USACERL Technical Mznuscrlpt M-89/1 i

September 1989

US Army Corps
of Engineers
Construction Engineering
Research Laboratory
A D-A 213 062

Electromagnetic Wave Propagation Through

Circular Waveguides Containing Radially
Inhomogeneous Lossy Media
by
Keith W. Whites

Propagation characteristics of an electromagnetic

(EM) wave inside a waveguide are greatly modified by
the introduction of dielectrics into the guide.
Frequency of cutoff, attenuation, and power flow
distribution are all properties of the EM wave that
are highly dependent on the physical structure and
composition within the guide. Shielding applications
take advantage of the large amount of attenuation
provided by a waveguide when the incident EM wave
has a frequency lower than the cutoff frequency of
the waveguide. In practice, small sections of
waveguides are inserted through the metal walls of a
shielded enclosure to construct an air passage that
does not compromise the shielding effectiveness of the
shelter. Typically, however, other fluids and materials
besides air must be transferred inside the enclosure.
Therefore, hoses are sometimes inserted through these
waveguides, which can reduce the shielding provided
by the waveguide above and below cutoff.
This work has investigated the EM propagation
through these loaded waveguides structures, placing
special emphasis on wave attenuation. The study was
divided into two problems: the modal coupling and
excitation of a finite inhomogeneous guide.
Solutions to the radially inhomogeneous infinite
circular waveguides have been obtained numerically
from an exact theoretical development beginning with
first principles. These inhomogeneous waveguides are
layered with circular, concentric annuli, possibly
having complex constitutive parameters. The modes
that were found to exist in these waveguides are
hybrid, meaning that they have both axial E- and H-
fieldq. The foundations and numerical solutions
developed were verified experimentally.

Approved for public release; distribution is unlimited.

89 9 28 083
The contents of this report are not to be used for advertising, publication, or
promotional purposes. Citation of trade names does not constitute an
official indorsement or approval of the use of such commercial products.
The findings of this report are not to be construed as an official Department
of the Army position, unless so designated by other authorized documents.

DESTROY THIS REPORT WHEN IT IS NO I o'(;ER NEEDED

DO NOT RETURN IT TO THE ORIG1I\ 4 TOR
 UNCLASSIFIED
SECURITY CLASSIFICATION OF THIS PACE
i or- Appro.ed
REPORT DOCUMENTATION PAGE (A,8No
I p Dire
0704 0F88
wjn 10 19R6

la REPORT SECURITY CLASSIFICATION lb RESTRICTIVE MARKINGS

UNCLASSIFIED
,'a SECk;RITY CLASSI~ICATION AUTHORITY 3 DISTRiBuTION, AvAILA8!LjrY OF REPORT
Approved for public release;
2b DECLASSIFICATION DOWNGRADING SCHEDULE distribution is unlimited.
4 PERfORMiNG ORGANIZATION REPORT NuMBER(S) S MON,TORANG 9(. ANZa>.OlN, REPOPT NM8E ?(S)

USACERL TM M-89/I1
7
6a *,AME OF PERFORMING ORGANIZATION bb OFFICE SYMBOL a NAME O VO%, 0R.NG jPGANZA' DN
U.S. Army Construction Engr (If applicable)
Research Laboratory CECER-EM
6( ADDRESS (City. State. and ZIP Code) b aDDRESS Cty. State and ZIP Code)
P.O. Box 4005
Champaign, IL 61824-4005

Sa NAME OF FUNDING SPONSORING Bb OFFICE SYMBOL 9 PROCUREMENT :NSTRUMENT IDEN'IF,(ATiON NUMBER

'PGCANIZAT'ON j If applicable)
HQUSACE [ CEEC-EE
Ic ADDRESS (City. State, and ZIP Code) 10 SOURCE OF FUNDNG NUMBERS
20 Massachusetts Ave, NW PROGRAM PRO;ECT TASK WORK ,NIT
WASH DC 20314-1000 ELEMEINT NO NO NO ACCESSON NO
4A] 6234 AT4J MA C59
. IInclude
( Security Classtation)
Electromagnetic Wave Propagation Through Circular Waveguides Containing Radially
inhomogeneous Lossy Media
'2 PERSONAL AUTHOR(S)
Keith Wayne Whites
13a T'PE OF REPORT I!30 TIME COVERED 14 DATE OF REPORT (Year Month. Day) 15 PAGE COUNT
FinalI FROM _ TO 1989, August 168
16 SUPPLEMENTARY NOTATION
Copies are available from the National Technical Information Service
Springfield, VA 22161
COSATI CODES 18 SUBJECT TERMS (Continue on reverse if necessary and identify by block number)
GROUP SUBGROUP electromagnetic wave propagation
o0 14 waveguides
inhomogeneous ]1)ssv mejia
A r Continue on reverse ,If necessary and identify by block number)

