Theory of Antennas of Arbitrary Size and Shape*

S.A. SCHELKUNOFF?, ASSOCIATE,I.R.E.

Sunurrclry-In this paper there are presented ( I ) a quite general method of antenna analysis; (2) a physical picture of transmirsion phenomenain antenm, based on this method; and (3) an expression for the input f f impedonce o antemu o any shape, whose transverse dimensions are small compared with the wavelength. In a brief historid sketch o the f antenna problem the factors which must be taken into consideration in solving the problem are discwed While in ordinary transmission lines the vdtage is proportional to the charge, this is not the case in antemu. The explanation lies in the fact that antenmu are multiple transmission lines(like wwe mdes)and not simple, that i, single-mode trammission lines. Ow presenttheory is s based on the voltageivrent equations since these appear to be consiakrably simpler than charge-went equations. The latter are considered only briefly, In the absenceof dissipation and in so far as the totalvdtage wave and the principu w e n t wave are concerned rndiation is strictly an end effect. In so far as the total w e n t and thetotalchargewavesare concerned radiation effects dismbuted are (nonmifom&) along the entire antenna In the first approximation,regardless of the shape of the wire the charge is proportional to the voltage and waves are sinusoidal, the e n t w wave having nodes while the voltage wave and the charge wave antinodes at the e&of the antenna. The second approximation depends on the shape of the longitudinal cross section of the antenna as well as on the sue of the transverse cross section. Our analysis is based on Ma.wells equations but the final results are quite simple and the physical picture growing out of this mathematics is attractive to an engineer. It is pemuhsible to think that a wave emergmg from a generator in the center of an antenna is guided by an antenna until it reaches its boundary sphere passing through the ends of the antenna and separating the antenna region from the external space; at the boundary sphere some energy passes the external space and some into is reflected back-a situation existing at the juncture between two transmission lines with different charactektic impedances. We may also think of the antenna as the wall of an electric horn with an aperture so wide that one can hardy see the horn itself-just like a Cheshire Cat: only the grin can be seen. In fact, the mathematics that we we is that horns. appropriate to wave guides and electric The antenna problem is stated in Section I and its histoly is briefly discussed in Section II; Section III contains a summary and a discussion of the results for antennas with uniformly distributed capacitance (conical antennas); Section IV is devoted to antennas with non-uniformly distributed capacitance; Section V presents a derivation of the formulas contained in Section III; Section VI reviews the induced-electromotive-force method of computingradiation and its we in thepresentproblem; Section VII is devoted to the current-charge equations; Section VIII is devotedtowauepropagationalongparallelwires;inSection I X an expression is given for the impedonce of an infinitely longcylindrical wire, and SectionX deals withan approximation needed in our discussion of the problem.

I. INTRODUCTION
Two Problem

It is beyond the scope of this paper to discuss adequately prior work on radiation from conductors of finite length and only a few representative papers w cited. Two problems l be l i have presented themselves.In Problem A the current distributioninanaotennaisgivenanditisrequiredtofindthefield and, hence, the external electromotive forces needed to produce the given current distribution. Problem B is the inverse of A: The distribution of applied or externalforces in an antenna is given and it is required to obtain the field and, hence, the current produced by the applied forces. Problem A has been solved rigorously and completely with the aid of retarded potentials. On the other hand, Problem B presents many difficulties and it is the engineers hard luck that he happens to be interested in just this problem.

Solution of Problem A is Usdd within Limits

The solution of Problem A is not altogether useless to the engineer. Theory is not the only source of information concerning the current distribution produced by given forces. For example, the current distribution can be determined experimentally, thus, it has been known long ago that on a t i wire hn the current is distributed almost sinusoidally (Fig. 1) and t i hs

Fig. 1-Current distribution in an infinitely thin perfectly conducting antenna of any shape is sinusoidal with current nodes the ends of at the antenna The distribution may be taken as the first approximation for thin antennas.

*Decimal classification: R120. Original manuscript received by the Institute, April 25, 1941. Telephone Laboratories,Inc. New York, N.Y.

e Bl

fact has been employed to obtain approximately the radiated power, the input impedance, and the field. If the length of the wire is in the neighborhood of one half of the wavelength (or one quarter of X if the ground takes the place of the other half), the results are fairly satisfactory from the practical point of view. But when the length becomes equal to 1 wavelength, then a more accurate solution becomes necessary. The theoretical radiation pattern may be still good enough (except in the former null directions) but the input impedance is comThere may be occasions,of course, when a more accurate solution would be desirable and, certainly, no one is likely to object to having one, provided it is simple enough.

Reprinted from the Proceedings o f t h e /.R.F., 29, pp. 493-521, vol. Sept. 1941. Copyright, 1941, by the Institute of Radio Engineers, Inc.

SCHELKUNOFF. A N T E N N A S O F ARBITRARY SIZE A N D SHAPE

1165

puted to be infinite.Whileinfinitymay be regarded as a legitimate hrst approximationtoalargequantity, it is a useless approximation. For free-space antennas of length nh, practicalapplicationsrequireasecondapproximationand this, in so far as theory is concerned, leads us back to Problem B.

Slow Convergence off Resonance

There is a weakness inherent in the method used by Page and Adams. This method employs spheroidal harmonics and offresonanceleads to complicated slowly and converging series. In theirnumericalcomputations,the authors limited themselves to spheroids about one-half wavelength long.13
The Importance of Shape

11. HISTORY PROBLEM B AND GENERAL OF COMMENTS METHODS FOR ITSSOLUTION ON

Two Methods of Approach and an Unproved Proposition a

Broadly speaking, we may distioguish between two methods of approach to the solution of Problems B. Some have based their work on Maxwells equations, that is, on the electromagnetic laws that are considered well established; while,otherwriters-repulsedbythecomplexity of such methods, ventured to start from a new and unproved premise. This new Proposition a may be formulated as follows: power losses due to radiation produce the same effect on transmission of electric waves along conducting wiresas losses due to hipation. This premise is inconsistent with Maxwells equations (see Section VIIl); and yet, with its aid some approximate results have been obtained. We shall return to this point later.
Spheroidal Antennas

An important point,however, is that while in thefirst approximation the current distribution is independent of the size of the transverse cross section of the antenna and of the shape of the longitudinalcross section and is sinusoidal for all conductors, in the second approximation the current distribution depends on both factors. It was RayleigH4 who, in discussing Abrahams was and the papers, to first point outthat resonant frequencies of finite wires are independent of the shape of the wires in the first approximation but not in the second. Rayleighs conclusion is borneoutby Englunds experiments16 and by our calculations. In fact, not only the resonant frequencies but other quantities as well are affected,inthesecondapproximation,bythesize of the transverse cross section of the antenna and the shape of the longitudinal cross section; consequently, way must be found a to take these factorsinto consideration.
Cylindrical Antennas

Thosewriterswhoprefer safe premises as apoint of departure have to look for some particular shape of conductors whichwould lenditselfmorereadilytomathematical treatment. Spheroids have been the first to attract attention. A b r a h d treated free oscillations on a perfectly conducting thin prolatespheroidandobtainedanexpressionforthe resonant frequencies or wavelengths. Recently, and Page Adamsand,then,Ryder in his as yetunpublishedthesis, have dealt with forced oscillations on spheroids.
M. Abraham, Die electrischen Schwingungen um einen stabf6rmigen kiter, behandeltnachderhiaxwellschen Theone, Ann. der Phys., vol. 66, pp. 435-412, 1898. 3L. K i n g , On the radiation Geld of a perfectly conducting base V. insulated cylindrical antenna over a perfectly conducting plane earth, and the calculation radiation resistance and reactance,Phil. Trans., of ser. A, pp. 381-422; November 2, 1937. 4G. M e , ElektrischeWellenanZweiParallelen&ahten, Ann. der. Phys., p. 248, 1900. Page and Norman I. A h , Jr., The electrical oscillations of aprolatespheroid,Paper I, Phys. Rev., pp.819-831; May 15, 1938. 6John R Carson, Propagation of periodic currents over non-uniform lines, Electrician, pp. 272-213; March 4, 1921. C. P.Steinmetz,The generd equations of the electric circuit-III, Proc. A . I . E.E., p. 255, 1919. *Ronald King, Telegraphists equationsat ultra-high frequencies, P h p i o , pp. 121-125; April, 1935. 9 ~ 0.Pedersen, Radiation fromvertical . a antenna flat over perfectly conductingearth, Ingenimidemkubelige Sknfter, ser. A, Nr. 38, 1935. E Siege1 und J. Labus, Scheinwiderstand von Antennen, Hochfrequenr. und Elektrwkwtik, Bd. 43, pp. 166-172, 1934. J. Labus, Rechnerische Ermittlung Impedanz von Antennen, des H0Cfifrequen.z.und Elektroakrrstik, Bd. 41, pp. 17-23, 1933. L. J. Chu and J. A. Stratton, Forced oscillations of aprolate spheroid, J a v . Appl. Phys., pp. 241-248; March, 1941.

A different method was chosen by K ~ n g . ~ Starting from an integralequation,heobtainedthesecondapproximation to the solution of Problem B for a thin cylindrical wire and his work could be extended to wires of other shapes. If anythmg, this method is more complicatedthan the one employed in the case of spheroidal antennas and it does not lend itself to any simple physical interpretation; one just has to take the final quantitative results. On the other hand, the calculations are carried for out cylindrical antennas which are of greater practical interest than spheroidal antennas.
Our Method and Its Advantages

Leigh

We also start with Maxwells equations but choose conical conductors. There are several advantages to this choice. In the first place, the functions to which one is naturally led represent waves on thewire ratherthan free oscillations. Consequently, the conditions existing off resonance can be studied just as readily as thosenearresonance.Furthermore, this means that no complications will arise if we break the wire at some point andinsert a resistance, or any impedance, for that matter. The difference between our method of treating the conical 13Sincecompletion of this paper, L.J. Chu and J. A. Stratton have published a comprehensive discussion of forced oscillations on prolate spheroids and dealt with the conditions off resonance as well as near resonance. 14Lord Rayleigh, On the electrical vibrations associated with ti hn terminated conducting rods, Phil. Mag., pp. 104-107; July, 1904. lSJ. A. Pollock, A comparison the of periods of the electrical vibrations associated with simple circuits, Phil. Mag., pp. 635-652; June, 1904. 16C. R. Englund, The ~ h l r a period of linear conductors, Bell l Sp. Tech. Jow., pp. 404-419; July, 1928.

1166

P R O C E E D I N G S O T H E IEEE. VOL. 72, NO. 9. SEPTEMBER 1 9 6 4 F

wire and the conventional method of treating the spheroid is precisely the difference existing between two possible methods of dealing with a finite section of an ordinary transmission line. On one hand, thevoltage and thecurrent in sucha section can always be represented as the result of interference of progressive waves traveling in opposite directions; and, on theother hand, thesame quantities can be represented in terms of harmonics corresponding to natural oscillations in the section of the line. The -first method is so much simpler than the second, whichgivesthe results as i n f i i t e series of partial fractions, that probably only few are even aware that the solution could have been found by the second method in the first place.
The Shape of Conductors May Be Taken into Account

The real part R , of Z, represents radiation and should properly be called the radiation resistance; it is the resistance of the outer spaceas seen from the ends of the antenna. Unfortunately, this name has been generally given to a quantity which t r sout to be equal to K 2 / R , . un
Methodr Based upon an Unproved Hypothesis

Another advantage is our choice of conical shape is that in thesecondapproximationtheeffects of theshape of the conductor become separated from the end effect or radiation. Consequently, the equations developed for conical wires can be amendedto take care of the shape effect.
Antenna as a Transmission Line

Finally, our method turns out to be consistent a with physical picture which is rather attractive to the engineer. Let us suppose that a wire is energized at the center. A spherical wave emerging from the generator is guided by the antenna until it reaches the limit of the antenna region, that is, the sphere passingthrough the ends of the antenna; there, someof the energy passes into the outer space and some is reflected back, a situation existing when one transmission line isjoined to another.
Antenna as an Electric Horn

The ideathat the ortransmission-line theory could be amended to takecare of radiation effectshadoccurred to ~teinmetzas long ago as 1919,in connection withhis studies of electric waves on parallel wires. He failed to obtain correct results because twice he was deceived by physical intuition. In the first place, he believed that radiation losses and dissipation losses produce the same effect on transmission ofwaves. In effect this wasa new proposition, which we havecalled Proposition a,that could be inconsistent with established laws and thus he was taking chances. It is perfectly true that each element of current-carrying wire radiates power; but at the same time it receives equal power the from surrounding medium due to the action of other elements plus power that is dissipated in that element.* In such circumstances, it is impossible to decide in advance just how radiation affects the current and voltage distribution on the wire.
The Hypothesis Is Wrong

In Section WI there is given a simple and straightforward proof that the voltage across a pair of parallel perfectly conducting wires i not attenuated. This proves that Proposition a is s not true as it stands. Nevertheless, we shall have an occasion to point out that there is some truth in this proposition.
Effect of Phase Velocity on Radiation

We may also t i k of the wire as the wall of an electrichorn hn with an aperture so wide that one can hardly see the horn itself. In fact, the mathematical analysis used by us is precisely the analysisappropriate to wave guides and electric horns. We end up with a picture of the antenna as a transmission line (Fig. 2) whose output impedance Z, represents the end effect.

