Surface Integral Equation Method

8/12/2019 Surface Integral Equation Method

1/28

Progress In Electromagnetics Research, PIER 52, 81108, 2005

SURFACE INTEGRAL EQUATION METHOD FORGENERAL COMPOSITE METALLIC AND DIELECTRICSTRUCTURES WITH JUNCTIONS

P. Yla-Oijala, M. Taskinen, and J. Sarvas

Electromagnetics LaboratoryHelsinki University of Technology

P.O. Box 3000, FIN-02015 HUT, Finland

AbstractThe surface integral equation method is applied for theelectromagnetic analysis of general metallic and dielectric structures ofarbitrary shape. The method is based on the EFIE-CFIE-PMCHWTintegral equation formulation with Galerkins type discretization. Thenumerical implementation is divided into three independent steps:First, the electric and magnetic field integral equations are presentedand discretized individually in each non-metallic subdomain with the

RWG basis and testing functions. Next the linearly dependent and zerounknowns are removed from the discretized system by enforcing theelectromagnetic boundary conditions on interfaces and at junctions.Finally, the extra equations are removed by applying the wantedintegral equation formulation, and the reduced system is solved. Thedivision into these three steps has two advantages. Firstly, it greatlysimplifies the treatment of composite objects with multiple metallicand dielectric regions and junctions since the boundary conditions areseparated from the discretization and integral equation formulation.

In particular, no special junction basis functions or special testingprocedures at junctions are needed. Secondly, the separation of theintegral equation formulation from the two previous steps makes iteasy to modify the procedure for other formulations. The method isvalidated by numerical examples.

1 Introduction

2 Step One: Presentation and Discretization of SurfaceIntegral Equations in Dielectric Subdomains

2.1 Presentation of EFIEs and MFIEs

2.2 Basis Functions


2/28

82 Yla-Oijala, Taskinen, and Sarvas

2.3 Discretization by the Galerkin Method and RWG BasisFunctions

3 Step 2: Enforcing the Boundary Conditions

3.1 Boundary Conditions3.2 Classification of Edges and Junctions

3.3 Dielectric Edge or Junction

3.4 Metallic Edge or Junction

3.5 General Metal-Dielectric Junction

4 Step 3: EFIE-CFIE-PMCHWT Formulation

4.1 PMCHWT at Dielectric Edge or Junction

4.2 EFIE and CFIE at Metallic Edges and Junctions4.3 EFIE and CFIE on General Metal-Dielectric Junction

5 Numerical Validation

6 Conclusions

References

1. INTRODUCTION

Composite structures of metallic and homogeneous dielectric materialshave many important applications e.g., in the radar technology,antenna design and microwave engineering. Surface integral equationwith the method of moments (MoM) [1] is a popular and powerfulnumerical method for the analysis of such objects. The method iswell documented for single and isolated metallic and dielectric objects.However, the treatment of the field problem of a general compositestructure with multiple material junctions remains challenging. Many

earlier papers on this topic only consider rotationally or axiallysymmetric objects, e.g., [27], or if general 3D objects are considered,[814], either the metallic surfaces and interfaces of dielectric objectsare not allowed to touch each other, or the question of the modelingthe surface currents and testing procedure at junctions is more or lessomitted. There are, however, several papers which discuss the general

junction problem in detail, e.g., [5, 15, 16], [17] pp. 443447, and [18].In treating the field problem of a composite structure with

the surface integral equation method, the main steps are the

choice of the integral equation formulation, discretizing the surfaceintegral equations by the basis and testing functions and enforcingelectromagnetic boundary conditions at interfaces and junctions. Inthis procedure, usually, the junction problem, i.e., treatment of the


3/28

Progress In Electromagnetics Research, PIER 52, 2005 83

basis functions and testing procedure at the junctions, is the maindifficulty, and it has been treated in several ways.

Some papers like [5, 10] and [18] use triangular half-basis functions

at the junctions to expand the surface currents and to test theequations. The continuity of the surface currents is enforcedby equating appropriate unknowns associated with the half-basisfunctions. However, in this approach the testing at junctions, withoutproducing unnecessary line integrals, easily becomes complicated andoften the topic is not discussed in detail or it is omitted. In addition,using different basis and testing functions on different parts of thesurfaces complicates the practical implementation. In [15] surfacecurrents at the junctions are expanded with so called multiplets, i.e.,

combinations of the original basis functions associated to a junction.But the testing at the junctions is again problematic because it cannotbe done directly with the multiplets.

A straightforward way to avoid complications with the surfacecurrent expansion and testing is the use of the (full) Rao-Wilton-Glisson (RWG) [19] basis functions in both operations. In [16]the surface currents are expanded by the RWG functions, but thetesting, enforcing the boundary conditions and the integral equationformulation are implemented by certain generalized testing currents

expanded in the RWG functions. Combining the testing, the boundaryconditions and the formulation this way makes the treatment ofthe junctions somewhat complicated and, in particular, the testingbecomes fully dependent on the used integral equation formulation.In [17], pp. 443447, the RWG basis functions are used both in theexpansion and in the testing while the boundary conditions and theenforcing of the integral equation formulation are handled with theNumber of Unknowns Reduction (NOUR) scheme. In NOUR first thehomogeneous parts of the composite body are considered as isolatedbodies and then they are merged together and the extra unknowns areremoved. This approach, however, becomes complicated with multiplemetallic and dielectric junctions and in [17] they are not considered.