Propagation characteristics of an electromagnetic (EM) wave inside a waveguide ,Irc-

g-citly modified by the introduction of dielectrics into the guide. Frequency of cutoff,
atca.uation. and power flow distribution are all properties of the EM wave that are highly
Lilependent on the physical structure and composition within the guide. Shielding applications
take advantage of the large amount of attenuation provided by a waveguide when the
incident EM wave has a frequency lower than the cutoff frequency of the waveguide. In
practice, small sections of waveguides are inserted through the metal walls of a shielded
enclosure to construct an air passage that does not compromise the shielding effectiveness of
the shelter. Typically, however, other fluids and materials besides air must be transferred
inide the enclosure. Therefore, hoses are sometimes inse,,td throm-,h these waveguides,
which can reduce the shii.lding provided by the waveguide above an i below cutoff.
- (Cont'd)

.,.,,-
T_ ) AS RpT E] DTC USERS Uii1idSb LI IL:U
',sP,'"b t ND ,,K.iA 22b TE E U-NE(In(lude 4rc7a Code) 2.. O'F E ,V,'.
"iii e P. Mann (217) 373-7223 CECER-IMO

' a U
O0 FORM 1473, ,: -AAPR,
'% 4 APrR uoe ht i f,l "T' , ,"J ,' yi A '" ; " ,'YA
Al )h , . u,. , i bS,, ,
ul... UNCLASSIFIED
a
 111

FOREWORD

This work was pcrformed for the Directorate of Engineering and Construction, Headquarters, U.S.
Army Corps of Engineers (HQUSACE), under Project 4A16234AT41, "Military Facilities Engineering
Technology"; Work Unit MA-C59, "Electromagnetic Pulse (EMP) Validation and Design
Recommendations for Command, Control, Communications and Intelligence Facilities." The HQUSACE
Technical Monitor was Mr. L. Horvath, CEEC-EE.

This project was conducted by the Engineering and Materials Division, U.S. Army Construction
Engineering Research Laboratory (USACERL-EM) in the EMP team. Mr. Ray McCormack is the EMP
team leader and the Principal Investigator for the Work Unit. The research was done in partial
fulfillment of the requirements for the dcgrcc of Master of Science in Electrical Engineering at the
Graduate College of the University of Illinois at Urbana-Champaign, 1988. The author tharks Professor
Raj Mittra of the University of Illinois for guidance and useful suggestions in the research. Support
form USACERL in both funding and facilities also is appreciated. Gratitude is expressed to Dr. W.
Croisant, Dr. C. Feickert, and Mr. M. Mclncrney, USACERL, for many stimulating conversations,
practical suggestions, and help in preparing the thesis.

Dr. Robert Quattronc is Chief of USACERL-EM. COL Carl 0. Mignell is Commander and
Director of USACERL, and Dr. L.R. Shaffer is Technical Director.

Ac;

+'
Lv~~it Codes
C..":
/
' 4dor

II,' ,,IA 7l

I
 iv

TABLE OF CONTENTS

CHAPTER PAGE

1. INTRODUCTION .......................................... 1

2. THEORETICAL FORMULATION ............................. 6

2.1 Scalar-W ave Function Method ................. ........... 7
2.2 Multiple Concentric Dielectrics ............................. .1
2.3 Two-Dielectric Matrix Construction .......................... 17
2.4 H ybrid M odes ........................................ 18
2.5 Modal Nomenclature and Designation ........................ 23
2.6 Nonzero W all Losses ................................... . 24

3. NUMERICAL IMPLEMEN-rATION ............................ 29

3.1 Root Finding by Muller's Method ........................... 30
3.2 MDCV Computer Program ............................... 32
3.3 Results for Two Dielectrics ............................... 34
3.4 Other Axial W avenumber Plots ............................. 38
3.5 Pow er Flow ......................................... 39

4. rNHOMOGENEOUS THREE-LAYERED DIELECTRICALLY-LOADED

W AVEGUIDES .......................... ............... 65
4.1 Lossless Three-Dielectric Waveguides ........................ 66
4.2 Complex and Backward-Wave Modes ........................ 69
4.3 Attenuation in Lossy Three-Dielectric Waveguides ................ 75
4.4 Power Flow in Lossy Three-Dielectric Waveguides ............... 78