21

This brings us to another point. Proposition a has nothing to do (at least directly) with the very large value that Steinmetz obtained for the total radiation from a parallel pair. He used the same method which had beenused before and which has been used since forapproximatesolutions of similar problems; but he apparently thought that sincethephase velocity of waves along the line very high he could make was it infinite, and assume, in making calculationsof radiation, that thecurrentsalongthe entire line were in phase. Thus, he found that the radiated power was proportional to the length of the line while in reality the power radiated a long lineis by independent of its length.4,i9 This discrepancy is due entirely to the effect of phase velocity of the current waves along the line and has n o w to do with the mannerin which radiation affects thecurrent distribution itself. For a discussion of other aspects of the problem of radiation from parallel wiresthe reader is referred toCarson.
181f the wire is perfectly conducting, the tangential electric intensity must vanish at the surface of the wire and the flow of power from the s wire or to the wire is 0. If the wire i imperfectly conducting the flow of power is into the wire and not out of it. John R. Carson, Radiation from transmission lines, Jour. A . I . E. E., p, 189; October, 1921. John R Carson,The guided and radiated energy in wire transmission, Jour. A.Z. E. E.,pp. 906-913; October, 1924.

Fig. 2-The input impedance of a conical antenna any sizei equal of s to the input impedance of auniform transmission linewith a certain output impedance Z,. input impedance of a t i antenna of The hn any shape may be similarly represented, except that the characteristic impedance will be variable.

From the point of view developed in this paper there difference between the end effect and radiation.

is no

S C H E L K U N O F F . A N T E N N A S OF ARBITRARY SIZE A N D SHAPE

1167

clearer This may made be situation an withby an analogy electriccircuitcomprised of aphysicalresistor,aphysical Pedexn9 and also Siege1 and Labuso* made use of Propinductor, and a physical capacitor in series. Such a circuit is osition a in their equations for a smgle wire but they based usually regarded as a simple electric circuit, with one natural their computationson more nearly the actual current distribufrequency or, takingthedecrement into account,withone tion in the wire. Thus they computed an approximately correct natural oscillation constant. In reality the circuit can osciltotal radiated power and only then they postulated the charlate ininfinitelymanymodes; it is onlybecausethe first acter of its distribution. natural frequency is verymuchlower thantherestthat in Pedersen has tried two ddferent hypotheses.Firsthe asordinary applications the circuit behaves if it were a simple as sumed that radiation loss is concentrated at the current anticircuit. node and then he assumed to be distributed uniformly along it The most prominent transmission mode of a given line will the antenna The current distributions calculated on either of be called the principal or the dominant mode. these two assumptions turned out to be nearly the same and checked fairly satisfactorily with a measured current distribuPrincipal Waves Guided by Two Coaxial Cones tion Naturally, the calculated expression for voltage distribution Let two coaxial conical conductors, having a common axis mustnecessarily be wrong (Section W ) but no voltage I; (Fig. 3), be energized at the common apex.Intuitivelyone measurements have ever been made. Furthermore, for comparativelyshort antennas thecurrentdistributiondoesnot markedly depend on just where the power is lost (see Section
Single Wires
x).

Whatevermay be saidaboutthemethod,Pedersensucceeded in obtaining better approximations to the solution of the antenna problem than the ones available at that time.

111. ANTENNAS UNIFORM WITH CHARACTERISTIC IMPEDANCES-GENERAL DISCUSSION SUMMARY AND Fig. 3-Cross
Perfect Conductivity

sections of infinitely long conical conductors and electric lines of force for principal waves.

Unless otherwise specifiedall conductors are assumed to be perfect. This assumption simplifies the mathematics and separates the effect of radiation on transmission ofwaves along a antennafrom the effect of dissipation. In thefirst apn proximation it is reasonable to suppose that the two effects are independent and can be superimposed.
Transmission M

Until recently whenever one thoughtelectric waves guided of by a pair of parallel wiresor by coaxial cylindrical conductors, one was apt to visualizeapicture of electriclines of force extendmg from one conductorto the other and lyingin planes perpendicular to them, that is, in equiphase surfaces. One was copscious that near the ends of the conductors the Geldwas somewhat warped; but one felt that the end effect was small and could be ignored. Thusone was concernedwithone configuration of lines of force, with one propagation constant, withonephasevelocity,withonecharacteristicimpedance, and with one pair of transmission equations, that is, with one transmission mode. A transmission line with a mgle tranmission mode will be called a simple transmission line. But physical transmission linesare multiple. They are capable of guidmg many types of waves, with different configurations of lines of force, with different propagation constants andwithdifferentcharacteristicimpedances.Recently, this fact has burst into prominence in connection with the waves in hollowmetal tubes; but theconventionallines are also multiplelines.Theonlyreason why in thepasttheywere regarded as simple lines is due to the fact that their transverse dimensions were so small compared with the wavelengths in which engineers happened to be interested that only the first transmission mode was quantitatively signhcant.

feels that if the cones were of infinite length, the wave would be that lines such the of electric would force follow the meridians of spheres concentric with the apex of the cones. There is indeed a wave, namely, the transverse electromagnetic spherical wave ( T J N ) for which this is true2; this wave turns out to be the most prominent and we shall callit the principal wave. If we imagine a homogeneous spherical conductor coincidingwithsomeequiphasesurface,thentheoutward-bound progressive wave will be reflected from it. Electric lines, however, will still follow the meridians.
Equations for Principal Waves

Even from elementary considerations is obvious thatin so it far as principal waves areconcerned,thedoublecone is a uniform transmission line. Consider, for example, the capacitance per unit lengthA B (Fig. 3(B)). The length the lines of of electric force and the circumference the conductor areboth of proportional to the distance r from apex; the hence, the capacitance remains unchanged. principal waves are just as easily treated rigorously.2 The seriesinductance L andtheshuntcapacitance C perunit
21S. A. Schelkunoff, Transmission theory of spherical waves, Tram. A . I . E. E., pp. 744-750,1938. UExcellent elementary discussions of principal waves on a double cone may be found in a paper by HOW^^^ and on page 183 of a paper by Carter.24 23G. W. 0. Howe, The nature of the electromagnetic waves employed in radiotelegraphyand the mode of their propagation, Elec. Rev., pp. 486-489; September 26, 1913. 24P.S. Carter, Simple television antennas,RCA Rev.,pp. 168-185; October, 1939.

1168

PROCEEDINGS OF THE IEEE. VOL 72, NO. 9.SEPTEMBER 1%

length and thecharacteristicimpedanceKarefoundtobe (for the double cone in Fig. 3(B))

we shall have

Io()

IocosB(I - r ) ,
V,
=

P J, L = - logcot -, s 2

c=

s E

+ log cot- 2

V o ( r )= VosinB(I- r ) ,

iflo.

(4)

9 Ji K = - logcot J. = 1201ogcot -, n 2 2 (1) where J, is the angleof the cone.25The phasevelocity of these waves is equal to that of light. The principal voltage and the principal current can then be expressed in the following general form

These are the equations for a spherical cavity resonator and are of no direct interest in our present problem. If the sphere were a perfect magnetic conductor, then the current I o ( / ) would have to vanish instead of the voltage and the voltage-current equationswould be I o ( r ) = I0sinB(I - r ) ,
V o ( r ) = V0cosB(I- r ) ,
Vo = -iKIo.

(5)

The voltage between the corresponding points A and A is defined here as the line integral of the electric intensity along anycurve joining A and A and lyingcompletelyinthe equiphase surface passing through and A. This definition is A in keeping with the usual definition of the voltage across a pair of parallel wires. This voltageis dfficult to measure except near the origin and may be regarded as an auxiliary variable that helps us to find measurablequantities such as the input impedance, current distribution, etc. Fig. 4 is a graph of K as a function of the reciprocal angle. For cones of small angle, 1/Ji is approximately equal to r / p ,

Weknow from experience that the current distribution in a t i wire is approximatelythat given by (5). This suggests that hn the impedanceof free spaceas seen from theoutput ends of the antenna is so highcomparedwithK that analmost complete reflection takes place. Later in this paper we shall actuallyprove that as Ktends to infinity, the currentand voltage distributions on a conical wire, on any wire for that matter, approach (5). If the spherical sheet at r = I had some finite impedance, we should have I o ( r ) = ~ ~ s i n B- r ) + IocosB(/ - r ) , (/

G ( r ) = ~ , c o s B ( I- r ) + ~ s i n B ( - r ) , I V, = -iKIo,
VO

iKIo.

(6)

The input and the output impedances would then be

v o z . =(0) -Io@) Iosinfll

-iIocosflI

+ i1Osin8I + IOCOSBI

Sphere of Discontinuity

n c c w n o c ~RADUYS

(4)

Fig.4-Thecharacteristicimpedance of conical antennas of types 3(B) as a function of the reciprocal angle cone shown Fig. in = a / l , where I is the length of the cone and a is ifs maximum radm.

Equations (6) would have represented the actual conditions in a conical antenna (Fig. 5), were it not for the fact that the space outside the boundary sphere S is a multiple transmission line with a set of transmission modes different from that in the antenna region. In particular, in free space there is no transmission mode which is even similar the principalmode to just discussedby us. For the latter, thefieldconcentration near the conductors (if theyare h isexceedinglyhigh, ) because the conductors can support high currents quite read-

where p is the radius of the cone at distance r from the apex, and 2 2r K 12010g - = 12010g -. (3)

Ji

When this ratio r / p is 100,thenK = 635 ohms; and for r / p = 1OO0, K = 913 ohms. It is hardly necessaryto point out that K varies much slower than the ratio r / p . Standing Principal Waves If at r = I we assumeareflectingsphere,thenthe amplitudes of progressive waves in (2) will be equal and standing waves will result. For a perfectly conducting sphere at r = I,
D

25Thequantity 1) is the intrinsic impedance of the medium between the cones; for air 1) is approximately equal to 120 n or 377 ohms.

Fig. 5-The cross section

of a conical antenna of length I and of the boundary sphere S.

SCHELKUNOFF: A N T E N N A S OF ARBITRARY SIZE A N D SHAPE

1169

ily. On the other hand, all radial currents in dielectrics are comparatively feeble and the magnetic intensity near A - B the axis outside S must be very small. Hence the energy carried by the principal wave from 0 to the boundary sphere S must travel thenceforward in different transmission modes and besides an ordinary reflection of the principal wave, secondary waves in the antenna region will be generated to match the field outside S. This is our picture of howthe end effect_..comes being. Radiation isone into part of this end effect; the reactive field associated with the secondary waves is the other part. The sphere of discontinuity is the aperture of the antenna regarded as an electric horn.
Free-Space Transmission Modes

(4

(B)

Fig. 7--Electrical lines for the second-order transverse magnetic spherical waves: (A) lines in free space; (B) lines in the presence of two coaxial conical conductors.

Before consideringsecondarytransmissionmodes in the antenna region, we shall review briefly the modes of transmission in free space. The principal mode in free space is characterized by electric linesof force whose shape is suggested in Fig. 6(A). The energy emitted by a very small doublet travels

metry,canberepresented as theresultant of anumber of waves traveling in the above-discussed transmission modes.