EFIE-PMCHWT is the usual formulation for general compositestructures [9, 1117] and [18]. In this formulation Electric Field IntegralEquation (EFIE) is applied on metallic surfaces and the Poggio-Miller-Chang-Harrington-Wu-Tsai (PMCHWT) formulation [20] is appliedon the dielectric interfaces, but it is not sufficient for removing theinterior resonances if the structure includes closed metallic surfaces. Inthat case, in addition, on those surfaces the Combined Field IntegralEquation (CFIE) [21] must be applied.

In this paper we treat general composite structures consistingof open and closed metallic surfaces and homogeneous dielectric


4/28


subdomains. All types of interface and surface junctions are allowed.The method is based on EFIE-CFIE-PMCHWT formulation toremove internal resonances (EFIE on open metallic surfaces, CFIE on

closed metallic surfaces and PMCHWT formulation on the dielectricinterfaces), and the numerical implementation is organized into threeindependent steps. Though the composite structures with junctionproblems are largely discussed in the literature, we believe that ourapproach in the presented generality is novel. The task division intothree steps is the following:

Step 1: Present and discretize the electric and magnetic fieldintegral equations in each homogeneous dielectric subdomainseparately. Perform the discretization by the Galerkin

method with the oriented RWG basis and testing functions.Step 2: Enforce the boundary conditions on the metallic surfaces,

the dielectric interfaces and at the junctions.

Step 3: Enforce the wanted integral equation formulation.

The first step leads to a system matrix of a diagonal block structure.The second step amounts to removing and combining unknowns leavingmore equations than unknowns. In the third step we reduce the numberof equations to that of the unknowns by combining, or removing, the

extra equations according to the applied integral equation formulation.In practice, we condense the last two steps into a few simple

bookkeeping rules by which the final system matrix is directlyassembled in an efficient way and without ever forming the matrices ofthe intermediate stages. In particular, the removed columns and rowsof the system matrix are never computed in practice. However, wefind it instructive to describe the treatment of the above three steps interms of the discretized system matrix, because from that descriptionthe simple rules for the efficient system matrix assembly can easily

been derived.A similar division of the task into three steps was presented

in [10] for penetrable structures without junctions; apparently thistask division was not suggested for the general case, because thepaper treats the junctions with half-RWG basis functions; anyway,a discussion on the testing and other details at general metal-dielectric

junctions is omitted in the paper.The procedure of this paper, i.e., why the discretization can be

performed in subdomains before applying the boundary conditions

and the integral equation formulation, can be shortly justified asfollows: The discretization of the EFIEs and MFIEs separately ineach subdomain by the Galerkin method only means replacing theoriginal equations in subdomains by the approximative equations


5/28


obtained by restricting the surface currents to the function subspacespanned by the basis functions and by orthogonally projecting theoriginal equations into the same subspace, see e.g., [22], Chapter 13.

Therefore, the boundary conditions and the formulation, withoutchanging their physical meaning, can be applied to the approximativeequations equally well as to the original ones. On the other hand, theapproximated equations are presented equivalently by the discretizedsystem equations, and accordingly, the boundary conditions and theformulation can be directly applied to the system equations.

2. STEP ONE: PRESENTATION ANDDISCRETIZATION OF SURFACE INTEGRAL

EQUATIONS IN DIELECTRIC SUBDOMAINS

We consider the time harmonic electromagnetic waves in a piecewisehomogeneous medium with the time factor eit. Let the computa-tional domain be divided into homogeneous dielectric, or penetrable,and perfectly conducting (PEC) domains D1, . . . , DKas in Figure 1.The constant electromagnetic parameters of the domains are denotedby j , j and j , where j = rj0 + ij/ denotes complexpermittivity. A PEC object is defined by setting j =. In addition,

let D0 denote the (unbounded) background, let Sj denote the surfaceof Dj , j = 0, . . . , K , and let Sjl denote the interface of domainsDj and Dl. The PEC domains are further classified as closed andopen domains. The latter ones represent thin metallic plates with zerovolumes.

Dn

Dm

Figure 1. Piecewise homogeneous medium with open (Dp) and closedmetallic objects (Dm, Dn). Metallic objects are denoted by shading.


6/28


2.1. Presentation of EFIEs and MFIEs

Let Epj , Hpj denote the primary or incident field of the sources in a

domain Dj . The total field in Dj can be expressed as a sum of the

incident field and the induced secondary field Esj , Hsj as

Ej = Epj +E

sj , Hj = H

pj +H

sj . (1)

We define the equivalent electric and magnetic surface current densitieson the surface Sj as

Jj = njHj and Mj =njEj , (2)

where nj is the inner unit normal ofSj pointing into the interior ofDj .

Ifj < , the induced electromagnetic fields, Es

j , Hs

j , generated bythe currents Jj and Mj, can be expressed at point rDj as [10, 15]

Esj(Jj ,Mj)(r) = 1

ij(DjJj)(r) (KjMj)(r) (3)

Hsj(Jj ,Mj)(r) = 1

ij(DjMj)(r) + (KjJj)(r). (4)

The integral operators Dj and Kj are defined as [10, 15]

(DjF)(r) = Sj

Gj(r, r)s F(r

)dS + k2j

Sj

Gj(r, r)F(r)dS, (5)

(KjF)(r) =

Sj

Gj(r, r) F(r)dS, (6)

wheresdenotes surface divergence of a tangential vector field in theprimed coordinates, and

Gj(r, r) =

eikj |rr|

4|r r|, (7)

is the homogeneous space Greens function andkj is the wave numberof domain Dj.

