IEEE TRAXSACTIOSB ON A S T E N S A S AND PROPAGATIOK, VOL. AP-19, NO.

5, SEPTEMBER 1971

Computation of Characteristic Modes

for Conducting Bodies

Absfract-A procedurefor computing the characteristic modes where the I j are coefficients to be determined. Subst,it,ut.ing
for conducting bodies of arbitrary shape is developed. The method (2) into [ 2 , eq. ( 1 3 ) ] and using the linearity of the oper-
is applied to conducting bodies of revolution and to wire objects,
ators, we obtain
andgeneral computerprograms are discussed. Illustrative ex-
amples of the computation of characteristiccurrentsand char- Wj
IjX = X, I j R Wj. (3)
acteristic fields are given for a cone-sphere, adisk, and a wire i j
arrow. Modal solutions using these modes are computed for repre-
To obtainsymmetricmatrices, we use the same Wi as
sentative antenna and scatteringproblems to illustrate convergence
of the solution as the number of modes is increased. testing functions. Taking the symmetric product. of (3)
with ea.ch Wi, we have
I. INTRODUCTIOK Ij(Wi,XJVj) = X, Ij(Wi,RWj}, i = 1,2,*. a . (4)
i j

A THEORY of characteristic modes forconducting

bodies has been developed by Garbace [ l ] and by
an alternative a.pproach by Harrington and 3Iautz [SI.
This can be written as t.he matrix eigenvalue equa.tion
[ X I C I ] , = X,[R][Zln (5)
A rather involvedprocedureforcalculating t.he modes where [I]" is the column matrix of the li, and
for wire objects was devised by Turpin [ 3 ] , but because
of computational difficulties he wa.s able to obta.in only [R] = [(Wi,RWj)] (6)
four or five modes on a wire. Also, he gave no indication [X] = [( W i , X W
( 7j)> ] .
of how to extend his method to solid bodies. In this paper
we give astraightforwardprocedureforcomputing the Equation ( 5 ) is a rea.1 symmetric weighted matrix eigen-
modes on conducting bodies of arbitrary shape. Thenum- value equation. I t s eigenva.lues X, approximate those of
ber of modes obtained is limited only by the speed and the operatorequation,and it.s eigenvectors [ l l n define
storage capabilities of the computer, not by inherent com- functions according to ( 2 ) which appr0ximat.e t,he eigen-
puta.tiona1 difficulties. We have computer programs which funct.ions of the operator equation.
implement this methodfor conducting bodies of revolution The correspondingmatrixapproximation to the com-
a.nd for wire objects. ?Ye also have progra.ms which use plex eigenva.lue [2, eq. ( 11)3 is
these modes for the analysis of radiation and scattering
[z][1]n = (1 + j L ) [ R l C l l n (8)
problems. A number of computations for bodies of revolu-
tion and wire objects aregiven to illustrate the modes and +
where [ Z ] = [ R j X ] . The matrix [ Z ] is known as the
their use inpracticalproblems. The general theoryis generalized impedance of the body [ 4 ] , and hasbeen
summarized in [ 2 ] . We use [ a ] as a reference for the basic evalua.ted in a number of cases [SI, [SI. If the W; are
formulas needed here. differentiable,aconvenientformula for the impedance
elements is [Z]
11. REDUCTION
TO M A ~ I EQUATIOXS
X

The reduction of opera.torequations to mat.rix equa-

t,ions can be effected in the usua.1 way by the method of
moments [ 4 ] . For the space of square-integrable vector (9)
functions A and B on S we use the symmet.ric product
where t.he primes den0t.e functions of the primed coordi-
(A,B)= # A - B ds. (1) nates, and
S