5. EXPERIMENTAL INVESTIGATIONS .......................... 111

5.1 Resonant Cavity Method Justification ........................ 112
5.2 Resonant Cavity M easurem,_nts ............................ 113
5.3 Theoretically Predicted Resonances .......................... 116
5.4 Finite Open-Ended Waveguide Measurements ................... 117
6. CONCLUSIONS AND SUGGESTIONS FOR FURTHER STUDY.........135

APPENDIX. MDCW COMPUTER PROGRAM LISTING...............139

REFERENCES..............................................156
 CHAPTER 1
INTRODUCTION

The propagation characteristics of an electromagnetic (EM) wave in-aide a

waveguide are greatly modified by the introduction of dielectrics into the guide. The

frequency of cutoff (where the wave begins propagation), the attenuation and the
distribution of power flow are all properties of the EM wave that are highly dependent on
the physical structure and composition within the guide. Shielding applications take
advantage of the large amount of attenuation provided by a waveguide when the incident
EM wave has a frequency lower than the cutoff frequency of the waveguide. in practice,
small sections of waveguides are inserted through the metal walls of a shielded enclosure to
construct an air passage which does not compromise the shielding effectiveness of the

shelter. Typically, however, other fluids and materials besides air must be transferred
inside the enclosure. To accomplish this, hoses are occasionally inserted through these
waveguides; hence, the shielding provided by the waveguide above and below cutoff can

be reduced. It is the aim of this work to investigate the EM propagation through these

loaded waveguide structures placing special emphasis on the attenuation of the wave.
This loaded waveguide problem may be separated into two main subproblems, (i)
the modal coupling and excitation of a finite inhomogeneous guide, and (ii) the modal
solutions and wave properties for an infinite inhomogeneous guide. These two parts are
related by the fact that once the excited modes are identified in the finite guide, the
attenuation and other attributes of the wave can be calculated using the infinite guide
solutions. Identifying the excited modes is a difficult task theoretically and is not
considered in this work. Experimentally, however, a large number of measurements were
performed on such an arrangement. Instead, this work emphasizes the wave properties not
directly dependent on the source excitation, but concentrates more on the inhomogeneous
 2

waveguide penetration itself and the characteristic properties of the geometry and loading

materials.
As a first approximation to the theoretical analysis of this problem, the waveguide

structure is modelled as an infinite waveguide containing layered concentric dielectrics

forming homogeneous annuli. In this vein, there has been much work done for the two

concentric dielectric problem in which a dielectric rod is inserted inside the guide.
Pincherle [1] was one of the first to consider this two-dielectric guide aid found solutions

for both the rectangular and cir,:ular gui6es. For the i'tter. only azimuthally symmetric
modes were considered. The two papers by Clarricoats [21 provide a more complete,
authoritative exposition on the two-dielectric guide and tiey also build the stage from which
many papers over the next decal". '1960's) on the topic of the rod-insert waveguide were

set. Recently, a number of papers coacerning the two-dielectric insert waveguide and
resonators have appeared by Zaki and Atia [3] and Zaki and Chen [4], [5]. In these papers,
more emphasis is placed on the nonazimuthally symmetric modes and their corresponding

field patterns and resonant frequencies for rd-loaded cavities of similar cross sections.
Until recently, very little if anything has been published for the cases in which more than
two dielectrics fill the waveguide. Bruno and I,-ridges [61 consider a variant of the three-
dielectric problem-- that of a two-dielectric lossless rod guide. Chou and Lee [7] analyze
the case of a perfect electrically conductin g (pec) waveguide with multiple coatings having
perhaps complex permeabilities. This is related to the problem at hand, but here, however.
lossy dielectric fillings rather than lossy m2gnetic coatings are relevant. In this work, the
theoretical formulation and subsequent numerical solutions for the waveguide filled with
any number of concentric, perhaps lossy, dielectrics will be presented. Both electric and
magnetic losses are allowed or any combination of pr's and e,'s; however, only electrical

losses will be considered here.