Next we write the needed integral equations on the boundarysurfaces Sj of the domains Dj , j = 0, . . . , K . Let r be a point inDj . Letting r Sj and using the definition of the surface currents,we get for the tangential components,

Epj +E

sj

tan

= njMj , (8)H

pj +H

sj

tan

= nj Jj , (9)


7/28


for all j, j = 0, . . . , K . By expressing the secondary fields in terms ofthe surface currents with the integral representations (3) and (4) in (8)and (9), and by using the tangential boundary values of the operators

Kand D, [23], we get the following equations on the surface Sj

1

ij(DjJj) (KjMj)

1

2njMj

tan

= (Epj )tan (10)

1

ij(DjMj) + (KjJj) +

1

2nj Jj

tan

= (Hpj)tan, (11)

for all j, j = 0, . . . , K . Equation (10) is called the Electric Field

Integral Equation (EFIE) and (11) is called the Magnetic Field IntegralEquation (MFIE). The PMCHWT formulation can be handled withEFIEs and MFIEs. For the CFIE formulation we, in addition, need toform the cross product ofnj and MFIE and get the equation

1

ijnj (DjMj) + nj (KjJj)

1

2Jj =njH

pj , (12)

which is called as nMFIE in this paper.

2.2. Basis Functions

For the discretization of the equations (10)(12) we use the Galerkinmethod and the Rao-Wilton-Glisson (RWG) basis functions [19]. Inorder to define the basis functions suppose that the surfaces are dividedinto planar triangular elements so that the elements on Sjl and Slj onthe opposite sides of the interface are equal and the meshes of thesurfaces match.

An RWG basis function fnassigned to an edgeenof the triangularmesh is defined as

fn(r) =

Ln

2A+n(r p+n ), r

T+n,

Ln

2An(r pn ), r

Tn,

0 otherwise.

(13)

HereA+n is the area of the triangleTn, Lnis the length of the common

edge en and pn is the free vertex of Tn. An RWG function isdisplayed in Figure 2.

Though the RWG function fnis not continuous on the surfaceSj ,it still has the following partial continuity property: the component of


8/28


Figure 2. An RWG function fn assigned to the edge en.

fn normal to the edge en is continuous across the edge, and on theouter boundary of the triangle-pair T+n Tn the component of fn

normal to that boundary vanishes and therefore is continuous acrossthat boundary. This partial continuity property of RWG functionsmakes it possible to expand surface currents in the RWG functionsand preserve the surface current continuity across edges and junctions.

We finish this section by orienting the RWG functions which willbe used in the discretization. Leten be an edge on the interface Sjlwhere only two domains Dj and Dl meet. Then two RWG functions

fn1 and fn2 are assigned to that edge and they are on the oppositesides of Sjl . We now orientate them so that they flow into oppositedirections, i.e., fn1 = fn2 see Figure 3. By this orientation theboundary conditions of the surface currents can be easily satisfied,as we will see later. If en is a junction edge, where more than twosubdomains Dj meet, then all the RWG functions assigned to en areoriented so that on the opposite sides of the interfaces meeting en theyagain flow into opposite directions, i.e, they all flow past the edge eitherclockwise or counterclockwise, as illustrated in Figure 3.

2.3. Discretization by the Galerkin Method and RWG BasisFunctions

In our approach the integral equations (10)(12) of the previous sectionare discretized and converted into matrix equations by the Galerkinmethod with the oriented RWG basis functions separately in eachdomain. Letfn, n= 1, . . . , N

(j) be the RWG basis functions definedon the triangular mesh on the boundary surface Sj of the subdomain

Dj , whereN(j)

is the number of basis functions on the surface Sj . Theunknown electric and magnetic surface current densities on the surface


9/28


10/28


+ij

Tm

fm(r)

Tn

Gj(r, r)fn(r

)dSdS, (16)

ZHJ(j)mn =

Tm

fm(r) Tn

Gj(r, r

) fn(r

)dS

dS

+1

2

Tm

fm(r) (nj fn(r))dS, (17)

ZEM(j)mn =ZHJ(j)mn and Z

HM(j)mn =

jj

ZEJ(j)mn , (18)

ZnHJ(j)mn = Tm

(nj fm(r)) Tn

Gj(r, r) fn(r

)dSdS

1

2

Tm

fm(r) fn(r)dS, (19)

ZnHM(j)mn = 1

ij

T+m

+

Tm

mm (njfm(r))

Tn

Gj(r, r)s fn(r

)dSdl

(20)

where in the line integrals in (20) over the boundary curves T+m andTm of the trianglesT

+m andT

m the vector mnis the unit outer normal

of these boundary curves. In addition,Tn = T+n T

n is the support

of fn and Tm = T+m T

m is the support of fm. For the numerical

evaluation of these matrix elements with the singularity extractiontechnique we refer to [24].

