin 30" 60" 0.07390 This methodof Crout is easily programmed for a digi-
300 4 cos and 4 sin 60 30 0.07428 tal computer. Even for large systems of linear equations,
loo0 15 cos and 15 sin 30 60 0.06763 it has been found to beefficient with respectto computa-
1000 15 cos and 15 sin 60 30 0.06762 tion time and memory storage requirements. (As each
element in the auxiliary matrixis calculated, it is stored
in the location previously occupied by the corresponding
VIII. REPEATEDSOLUTIONS FOR DIFFERENT
element of the original matrix.)
ANGLESOF INCIDENCE The incident field E,i(O, z) enters into the problem
Most of the computation effort in the linear-equation only in the right-hand sideof the linear equations. Thus,
technique is concernedwith the integrations over the it is straightforward to solve for the currents induced on
surface of the conducting body and the solution of the a wire by any type of incident wave. Section I X de-
system of linear equations. Once a solution has been scribes the results obtained when the incident field is
obtained for the currents on a given wire with a given generated by a short dipole parallel with a slender wire,
incident field, very little additional effort is required to as a function of the distance between the wire and the
obtain the solution for a new incident field or a new dipole.
angle of incidence. Theintegrations need not bere-
peated, since they depend only on the wire length and Ix. SCATTERING BY WIRE I N NE.4R-ZONE FIELD
radiusandthefrequency of theincidentwave. The Figure 4 shows the current distribution induced on a
matrix inversion for the linear equations need not be wire when the source of the incident field is an electric
repeated,sincethe coefficients dependonlyonthese dipole parallel with the wire. If the dipole is a t a dis-
same factors. Thus, itwould be inefficient to repeat the tance d from the wire, as shown in Fig.4, the tangential
entire calculation for each new angle of incidence. component of the incident field is given by
This can be illustrated with a simple example. Con-
Ke-jkT
siderthesolution
equations:
of the followingsystem of linear
+
Egl(O,Z) = -[(l + j k r ) ( 2 z 2 - d 2 ) k2r2d2] ( 3 7 )
r5
Cdl + C d z = El 7) (2 where
C2111 + C2212 = E2. (28) r = dzz + d2. ( 38)
In solving such a system by the method of Crout [18], For the solutions shown in Fig.4, the dipole strength
one first calculates an "auxiliary" matrix C' whose ele- was adjusted to maintain the incident field E,' a t unit
ments aredefined as follows: strength at the center of the wire as the distance d was
Clll = c11 ( 2 9 ) varied. It may be noted in Fig. 4 that the current distri-
bution approaches that for an incident plane wave as
C12' = C12/Cll (30) the distance to the dipole increases. I t is interesting that
Czl' = C2l ( 3 1) the current distribution in the plane-wave case differs
c22f = c22 - C2llC12'. ( 3 2 ) noticeably from the cosine current generally assumed
for the half-wave wire in the variational solution.
The preceding calculations for the auxiliary matrix need I t is believed that considerableeffort would bere-
be carried out only once, since the elements of the ma- quired to set up a variational solution for the straight
trices C and C' depend only on the scattering structure. wire near a dipole source. In particular, a trial function
The elements E1 a n d E2 in ( 2 7 ) and ( 2 8 ) represent the would haveto befoundforthecurrentdistribution
incident field intensity a t certain points along the axis which would permit a reasonably accurate representa-
of the wire. For each new incident field (or each new tion of the true current function and at the same time
angle of incidencein theplane-wavecase),onefirst would contain not more than about four unknown con-
calculates stants. On the other hand, a minimum of effort is re-
El' = EI/CI{ (33) quired with the linear-equation formulation for each new
type of incident field considered.
Ez' = (E2 - Cz{E?)/C22'. (34)
X. SCATTERING BY A FINITE DIELECTRIC ROD
The solution for the system of linear equations is then
given by The linear-equation technique applies equally well in
calculating the scattered fields of a dielectric body. If
1 2 = E; (35) the permeability of the body is the same as thatof free
11 = El' - Ez'C14. (36) space, the scatteredfield may be generated by an equiv-
1965 Richmond: Computer Solutions of Scattering
Problems 803

0.016
I I I
Frequency: 9.5 Gc /
Radius Of DielectricRod! 0.0625" /
0.014 - Rod Material: Polystyrene
/
/
DielectricConstant: 2.54
-
--- Linear-Equation Solution
Physical Optics
/
/

o'o'2- Approximation //
Experimental
Results

WireLength: t = 0.5X
WireRadius : a = 0.005X

0
Rod Lenqth Ilnches)
0
0 0.05 0.10 0.15 0.20 0.25 Fig. 5. Calculated and measured echo area of dielectric rod as a
-
2
X
function of length.