Because t,he mode currents are real, we use a set, of real exp ( - j k I r - r' 1)
+2 =
functions Wi as expansion funct.ions for Jn, i.e., 4~ I r - r'I
The Hermitian parts of [ Z ] are its real part [ R ] and its
imaginary part [ X ] , obtained in the convent.iona1 way.
I n particular, t,he e1ement.s Rij are given by (9) wit,h $2
Manuscript received October 13, 1970; revised March 1, 1971. replaced by
This work was supported in part by Air Force Cambridge Research
Laboratoriesunder Contract F19628-68-C-0180 and in part by
the XSF under Grant GK-4227.
sin k I r - r' I
(11)
The authors are with the Department. of Electrical Engineering, " = j4T I r - r' I
Syracuse University, Syracuse, X. Y. 13210.
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.
630 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, SEPTEMBER 1971

and theelements Xij are given by (9) with 9~ replaced by

zl

Numerical evaluation of these elements is considered in

~ 5 ~ 7 1 .
The matrix equivalents of t.he orthogonality relation-
shipsfor t.he characteristiccurrents, [a, eq. (IS)], are
also of interest. For example, that for R is

{Jm,RJn)= (XIi,mFVi,R
i j
li,nWj)

= Ii,mIj,n{JVi>RWj) Fig. 1 . Coordinate system for bodies of revolution.

i,i

OF MODES
111. EVALUATION TEN 2.673 2.682
TEE 21.60 21 .a2
TMal - 1.082 - 1.096
We next discuss solution of the matrix eigenvalue equa- TM03
-290.7 -284.4
t.ion TE1:
21.66 21.60
TM 12 -11 .oo -11.30
TE:: 21.60 21 .56
[XlCIl = XRlCIl (15) TXII~? - 11 .oo -11.26
xvhich is (5) with the subscript 72. dropped for brevity.
Theconventionalmethod for reducing (15) to asym-
metric unweighted eigenvalue equa.tion requires [ R ] to be -11 = Thediagonalmatrix [p] is thenpartitioned as
positive definite [ S I . I n our problem [ R ] is positive semi-
definite in theory, but. because of numerical inaccuracies
it -is actually indefinite, with some small negative eigen-
values. We therefore modify the conventional method as
follo~~-s. n-here [p11] contains all pi not considered zero. We also
Let [ C ] be t,he orthogord matrix n-hich diagonalizes partitiontheothermatricesin (17) conformablywith
R according to
[PI, i.e.:

L... ... . .. ...A

where the p iare theeigenva.luesof R ordered pl 2 pB 2 - a .

Premultiplying (15) by [01, and using (16), we obtain

[ B X 0 ] [ 0 I ] = X[p][BI]. (17)
Only the larger pi can be considered accurate, and we set
all pi < Mpu, equa.1to zero, where M is some small number
set by our Authorized
estinmt.ed accuracy
licensed ofto:[R].
use limited (We
Universiti usually
Malaysia Perlis.take
Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.
AARBINGTON A N D MAUTZ: CONPETATIOK OF CHARACTERISTIC MODES 631

-2 --
(a) AI = - 0.0171 (b) A2 = 0.524

0
2 I I J t 2 1

-2 t ( c ) A3 - 1.27
-21 (d) A4 = 2.08

2 1

-2 t ( e ) X 5 = 5.62

Fig. 2. Characteristic currents for cone-sphere, length 1.36 wavelengths, sphere diameter 0.4 wavelengths.
Six lowest order rot,ationally s p > m e t r i cmodes are shown.

where the definition of [B] and [y] is obvious by com-

paring ( 2 6 ) with (25). The eigenvalues of ( 2 6 ) are the
smaller eigenvalues of ouroriginal (l5), and the eigen-
vectors of ( 2 6 ) give the correspondingeigenvectors of
(15) according to

n-here [SI is the identit)- ma,trix.

To diagonalize the realsymmetricmatrices [R] and
[ B ] we used the Jacobi met.hod [SI, available in t.he IBM
System/SGO scientific subroutine package (EIGEN). We
have also used the Givens-Householder met.hod [SI, but.
the accuracyof the dominant. X, as checked bytheRayleigh
quotient, was not as good as for the Jacobi method. Ot,her
methodstried were solution of [R-lX][I] = X[I] and
(l/X)[I] = [X-lR][I] by the QR doublest,ep met.hod
[lo], but again the accuracy ~ ‘ 5 1 snot as good as for tohe
Jacobi method. Xlso, as a numerical check, we used t.he
GramSchmidt procedure to calculate a set of ort.hogona1
eigenvectors corresponding to pi = 0. The complete set
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.
632 I E E E TRANSACTIONS OX ANTENNAS AND PROPAGA4T.IOR, SEPTEXBER 1971