The vectorsIJ(j) andIM(j) in (15) are the coefficient vectors ofJj

and Mj , including the coefficients (j)

1

, . . . , (j)

N(j)

and (j)

1

, . . . , (j)

N(j)

,

and VE(j), VH(j) and VnH(j) are the excitation vectors

VE(j)m =

Tm

fm(r) Epj (r)dS, (21)

VH(j)m =

Tm

fm(r) Hpj (r)dS, (22)

VnH(j)m

= Tm

(nj f

m(r)) Hp

j(r)dS. (23)

Abovem, n= 1, . . . , N (j), runs through all basis and testing functionson the surface Sj and all matrix blocks in (15) are square.


11/28


By repeating above testing procedure for all boundary surfacesSj of the domains Dj, j = 0, . . . , K , we obtain the following blockdiagonal matrix equation

Z(0) 0 0

0 . . . 0

0 0 Z(K)

I(0)...

I(K)

=

V(0)...

V(K)

, (24)

where

Z(j) =

ZEJ(j) ZEM(j)

ZHJ(j) ZHM(j)

ZnHJ(j) ZnHM(j)

, (25)

I(j) =

IJ(j)

IM(j)

(26)

V(j) =

VE(j)

VH(j)

VnH(j)

. (27)

This is the initial discretized system matrix equation, whichpresents only the (approximative) integral equations (10)(12) ineach subdomain and not the entire scattering and transmission fieldproblem, because the interactions between the dielectric domains havenot yet been taken into account via the boundary conditions. Alsoextra equations must be removed by applying the the integral equationformulation. We start now carrying out those two tasks.

3. STEP 2: ENFORCING THE BOUNDARYCONDITIONS

The solutions of the integral equations (10)(12) of Section 2.1 mustsatisfy electromagnetic boundary conditions on the domain interfacesand at the junctions. In the previous sections these solutions, namelythe equivalent surface current densities, were defined domain-wisewithout taking into account the boundary conditions. Due to the theseconditions, the discretization of the equations leads to system equationswhich include linearly dependent unknowns, related to J and M onthe interfaces of dielectric domains, and zero unknowns related to Mon the surfaces of metallic objects. In this section we find definite ruleshow the extra unknowns and basis functions are removed by enforcingthe boundary conditions.


12/28


3.1. Boundary Conditions

The formulation of Section 2.1 includes two kind of surfaces, metallicand dielectric ones. A surface of a PEC object is called a metallic

surface and an interface of two homogeneous penetrable domains with < is called a dielectric surface, or interface. Note that theinterfaces of homogeneous non-metallic domains are called dielectricsurfaces although the domains may be magnetic, too, i.e., r > 1.The metallic surfaces are further classified as closed and open surfaces,depending whether the surface circumvents a 3D body or not.

The boundary conditions of the electromagnetic fields directlydetermine how the surface currents behave on the metallic anddielectric surfaces. The boundary conditions are

1. nEand n Hvanish on metallic surfaces,

2. nEand nH are continuous across dielectric surfaces.

The first boundary condition immediately yields the important result:the magnetic current M =n E vanishes on metallic surfaces. Inthe following subsections we study what other properties the boundaryconditions imply for the surface currents, in particular, at junctions ofvarious types.

3.2. Classification of Edges and Junctions

For treating the boundary conditions and the integral equationformulation, we classify the edges and junctions into the following threecases. Letenbe an edge on the triangular mesh. It is called a junction,if more than two domains or surfaces meet at en. Otherwise, we callen a single edge, or just an edge. We classify edges and junctions enas follows:

1. Dielectric edge or junction: en lies on an intersection of two ormore dielectric surfaces but does not meet any metallic surfaces.

2. Metallic edge or junction: en lies on intersection of open or closedmetallic surfaces but does not meet any dielectric surfaces.

3. Composite metal-dielectric junction: enlies on an intersection of atleast one open or closed metallic surface and at least one dielectricsurface.

Next we consider the boundary conditions in these three cases of edges

and junctions.


13/28


3.3. Dielectric Edge or Junction

Let us next consider the surface currents and their expansions in thebasis functions on a dielectric interface Sjl between two dielectric

domains Dj and Dl. The boundary conditions directly imply that

Jj =Jl, and Mj =Ml. (28)

This follows from the obvious fact that the surface currents are relatedto the fields by J = n H and M = n E and the tangentialcomponents of the fields are continuous across a dielectric interfaceand the normal vectors n point into opposite directions. Because thetwo basis functions assigned to an edge on Sjl and on the opposite sides

ofSjl , flow into opposite directions (orientation of the basis functions),in the surface current representations they must have equal coefficients.This is illustrated in Figure 4.

1

2

1 2

34

Figure 4. Combination of the unknowns associated to the electricand magnetic basis functions assigned at an edge on the interface oftwo dielectric domains and on the junction of four dielectric domains.Dashed lines denote dielectric surfaces and curved arrows indicate thebasis functions with the same unknown coefficient. Numbers indicatedielectric domains.

This analysis generalizes to the dielectric junctions as follows.Let en be a dielectric edge or junction where two or more dielectricdomains meet. Because the components ofE and H, parallel to en,are also tangential components on the surfaces which meet at en, thosecomponents are continuous across the edge en into all directions. Thisimplies that the components of the surface currents Jj = njH andMj =njE, normal to the edge en, are continuous across the edge

on an interface Sj which meets en. This is the (partial) continuityproperty of the surface currents on the dielectric edges and junctions,i.e., Kirchoffs laws for the surface currents. Furthermore, because ofthis continuity and (28), the normal components of all currents Jj and


14/28


15/28


the columns of the system matrix, the fields of the basis functions mustbe presented by only one of them, say that off1, and 1 2 can beconsidered as a new single unknown.