Fig. 4. Currentdistributioninduced on straightwire by parallel

electric dipole. Let it beassumed that a plane wave is normally inci-
dent on the dielectric rod with the incident electricfield
alent electric current density J radiating in free space, vector parallel with the axisof the rod. In this case, the
where total field EE(plz ) must be an even function of z. With
the aid of Maxwell's equations, it can be shown that
J =~ W ( E - eo)E. (39) the total field in the rod can be expanded in a mode
series as follows:
E represents the total electric field intensity in the di-
electricbody,given bythesum of theincidentand
scattered field intensities as follows:
E = Ei + Ea. (40) where Jo(x) represents the Bessel function of zero order
and L' = L/X. In a rigorous solution, an infinite number
The techniques may be illustrated by considering the
of modes would be employed in (44),but here we shall
problem of scattering by a dielectric rod of finite length.
include only a finite number N . The summation in (44)
From a study of the rigorous solution for a plane-wave
is substituted for E,(O, z ) and E,(p, z') in (41).
incident on a dielectric rod of infinite length, it may be
The integral equation (41) is enforced a t N points
deduced that the fields in a slender rod are nearly inde-
z = zm on the axis of the rod (between z = 0 and z = L / 2 )
pendent of the angle 4. Onthe axis of the rod the
to obtain N linear equations for the mode amplitudes
tangential scatteredfield is given, with the aidof (39) by
En.The integrals in (41)may be evaluated numerically
with the Newton-Cotes formulas, but it must berecog-
EZa(O.z) = 0.5(Er - 1) l u E z ( p ,z')F(r, P ) dpdz' nized that a singularity occurs a t z' = z and p = 0 where
-LIZ
r goes through zero. The difficulties involved in handling
= E,(O, 2) - E,"O, 2 ) (41) this integrable singularity may be avoided by changing
the limits of integrationto 6<p<a where 6 is much
where a denotes the radius of the rod, L is the length, smaller than the radius a. The solution then will cor-
eI is the relative permittivity, respond to that for a dielectric rod having a tiny hole
e- jkr of radius 6 drilled along its axis. If 6 is much smaller
F ( r , p ) = p[(l + j k Y ) ( 2 Y 2 - 3 p 2 )+ k2p*r2] - (42) than a, the volume of dielectric material thus removed
Y5 is small compared with the total volume and the error
and fromthissourceshouldbe negligible. The change of
variables dz' = p sec2 8'8' is found t o be helpful.
r = z/p2 + (2' - 2 )
'
. (43) Figure 5 shows the echo area calculated in this man-
804 PROCEEDINGS O F THE IEEE
nerfor polystyrene rods of +-inch diameter a t afre-
quency of 9.5 Gc. It may be noted that the calculations Laboratory, whoseassistance
show excellent agreement with the experimental mea-
surements which are also presented in Fig. 5 . The double
integration (on p and e’) in (41) was performed with the
fifth-order Sewton-Cotes formula.
Comparatively poorresultsareobtainedfromthe
physical-optics solution for this problem, shown by the
dashed curve in Fig. 5 . In the physical-optics solution,
the field in the rod is taken to be the same as that in a
rod of infinite length. The scattered fieldis then con-
sidered t o be the field generated by the equivalent cur-
rent (39) radiating in free space. This currentis assumed
t o exist only in the region occupied by the finite dielec-
tric rod. Evidently, the field in the finite rod differs sig-
nificantly from that in the infinite rod.
XI. CONCLUSIONS
In recent years, we have begun to exploit the high-
speed digital computer to obtain accurate solutions for
the fields scattered by bodies of complexshapes.Ap-
proaches are now feasible which overcome the limita-
tions inherent in the method of separation of variables,
thevariationalsolutions,physicaloptics,geometrical
optics, and the geometrical theory of diffraction.
Inthemethodconsideredhere,asystem of linear
equations is obtained by enforcing the boundary condi-
tions a t a finite number of points on the surface or in
the interior of the scattering body. The solution of this
set of equations yields the current distribution induced
on the conducting surface.
This paper surveys the recent progress in this area
and discusses several techniques for increasing the effi-
ciency of the calculations and extending the maximum
dimensions of the scatteringbodies that can be handled.
Among these techniques are the change of variables for
integration,expansion of thecurrent in aseries
modes, and interpolation.
Thesetechniquesareillustratedbyconsidering
comparatively simple example: the slender wire of finite
length. Numerical results are shown for the current dis-
tribution induced on such wires and for the backscatter
and bistatic echo area. Calculations are illustrated
a half-wave wire near an electric dipole which acts as