Seconda?y Transmission Modes in the Antenna Region

Suppose we have a free-space wave like that shown in Fig.

6(A) and suppose we insert a t i conical conductor coaxial hn

(A)

(B)

Fig. 6--Electric lines for the first-order transverse magnetic spherical waves: (A) lines in freespace; (B) lines in thepresence of two coaxial conical conductors.

outward in this mode. The electric field has two components, theradialcomponent E, and themeridiancomponent E#. The former is proportional to cos e and the latter to sine, where 0 istheanglewiththeaxis of thewave.The radial displacement current flows in one direction in the northern hemisphere and in the opposite direction in the southern. The two hemispheres play parts of the conductors a two in transmission line.Of course, the radial dqlacement current is not distributed uniformly within each hemisphere; the current density is maximum along the axis of thewave.We should note, perhaps, that while E, varies ultimately as rF2, where r is the distance from the doublet, the radial current density per unitsolidangle and hencethe total radialcurrent in each bemisphereare independent of thedistance,exceptforthe phase factor e-@,. Thelines of electricforcecorresponding to thesecond transmission mode are shown in Fig. 7(A). Higher transmission modes will have still more sets of closed loops. Mathematically all these modes are represented by zonal harmonics; the radial electric intensity is proportional to Pn(cosB)and the meridian intensityto ( d / d e ) Pn(cose). Any field outside the antenna region, having circular s y m -

with the axis of thewave (Fig. 6(B)). Electric lines of force must be perpendicular the conductor; but since the meridian to intensity near the axisis small,,we do not expect a very radical change in the configurationof the linesof force and we expect that this configuration will resembletheoneshown in Fig. 6(B). This is indeed the case. The major difference is that in the presence of the conical conductor the radial electricintensity (and hence the radial current density) is highest not at the surface of the conductorbut at some distance fromit. Right at the surface of the cone, the radial intensity vanishes. As the angle of the cone gets smaller, the field distribution becomes morenearlylikefree-spacefield distribution exceptinthe ever-diminishing region adjacentto the cone. Fig. 7(B)showshowthe second26mode in the antenna region is related to the second mode in free space. Incidentally, i the antenna is energizedat its center, this mode and a l f l other even modes do not appear. This is because the conduction currents associated with them flow in opposite directions in the two halves of the antenna, while the currents produced by the generator must flow in the same direction.
The Total Voltage Associated with Any Secondary Wave in the Antenna Region Is Zero

As in the case of the principal wave we define the voltage between two points on the upper and lower cones as the line integral of the electric intensity along any path joining the point of theupper cone to thepoint of thelowercone, provided the path is situated in the equiphase circuit passing through the two points.27It is shown in Section V that for all secondary waves in the antenna region this voltage is equal to zero
261t is reallythe third if we counttheprincipal; but it i more s Convenient to designate tbe principal mode in the antenna region as the 0 t h mode. If the points are not situated in the same equiphase surface, then the voltage between them is not defined.

1 70 1

PROCEEDINGS OF THE IEEE, VOL. 72, NO. 9, SEPTEMBER 1484

Conduction Currents Accompanying Secondary Waves Vanish at the Origin

It is shown in Section V that the conduction current accompanying any secondary wave vanishes at the origin I m ( 0 ) = 0. (9) Near theorigin I,,,(r) varies as rrn+lrA where, approxi, mately, A = 120/K. Thus the effect of the secondary current waves on the total currentis rather unimportantnearthe origin, but becomes more pronounced near the output terminals of the antenna.
Voltage-Current Equations

We can now write the complete voltage-current equations for a perfectly conducting transmitting antenna energized at its center in the following form:
V(r)= b ( r ) ,

I ( r ) = I o ( r )+ f ( r ) ,

(IO)

where the principal voltage-current waves are given by and (6) the total secondary current f ( r ) is the sum of odd secondary current waves

the transmission equations in the voltage-current form or in thecharge-currentform; but these two possibleformsare quite different in the case of multiple transmission lines unless one mode predominates overall others.Thecharge-current equations aremore complicated than the voltage-current equations and are here considered (see Section VII) only because of their bearingon the idea that radiation takes placecontinuously along the antenna. In that section we shall show that the effect of radiation on the distribution of the total charge (but not the total voltage) and the total current in an antenna may be represented by a continuously distributed series resistance and a continuously distributed series inductance in addition to the normal inductance of the antenna regarded as a simple transmission line.This coupled with the fact that for comparatively short antennas the precise distribution of ths resistance has but little effect on the current distribution, explains why equations obtained by Pedersen, Siegel, Labus and others turn out to be fairlysatisfactoryapproximations to the antenna solution, provided we replace their in equations V by a quantity proportional to q. On thewhole,however, we find the voltage-current equations much simpler; in and these equations, radiation may be represented as a terminal impedance.
Secondary Waves in an Antenna Affect the Amplitudes Principal Waves in a Way an Output Impedance Would
of the

I ( r ) = ~ ~ (+ 1 3)( r ) + 1 5 ( r ) + r

... , f(O) = 0.

(11)

The total voltage wave consistsof just two principal waves, of which the second represents the effectof incomplete reflection at the boundary sphere S. As the result of this incomplete reflection, the voltage maximum doesnot occur at r = 1. This is the only effect of radiation on thevoltagewave; no "attenuation" is introduced into the voltage wave. The total current wave, on the other hand, is affected more radically.
Imperfect4 Conducting Antennas

Consider the principalwaves in the antenna as given by (6). Substituting from (10) into the expression the for output impedance Z , and taking its reciprocal, we have

Thus the output admittanceconsists of two admittancesin parallel. Transmission-line diagrams (Fig. 8) represent above the relationship graphically. The current- f( I) is that part of the

It will be seen that f( r ) is rather small compared with r ) Io( and it is natural then to assume that the imperfect conductivity of an antenna will marufestitselflargelythroughthe principal wave which will become I o ( r ) = -iIosinhI'(l - r )
G ( r ) = -iKI,coshr(l-

+ IOcoshI'(1- r ) , r ) + KI0sinhr(l- r ) .

(12)

--

Ie (1)

-- I

The propagation constant I' is given by R r = -2K + iB, (13) where R is the resistance (of both cones) per unit length.
The Charge Is Not Generally Proportional to the Voltage

Fig. I-The relation between principal, secondary, and total currents at the ends of an antenna

It is evident that forprincipal waves,the electriccharge

qo( r ) per unit length is proportional to the voltage Vo(r ) =

V ( r ) . The charge q m ( r ) associatedwith a secondary wave, being proportional to the derivative of the current, does not vanish while the corresponding voltage does; this means that the total charge is not proportional to the total voltage. In a simple uniform transmission line the charge is proportional to the voltage and it does not matter whether we write

principal current whichisdiverted into a shunt admittance. Whenever a capacitor (or any impedor) is shunted across a pair ofwires,we can look upon this capacitor as a practical meansforproducing a local fieldsuperimposeduponthe normal field surroundmg the pair of wires. Broadly speaking the capacitor is an irregularity in the transmission line. Sudden termination of thewires in the antenna case is also an irregularity producing a field which is superimposed upon the normal field of the principalwaves guided by the antenna. The totalcurrent I(1) maybedifferentfromzero. For instance, if the tops of the conical conductors large, are appreciable current may flow over the edge; or, if the antenna

SCHELKUNOFF:A N T E N N A S OF ARBITRARY SIZE A N D SHAPE

1171

itself is thin,end cpacitancesmay be provided by a number of wires spreading fanwise from the ends of the antenna wires. However, i the cross section of the ends of the antenna is f small and no loading is provided, then thetotal current is zero and

This is the case with which we are specifically concerned in this paper. The output Impedance of a Transmitting Antenna In Section V we shall prove that if the characteristic impedance of the antenna is largeand if the total currentat the ends of the antenna is zero, then we have approximately I o F( L) - iG( L) , IO K where the phase length of each coneis28
-=
? W A X LCYCTW, L. Y IlAOIANS

Fig. 9-The real and the imaginary parts of the transformedoutput impedance K z / Z , = G(L)+ iF(L), where L = 2rl/h.

(16)

zi=
The functions G( L) and F( L)are given by the series

F2+ G2 sin2L + Fws2L - 2K

F sin2L+-sin2L+K

F2+ G2 ws2 K2

The Care o an Infinitely Large K f

,~~m+~,2(L)~2=+~,*(L).

(18)

Presently we shall givemuchsimplerexpressionsforthese functions. Substituting in (7,we have for the output impedance

If K is infinitely large, then I o = 0 and the current distribution in the antenna becomes V ( r ) = Vocoslp(I- r ) , (23) Z ( r ) = Iosinlp(f - r ) . From (22)we find that the input impedance tends asymptotically to

z, G ( L ) K2iF( L ) = +
-= K2

It is worth noting that at a distance of $ wavelength from the terminals, this impedance appears as

z t

G( L ) + iF(

L).

(20)

Since the input current tends to Io sin L, the complex flow of power is \k = f [ G ( L ) iF(L)-fiKsin2L]Ii. (25) The preceding expressions are the first approximations to the antenna equations.

The graphsof the real and imaginary parts of this transformed impedance areshown in Fig. 9. The Input Impedance The input impedance of the antenna is obtained from the usual transmission-line equations directly from (7); thus we or have (G+iF)cos(L-:)+iKsin(L-:)

Formulas for G(L) and F(L) The above asymptotic formulas provide a very convenient methodfor obtaining G and F functions. Since(25) represents the power flow when the current is sinusoidal with a node at the end of the antenna, G( L) must be the so-called radiationresistance as referred to themaximumcurrent. This resistance is independent of theshape of the antenna and hence is equal to that of a cylindricalwire. The latter can and has been computed either by the Poynting flux method or by the induced-electromotive-force method(see Section VI). The latter method can also be used for computing F(L), except that this time it is necessary to make calculations for conical antennas. As will beseen in Section N,antennas of other shapes than conical will havean extra reactivecomponent whichmustbe attributed to the nonuniformity of theline rather than to the end effect. Thus we have obtained the following expressions G ( L ) = 60(C + log2L - Ci2L) + 30(C + logL - 2Ci2L +Ci4L)cos2L + 3O(Si4L - 2Si2L)sin2L,

Zi= K
KCOS(L=K

); + i ( G +

iF)sin(L-

5)

GsinL i(FsinL - KcosL) (Ksin L + Fcos L ) - iGcos L

Separating the real and the ima%inary parts, we obtain

*It is hoped that our use of L to designate the phase length and the inductance will not lead to confusion. This use of L in the second sense is only occasional and it will be obvious from the context when it is meant to designate the inductance.

1172

PROCEEDINGS OF THE IEEE, VOL. 72, NO. 9,SEPTEMBER 1 8 94

F(L)

6OSi2L

+ 30(Ci4L - logL - C)sin2L

(26)

-3OSi4Lcos2L,

affectsresonantlengths. Section IV.

This factor will beconsideredin

where C = 0.577 . . . is Eulers constant. The method just outlined be successfully can used for obtaining second approximations to antenna problems when terminalconditionsareotherthanthoseconsidered in this paper.
Resonance

Resonance in Lecher Systems

The shortening effect just discussed exists also in Lecher systemsandforthesamereason. A suddendiscontinuity introduces a terminal capacitance. Consider a Lecher system short-circuited with a metal diskat one end and openat the other.(Fig. 11).The open end is not electricallyopen and

From (22) we observe that the input reactance will vanish when

Inasmuch as we are concerned here with approximations as far as the first powers of the characteristic admittance 1/K, we approximate (27) by 2F tan2L = -K . (28) The solution of the above equation is

Hence, the input reactance of a conical antenna vanishes when

-= 41

kh

1-

2F(?) nkK

= I - 120Sikn + 60(-)ktSi2km nkK

Fig. 11-A Lecher system comprised of two parallel wires, short-circuited at the left end with a metal disk. A parallel combination of two admittance boxes on therightrepresents the endeffect;the conductance box represents radiation and the second box the end capacitance. It is assumed that s is substantially smaller than 1.