The above analysis generalizes for junctions of an arbitrarynumber of open metallic surfaces inside a homogeneous medium. Notealso that we do not need the basis functions which are assigned to aboundary edge of an open metallic surface and lying completely insidea homogeneous dielectric domain, because Jhas no component normalto such a boundary edge. Therefore, those electric basic functions andunknowns associated with them can be removed. In summary, we getour second rule:

Rule 2: At a metallic edge or junction meeting no dielectric

surfaces, all the magnetic basis functions in the expansion ofM,with their unknowns, must be removed. In addition, if all surfacesassigned toen are open metallic surfaces and en lies completely inthe interior of a homogeneous dielectric domain, one of the electricbasis functions in the expansion ofJmust be removed.

For the system equations this rule means that those columns of thesystem matrix, which are associated with the removed unknowns, willbe removed. Rule 2 is illustrated in Figure 5.

3.5. General Metal-Dielectric Junction

Next letena general metal-dielectric junction. Because Mvanishes onmetallic surfaces, then at en the component ofM, normal to en, alsovanishes, and so the magnetic basis functions assigned at en and theunknowns associated with them are removed. Again due to the factthat the magnetic field is not continuous across open metallic surfaces,Jmust have two independent values on the opposite sides of an open

metallic surface. This implies that in the expansion ofJ

the unknowncoefficients of the basis functions assigned to en on the opposite sidesof metallic surfaces should be considered as independent unknowns,too. By combining this with the previous Rule 2, we obtain our thirdrule:

Rule 3: In the expansions of J the unknown coefficients ofthe electric basis functions assigned to a general metal-dielectric

junction between two metallic surfaces must have the samevalue and the corresponding unknowns must be combined to a

single unknown. In the expansions ofM the magnetic basisfunctions, with their unknowns, assigned to a general metal-dielectric junction must be removed.


16/28


1 1

1

1 2

3

1 2

34

1

1

1 1

11

Figure 5. Basis functions and their unknowns at metallic junctions.Solid lines denote open metallic surfaces, closed metallic domains aredenoted by black and the unknowns associated to the electric basisfunctions denoted with dashed lines are removed. Numbers indicatedomains. Note that the unknowns can not be combined in the threecases at the top.

For the system equations this rule means that those columns, which areassociated with the unknowns to be combined, will added together, andthose columns are removed, which are associated with the magneticbasis functions to be removed. Figure 6 illustrates treatment of the

electric unknowns at composite metal-dielectric junctions.

4. STEP 3: EFIE-CFIE-PMCHWT FORMULATION

After having removed the extra unknowns from the system equationsby enforcing the boundary conditions, the remaining system has moreequations than unknowns. In order to get a well-defined system,the number of equations has to be reduced to that of the remainingunknowns. The way how this reduction is done depends on theintegral equation formulation. In this paper we apply the EFIE-CFIE-PMCHWT formulation, where EFIE and CFIE are applied on themetallic surfaces depending whether the surface is open or closed, and


17/28


18/28


together, respectively. This, in turn, corresponds to combining therows of the matrix equation.

4.2. EFIE and CFIE at Metallic Edges and JunctionsFirst let en be an open metallic edge or a junction of open metallicsurfaces. By Rule 2we have already removed all unknowns associatedwith the magnetic basis functions assigned to en. Since we apply EFIEon open metallic surfaces, now we also remove all MFIEs and nMFIEs,which are tested with the basis functions assigned to en, from thesystem equations.

Ifen is completely inside one dielectric domain, byRule 2we havealso removed one of the unknowns associated with the electric basisfunctions assigned toen. Thus, now we, in addition, remove that EFIEfrom the system equations which was tested with the correspondingelectric basis function. Thereafter, if en is an edge or a junctionwhere M open metallic surfaces meet inside one dielectric domain,there remain M 1 electric equations for M 1 (combined) electricunknowns associated with en.

Next let en be a single edge on a closed metallic surface or at ajunction of a closed metallic surface and at least one open metallicsurface. Again by Rule 2we remove all the unknowns associated withthe magnetic basis function assigned toen. The CFIE formulation [21]atenmeans that the EFIE and nMFIE tested by the remaining electric

basis function and denoted by EFIE(l)n and nMFIE

(l)n , respectively, are

combined asEFIE(l)n + (1 )lnMFIE

(l)n , (30)

where l =

l/l and is a coupling coefficient, which we set = 1/2. Accordingly, there remain one (combined) system equationand one (electric) unknown aten and we have also removed the MFIE,

which is tested with the removed magnetic basis function, from thesystem equations.

4.3. EFIE and CFIE on General Metal-Dielectric Junction

Finally we consider the formulation at a general metal-dielectricjunctionen. Also here byRule 2we have already removed the magneticunknowns associated with the magnetic basis functions assigned toen.Accordingly, we remove also all MFIEs tested by those magnetic basis

functions.Ifen does not meet any closed metallic surfaces, then the EFIEs

assigned to en are combined between any two metallic surfaces by


19/28


summing the corresponding EFIEs together

M

m=1

EFIE(m)n , (31)

whereMis the number of dielectric domains between the two metallicsurfaces. In that case we remove also all nMFIEs tested by thoseelectric basis functions associated to en.