of eigenvectors [ I ] , , was orthogonal to within the accuracy

of computation.
Finally, the characteristic currents Jn are given by ( 2 ) ,
where I i are the components of [I],,, and the character-
istic fields are given by
E, = -LJ, = - IiLWi. (2s)
i
(a) a1 = - 0.0171
HoJTever, rather than eva1uat.e LWi numerically, we use
reciprocity as discussed in [ a ] . For this, we place the unit. e=o
current element, I1 = urnat thepoint of mea.surement., and
compute the nzth component of E , as the following equa-
tion

E,-u,,, = Qp Em*J, d~
S

= Ii # Ern-Wi ds. (29) (c) Xg = - 1.27

i s

The final integrals are measurement, coefficients with re-

spect to the W i . If the radiation pattern is desired, we
place urnon the sphere at infinity and compute the mth
component of the radiation field as

-.iw
En-urn= -exp ( -jkrm) IiR; (30)
4srm i

n-here R ; are the plane-wave measurement coefficients

(e) .A5 = 5.62 (f) a6 = - 16.9
Fig. 3. Characteristicgainpatterns for cone-sphere,length 1.30
R; = Qp W i - u , exp ( -jkrn*r) ds. (31j diameter 0.4 wavelengths. Six lowest order
m--avelengths, sphere
S shown. are modes symmetric
rotationally

The notat.ion of ( 3 0 ) and ( 3 1 ) is discussed in [2].

element (9j is zero. The sets ( 3 2 ) and ( 3 3 ) can therefore
TO BODIES
IV. APPLICATION OF REVOLUTION
be ireated independently. The impedance matrix for thc
A general computer program for calculating the char- set (32') has the block diagonal form
acteristic modes onconducting bodies of revolution of -.

1
CZPl COl *
arbitrary shape has been developed [ l l ] . An outline of
this pr0gra.m and some illustrativeexamples of its use COl CZll COl * ' .
a.re given here. Dl =
Fig. 1 defines the coordinate system used for bodies of
revolution. The body S is generated by rotating the con-
E01 LO1 CZJ..-
tour C about, t.he z axis. The coordinates on S are t (length ... . . . . . . .. .
variable along C) and 4 (angle of rotation from the x
axis). The spherical coordinates of a field point, are r,e,+. where, for H = 1.2,---,
The current J o n S has two componentsJ t and J+. Letting
u t and ug denote unit vectors in the t and 4 directions,
we can choose two sets of real expansion functions CZJ = [ [ZJlftl
[Z,'+l

[zn"] [Zn"]
] . (33

) 71.4,u + f i ( t )sin?@)
{ u r f i ( t ) , u t f i ( tcos (32)
The submatrices on the right-hand side of ( 3 5 ) are com-
a,nd puted using the expansion and testing functions in Table I.
The orthogonality between sets ( 3 2 ) and ( 3 3 ) , as well
f U g f i ( t ) , u t f i ( t ) sin ?14,- u + f i ( t ) COS ?zO} (33)
as the bIock diagonal form of (34), is a consequence of
where i and 71. are positive integers. These two sets are [ 3 , eq. ( 3 3 ) - ( 3 7 ) ] . This is equivalent. to the statement
sufficiently general to represent a n a.rbitrar; J on S if tha.t a current of the form J = u J t cos 114 uOJ+sin 119 +
thefi(t) form a complete set in the t domain. If the testing produces a tangential electric field on S of the same form.
+
function W i is from the set. ( 3 2 ) , and the expansion func- E,,, = utEt cos 71.9 ueE+sin 71.4, and that a current of
tion Wj is from the set( 3 3 ) , then the resulting impedance the form J = u J t sin 714 - u J + cos 114 produces a tan-
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.
HARRINGTON AliD MAUTZ: COMPUTATION OF CHARACTERISTIC MODES