(30) If it were notfortheendeffect,theresonantlengths of antennas would be given by 21 = kh/2. The end effect makes antennas resonate when they are somewhat shorter than kX/2. For the first three resonances the shortening effect shown in is Fig. 10. For higher resonances the shortening is substantially equal to 45h/K, when k is odd, and 15h/K when k is even. In antennas of othershapesthanconical,anotherfactor

the impedance across it represents the power loss by radiation and the capacitance. end If s is interaxial the separation between the wires and a is their radius, then in the neighborhood of the principal resonancewe have approximately 120(s - a ) G=- 120n2s2 K2h2 K = 12010g (31) a K2u where u is velocity the of light. These values been have computed by the method of the induced electromotive force fromthesinusoidalcurrentdistribution,taken as thefirst approximation to the true distribution. In this case it is easy to show that for the principal resonance h = 41 + 4uCK. (32) Substituting from (31), we have 480(~ a ) h=41+ (33) K For this case, Englund16 has obtained the following relation experimentally h = 41 + 12.4, (34) forwavelengthsfrom 400 to 750 centimeters. In his setup s = 10.1centimeters, a = 0.635centimeter,and K = 332 ohms. Substituting these values in (33) we obtain X = 41 + 13.7. (35) Some of thediscrepancybetweenthemeasuredandthe calculated values is probably due to the fact that the diameter of the short-circuiting disk was only 15.5 centimeters whereas in computing C we have assumed the disk to be large enough for the current in it to produce the same effect on the Lecher system as the image of the system in the disk. Some of the
C=

C ~ Y T E R R T I C IUCEMYCE IN W m

Fig. 10-Deviation of the resonant length of conical antennas from 21 = k h / 2 .

S C H E L K U N O F F . A N T E N N A S OF ARBITRARY SIZE A N D SHAPE

1173

discrepancy (to theextent of a few millimeters) be must ascribed to the second-order errorsin the computed value and in measurements. The more precise value C contains small of terms depending on the length I of theLechersystem; but Englunds formula does not include such a term.
Input Resistance and Reactance Curves

The input impedance of conical antennas may be computed from (22). This impedance depends on the characteristic impedanceand on thelength of the antenna. In Fig. 12 the

/a

Fig. 13-The input resistance of conical antennas.

plIN I A O I A N S
Fig. 12--The input impedance of conical antennas as a function of 2 d , 4 and K. Solid curves represent the real component and the dotted curves the i m a g i n a r y .

resistance and the reactance are plotteds functions of 2nl/A, a where 21 is the length of the antenna in freespace.The characteristic impedance K is the parameter; it is defined by (1) and plotted in Fig. 4. For a vertical antenna of length I above a perfectly conducting ground we can use the same set of curves and only divide by two the ordinates andthe characteristicimpedance. For example,the maximum resistance of an antenna in free space, with K = 1000, is about 5000; for a vertical antenna of the same size, K = 500, and the maximum resistance is2500. In Fig. 13 the input resistancealone is shown. It will be observed that the input resistance depends very markedly on K in the region (shownon a larger scalen Fig. 14) around the i second resonance, or more generally in regions around even resonances. The reactance curves are shown separately in Fig. 15.
Current Distribution

Fig. 14--The input resistance of corucal antennas in the neighborhood of the second resonance.

radiation pattern are not very sensitive to the changes in the current distribution. Radiation patterns will be affected seriously only in those directions in which radiation is small. The current distribution is given by (10) and (11). In Section V we obtain the following approximate expressionfor a typical secondarywave: 30a(4m + 3 ) L r 2 m + 1 ( r ) = Io ( m + 1)(2m + l ) K

In practice, precise knowledge of current distribution is of lesserimportance than knowledgeofthe input impedance. This is because directive the gain of antennas and their

1174

P R O C E E D I N G S O T H E IEEE, VOL. 72, NO. 9, SEPTEMBER 1964 F

r/ I

'A
Fig. 15-The input reactance of conical antennas.

Fig. 17-Secondary current waves for K = l 0 and I = h/2. The o0 in phasewiththedominant solidcurvesshowthecomponents current 1 sin/3(I - r ) , and the dotted curve shows the quadrature , components.

..

..

D l O T I N C f IN WAVfLfffiTMO

(r/A)

Fig. 16-Theamplitudes of thefirst,third, and fifth ( m = 0,1,2) secondary waves as functions of the distance from the center of the antenna

First of all let us consider individual secondary waves. Near r = 0, the amplitude of the ( 2 m + 1)st current wavevaries nearly29 as r2"'+'. Fig.16 illustrates theway in whichthe amplitudes of the lst, 3rd, and 5th ( m = 0,1,2) vary with the distancefromtheorigin. If this distanceislessthan wavelength, onlythe 1st waveis important; at distancessmaller than wavelength only the 1st and the 3rd3' are important. The maximum amplitudes of secondary current waves depend on the characteristic impedance andon the length of the antenna. Fig. 17 showsactualsecondarycurrent wavesfor K = loo0 and I = A/2. Of the components in phase (solid curves)withthedominantcurrent I, sin@(, - r ) , onlythe first is important except at r = 1; of the quadrature components thefirst two aresufficient. Both componentsare inversely proportional to K . In Fig. 18 the solid curve represents the amplitude of the total current, the dash curve shows amplitude the of the component in phase with Io, and the dash-dot curve represents theamplitude of the quadrature component. In this figurethe current does not quitevanishattheend of the antenna; this is because near the very end of the antenna that

Fig. 18-The total current in the antenna of length 21 = A; K = 1 o . oO The solid curve represents the amplitude of the total current; the dash curve represents the amplitudeof the componentin phase with 1 and the , dash-dot curve the is amplitude of the quadrature component.

part of the total secondary current which is in phase with the dominant current term Io sin @( - r ) is determined bya very I large number of secondary-current waves and in computing our curve we have taken into account only two. This situation is closely related to very slow convergenceof the series representing F ( L ) in(18)and it is understandable on physical grounds. The field distortion in the immediate vicinity of a sharp endmustbemuchgreater than elsewhere and more terms will be needed to represent the field accurately. Ifwe dissolve both the current and the charge near the end of the antenna into two quadrature components,thenit becomes evidentthat theslope of thecurrentcurvedepends on the charge at the endand must be quite large. Thus near the ends of the antenna, the current approaches zerovery abruptly. In Fig. 19 the total current in the antenna is compared with the principal current. The difference between the real parts is seen to be quite small; but the difference between the imaginary parts is relatively large, except near the center.
Minimum Amplitude

29More accurately, the amplitude varies as (2m + 1 + 12O/K)th power of r. m'Ile even-order wavesare absent when a generator is the center. at

We shall now find the ratio Zrmn/Zmax the first minimum of amplitude to the first maximum amplitude (counting from the generator). The first minimum relatively close to the generais

SCHECKUNOFF:A N T E N N A S OF ARBITRARY SIZE A N D SHAPE

1175

and Leff = 3.64. Replacing t e ground by the image of the h antenna, we take the characteristic impedance of the corresponding free-space antenna" as 1080 ohms. The ratio Zmin/Zmax is computed to be 0.131. As nearly as we can read form Pedersen's picture, enlarged by King,' the experimental value is 0.132. The ratio calculated by King is approximately 3/34 = 0.0882; this is lower than the experimental value by about 33 percent,thedifferencebeingconsiderablylarger than the experimental error. We can offer no explanation of this discrepancy. King's method is rigorous; and the accuracy of his approximation would seem to beof the same order of magnitude as ours. Momson and Smith32 havemeasuredthe current distribution in a tower 6 feet and 6 inches square 400 feet long. and The diagonal of the cross section is 9.2 feet and rather arbitrarily we have chosen 2 a = 8 feet as the diameter of an equivalent circular tower in our computation of the average characteristicimpedance.Thus wehave obtained K = 516 ohms (for free space;258 ohms for the actual tower above the ground). Thephaselength of the antenna is L = 3.7 and G ( L ) = 134. The calculated ratio is 0.26 and the measured ratio as read from the picture32is 17.5/75 = 0.233.
Transmitting Antennas Fed at an Arbitrary Point and Receiving Antennas

r/ I

Fig. 19-Curves for the total currentand the principal current.

tor where the secondary current wave is very small. The first maximumis farther away where the secondarycurrentis greater; but there the principal current is large. Thus we can obtain the desired information fairly accurately from the principal current

(37)

The maximum value of t i ratio will occur where sinp(fhs r ) is nearly unity and where c o s g ( f - r ) is, therefore, nearly zero. Hence, the maximum value of (37) is nearly unity. The minimum value will occur where cosg(1- r ) is nearly unity. In the v c n t of t i point, we can find a value of B ( l - r ) iiiy hs for which the real part of (37) vauishes; thus, G / K is nearly the minimum of I o ( r ) / I o and

Most of the foregoing formulas have been derived specifically for free-space antennas fed at the center (or for vertical antennas directlyaboveground).Themethoddescribed in Section V is applicable equally well to other cases. Here we shall limit ours to a few general remarks. If two equal electromotive applied are at two points equidistant from the center, then the current distribution will still begiven in the form (10) and (11).However,the ratio I o / I o will have to be determined anew as a function of the distance from thecenter as well as a functionof K and L. If two equal but oppositely directed electromotive forces are applied at points equidistant from the center, then none of the current components in (10) and (11) will be present in the new expression for the current. This is because all the terms in (10) and (11) correspond to currents flowing in the same direction in the two halves of the antenna. The new expression will contain its own principal wave, principal antisymmetric wave, and a set I*( r ) , 14(r), . . . of antisymmetric secondarywaves. An electromotive force Y applied at some point can always be regarded as the resultant of two pairs of forces(V/2, V/2) and (V/2, - V / 2 ) applied at points equidistant fromthe center. The samegeneralmethod can also be used to developa theory of "end-fed" wires by considering a single cone, instead of a double cone.

Iv. ANTENNAS WITH VARIABLE CHARACTERISTIC IMPEDANCES

The Problem
Conical antennas support spherical waves regardless of the magnitude of theconeangle and theirtheory is relatively

There areno measurements of current distribution in antennas with uniform characteristic impedance. However, it is of interest to compare the measurements on nonuniform antennas with values computedfor uniform antennas.In the case of the Copenhagen antennh9 a vertical antema directly above the ground, the weruge characteristic impedanceis 540 ohms
1 76 1

Section IV for a method of computing average i pedances. m 32J. F. Morrison and P. H. Smith, "The shunt-excited antenna," PROc. LRE, VOL 25, pp. 6734%; J ~ n 1937. e "By "equal" we mean equal amplitudes and equal phases, so that at all times the forces act in the same direction.
31See
PROCEEDINGS O T H E IEEE. VOL. F
72,

NO. 9, SEPTEMBER 1964

simple.Antennas of othershapesaredefinitelymorecomplicated.However, if theirtransversedimensionsaresmall enough, they support approximately spherical waves and approximatesolutionsarerathersimple. We obtain these approximations largely on the basis of the physical picture implied by thetheory of conical antennas rather than by a direct mathematical analysis.
Principal Waues

average characteristic impedance is 21 K, = 12010g -, a where a isthemaximumradius of theantenna.These averagecharacteristicimpedancesareshowninFig. 20 as functions of l / a .

Imagine a setof spheres concentric with the generator. Each small segment of the antenna may be regarded as a section of a cone and wemay write the following approximate expressions for the distributed series inductance and shunt capacitance per unit length: L = - l o g - = - l o P- , P 2 g
5 7

$ 5 7

2r
P

(39)
I/a

The transmission equations for the principal waves will be

Fig.20-Theaveragecharacteristicimpedance: (1) cylindrical antenna, (2) spheroidal antenna, (3) antenna shape ( 5 ) shown in Fig.2.