Ifen lies on a closed metallic surface, then the EFIEs and nMFIEsassigned to en are combined between any two metallic surfaces bysumming the corresponding CFIEs as

Mm=1

CFIE(m)n , (32)

where M is again the number of dielectric domains between the twometallic surfaces. Due to Rule 3we in both above cases arrive at thesame reduced number of equations and unknowns at en.

5. NUMERICAL VALIDATION

In this section the developed method is verified by consideringnumerical examples. We consider scattering by inhomogeneousdielectric and composite objects and analysis of mlcrostrip anddielectric resonator antennas. The example cases are illustrated inFigure 7.

First consider an inhomogeneous dielectric sphere made of twohemispheres illuminated with an axially incident plane wave. Therelative electric permittivity of the hemisphere at the top, r1, is fixedas 4 and the relative electric permittivity of the second hemisphere atthe bottom,r2, is varied from 4 to 8. The electrical size of the sphereisk0r= 1, wherek0is the free space wave number (wave number of theexterior) andr is the radius of the sphere. Figure 8 shows the electricand magnetic surface current densities on the surface of the spherealong a circumferential arc. Along the arc, two current components,J, J and M, M, normalized with the incident magnetic field andelectric field, respectively, are displayed similarly as in [26] and [25]. Inthe case r1=r2, the solution of our method is compared to the Mieseries solution and to the numerical solution of a homogeneous sphereand the results show a good agreement.

Then we consider the same inhomogeneous dielectric sphere,but in this case the second hemisphere at the bottom is PEC.Figure 9 shows the equivalent electric and magnetic surface currents


20/28


1

1

2

1

1

Figure 7. Example geometries from left to right and fromtop to bottom: An inhomogeneous dielectric sphere made of twohomogeneous hemispheres, a composite sphere made of a dielectrichemisphere and metallic hemisphere, a rectangular microstrip antennaand a hemispherical dielectric resonator antenna. The spheres areilluminated with a plane wave from the top (from the side of r1)and the antennas are excited with coaxial lines.

on the surface of the sphere similarly as in Figure 8. The currentsare calculated with the EFIE-PMCHWT and CFIE-PMCHWTformulations.

Consider next a rectangular microstrip antenna with a finiteground plane and substrate [27]. The size of the patch is 30 20mmand the size of the ground plane is 50 40 mm, both centered tothe origin. The antenna is fed by a coaxial line with inner radiusof 0.22 mm and outer radius of 1.4 mm at point (0,2.5, 0) mm. Therelative epsilon of the substrate is 10 and the height of the substrateis 6.35 mm. Figure 10 shows the calculated input impedance of theantenna as a function of the frequency. The same antenna is consideredalso in [28] with an infinite ground plane and with a lossy (infinite)substrate (r = 10.2, tan = 0.001). Figure 10 shows the results of thepresent method with a lossy substrate, too. For comparison the resultsobtained with the multilayered medium approach of [29] and [30] arealso presented. In that method, the ground plane and the substrate aremodeled as planar infinite layers with an appropriate layered mediumGreens function. We repeated the calculations of Figure 10 with alarger 100 80 mm ground plane and substrate, too, using the methodof this paper. Figure 10 shows that the results calculated with the largesubstrate and ground plane agree better with the results obtained withthe layered model of [29] and [30] than the results calculated with the


21/28


Figure 8. The electric and magnetic current densities on the surfaceof an inhomogeneous sphere made of two hemisphere with r1 = 4 andr2= 4, 5, 6, 7, 8, illuminated with an axially incident plane wave.

original dimensions. Thus, we may conclude that the finite size of thesubstrate and ground plane has a clear effect on the computed inputimpedances.

Then we consider a hemispherical dielectric resonator antenna(HSDRA) [31, 15]. The hemisphere of radius r = 0.0254 m is placedover a finite PEC plane of size l1l2, where l1 = l2 = c r with aconstant multiplierc, both centered to the origin. The relative epsilonof the sphere is 8.9 and tan = 0.0038. The antenna is fed by a 50 Ohm


22/28


Figure 9. The electric and magnetic current densities on the surface

of an inhomogeneous sphere made of two hemisphere with r1 = 4 and2=, illuminated with an axially incident plane wave.

coaxial line of inner radius 0.75 mm placed at point (0, 0.0174, 0)m.Height of the probe is 0.0152 m. Figure 11 shows the input impedanceof the antenna as a function of the frequency with c = 2 and c = 3.The results compare reasonable well with the ones presented in [31]and [15]. The differences may be explained by the fact we apply finite

ground plane whereas in [31] and [15] the ground plane is consideredby mirroring as an infinite one. Figure 11 shows that the results ofthe present method approaches the ones of [15] when the size of theground plane is enlarged.