-2 -2 --

2 2--

"t (c) A 3 = - 3.74

I (d) X4 = 5.05

2 2

o re a
e

"tI J/
-2

I C

( e ) .A5 = - 13.1
-2 t ( f ) .A6 = 17.9

Fig. 4. Characteristic currents for conesphere, lengt,h 1.36 wavelengths,spherediameter 0.4 xavelengt.hs.
Six lowest. order sin Q, cos Q modes are shown.

gentialelectric field on S of the sameform, E,,, = The impedance matrix for t.he set ( 3 3 ) has t.he same
utEt sin 124 - u+E, cos nd. form as (34) except that [ZO"] is replaced by [Z@].
The impedance elements can now be evaluated in terms The table of expansion and test>ing functions must now
of those used inearlier work [5]. Lettingthecareted be chpnged by repla.cing all cos 11.9fact,ors by sin ?IC$ fac-
matrices denote those of [5, eq. ( 3 9 ) ] ,we obtain tors, and all sin n.4 factors by -cos 77.9factors. The rela-
tionship of these new impedance matrices to those of [ 5 ]
[Zoq = [Bofq (36)
is now given by

[ZO"+] = [Bo+@] (39)

and by ( 3 7 ) for -72 > 0.

This result is obtained by expressing sin n4 and cos n4 in
terms of exponentials. For example, the elements of [Z,t+]
For computation, we specify 2N + 1 nearly equidisbant
points t O , t l , - - - , t ? ~along C, with to at. the beginning and
are given by t 2 at,~ theend of C . Thefi(t) a.re taken as trianglefunct.ions
iZat+)ij= ( u t f i ( t )cos ~ 4Zu,fj(t)
, sin n 4 ) divided by the radius, each extending over four intervals
( t j - tj-l). For expansion, the fi(t) are approxima.t,ed b y
four pulses, and fortestingtheyareapproximatedby
fourimpulses. The details are given in [SI. Because of
1 . . theseapproximations, the impedancematrices a.re not
= - (2,V) i j . (38) exactly symmetric, as theywould be if evaluat.ed exactly.
?j
We eliminated t.his asymmetry by averaging correspond-
dimi1a.r derivations yield the other submatrices of ( 3 7 ) . ing off diagonal elements of[Z]. A number of computa-
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.
. \

634 I E E E TRANSACTIONS OX ANTEXKAS AND PROPAGATION,SEPTEMBER 1971

tional checks shom-ed that this averaging had no noticeable

affect in radiation and scatteringproblems.
The computerprogramforbodies of revolutionwas
first run for sphericalconductors, and the results com-
pared t.o t,he exact modal solut.ion. Table I1 shows a com-
parison of t>heexact X (computed fromsphericalmode
theory) to the approximate X (computed fromour general
program) for a few of the modes. The computations a.re t
for a sphere of radius 0.2 wavelength using 9 expansion (a) hl = - 1.55

functions for J t and 9 for J 6 (the [ Z ] is 18 X 18). These

numbers are smaller than usually used. Sources of error
in the computations are 1) the circular contour is approxi-
mated
. .
by 20 straight. line segments, 2) thecurrentis
approximated by triangle functions divided byra.dius, and
3) approximations are introduced into the evaluation of
[ Z ] . The a.ccuracy of t.he approxima.te X decreases as the
exact I X I gets larger, but, such modes are less important
inradiationandscattering problems. The accuracy of (c) xg
t-
= 3.74
t
(dl X4 = 5.05