If p - 0, L and C become increasingly more constant as r , varies. We expect,therefore,thefollowingasymptoticsolutions of (40)

V (r ) = { ( Ae-JBr+ Be@),

where the characteristic impedance K ( r, p ) is now a slowly varying function of the distance from the generator
K ( r , p ) = : E l o g - =2 r
P

12010g--. 2r
P

(42)

It is interesting to note that (41) are exactly the approximate expressions obtained by Brillod4 directly from the secondorder differential equations resulting from elimination of either V or Z from (40).
Average Characteristic Impedances

As K , approaches infinity, difference the ( K ( r ,p ) K , ) / K , approaches 0 and the transmission line becomes more nearly uniform. Consequently,we may regard the uniform line with the characteristic impedanceK, as a first approximation to the given nonuniform line. The second approximation will contain terms depending on the reciprocal of K , and these termsarelargeenoughtobe of importanceinengineering applications. For example, sections of transmission lines with variable K ( r , p ) do not resonate at the same frequencies as uniform sections of equal lengths. The effect is of the same order of magnitude as theeffectduetoradiationand may either aid or oppose it. We can compute this effect if we find the second approximation to the input impedance of nonuniform lines; but there exists a somewhat simpler method which will be employed in the following section.
Variable Capacitance and Inductance Affect Resonance Conditions in Finite Sections of Nonuniform Transmission Lines

We shall define the average characteristic impedance K , as follows

K,

/K( r , p ) a?.
0

Multiplying the first equation in (40) by I* and the conjugate of the second by V , adding the results and integrating from 0 to I, we have
V ( I ) I * ( l ) - V(0) I*(O) i w / d [ C W * - LIZ] dr. (47) =

Applying this formula to a cylindrical wire, we obtain

K,

12O(log 7 - 1) ; 21

this is substantially equal to the characteristic impedanceof a cylindrical antenna as given by Peder~en.~ For a spheroidal antenna, the average characteristic impedance is
K,
1201%

If thesection of theline is eitherelectrically open or short-circuited at both ends, theleftside of this equation vanishes and we have

I 0,

(45)

where a is the radius at the base of the antenna. For an antenna of the shape (5) among those in Fig. 2, the

J. C. Slater and N. H. Frank,McGraw-Hill Book Company, New YO&, N.Y.,(1933), pp. 147-148.
X H E L K U N O F F : A N T E N N A S OF ARBITRARY SIZE A N D SHAPE

This is merely an expression of the well-known fact that at resonancetheaverageelectricenergyandtheaveragemagnetic energy are equal. Furthermore, Rayleigh has shown that the first-ordererrors in distribution of V and Z result in second-order errorsin the resonant frequencies. Thus, (48) can be used for an approximate computationof resonant frequencies. First of all, however, we shall obtain two special forms of
I177

(48), especially adapted to the when case 1 LC = - = constant.

V2

but the second gives

(49) 161 -=-1 II I

4 41 - 2.35, - = 1.53, - = 2.41. (58) 2s1

h

8 2

Taking ti relationship into consideration, we rewrite (40) as hs follows:

h2

I-, 4

TITL

Substituting from (50) in (48), we have

In order to obtain apTheforegoingformulasareexact. proximate formulas, we shallassumesinusoidalvoltage-current distributions. T h u s , for the first resonance with one end short-circuited and the other open (let the end at r = 0 be short-circuited and the end at r = I open), we have nr I ( r ) = Zcos -, V ( r ) = Vsin 21 21 (52) Substituting in ( X ) , we obtain

The value of 2d/A as given by the exact theory should be the first root of Jo( x) = 0; this root is 2.40 . . . In connection withantenna problems L becomes infinite in the approximateequations (39) but not rapidlyenough to cause any trouble.U i g these equationswe obtain the followsn ingresultsfor wiresenclosedwithinfictitious reflecting spheres, with their centers mid-points of the wires and with at their surfaces passing through the endsof the wires. For cylindrical wires Si n 120Si n 70.74

x= -

n log - - 1

I=--=--

nKa

Ka

For thin spheroidal wires

x= where
X=

Sin

+ 0.5Si2n = 120(Sin
x log -

+ 0.5Si2n)
TKa

cs7 c o
c d r

I a

nr dr

x=

nr jdccos 7 dr

97.82 --

jdcdr

(54)

I=

Similarly, the resonance both open, for first with ends I ( r ) = I sin ( a r / l ) and

Ka - S i 2 s - 0.5Si4n I 27 log a

12.84 --

Ka

(60)

For wires with diamond-shaped longitudinal cross section Si x 70.74 x = - 712 1 = -- Ka log a Si n - 0.5 Si 2n =-43.66 I21 Ka (61) n log a Resonance in Very Thin Conical Antennas

For the first resonance with both ends short-circuited, we have

It will beobservedfrom (39) and (42) that L isdirectly proportional and C inversely proportional to K ( r , p). Evidently, the approximations corresponding to x and 5 on the one hand and to and [ on the other are not necessarily the same. In fact, in some extreme cases in which either L or C becomeinfiniteatoneendtoorapidly,one of theabove formulas becomes entirely useless. For example, in a cylindrical cavity35 forcylindrical waves C isdirectly proportional and L inversely proportional to the distance r from the axis. For the principal resonance with voltage maximum at r = 0 and 0 voltage at r = I, we have

We are now ready to consider the relations between lengths of antennas of various sizes and shapes and resonant wavelengths, under the condition that K is very large. Subsequently, we shall obtain approximate expressions for theinput impedance of nonuniform antennas from which resonant lengths can be computed for antennas with moderate characteristic impedances. In conical antennas L and C are distributed uniformly and the resonant wavelength affected only by the end reactance. is In so far as the first power of 1/K is concerned this effect is given by (30) which becomes

where

x and x

h TK 21 _ -I - - = F(n) x 7lK

_ - 2F(S) 41 -I--=

+ 60Si2n = I - - 97.82 TK K 1 - 60Si2n - 3 0 S i 4 ~ I - - 12.84 = TK K

1-

120Sia

are given by (54). The first formula is useless for the first and the second resonances, respectively.

(62)

35Asection of a circular cylinder between two parallel planes.

1178

PROCEEDINGS OF THE IEEE.

VOL. 72, NO. 9. SEPTEMBER 1984

ResonanceSpheroidal Very in Thin Antennas

In spheroidal antennas we have to consider two effects: The end reactance and thenonuniform distribution of L and C . Ultimately, as Kincreasesindefinitely,the two effects become additive. Thus for the first resonant wavelength, we have

thin cylindrical wire the equivalent spheroid would have to be very fat in comparison. In fact, ifwe let the radius of the cylinder approach zero, then theratio of the base radiusof the equivalent spheroid to the radius of the cylindrical wire will approach infinity.
The Input Impedance of Antennas with Variable characteristic Impedance

Hence, in thin spheroidal antennas the deviation of the resonant length 21 from h / 2 is proportional at least to the square of l/K. This result a r withAbrahamsformulawhich, gm when expressed in terms of K,, is 41 5040 -= x l - ( K , + 83)2 For the second resonance we have

Wehave already stated that in thefirstapproximationa nonuniform line may be regarded as a uniform line with a characteristicimpedanceequal to theaveragecharacteristic impedance. The goodness of this approximation depends on the relative deviation of K( r, p ) from K , . Even when these deviations are small for sections only afew wavelengths long, they will be prohibitively large for really long sections. For example, the average characteristic impedance of an infinitely long cylindrical wireis infinite; but the input impedance of a cylindrical wire extending to infinity on both sides of a generator iscertainly not infinite. As amatter of fact(see footnote 20), for t i wires this input impedance is approxihn

mat el^^^

Resonance in Very Thin Cylindrical Antennas

K(0)

Similarly at the first resonance in cylindrical antennas, we have

h 12010g 2a - 207.

(69)

This impedance is represented by curve a in Fig. 21. Curve b

_41 -I--x

2F(3 TKa

x=l--= 60 Si 277
TKa

27.08 1--.
Ka

At the second resonance we shall have

ResonanceVery Antennas in Thin with Longitudinal Cross Sections

Diamond-Shaped
Fig. 21-The input impedance of an infinitely long cylindrical wire. Curve (a) is a plot of equation (69); curve (b) is a plot of the more accurate equation(165).

For these antennas we have 41 27.08 -=I--= 60Si2.n 1 --, h VKa TKa 21 _ -1 + -

120Sis - 120Si277 + 30Si4s 30.82 =1+-, nKa Ka

for the first and second resonances, respectively.

Resonant Conditions Depend on the Shape of the Longitudinal Cross Section of theAntenna as Well as on the Size of the Transverse Cross Section

is a plot of a more accurateequation (165). Fig. 22 is a plotof K(0) over a wider range of radii. In practice, however, antennas are never very long and we can treat them as transmission lines slightly with variable characteristic impedances. One solution for such lines, based on Picards method of integratingdifferentialequations,has been obtained by Carson.6 Adapting his solution to our problem and retaining only the first order of correction terms,we obtain the following expression for the input impedance:

In its effect on the resonant length the average characteristic impedance K, represents average the size of the cross section of the antenna. Since K, depends on the logarithm of somemeanradius,theresonantlengthvaries rather slowly with the size of the cross section of the antenna. Thesecond factor is the shape of thelongitudinal cross section. Page and Adams have supposed a cylindricalwire that is equivalent to a somewhat fatter and somewhat longer spheroid. On geometric grounds ti appears reasonable; but hs our computations do not support the assumption. For a very

Zi= K,

G s i n L i [ ( F - N ) s i n L -(K, - M ) c o s L ] [(K, + M ) s i n L + ( F + N)cosL] - iGcosL

36Thereis a reactance in shunt with the resistance (69); this reactance depends on the length of the segment over which the electromotive forceis applied, and has little effect on the total impedance unless the segment is very short. Theoretically, the terminals of the generator could be brought so closetogether as to short-circuit it; but, in practice, this is not done.

ANTENNAS SCHELKUNOFF: SIZE

OF ARBITRARY

A N D SHAPE

1179

Diamond-shaped antennas can be treatedmore accurately if we takecognizance of thefact that thefirst half of the antenna has a uniform characteristic impedance and the second half, nonuniform. Thus wemay obtain the input impedance into the second half from L G ~ h - + i (F-N)~in--(K,-M)cos2 2 . 2 Z, = K , [ ( K , + M ) s i n -L ( F + N ) c o s - 2 -iGcos- L 2 + 2

" [

"1

"1

! !

(76) and, then,computethe input impedance into thefirsthalf using (76) as its terminating impedance.The functions M and N to be used in (76) are38 M(L) N(L)
= 60log2

Fig. 22-'lhe input impedance of the infinitely cylindrical long wire. For 0.1 < X/20 5 3, this curve has been computed by numerical integration from (164);for X/2a 2 500, the curve has been wmputed from the approximate equation (165); the then, two curves have been freely joined together.

+ 60(C + logL - Ci2L)cosL -6OSi2Lsin L ,

21 12010g - . a

6O(Si2L - 2SiL)cosL -6O(C + logL + Ci2L - 2CiL)sinL, (77)

where M and N functions are defined by M( L ) = PI'[ K , - K( r , p ) ] sin2Prdr,

0
0

K,

Input Impedance Curves for Cylindrical Antennas

N(L)

'/'[ K'
R i=

- K(r'

cos2Brdr'

In Figs. 23 and 24 the input resistance and the reactance of (71) cylindrical antennas in free are space shown as functions of

The real and imaginary parts of the input impedance are KaG(Ka+N~in2L-Mws2L) G2cos2L+[(K,+ M)sinL+(F+N)cosLI2

x. K,[~(G2+F2+M2-N2-K~)sin2L+(MN-K,F)cos2L+(MF-K,N)] =
G2cos2L+[(K,+M)sinL+(F+N)cosL]2

(72)

For cylindrical antennasM and N functions become M(L) N(L)

= =

60(log2L - Ci2L + C - 1 + C O S ~ L ) , 6O(Si2L - sin2L). (73)

I / X for different values of K . In Fig. 25 we havethe input resistance at resonance (.X . = 0) in the vicinity of I = X/4. , Fig.shows 26 resonant the impedance in &e vicinity' of
I

For antennas in freespace,with a rhombic longitudinal cross section, or for vertical antennas of triangular shape3' of base radius a, above a perfectly conducting ground, have we 60(C + log2L - Ci2L)(1 - cos2L) -6OSi2Lsin2L, N(L) = 6OSi2L - 60(C + log2L - Ci2L)sin2L, 21 K , = 12010g - . (74) a For spheroidal antennas we obtain M(L)
=

M(L)

60(C + logL - Ci2L) +30(Si4L - 2Si2L)sin2L +30(C

+ log2L - 2Ci2L + Ci4L - 1 ) w s 2 L ,

6OSi2L 30(Ci4L - logL - C - 21og2)sin2L -3OSi4Lcos2L: M(L) = G(L) - 6010g2 - 30(1 - 10g2)Cos2L, N(L)
=

k
Fig. 23--The input resistance of cylindrical antennas in free space. the For vertical antennas over a perfectly conducting ground divide ordinates and 2. , by K

N(L)

= F(L)

- 6010g2~in2L.