23/28


1.8 1.82 1.84 1.86 1.88 1.9 1.92 1.94 1.96 1.98 20

20

40

60

80

100

120

140

Frequency [GHz]

Z

Real(Z)

Imag(Z)

Real(Z) - large

Imag(Z) - large

Real(Z) - infinite

Imag(Z) - infinite

1.8 1.82 1.84 1.86 1.88 1.9 1.92 1.94 1.96 1.98 20

20

40

60

80

100

120

140

Frequency [GHz]

Z

Real(Z)

Imag(Z)

Real(Z) - large

Imag(Z) - large

Real(Z) - infinite

Imag(Z) - infinite

Figure 10. Input impedance of a rectangular microstrip antenna fedby a coaxial line as a function of the frequency. Real part is denotedby a solid line and imaginary part is denoted by a dashed line. Atthe upper figure the substrate has r = 10, tan = 0 and at the lowerfigure r = 10.2, tan = 0.001. The results obtained by the layered

model [29, 30], are denoted by circles and the results calculated with alarge 100 80 mm ground plane and substrate are denoted by stars.


24/28


Figure 11. Input impedance of a HSDRA fed by a coaxial line asa function of frequency. Real part of the impedance is denoted by a

solid line and imaginary part is denoted by a dashed line. The resultsobtained with the larger ground plane of size 3r 3r are denoted bycircles.

6. CONCLUSIONS

In this paper electromagnetic analysis of complex three dimensionalstructures made of composite metallic and homogeneous dielectric

regions is considered with the surface integral equation method. Themethod is based on the EFIE-CFIE-PCMHWT formulation and theGalerkin method with the oriented RWG basis and testing functions.An efficient three steps numerical procedure is presented. As thefirst step, the discretization and testing of the integral equationsare performed individually in subdomains. In the second step, theelectromagnetic boundary conditions are enforced by removing zerounknowns and combining linearly dependent unknowns. As the laststep, the number of equations is reduced by enforcing the wanted

integral equation formulation.The great benefit of the our approach is that it separates theexpansion of the surface currents in the basis functions, the testingand the integral equation formulation from the enforcement of the


25/28


boundary conditions, thus making the treatment of junctions muchmore straightforward and easier. In particular, no special junctionbasis functions or special testing procedures at junctions are needed.

Also separating the formulation from the first two steps, makes it mucheasier to modify the procedure for other integral equation formulations.In practice, the procedure can be condensed into simple bookkeepingrules by which the final system matrix can be directly assembled inan efficient way. The presented approach is validated with numericalexamples.

REFERENCES

1. Harrington, R. F., Field Computation by Moment Methods,Macmillan, New York, 1968.

2. Medgyesi-Mitschang, L. N. and J. M. Putnam, Electromagneticscattering from axially inhomogeneous bodies of revolution,IEEE Trans. on Antennas and Propagation, Vol. 32, No. 8, 797806, 1984.

3. Govind, S., D. R. Wilton, and A. W. Glisson, Scattering frominhomogeneous penetrable bodies of revolution, IEEE Trans. onAntennas and Propagation, Vol. 32, No. 11, 11631173, 1984.

4. Huddleston, P. L., L. N. Medgyesi-Mitschang, and J. M. Putnam,Combined field integral equation formulation for scatteringby dielectrically coated conducting bodies, IEEE Trans. onAntennas and Propagation, Vol. 34, 510520, April 1986.

5. Putnam, J. M. and L. N. Medgyesi-Mitschang, Combined fieldintegral equation for inhomogeneous two- and three-dimensionalbodies: The junction problem, IEEE Trans. Antennas andPropagation, Vol. 39, No. 5, 667672, May 1991.

6. Goggans, P. M., A. A. Kishk, and A. W. Glisson, Electromag-netic scattering from objects composed of multiple homogeneousregions using a region-by-region solution, IEEE Trans. Antennasand Propagation, Vol. 42, No. 6, 865871, June 1994.

7. Yla-Oijala, P. and E. Somersalo, Computation of electromagneticfields in axisymmetric RF structures with boundary integralequations, Journal of Electromagnetic Waves and Applications,Vol. 13, 445489, 1999.

8. Sarkar, T. K., S. M. Rao, and A. R. Djordjevic, Electromagnetic

scattering and radiation from finite microstrip structures, IEEETrans. on Microwave Theory and Techniques, Vol. 38, No. 11,15681575, Nov. 1990.

9. Rao, S. M., C. C. Cha, R. L. Cravey, and D. Wilkes,


26/28


Electromagnetic scattering from arbitrary shaped conductingbodies coated with lossy materials of arbitrary thickness, IEEETrans. on Antennas and Propagation, Vol. 39, No. 5, 627631,

May 1991.10. Medgyesi-Mitschang, L. N., J. M. Putnam, and M. B. Gedera,Generalized method of moments for three-dimensional penetrablescatterers,J. Opt. Soc. Am. A, Vol. 11, 13831398, April 1994.

11. Chen, Q. and D. R. Wilton, Electromagnetic scattering by three-dimensional arbitrary complex material/conducting bodies,IEEE AP-S Symposium 1990, Vol. 2, 590593, 1990.

12. Arvas, E., A. Rahhal-Arabi, A. Sadigh, and S. M. Rao, Scatteringfrom multiple conducting and dielectric bodies of arbitrary shape,

IEEE Antennas and Propagation Magazine, Vol. 33, No. 2, 2936,April 1991.

13. Shin, J., A. W. Glisson, and A. A. Kishk, Analysis ofcombined conducting and dielectric structures of arbitrary shapesusing an E-PMCHW integral equation formulation, IEEE AP-SSymposium 2000, Vol. 4, 22822285, 2000.

14. Donepudi, K. C., J.-M. Jin, and W. C. Chew, A higherorder multilevel fast multipole algorithm for scattering frommixed conducting/dielectric bodies, IEEE Trans. Antennas andPropagation, Vol. 51, No. 10, 28142821, October 2003.