computation increases as the numberof contour se'gments

is increased. Plots of some of the approximate mode cur-
rents, compared to the exact mode currents, are given for
the sphere in [7].
The generalprogram was next run foranumber of
bodies of revo1ut.ion with contours of different shapes. For
representative computations, consider a cone-sphere body,
formed by a cone of 10" half-angle smoothly joined to a
sphere of diameter 0.4 wavelength(approximatetotal
(e) X5 = - 13.1 (f) X6 = 17.9
length 1.36 wavelengths). The contour is approximated
b y 40 straight-line segments of equal length, and 19 ex- Fig. 5. Characterist.ic gain pat.terns for a cone-sphere, length
1.36 wavelengths, sphere diameter 0.4 wavelengt.hs. Six lowest
pansionfunctionsare used for J t and Jd (the [ Z ] is order sin +, cos @ modes are shorn-n.
38 X 38). Fig. 2 shows the six lowest. order characteristic
currents for the rotationally symmetric modes ( n = 0 ) ,
plotted versus the contour length variable, starting at the of scattering due to a plane-n.ave xsially incident on the
tip and ending on t.he sphere. These modes are the ones body. Sou- each mode has boTh a J , (plotted by x) and
which are usedin the solution of radiationfromrota- a Jd (plotted by squares), 15-here J , varies as cos and +
t,ionally symmetric apertures. The x denote i-directed cur- J* varies as sin +. The graphs arefor J r in the 4 = 0 pla.ne,
rents, and the squares denote +directed currents. In the and J6 in the 4 = a 2 plane. The small oscillations in Jd
rotationally symmetric case, if J t i s nonzero, then J+ = 0: are due to inaccuracies of computation. The true Jd is
and vice versa. Note that the first three modes have only probably that obtained by smoothing out the oscillations.
J , and are similar to st.raight wire modes. The next two Again thecurrentsare normalized so that their mean-
modes have only Jd,and are similar to wire loop modes. square value over the surface is unity.
The value of X in each case is listed under the correspond- Fig. 5 show the characteristic gain patterns for the six
ing graph. The currents are normalized so that their mean- lowest order cos 4>sin + modes of the cone-sphere. They
square value over the surface is unity. are plots of t h e normalized radiationintensityversus8
Fig. 3 shows t.he characteristic ga.in patterns for the sis from the corresponding mode currents of Fig. 4. (Aga.in,
Iowest order rotat,ionally symmetric modes of the c o n e tic marks correspond to a gain of 2 . ) S o w the ra.diation
sphere. They are plots of the normalized radiation inten- field for each mode has tn-o components, EOwhich varies
sity.versus 8from the correspondingmode currents of as cos +, and E+ which varies as sin +. The gra.ph labeled
Fig. 2. (Tic marks correspond t,o a gain of 2 . ) The first. ge is the 8-polarized ga.in in the +
= 0 plane, and that
t,hree modes are due to J t and ha.ve only an Eo in the radia- labeled go is the +-polarized gain in the = a//2 plane. +
tion field. The next two modes are due t.o Jd and have To illustrate computations for a bod)- of another shape,
only an E6 in t,he radiat.ion field. The three-dimensional Fig. 6 shows similar results for a disk of radius one-half
pattern is obtained by rotating each plot. about 6he Z axis I\-avelength. Shonn are the three lowest order character-
(vertical axis). istic currents and characteristicgain patterns for the COS +,
Fig. 4 shows the six lowest order characteristic currents sin + (12 = li modes. The interpretation of the two com-
+
for t.he GOS +, sin modes ( n = I ) of t.he same cone- ponents of J ! and the two components of g, is the same as
sphere. These are themodes n-hich are used in the solution discussed for the cone-sphere. The computations were
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.
RARRINGTOS A N D MAUTZ: COMPTJTATIOK OF CHARACTERISTIC MODES 635

Ram= Ii J J Wi-u, exp (-,jkm.r)C ~ S (41)

i s

where urnand k, are the polarization and propagat>ion

vectors, respectivelyl of the measurement plane wave. In
evaluating (40)and (41) for bodies of revolution, we use
four impulse approximation to Wi,as discussed in [SI.
Once V n iand Rnmareevaluated, the radiation field i s
given by [ Z , eq. (437. For plane-wave scatkering problems,
n-e evaluatethe plane-wave excitation coefficients R,i,
r2,eq. (48)J by (41) witahu, and k, replaced by ui and
ki, the polariza.tion and propagation vectors of the inci-
dent plane wave. Once Rniand Rnmare evaluated, the bi-
static radar cross section is given by [Z, eq. (51)].
Fig. 7 illustrates convergence of the modal solution for
the cone sphere (10" half-cone angle, 0.2 wavelength

Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.
636 IEEE TRANSACTIOKS ON ANTENNAS AND PROPAGATION, S E P T E ~ E R1971

0 c
0 0
m m

(a) two modes

0 C
B
m m

( c ) four modes (d) f i v e modes

Fig. 7. Co~~vcrgence
of modal solution for cone-sphere excited by voltage across slot at cone-to-sphere junction.
Solid curves are matrix inversion solution, 1: are modal solutions.