(75)

371nverted conical antennas.

'*In (76) G , F, M,and N are functions of L , as indicated by the fiist term in the numerator, while theargument of thesines and wsines is L / 2 .

1180

PROCEEDINGS O THE IEEE, VOL. 72. NO 9, SEPTEMBER 1% F

Fig. 24-lheinputreactance of cylindricalantennas in freespace. For vertical antemas over a perfectly conductingground divide the ordinates K, by 2.

In this latter case, it isknown that the base insulator is responsible for about 30 micromicrofarads in shunt with the antenna; this capacitance is more than sufficient to account for the difference of 2$ per cent between the measured and computed values.On the other hand, the tower is tapered near the base and consequently has less capacitance than it would have had if it were not tapered; we feel confident that if al l these factors aretaken into considerationthediscrepancy betweenthetheory and experiment would be quite small. Hence, it is unnecessary to postulate the existenceof a lumped series inductance of 6.8 microhenrys and alumpedshunt capacitance of 200 micromicrofarads at the base of the antenna in order to explain the measurements. These values were assumed, but not accounted for, by Morrison and Smith in order to bring their measurements into agreement with formulas from Siege1 and Labus. We now believe that the antenna theory is in such a state that accurate results can be calculated if all visible factors, such as base capacitances and antenna shapes are taken into consideration. The imperfect conductivity of the ground does not appear to affect the results. For someunexplainedreasontheimpedancecurvesfor cylindrical antennas publishedbyKingareconsiderably in error in some regions. Thus, when log21/a = 10.58 so that K, = 1150, Khg obtains& R,, = 16,000 and X,, = 9O; OO our values are R , = 5500 and X,, = 3000. Available measurements of R,, support our results (Fig.26).
The Input Impedance for 1 = h / 4 and I
= h/ 2

When 1 = X/4 and

= A/2,

we have, respectively,

CWI~CTCIISTK

IYCCDUCL

IN

onus

Fig. 25-The resonant impedance of cylindrical antennas a a funcs tion of K, when I is in the vicinity of h/4.

z, = K a [ K - M ( n ) l [ G ( r ) - i F ( r ) - i N ( a ) ] . (78)
G Z+ ( F + N ) * The quantity G ( s / 2 ) = 73.13isthe input resistance of a half-wave antenna for K, = 0 ,that is, for an infinitely t i 0 hn
10

IS

Fig. 26-The resonant impedance of cylindrical antennas is a funcn tion of K, when I is i the vicinity of h/2.

c 10 u t

. a

Z T

I = h / 2 . The points (exceptthe first) arevaluesobtained

experimentally Feldmadg for h = 9 meters. by The fiist point u 5 is the value obtained by Momson and Smith32 for the square 2. tower. Resonant lengths of cylindrical antennas are short ofA/2 and h by the percentages shown in Fig. 27; curve 1 is for the principal resonance and curve 2 for the second resonance. The points above curve 1 havefrom beenEnglund,16 taken those 2 above by with haveFeldman been curve one supplied 2 300 um ~ m m o o s o IN OMMI o UOQ ODD c**-n*snc exception, the first point from the paper by Fig. 27--Theresonantlengthofcylindricalantennas i freespace. n and Smith. 2 resonance. Curve 1 is principal for Curve the is for the second resonance. The ordinates representtherequiredshortening in per

I W C -

39Feldmans measurements were made for vertical wiresdifferent of sizes; we have doubled his results to obtain the values corresponding free-space tocondition. the

cent.

@Kings values have

been doubled to obtain the free-space figures.

SCHELKUNOff: A N T E N N A S Of ARBITRARY SIZE A N D SHAPE

1181

wire. For any finite value of K,, the input resistance depends also on M ( n / 2 ) , that is, on the shapeof the longitudinal cross section of the antenna. For example, cylindrical for and spheroidal antennas M ( n / 2 ) isequal,respectively,to -21 and 41. Hence, when 1 = A/4, the input resistance of cylindrical antennas is somewhathigher than 73.13 and the input resistance of spheroidal antennas is somewhat lower than 73.13. When 1 = A/4 and K, = w , the input reactance of cylindrical antennas is 30 Si 2n = 42.5 and that of spheroidal antenias is 0. For cylindrical antennas with finite characteristic impedances the input reactance is somewhat higher than 42.5 and for spheroidal antennas it is still zero. These results do not include terms depending on l/K,. From Abrahams expression for the resonant frequencies and from (70), we find that theapproximatevalue of the input reactance of thin spheroidal antennas, when 1 = A/4, is 7900(K, - 41)/( K, + 83)2.
Bent Antennas

When these functions have been computed, the input impedance is obtained from (70).
Loop Antennas

If an antenna is bent into a complete loop and the ends are joined together, the procedure for calculating G and F functions is the same as that outlinedaboveexcept that the sinusoidal current distribution mustbechosen to have an antinode instead of anode at the far end of the antenna. Naturally, M and N functions are not affected by the conditions at the far end.
Current Distribution

Our method of antenna analysis is applicableto bent anten-

nas (Fig. 28). From the theory of principal waves guided by

In order to findthecurrent distribution in nonuniform antennas we should calculate the principal current distribution from Carsons equations,6 assumingthat the transmission line is terminated by the output radiation impedance as given in this paper, and superimpose, on it thesecondary current distribution. The latter can be obtained from (11) and (36). Knowledge of current distribution is needed largelyi comn puting radiation patterns, and the& are affected by the ground conditions to a much greater extentthan by the changesin the current distribution. It is doubtful, therefore, that the labor expended in calculatingcurrentdistributions,light as itis, would be commensurate with the practical value the results. of However, the currentdistribution can be measured and, hence, used for checking the theory.

V. ANTENNAS UNIFORM WITH CHARACTERISTIC IMPEDANCES: MATHEMATICAL ANALYSIS

A set of equations for transverse magnetic sphericalwaves and its general discussion may found in previous paper^.^',^^ be This set will now developed be to suittheneeds of our particular problem. In thefirstplace we shallassume that d / a @ = 0. It may appear that this assumption will restrict the analysis to coaxial cones; in reality, however, the solutions of the restricted equations canbe used for the more general case of two conical wires inclined to each other. Thus, if d / d @ = 0 our fundamental equations become
Eo = - -, ae

Fig. 28-A bent antenna.

thin diverging conical wiresz1we observe that the capacitance between two elementsat A and A dependsonly on the distance between them and on their radii. For equal radii, we shall have then
(79) For unequal radii, a is the geometric mean of the actual radii. From (79) we can obtain the average characteristic impedance; then from (71) we compute M and N functions which are needed to correct transmission equations of the principal waves for the nonuniform distribution of the inductance and capacitance. The function G ( L ) is obtained from (25): First we compute the radiated power on the assumption that a = 0 and that the current distribution is sinusoidalwithcurrent nodes at the open ends of the antenna; then, we divide this power by f of the square of the current amplitude. This calculation can be carried out by either of two methods: (1) ThePoynting flux method and (2) theproduced-electromotive-force method. Only the second of these methods can be usedforcalculating F ( L ) ; moreover, in t i calculation a hs must be assumed not merely infinitely small but also varyins with x in such a way that d ( x ) / a remains constant.

av

iocr2E, = n ( n

+ 1)A,

where n is a constant depending upon the boundary conditions at the surface of the antenna and the flux function A is
A + = f i [ J * + l / Z ( B r ) - iN+1/*(B4]( 8 ) , T

Thefunction A definestheoutward-boundprogressive wave while A - the standing wave finite at the origin r = 0. The function V(r, 8 ) is the potential defined as the line integral of E, along any path beginning on some fixed radius 8 = 8 and containedentirelyintheco-ordinatesphere of radius r. The function T ( 8 ) is a solution of Legendres
41W. L. Barrow, L. W. Chu, and J. J. Jansen, Biconical electromagnetic horns, h c I.R.E., 27, pp. 769-779; December, 1939. o. vol.
+

1182

P R O C E E D I N G S OF THE IEEE. VOL 72, NO. 9. SEPTEMBER 1%

= plogcot

-,
2

if n

0.

(82)

The function P ( x) is so defined that , P,(1) = 1.

(83)

where 8, and 0, are the angles made by typical radlus with the axes of the conductors. If the angle between the axes is 5, we have approximately pP,( -cos+,) + qPn( -cos{) = 0, pP,( -cos{)

If n is not an integer, P,(x) has a logarithmic singularity at x = -1. If n is an integer, P n ( - x ) = (-),Pn(x) and (82) is not the most general solution of Legendres equation. There is a second solution Q,(x) which singular two is at points x = kl. The boundary conditions are: the electric intensity E, must vanish at thesurface of the antenna andthefieldmustbe finite. If the conical conductors are of finite length, the entire space must be divided by co-ordinate spheres regions: (1) into free from conductors, (2) containing only one conductor, and (3) containing two conductors. In the region free fromconductors n must necessarily be an integer because for nonintegral values of n Legendres equationpossesses no solutionsfiniteforallvalues of 8. The function Q,(COS~) being singular at 0 = 0, n is also out of the picture. Thus in such regions the complete solution is a series of functions P,(cos e ) correspondingtodifferent integral values of n. In the region containing only one cone with its axis along 0 = 0, the coefficient q in (82) must vanish because P,(cos~) is infinite where 8 = n. Thus (82) becomes r(e)=pP,(-cose). (84) Thesingularity of this function at 0 = 0 is, of course,excluded by conductor. the If is angle the made the by generators of the conical conductor with the axis, then E, will vanish on the conductor if

+ qP,( -COSJlz)
=

= 0.

(92)

The equation for n will then become P,(-cos+,)P,(-cos+,) [Pn(-cos5)]2. p

=

(93) (94)

For equal cones, we have P,(-cos+) = kP,(-COS{),

Tq.

Inasmuch as withreference to a single co-ordinate system T(B,, e,) is a function oftwo spherical co-ordinates 8 and (9 thefield, if wanted,mustbecomputedfromthegeneral equations2 and not from (80). The boundary condition E, = 0 on the surface of the antenna will also be satisfied (see (80)) if E, vanishes everywhere; this happens when n = 0. In t n casethe nontrivial h solution is singular for two values of B so that the wave can existonly in thepresence of two conductorsexcludingthe corresponding from field. radii the This is the socalled principal wave and its properties have alreadybeendiscussed in detail2 In theregioncontainingtwoconductors,thetransverse voltage V , ( r ) between the conductors, corresponding to any transmission mode except the principal, vanishes, V ( r ) = 0. (95) This is true because V( r ) is the difference between the values of V ( r ,0 ) on thetwo conductorsandthesevalues,being proportional to T( e), vanish with T( e); and it is only for the principal wave that T ( 0 ) does not vanish on the conductors. Hence, the total transverse voltage V ( r ) between two conductors depends entirely on the principal wave. Let Zi(r) be the electric current in the upper cone of the double cone and:Z r ) the corresponding currentin the lower ( cone (Fig. 5). Assuming that the upward direction is positive, we have dA Z i ( r ) = . 2 a r s i n + H , ( r , + ) = -2nsin+-,

P, ( -cos +) = 0.

(85)

This is the equation for the constant n. In the region containing two conductors, one with its axis along 0 = 0 and the other along J = n, we must have T ( + , ) = T ( a - 4,) = 0 and therefore
P P n ( - c o S + , ) + qPn(cos+1) = 0, pPn(cosJI2) + qPn(-cos+2)
=

0.