15. Kolundzija, B. M., Electromagnetic modelling of compositemetallic and dielectric structures, IEEE Trans. on MicrowaveTheory and Techniques, Vol. 47, No. 7, 10211032, July 1999.

16. Shin, J., A. W. Glisson, and A. A. Kishk, Modeling ofgeneral surface junctions of composite objects in an SIE/MoMformulation,2000 ACES Conference Proceedings, 683690, 2000.

17. Chew, W. C., J.-M. Jin, E. Michielssen, and J. Song, Fast and

Efficient Algorithms in Computational Electromagnetics, ArtechHouse, Boston, 2001.

18. Carr, M., E. Topsakal, and J. L. Volakis, A procedure formodeling material junctions in 3-D surface integral equationapproaches,IEEE Trans. on Antennas and Propagation, Vol. 52,No. 5, 13741379, May 2004.

19. Rao, S. M., D. R. Wilton, and A. W. Glisson, Electromagneticscattering by surfaces of arbitrary shape, IEEE Trans. Antennasand Propagation, Vol. AP-30, No. 3, 409418, May 1982.

20. Mautz, J. R. and R. F. Harrington, Electromagnetic scatteringfrom a homogeneous material body of revolution,Arch. Elektron.Ubertragungstechn. (Electron. Commun.), Vol. 33, 7180, 1979.


27/28


21. Mautz, J. R. and R. F. Harrington, H-field, E-field andcombined-field solutions for conducting bodies of revolution,Arch. Elektron. Ubertragungstechn. (Electron. Commun.), Vol. 32,

157164, 1978.22. Kress, R., Linear Integral Equations, Springer-Verlag, Berlin,Heidelberg, New York, 1989.

23. Colton, D. and R. Kress,Integral Equation Methods in ScatteringTheory, John Wiley & Sons, New York, 1983.

24. Yla-Oijala, P. and M. Taskinen, Calculation of CFIE impedancematrix elements with RWG and n RWG functions, IEEETrans. Antennas and Propagation, Vol. 51, No. 8, 18371846, Aug.2003.

25. Rao, S. M. and D. R. Wilton, E-field, H-field, and combinedfield solution for arbitrarily shaped three-dimensional dielectricbodies,Electromagnetics, Vol. 10, 407421, 1990.

26. Umashankar, K., A. Taflove, and S. A. Rao, Electromagneticscattering by arbitrary shaped three-dimensional homogeneouslossy dielectric objects,IEEE Trans. Antennas and Propagation,Vol. AP-34, No. 6, 758766, June 1986.

27. Pang, Y.-H. and R-B. Wu, Analysis of microstrip antennas with

finite-sized substrate, IEEE APS Symposium, Vol. 2, 814817,Boston, USA, 2001.

28. Chen, W., K.-F. Lee, and R. Q. Lee, Input impedance of coaxiallyfed rectangular microstrip antenna on electrically thick substrate,Microwave and Optical Technology Letters, Vol. 6, No. 6, 387390,1993.

29. Yla-Oijala, P., M. Taskinen, and J. Sarvas, Multilayeredmedia Greens functions for MPIE with general electric andmagnetic sources by the Hertz potential approach, Progress in

Electromagnetic Research, PIER 33, 141165, 2001.30. Yla-Oijala, P. and M. Taskinen, Efficient formulation of closed

form Greens functions for general electric and magnetic sourcesin multilayered media, IEEE Trans. Antennas and Propagation,Vol. 51, No. 8, 21062115, Aug. 2003.

31. Junker, G. P., A. A. Kishk, and A. W. Glisson, Input impedanceof dielectric resonator antennas excited by a coaxial probe,IEEETrans. on Antennas and Propagation, Vol. 42, No. 7, 960966, July1994.


28/28


Pasi Yla-Oijala received the M.Sc. degree in 1992 and the Ph.D.degree in 1999, both in applied mathematics in the Universityof Helsinki, Finland. Currently he is working as a researcher

in Electromagnetics Laboratory, Helsinki University of Technology,Finland. His field of interest includes numerical techniques incomputational electromagnetics based on the integral equationmethods and antenna simulations with the method of moments.

Matti Taskinen received the M.Sc. degree in 2002 in appliedmathematics in the University of Helsinki, Finland. Currently he isworking towards the Ph.D. degree in Electromagnetics Laboratory,Helsinki University of Technology, Finland. His field of interestincludes computational electromagnetics, especially integral equationmethods and layered media.

Jukka Sarvas received the M.Sc. degree in 1968 and the Ph.D.degree in 1972, both in mathematics, in the University of Helsinki.He has been in the University of Helsinki, in the Helsinki Universityof Technology, and 19821984 in Outokumpu Co. In 19882002he was the director of the Rolf Nevanlinna Research Institute ofApplied Mathematics and Statistics in the University of Helsinki.He has been a visiting researcher 19741975 and 19791980 in the

University of Michigan, and 19992000 in the University of Illinoisat Urbana-Champaign. Since 2002 he has been a professor incomputational electromagnetics in Electromagnetics Laboratory of theHelsinki University of Technology. His main interests include the fieldcomputing with integral equations and the fast methods.

Surface Integral Equation Method

Documents

Transcript of Surface Integral Equation Method