sphere radius, 1.36 wavelengths total length) excited by plane), and the squares are the modal H-plane solution
a voltageapplied acros3 a narrowslot at the cone-to- (4 pola.rized, 4 = a/2 plane). The modes used are the
sphere junction. The solid curve in each figure is the radia- cos 4, sin 4 (?) = 1) modes of Figs. 4 and 5. The modal
tion gain pattern obtained by matrix inversion [5], and solution of Fig. S(a) uses the tu-o lowest, order modes,
the z are t.he corresponding moda.1 solution. The modes Fig. S(b) the fourlowest,Fig. S(c) the six lowest, and
used are the rotationally symmetric ones having only J t Fig. S (d) the eight lowest. Sote that a plane wave excit.es
on S (Fig. 2) and EOin the radiation field (Fig. 3 ) . These all n = 1 modes, the six lowest order modes being most.
are the only ones n-hich a.re excited by an E , on S. The important for this particular body. The sca.ttering pattern
modalsolution of Fig.7 (a) uses the tw-o lowest order has fully converged when eight modes are used.
modes, and is nea.rly zero. This could ha.ve been predicted We have also considered the convergence of modal solu-
from the fact. that the txo lon-est characteristic currents, tions for bodies of other shape, with similar results. For
Fig. 2 (a) and (b), have almost zero amplitude at thecone- example, for the disk of 1 wavelength diameter (charact.er-
to-sphere junction. The modal solution of Fig. 7 ( b ) uses istic currents and pa,tterns given by Fig. 6), the modal
the three Ion-est order modes, and shows that. the gain plane-wave scatteringpattern was almost the sameas
pattern is predomina.Iltly that. of thethird mode. The that of the lowest order mode, which n-as resonant,. The
modal solution of Fig. 7(c) uses the four lon-est order J t scattering pattern using two modes \vas almost identical
modes, and tha.t. of Fig. 7 ( d ) uses the five lowest order to the matrix inversion solution.
J t modes. Note that. the gain patt.ern is essentiallv fully If a conducting body of revolution is excited by a non-
converged when five modes are used. axially propagating plane xa.ve, the 97. = 0 modes, n = 1
Fig. S illust>ratesconvergence of the modal solution for modes and higher order sin ~ 4 cos , 774 modes are a.11 ex-
the same cone-sphere scattering an x-polarized plane u-ave cited. If the bodyiselectricallysmallindiameter (say
axially incident on t.hc cone tip. The solid curves in each less than 1/4 wavelength),only the 11 = 0 and n = 1
figure are t,he bistatic radar cross sections ( E plane and modes contribute significantly t.o the sca.ttering patt.ern.
H plane) obtained by matrix inversion [5]. The .T in each Similarly, in a,n aperture antenna problem, if the excita-
figure are the modal IGplane solution (e polarized, 4 = 0 tion is not rotationa.lly symmetric,modes other than n = 0
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.
HARRISGTOS AXDYAUTZ:COMPCTATIOX O F CHARACTERISTIC MODES 637

(a) two modes (b) four modes

(c) s i x modes (d) eight modes

Fig. 8. Convergence of modal solution for scattering by cone-sphere excited by axially incident, plane wave. Solid
curves are mat.rix inversion solution, being E-plane patternand u+ H-plane pattern. Iand squares are cor-
responding modal solutions.