(86)

a+

Eliminating p and q, we have the equation for n,

In(r) = 2 n r s i n + H , ( r , r - 4 ) = 2 n s i n + -( aA

ae

e-n-$

.
(%)

If the cone angles and + 2 are small their cosines are nearly equal to unity, P,(cos+,) = P,(cos+,) = 1 and (87) becomes approximately P,(-cos+,)P,(-cos+,) = 1. (88) If thecones are equal = q2 = P , ( - c o s + ) = *P,(cos+),

+,

For all values of n > 0 the function ~ N , + , , , ( x ) is infinite at x = 0; therefore, the proper expression for A is A - (see (81)) and

+ then
and p
Tq.

(89)

For small values of +,this becomes P,(-cos+) = f l .

(0 9)

Since for n > 0, the function &J,+1,2(x) vanishes at x we have

0,

If the two cones are inclined toward eachother and if their angles and q2 are so small that the proximity effect is negligible, we can express the total field as the resultant of two fields, associated with the individual cones, and the function

+,

I i ( 0 ) = Zn(0)
At the apex the electric current principal wave.

0.

(98)

is determined solely bythe

S C H E L K U N O F F . A N T E N N A S O ARBITRARY SIZE A N D SHAPE F

1183

Thus, we have justified equations (10) and the equivalent circuitfor double cones of equallengththat follows from these equations. It was shown there that the first approximationforthe output impedance (19) can be obtainedinthe form (26) by a method quite different from the present and yielding the results in their simplest form. On theother hand, the present method permits the use of successive approximations and is quite general. For this reason, we are justified in developing it still further. In the case of a double cone (Fig. 5 ) different transmission modes are determined by the roots of (89) or, for small cone angles +, by ( 0 .The latter equationisa verygood ap9) proximation even if is as large as 0.1 since cos = 1 - f ~ . will differ from unity by less than 1 per cent and Pn(l)= 1. From the theory of Legendres functions, we have

When n is aroot of (102) then by virtue of (89) the corresponding field function T ( 0 ) is proportional to
%m+l(e)= $[P2m+l+A(~S0) P2m+l+A(-c0se)], (109) where A = - 120 K The derivatives of the L functions at B = and 0 = n - $ are approximately

dL - = (-+ )

d L ( n - +) - SinnA - 120 d+ m+ $ K+

P,(-cose)

a=O

sinnn

( - ) r ( n + a + 1) [Ilogsin r ( n - a + 1)

3 e

(111) Hence,theelectriccurrents in the antenna, associatedwith these waves are nearly inverselyproportional to the characteristic impedance and they flow in the same direction at points equihstant from 0. When n is a root of (104), the function field T ( 0 ) is proportional to

where
q X

+ 1) + 1)

+ ( - x - 1) = + ( x )
9

+ ncotnx.
(100) to a = 0 is

L2m(0) In this case

i[p2m+A(cose) + p2m+A(-cose)]. (112)

For smallvalues of $ thetermcorresponding dominant and we have approximately 2 P,(-cose) = - sinnn a +cosnn. Hence, the equation
(102) P,(-cos$)
=

and the electric currents at points equidistant from 0 are in opposite directions. For smdl values of A , we have approximately
P,+,(coSe) = p,(cose)

(101)

+2A P,(cosO)logCos

~,+,(-cose)

(-)p,(cose)

becomes

s;=
where K is characteristic the impedance the to principal waves.* Similarly, the equation P,(-cos$) = -1 becomes

(-)(n + a)! 1 a!a!(n- a ) ! n + a

(-

1 + +n+a-l

... +n + l

)sin2az, e
1

s;=
(104) 1

a=l

(-)(.+a)! a!a!(n- a ) !

(-n + a
n + l

1 + n + a - 1

... +-

e ) cos27.
(114)

As the characteristic impedance K increases indefinitely, the

roots of (103) approach
120 n = 2 m + 1 + -, rn=O,I,2;.-. (106) K Likewise, the roots of (105) approach 120 n = 2m -, m = 0,1,2, . (107) K The roots of the characteristic equation (85) for a single cone approach 60 n = m + -, m = 0 , 1 , 2 , . . .. K (108)

Also, A

- 0, we have ,
L,(q
--*

e - .

(115) Thus, the characteristicfunctionsfor theregioncontaining two cones approach the characteristic functionsfor free space except that free space cannot support the principal wave. This property maybeused .for obtaining the second approximation to the solution of the antenna problem. We shall write general solutions appropriate to the free space, and to the antenna region within the sphere of radius I (Fig. 5). The two solutions must satisfy certain continuity conditions at the boundary sphere and it is in ti matching operation that hs (115) is useful.

P,(cOse).

1184

P R O C E E D I N G S OF T H E IEEE, VOL. 72, NO. 9, SEPTEMBER 1%

In the antenna region we can write

Hence, in order to find bk only have to expand sin cos e) we ( j?I into a series of Legendres functions. This expansion is known to be
m

(116) where J ( r ) is the total current in the upper cone, associated with all waves except the principal. We shall call this current complementary. The summation in (116) is extended over the roots of (89), excepting n = 0. In the free space we write

m-0

Substituting in (122) and comparing the result with (123) we obtain

At r

I, C must be continuous
X

k=l

But as K + 00 and J, + 0, we have (115) and, therefore, in the limit the coefficients a, and bk must be correspondmgly equal. The coefficients b, can be easily determinedfor the limiting case K = 00. In the first place f ( r ) 4 0 as K -P co. This can be shown by substituting from (111) into (116) and obtaining

Theonlytermwhichwouldnotapproachzeroistheterm corresponding to m = 0 in (107). This term is entirely absent, however, in the symmetric case of Fig. 5. In the case of cones of unequal length this term would be present and we would exclude it from I ( r ) and consider it separately as the principal antisymmetric w e n t wuw. In either case, the conclusion is that as K -P co, the current distribution in the antenna ten& to become strictly sinusoidal. If the current distribution is known, thefieldcanbe determined by the retarded-potential method. For a sinusoidal distribution of amplitude I,, with current nodes at the ends of the antenna, we have

Letting in (119) a , = b2m+l,we obtain theasymptotic value for the complementary current given by (16) and (18). as Theabovematching of radialelectricfields in thetwo regions can be regarded as the second step in a sequence of successiveapproximations. If we were to compute E@,we would discover thatit is discontinuous at the boundary sphere r = 1. By adding a proper expression to the external field, we can make E# continuous again and after matching E, for the second time we shall have a more accurate expression for the complementary current. In all these higher approximationswe no longer can make the coefficients u, and b, equal but have to expand the functions to be matchedin series of Legendres functions, Ln(e) or P (cos e) as the case may be , These and, functions therefore, are the coefficients of expansions are expressed by definite integrals.

VI. THEINDUCED-ELECTROMOTIVE-FORCE
METHOD OF COMPUTING RADIATION AND THE RADIATION PARADOX
The induced-electromotive-forcemethod43 computing of radiation consists in obtaining the work done (and thus the energy contributed to the field) in driving given electric currents against the electric forces produced them. This method by furnishes not only the power lost by the source in radiation but also the average reactive power, is, average that the interchange of energy between the field and the generator. For a linear current filament situated along the z axis and extended from z = - I to z = I , the complex power \k is

c = 2~rsineH+
=

- i ~ , ( e - ~ B r c o s - / te-JSr1 - l1 e - J B ~ ), (120) ~ z

wherer1 and r, are distances from the upper and the lower ends of the antenna. At great distances from the antenna this becomes

c = iI,[cos(pcosB)

- cos/II]e-@.

(121) where E, is the electric intensity parallel to the filament at the

421t should be noted, however, that Lo(8) not a member of the is orthogonal set [ L.,(8)] but dLo(8)/d(8) is a member of the orthoge nal set [dL,(B)/dB]. 43For a brief history of this method see S. A. Schelkunoff, A generalradiationformula, h c I.R.E., 27, pp. 660-666; Oco. vol. tober, 1939.

From Maxwells equations we have

1 a c 2niwcr2E, = -- = igl sin (PIcos e) e-@,. sme ae

On theother hand, at greatdistancesfromtheantenna (117) becomes

S C H E L K U N O F F : A N T E N N A S O F A R B I T R A R Y SIZE A N D SHAPE

1185

In viewof (23) and (25), G ( L ) and F ( L ) can be obtained correctly in this manner. In (126) the quantity -E,(z) is the applied electromotive force perunitlengthand,therefore, - E,(z)/Z(z) isthe radiation impedance per unit length. If the antenna is perfectly conducting,t i impedance is 0 along the entireantenna hs except in the region from z = - s to z = s, occupied by the source of energy. The conclusionis not startling, since afterall energy does come out of its source. What we have just said does not vitiate our previous assertion to theeffect that an approximatevalue of 9 can be obtained from an approximate value of Z(z). It is true that if Z(z) istaken to beasinusoidalfunction of z, then E, is different from zero and it would appear that there exists a nonvanishing distributed radiation impedance; butthis means Since action equals reaction, we have merely that the power given out by the point generator is approximatelyequaltothepower given outbyacertain V+ E,(z)dz = 0, (129) continuous distribution of generators along the entire antenna. -S Z(z) isreplacedbyamore where V is the applied electromotive force, that is, the force If thesinusoidalfunctionfor accurate function, then E, wl become il smaller along the transferring the charge from one wire to the other, while the greatest part of the antenna but it will become larger in the integral is the counterelectromotive force of the field. Hence, interval from z = - s to z = s. What the discussion in con(128) may be rewritten as nection with ( 7 ) tells us is that if K is large, the more accurate 9 = in*. (130) expressionfor 9 will notdiffer verymuchfromthe one obtained on the basis of sinusoidal distribution of current. In order to find 9, is necessary to compute the current Z it On the other hand, it is perfectly obvious that in the case of flowing through the generatorin response to the electromotive an antenna fed from a point generator the hstributed radiaforce V-a major problem in itself. tion resistance - E,(z)/Z(z), obtained from sinusoidal distriThe approximate method of computing 9 is based on the bution, should not possess any significance sinceis bound to it followingfacts: (1) In thin wiresthecurrent Z(z) isnearly approach zero as I ( z) is made to approach its exact value. sinusoidal, (2) the field and, in particular, E,(z) can be Feldman has communicated to the author that by taking the calculated from Z(z) by the retarded-potential method, and real part of - E,(z)/Z(z), computed on the assumption that (3) the error in the field becomes smalleras the error in I( z ) I( z ) is sinusoidal, and regarding real part as a distributed this becomes smaller. This method has already been applied sucseries resistance addedby radiation tothe ohmic resistance, he cessfully to find the approximate power radiated from thin computed the current and charge distributions along a long cylindrical antennas. wire.Thenhemeasuredtheshape of the minima of the In the c s of conical antennas we have ae current and of the charge. The measured values agreed well 1 1 with the computed values. = E r ( r ) z * ( r ) dr, (131) 2 -I Now, there is no question that there exists a distributionof seriesimpedance that will representcorrectlytheeffect of for the average work done order to maintain electric current in radiation on thetotal current andthetotalcharge in the in the lateral conicalsurface. To ti we mustadd another hs antenna (Section VII). What is surprisingisthat this resisintegral if the top and bottom surfaces of the cone are large tance has been obtained fairly accurately as the real part of enough to make a noticeable contribution to9. is from this It - E,(z)/Z(z). After all, i this were acorrectmethod f of integral that we have computed F ( L ) as givenby (26). For finding the equivalent distributed resistance, should expect we ti antennas G ( L ) is independent of the shape of the antenhn that its exact value will be found when the exact valueof I ( z) nas and can be obtained equally well from (126). In order to is used.Andyet if theexactvalue of Z(z) i used,then s carry out the necessary computations, the following expres- E , ( z ) / Z ( z ) must be equal to zero because this is precisely sionsforthefieldproducedby an infinitely thin filament the boundary condition fromwhich the exact current must be supporting sinusoidal current have been used obtained! Z(Z) = Isin@(/- z), z > 0;

surface of the latter. If the filament consists of two perfectly conducting wires, one extending from z = - 1 to z = - s and the other from z = s to z = I , and if the generator between z = - s and z = s consists of some means for transferring an electric charge from one wire to the other, then (126) becomes 1 s 9= E,(Z)Z*(Z)dz (127) 2 --s since E, = 0 on the surface of a perfect conductor. Furthermore, if we assume that in the short interval(-3, s) occupied by the source energy the currentI( z) is equal to a of constant I , (127) becomes