the source of the field.
This linear-equation technique is alsoapplicable t o
dielectricbodies. This is illustratedbyconsidering
plane-wave incident on a dielectric rod of finite length.
The resultsshow excellent agreement with experimental
measurements.
ACKNOWLEDGMENT
The author wishes t o express his sincere appreciation
for the extensive digital computations pravided by The
Ohio State University Computer Center in support of
this investigation. The experimental measurements of
backscatter from finite dielectric rods were carried out
by S. A. Redick of The Ohio State University Antenna
is gratefullyacknowl-
edged .
REFERENCES
R. F. Harrington, Time-Harmonic Electromagnetic Fields. New
York: McGraw-Hill, 1961, pp. 232-236.
G. Sinclair, “The patternsof antennas located near cylinders of
elliptical cross section,” Proc. I R E , vol. 39, pp. 660-668, June
1951.
____.
J. A . Stratton, ElectromagneticTheory. New York:McGraw-
Hill, 1941, pp. 563-572.
K. M . Sieeel. F. V. Schultz. B. H. Gere. and F. B. Sleator. “The
theoreticai and numerical determinatio’n of the radar cross sec-
tion of a prolate spheroid,” IRE Trans. on Antennas and Propa-
gation, vol. AP-4, pp. 266-275, July 1956.
P. M. Morse and H. Feshbach, Methods of Theoretical Physics.
New York: McGraw-Hill, 1953, pt. I, pp. 655664.
K. K. Mei and J. G. Van Bladel, “Scattering by perfectly-con-
ducting rectangular cylinders,” IEEE Trans. on Antennas and
Propagation, vol. .A!-11, pp. 185-192, March 1963.
M.G.Andreasen,Scatteringfrom parallel metalliccylinders
with arbitrary cross sections,” IEEE Trans. on Antennas and
Propagation, vol. bP-12, pp. 746-754, November 1964.
J. H. Richmond, Scattering by an arbitrary array of parallel
wires,” IEEE Trans. on Micrmzw Theory and Techniques,vol.
MTT-13, pp. 408312, July 1965.
-, “Scatteringby a dlelectric cylinder of arbitrary cross
section shape,“ IEEE Trans. on Antennas and Propagation,
vol. AP-13, pp. 334-341, May 1965.
C.R.Mullin,R.Sandburg,and C. 0. Velline, “A numerical
technique for the determination of scattering cross sections of
infinite cylinders of arbitrary geometrical cross section,” IEEE
Trans. on Antennas and Propagation, vol. AP-13, pp. 141-149,
January 1965.
E. M. Kennaugh, “Multipole field expansions and their use in
approximate solutions of electromagnetic scattering problems,”
Ph.D. dissertation, Dept. of Elec. Engrg., The Ohio State Uni-
versity, Columbus, December 1959.
hl. G. Andreasen. “Scatterine from bodies of revolution bv an
exactmethod,” I E E E T r a 6 . on Antennas a k Propugahon,
vol. AP-13, pp. 30$310, March 1965.
P. C. Lt’aterman, Exactradar crosssection of axisymmetric
conducting targets,” this issue, page 805.
F. V. Schultz, G. M. Ruckgaber,S. Richter, and J. K. Schindler,
“The theoretical andnumerical determination of the radar cross
section of a finite cone,” School of Elec. Engrg., Purdue Uni-
versity, Lafayette, Ind., Rept. TR-EE64-14, Doc. AD 606 828,
August 1964.
A. Baghdasarian and D. J. Angelakos, “Solution of circular loop
antennasandscatteringfromconducting loops bynumerical
methods.n this issue. Daee 818.
of [16] A. Sommerfeld, Part&l‘Diferential Equations in Physics. New
York: Academic Press, 1949, pp. 29-31; 159-164.
[17] H.Margenau and G. M. Murphy, The Mathematics of Physics
a and Chemistry. S e w York: Van Nostrand, 1956, pp. 476-479.
[18] P. D.Crout, “A short method for evaluating determinants and
solving systems of linearequationswithreal or complex co-
efficients,” A I E E Trans. (Supplement),vol. 60, December 1941,
pp. 1235-1241.
[19] R. G. Kouyoumjian, “The calculation of the echo area; of per-
for fectly conductingobjectsbythevariationalmethod,Ph.D.
dissertation, Dept. of Elec. Engrg., The Ohio State University,
Columbus, December 1953.
[20] C: T.=Tai, “Electromagneticback-scatteringfromcylindrical
wlres, J . Appl. Phys., vol. 23, pp. 909-916, August 1952.
[21] R. W. P. King and T. T. W u , The Scattering and Diffraction of
a Waves. Cambridge, Mass.: Harvard University Press, 1959, pp.
153 and 161.
[22] Y. Y. Hu, “Back-scattering crosssection of a center-loaded
cylindrical antenna,“ IRE Trans. on Antennas and Propagation,
vol AP-6, pp. 140-148, January 1958.
[23] J. H. Van Vleck, F. Bloch, andM.Hamermesh,“Theory of
radar reflection from wires or thin metallic strips,”- . J . Agpl.
_.
Phys., vol. 18, p. 274, 1947.
[24] R. LV. P. King, Theory of Lineur Antennas. Cambridge, Mass.:
Harvard Universitv Press. 1956. ch. 4.
[25] S. H. Dike and D . b . King, “Absorption gain and back-scatter-
ing crosssection of the cylindrical antenna,” Proc. I R E , vol.
41, pp. 926-93:, July 1953.
[26] K. Lindroth,: Reflection of electromaneticwavesfromthin
metal strips, Trans. Roy. Inst. Technol. ?Stockholm),no. 91,1955.
[27] E. Hallen,“Exactsolution of theantennaequation,” Trans.
Roy. Inst. Technol. (Stockholm), no. 183,1961.