modes nil1 be excited. Again, if the body is of electrically metric by averagingcorresponding off diagonalimped-
small diameter, usually only the 91 = 0 a.nd n = 1 modes ance elements.
contribute significantly to the radiation pattern. Computations of characteristic currents and character-
istic field.. have been made for a number of wire object.s,
TI. BPPLICATIOS TO W I R E OBJECTS and these modes have been used inmodal solut.ions to
demonstrate convergence in radiation and scatteringprob-
computer pr0gra.m for the computationof the general- lems. As an example, consider the wire a,rrow shown in the
ized impedance matrix of wire bodies of arbitrary shape central insert of Fig. 9. The parameter a is 1/4 wavelength,
is available [SI. This impedance matrix can be used in a and the wire diameter is 0.004 wavelength. The graphs of
slightly modified version of the bodies of revolution mode Fig. 9 shon- the six lowest order characteristic currents,
program [11] toobtainthecharacteristiccurrentsand plotted a.s a function of the contour variable, st.arting at,
fields of wire objects. A discussion of the method andsome the tip and ending at the tip. Sote that the modes are
representative computations are given in this section. either symmetric (even) or asymmetric (odd) about, the
X wire in space is specified by a number of points along midpoint with respect to the wire length variable. This
its a.xis, plus its diameter. There may be more than one could bepredicted from the symmetry of the geometry
n-ire present, and these wires may have free ends, may of the u-ire. We did not use this symmetry in the compu-
be closed on themselves, or may be joined together. The tations because the program is writtenfor 0bject.s of
expansion functions and testing functions are the same arbitrary shape. The currents were normalized by choos-
(Galerkin’s method), and are chosen t o be triangle func- ing their maximum amplitude to be unity.
tions extendingoverfoursubsections of the wire. For Fig. 10 shows the characteristic gain patterns for t.he
expansion, the triangles are a.pproxinlated by four pulses, sixlowest ordercharacteristic fields of the wire arrow.
and for testing they are approximated by four impulses. The a.rron- is considered to lie in the x-z plane, with it.s
Because of theseapproxinmtions, the impedance matrix axis a.long the z axis. The two patterns shown are the gain
is not exactly symmetric, as it should be. I t is made sym- in the x = 0 plane (labeled gI) and the gain in the y = 0
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.
. .
638 I N R E TRANSACTIONS ON ANTENNAS AXD PROPAGATIOW, SEPTEMBER 1971

11 ( b ) X, = 1.65

-l t

-1 -- -I --
( e ) d5 = 9.96 (f) .A6 = - 13.1
Fig. 9. Characteristiccurrents for wire arrow, with a 0.25 wavelength and wire diameter 0.004 wavelength.
Six lowest order modes are shown.

plane (labeled gl). The g, pattern is always 8 polarized, VII. DISCUSSIO~;

a.nd the gz pattern is t$ polarized for even currents and
e polarized for odd currents. The sca.le is linear, with each The method of computing characteristic modes devel-
interval betn-een tic marks representing an increment, of oped in this pa.per can be used for conducting bodies of
two in gain. The gain patterns in other planes, such as arbitrary shape,provided the body is not, electrically large.
the x-y plane, are not simply related to those in the .T-Z General computer programs for bodies of revolution are
and y-z pla.nes. For completeinformation, some sort of available for determining the modes, and for using them
three-dimensional presentation of the gain patterns would in modal solutions for antenna and smttering problems
be desirable. [ll]. Computer programs for determining the generalized
Modal solutions for nire a.ntennas and wire scatterers impedances for wires of arbitrary shape are also available
mere also ma.de a.nd compared tothematrix inversion [SI. Computerprograms for bodies of arbitra.ryshape
solut.ion [4], [SI. As modes were added, the modal solu- have not been written. primarily because of the difficulty
tion converged to the matrix inversion solution in about of specifying coordinates on a three-dimensional surface S.
t.he same way as for bodies of revolution. In the antenna The characteristic currents are relatively easy to graph
problem, given a. voltage source at some point along the because they are real and exist on a surface. For example,
nire, t.he modes are excited in proportion to their current on bodies of revolution there are at most two real com-
amplit.udes at the point. of excitation. By usingseveral ponents J t and J , which, for eachcos n+, sin 7 1 4 variation,
voltage sources at. several points, theexcitation of a num- can be expressed as a, function of the single contour length
ber of modes could be completely controlled.In the scatter- variable. Similarly, on wire objects, there is a single real
ing problem, the excitation of modes could be controlled current I which is a function of the \\-ire length variable.
by placing lumped loads along the wire. However, if the In contrast, the characteristic fields are complex and de-
loads were restricted to he passive, only partial control fined over three-dimensional space. Even on the radiation
of the mode excitation would be possible. sphere n-e ha.ve four real functions to describe, namely,
Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.
YhUTZ:
HARRIKQTOX AXD COMPUTATIOK OF CHARACTERISTIC MODES 639