--I

IS

-1

= Zsin@(l+z),

z < 0;
e-iflr2 e-iflrl

cos@/ -

-r1

r 1

VII. CURRENT-CHARGE EQUATIONS

The customary form of transmission equations for a line with distributed series resistance R , series inductance L, and shunt capacitance C is dV dl (133) - = - ( R + i w L ) Z , - = -iwCV.

dx

dx

2rrpH, = -iZ(e-flrcosPI - ie-prl - fe-iBr2). (132)

In theseequations rlr r , and r, are,respectively,thedistances from the lower end, the center, and the upper end of the antenna; p is the distance from the z axis.
18 16

Since the electric charge per unit length is q = CV, the above equations can be written also in the current-charge form

dZ - =
dx

- ioq,

provided, however, that C is independent of x . The second

P R O C E E D I N G S O T H E IEEE. VOL. 72, N O . 9, SEPTEMBER 1984 F

equation of this set expresses law the of conservation of electric charge. For a conical antenna (for any antenna for that matter) the second equation is satisfied automatically dl - = -ioq. dr (135) Differentiating ti with respect to r, we have hs

In the neighborhood of the generator this becomes approximately

and, near the ends, R=

5760 n4(1 - r )
=-

59 I- r'

It has been shownin Section V that the total current I ( r ) is the sum of the principal current Z,(r) and an infinite number of complementary currents corresponding to characteristic values of n

The approximate expression for R obtained from -E,( z ) / I ( z ) , on the assumption that Z(z) is sinusoidal, is 1201 R= (1' - r z ) ' f i s expression agrees quitewell with ours (Fig. 29). However,

Z(r) = lo(') + Z Z A r ) .
n

(137)

Each component of the current satisfies the following equation2'

(138)

Substitutingthesecondderivativefrom (136), we obtain

this equation into

indefinitely, the all We have seen that as K increases complementary current waves approachzero;consequently the current-charge equations for theantenna become substantially the equations of a uniform transmission line. Z I ( r ) i f s known, we can determine RC and LC so that (139) would become identical with (134). Assuming that C is independent of r and is equal to the capacitance for the principal wave4 --1 _ C= J, vK' log cot2 and comparing (134) and (139) we obtain

DIIIA*U

IN Y C I L R S

Fig. 29-Distribution o radiation resistance in thecurrent-charge f transmission equations. The calculated curye is for 21 = A/2; the experimental pns obtained from Feldman, are for 21 = 2.5A. i t,

there are no theoretical objections to (143) a thereareto s (146). If the antenna is so short that Bl is substantially smaller than unity and if weneglectthosecomponents of the total current which depend on K , then approximately R=
1607r2I2 ( I - r)h2

Taking Z(r) and I,( r ) from the equations contained in the main section of the paper we can replace (141) by the following approximate expression:
m

The total radiatedpowercalculatedfrom (147) isfound to agree with the well-known expression for the power rdated by a short doublet.

6 0 C (4m ~
R+ioL=iBK+
m=O

s i n B ( I - rF+ F ) - iG

cos/3(I-r)+10

1
+ 3608

+ remainder.

(142)

In order to obtain an expression for R when I = A/4, we need only one term in the numerator and we may retain only the principal term in the denominator. Thus we shall have
- cosj3r

Besides introducing a distributed resistance into the rent-charge of the radiation modifies the distributed inductance. Thus for Bl = r / 2 we find avvroximately (retaining only the first te& i tie summationj n Sin p r -- cosgr
Br

R= l(Br)2cos/?r

(143)
wL = BK

(148)

"We must bear i mind that nothing can be deduced from (139) n concerning the nature o C. In this respect the total-current-total; f charge equations are unlike the principal-current-principal-charge equations. length unit
OF ARBITRARY SIZE A N D SHAPE

(Bd2cosBr

The first term of this expressionrepresentstheseriesreactance per conical of anlong infinitely antenna while the
1187

NTENNAS SCHELKUNOFF.

second term is the end effect. N a the center of the antenna er (148) becomes

wL = BK + 1208. (149) Since the wires are perfectly conducting, E, must vanish on thesurface of eachwireexcept at points of application of In the neighborhood of the ends of the antenna, the added electromotive forces. Ihus, we have reactance depends appreciably on other terms in (141); there the series converges slowly and even the approximation for the reactive part used by us in (142) is not good. It is worth observing that if the series inductance is obtainedfromtheimaginary part of - E , ( z ) / I ( z ) assuming I ( z ) to be sinusoidal, the result where the subscripts designate the values the corresponding of 120r functions on the wires. Subtracting, we have wL= - 8r tan (150) l2 - r2 is very disappointing and may be regarded as an additional reminder that thecomparative success withtheresistance must have been purely coincidental. Thus fromatheory in which antenuas are regarded as multiple transmission lines, we have obtained two restricted Thus, the voltage across parallel perfectly conducting wires and theories. In the first of these radiation appears as a terminal the dfference o vector potentials satisfy theequations of a f impedance and in the second it appears as a distributed series uniform nondissipative transmksion line. The phase velociry is impedance. There is no inconsistency between these two views the velocity o light. f and both are valid provided they are suitably qualified. Thus On the other hand, the electric current I(z ) does not satisfy in the second theory, we think in terms of electric charge and thetransmission-lineequations.Thevectorpotential n is electric current and must not assume the usual transmissiondetermined by the completecurrent distribution line relations between the voltage and the charge. In the first theory, on theother hand, allthe usual transmission-line relations hold but the voltage, the charge, and the current in the transmission representing line the antenna correspond where r is the distance between the points z and z1 and the respectively to thevoltage,the principal charge,andthe integration is extended over both wires. If the distance beprincipal current in the antenna tween the wires is small compared with the wavelength and with the length of the wires and if the wires are energized in push-pull so that .the currents in the wires are equal and WII. WAVE PROPAGATION ALONG oppositelydirected,then n, - 112 is substantiallyproporPERFECTLY CONDUCTING PARALLEL tional to the electric current at the corresponding point and WIRES (155) turns into the usual engineeringequationsgoverning transmission of waves on parallel wires. Consider a pair of parallel perfectly conducting wires (Fig. The current in two ti infinitely long perfectly conducting hn 30), energizedbyanynumber of point generators.The parallel wires, energized in push-pull can be expressed in the electromagnetic field on or outside the wires can be expressed following form45

form in the following general E = - iwpA - grad Y , H = CurlA, divA y = -iwc

where V,

electromotive force distributed in push-pull in two short segments of length 2 7 , one on each wire, a = the radius of the wires,

= an

where A is the retarded magnetic vector potential and Y the retarded electric scalar potential. I the yires are so energized that theelectricconduction f current is strictly longitudinal, and if the wires are thin, the current is always at least approximately longitudinal; the only nonvanishing component of A is the one parallel to the wires A, =A, = 0, A , = n. (152)
In this case we obtain from (151)

Fig. 30-A diagram showing two perfectly conducting parallel wires.

In order to obtain this equation as well as equation (163) we express the pulse function, representing the applied electromotive intensity - E z , as a contour integral; theq we obtain themagnetic intensity H+ as the quotient of -E, andthe radial impedancz; and finally we compute the ament from I ( t ) = ZnoH,(n, z ) .

1188

P R O C E E D I N G S O T H E IEEE, VOL. 72, NO. 9, SEPTEMBER 1% F

s = the interaxial separation between the wires, z = the distance from the mid-points of the generators, and C = the imaginary axis in the y-plane indented around y = k i p (Fig. 31(A)).

at y = Si03 is n/2. The only Thephase of approximation involved in (157) isthat the radius of the wires

/ -

produced by the principal currentwave. The accuracy of such expressions increases increasing with characteristic impedance K. Let a pair of parallel wiresof radm a , whose interaxial separation is s, be open at both endsand beenergized in push-pull; then, in free space, the radiation can be represented by parallel combinations conductance G and suscepof tance B = OC as shown in Fig. 32, where

G-

607r(s*

- a)

A*

, c=

6O(s - a )

vK2

I*?

120n(s - a ) B=wC= K ~ A We omit the details of calculation.

(A)

1
(B)
Fig. 31-Contours of integration.
I(Z) =I(z)

Fig. 32-The equivalent circuit for two parallelwires, open at both ends and energized in push-pull.

besmallcomparedwiththeir interaxial separation; consequently, the proximity effect is not included in (157). If z is positive, the contour (C) can be deformed into (C) as shown in Fig.31(B).The integration around the point y = -is which happens to be a pole as well as a branch point of the integrand, yields the following term

If the wires are imperfect conductors, the fist equation of the set (155) becomes

This principal current wave is unattenuated and is substantially independent of T if 2777/A is small compared with unity. The remainder

- I,(z)

(159)

where 2, and Z, are the surface impedancesof the wires. The added terms affect the voltage across the pair continuously. Theprincipaleffectisdue to dissipation of energyinthe wires; but since the current distribution is affected by radiation, the latter also will produceacontinuouslydistributed effect on the voltage. The latter is, however, a second-order effect;besides, it is concentratedlargely in theimmediate vicinity of the ends, even though mathematically speaking it extends over the entire length.
IX.

is a purely local wave in the neighborhood of the generator and it represents the effect of radiation as well as the local reactance dependmg on T . Radiation introduces a resistance in parallel with the applied electromotive force. If the interaxial separation is small compared the with wavelength, this resistance approxiis mately K2A2 R= 60n2s2 If thewires are of finitelength,theexpressionforthe current becomesvastlymorecomplex.Theresult,however, will consist of two parts, one of which will representthe principal current wave and the other local waves in the neighborhood of the discontinuities, that is, ends of the wires or generators. It is fortunate, however, that the impedances introduced by radiation at the points o discontinuity can be f calculated with good engineering accuracy without setting up exactequations but using in the proper mannerthefield

ON THE IMPEDANCE AN INFINITELY OF LONGPERFECTLY CONDUCTING CYLINDRICAL WIRE

It is readily shown4 that the current in an infinitely long perfectly conducting cylindrical of radius a is wire

ipab I(z)= -

~ Y T K ~ a m ) - j (

qT (C) y / m K o

( -i

eyzdy,

a / m )

(163) where V, is the applied electromotive force uniformlydistributed over a section of length T and C is the contour shown in Fig. 31(A). Deforming (C) into (C), we can show that the real component of the input admittance is substantially independentof T

SCHELKUNOfF A N T E N N A S Of ARBITRARY SIZE AND SHAPE

1189

where a + i/3 is thepropagationconstant. If al is small compared with unity, then we have approximately

After some tedious transformations we have obtainedthe following approximate expression (when q = 120r) for ti hn wires 120M K(0) = log 2 3.2 - (log 2)2 1+-2M 4M2

For a nondissipative line with a terminal impedance the far end, we have Z,cosfil+ iKsin81 Zi= K Kcos81-k iZ, sin81 Multiplying the numerator and the denominator K / a l and comparing with (169) we obtain

Z,at

of (168) by

M = log-2 r a

x + flog2 - c,
4

where C = 0.577 * * * is E d d s constant. AS u approaches

= 1201%

0, (165)

x 5 - 207. (166)
~

where R is the resistance per unit length of the dissipative line. The same input impedance will be obtained, of course, if the nondissipative line is electrically open and at the point wavelengthfromthe open end there is an impedance Z, in series with the line, defined by

AN APPROXIMATION OF A D I S S I P A TRANSMISSION LINEBY A NONDISSIPATIW LINE WITH AN EFFECTIVE LOAD AT THE LAST NODE ANTINODE OR

x.

N T

The input impedance of adissipativeline electrically open at the far end, is Zi = Kcoth(a + i f i ) l ,

of length I,
(167)

It is to be expected, therefore, that the electric currents in these three different cases will be approximately the same in the neighborhood of thegenerator. If the line is about f wavelength long, the minimum current will OCCUT close to the generator and its measured value will not enable us to decide which, if any, of the three above-discussed distributionsof loss happens to be the true distribution. At the current antinode, the electric currentsin the three cases differ by larger absolute amounts;these differences are, however, small compared with the total current.

P R O C E E D I N G S OF T H E IEEE, VOL. 72, NO. 9, SEPTEMBER 1984