and Em have a conjugate syrnmctr!’. 1:or example, for

bodies of revolution, &(a - 8) = (-l)zEo*(0)and
E+(T - 8 ) = ( - l)n+lE,*(0). Allsucll information is, of
course, included in thegeneral comput ~r output , and could
be plotted if desired.
The characteristic modes of bodiw o f arbitrary shape
have most of the properties of the classical spherical modes,
(a) A 1 = 1.64 and ca.n be used in much the same ~ a y Because
. of their
orthogonality properties, they are convcnient to use for
B
m problems ofsynthesis
analysis, optimization
and of radia-
tion and scattering. Yon- that we have an efficient way of
calculatingthese modes, thctheorycan be applied to
practical problems. inyolving antennw and scatterers of
:vbitrary shape.

REFEREKCES

(c) A3 = 5.81 (d) A4 -- 6.81 I ] ] 11. J. Garbacz, *.A generalized e s p a t 1 h l for radia.ted and
scattered fie!ds,” Ph.1). dixserlation, Ohio Stale Univ., Colum-
bus, 1968.
:a] R. F. Harrington and J. R. hfaut.z, “Theory of characterist.ic
modes-for conduct.ing bodies,” this issue, pp. 622-628.
1 3 ) It. TI. Turpitl, “Basis transformation, least square, and char-
acteristic mode techniques for thin-wire wattering analysis,”
Ph.1). dissertation, Ohio State Univ., (:olr~mbns,1969.
141 It. F. Harrington, Field Computation b y :Ilommmt Methods.
Y e a York: Marmillan, 1968.
i j ] J. R . XImltz and R . F. ITarrillgton, ‘ ‘ I ~ a d i : ~ land i o ~scattering
~
froin bodies of revolution,” J p p l . Sci. R e s . , v o l . 20, June 1969,
pp. 405-435.
(e) A s - 9.96 (f) A b - 13.1
[SI H. 13. Chao and B. J. Strait, “Computer programs for radia-
tion and scattering by arbitrary configurations of bent. xire?,”
SyracuseUniv.,Svracuse, N. Y . , and Air Force Cambridge
Fig. 10. Characteristic gain patterns for wire arrow, with n 0.?5 Res. Labs., Sci. Rep. 7, Contract F19628-68-C-0180, Sept. 1970.
wavelength and wire diameter 0.004 wavelength. Patter113 111 li] R . F. Harrington and J. R. Mautz, “Theory and computation
z = 0 plane are labeled g,, those in y = 0 plane gw. Six lowe-t of Characteristic modes for cond1lcting bodies,” Syracuse
order modes are shown. Univ., Syracuse, X. Y., and .4ir Force Cambridge Res. Labs.,
Sci. Rep. 9, Contract F19628-6&C-0180, 1)er. 1970.
181 J. H. Wilkinson, The dlgebraic Eigmnl!du~Problem. New
the real and imaginary parts of EO and Em.In this paper York: Oxford, 1967, p. 34.
191 A. Ralston and H. Wilf, Ma-!henmtim/ :Ifelhods .for Digital
n-e have plotted mode patterns in terms of I/O = I EO Computers. Wiley, vol. 1, 1960, ch. 7, and vol. 2, 196i, ch. 4.
or g9 = I E, 12? which shows only part of the picture. For [IO] Program in Commun. Alss. Comput. J f a c h . , vol. 11, Dec. 1968,
pp . 820-826.
example, note that all plots of g are symmetrical a.bout 1111 J. R. Riautz and R. F. Harrington, “Computer programs for
6 = ~ j 2 even
, though the bodies do not have the corre- characteristic modes of hodies of revolution,” Syracuse Univ:,
dyracnse, K . Y . , and Air ForceCambridge Res. T.abs.. Scr.
spondingsymmetry. However, the field components EO Rep. 10, Contract F19628-6RCP0180, 5:ln. 1071.

Authorized licensed use limited to: Universiti Malaysia Perlis. Downloaded on July 06,2023 at 04:28:33 UTC from IEEE Xplore. Restrictions apply.