Solutions to this loaded waveguide problem hinge on finding roots of a
characteristic equation which is derived from first principles. The roots of this equation are
 3

the axial wavenumbers, kz, for a given azimuthal variation. There are an infinite number of

these roots for each azimuthal variation and developing a method by which a root can be

assigned to a mode is a large part of the total solution process. One method of identifying
roots in the inhomogeneous case, and one which is used extensively in this work, is to
"trace" the root in the complex plane as some parameter is continuously varied. The

starting point for this tracing process is chosen where the mode type and root are known,
usually the homogeneous guide case, and then the k, value is continuously computed while

some physical parameter is incremented, such as the radius of the dielectrics, the
constitutive parameters of the dielectrics, the frequency, or perhaps the amount of loss.
This tracing process becomes especially important, and also difficult, when there is much
loss, since the roots are then located somewhere in the total complex plane. In this work,
many examples will be given when very high dielectric losses are present, from which the
attenuation and axial power flow, among other things, will be examined. In particular, as

discussed at the beginning of this chapter, the three-layered, dielectrically-loaded

waveguide is very important in this study. An in-depth investigation into the attenuation

for this inhomogeneous guide will be performed in Chapter 4.

During this k, tracing process, for reasons to be discussed throughout this thesis
beginning in Section 3.3, backward-wave regions may develop for these dielectrically-
loaded waveguides. A backward-wave region is one in which the wavefronts are traveling
in the opposite direction to the net power flow. ("Net" power flow is a necessary statement
since it will also be shown that the power flow can have differing signs in different
dielectric annuli.) Clarricoats and Waldron [8] were the first to predict the existence of

such a phenomenon in the two-dielectric circular guide. This topic was extensively studied
after that with Clarricoats publishing two excellent papers on the topic [91, (101.
Experimental verification for the existence of these backward waves, in addition to the
theoretical development, was carried out by Clarricoats and Birtles [I 1I with a resonant
cavity technique and by Clarricoats and Slinn [12] using a slot in a loaded waveguide.
Both techniques showed very good agreement with the theoretically expected values. An

extensive list of applications for the rod-loaded waveguide backward-wave structure was

discussed by Waldron [13]. In this thesis, the existence of these backward waves will be

verified for the two-dielectric case and also intensively studied in the three-dielectric guide.

A new modal designation scheme for backward-wave modes will be introduced,

beginning in Section 3.5, that is more consistent with other properties of the wave, in

addition to the k., such as net power flow, boundedncss of the wave and continuity in the

k. trace. This new scheme relies on the introduction of a small amount of loss into the

dielectrics in order that the proper k, root can be chosen. Until the recent advent of

sophisticated Bessel function computer subroutines which can compute values for complex

arguments, this could not be done very easily. Along with this new modal designation

scheme comes a slightly revised look at the cutoff frequency for a mode. The issue at hand

concerning this topic is whether k, equals zero for cutoff, the phase and group velocities

are infinite and zero, respectively at cutoff, none of these, or maybe some combination.

The results of this study suggest that the frequency for which the group velocity is zero is a

more fundamental concept for cutoff than kz=O and, correspondingly, the phase velocity

becoming infinite.

In addition to backward waves forming during the k. tracing process, regions

having complex k, in the form of P3-ja can also exist, even in lossless media! Normally,

this would not be expected since wave attenuation in lossless media with finite power flow

would seem to violate conservation of energy principles. Clarricoats and Taylor [14]

predicted the existence of these complex modes in the two-dielectric waveguide. It is

pointed out there, and by Chorney [15], that these complex modes always exist in pairs

having complex conjugate axial wavenumbers so that zero net power flow is preserved

across any transverse plane of the waveguide. Therefore, conservation of energy is

enforced since no net axial power flow is present. This does not imply, however, that

locally there is no axial power flow. It will be seen later in this thesis that the axial power
 5

flow is mostly nonzero in the cross section of the guide and only the net power flow goes
to zero. Chorney also states that not only is the real power flow zero, but the complex

modes also carry no net reactive power. Laxpati and Mittra [16] also note the zero axial
power flow condition for the complex modes in lossless media and further claim this to be
a condition for these modes to physically exist in the waveguide and not merely as
"spurious," nonphysical modes. Recently, Kalmyk, Rayevskiy and Ygryumov [17] have

presented experimental evidence showing the existence of these complex modes in a

circular waveguide.
Although no net power flow is carried by the complex modes in lossless media,
once losses are introduced into the dielectrics, the relative phase difference between the two
coupled, complex modes will make net power flow possible. From a shielding application

standpoint, the amount of attenuation in "hese complex modes becomes an important issue.
The final topic considered in this thesis is the experimental measurements for the
inhomogeneous circular waveguide. There were two types of measurement arrangements

used here which applied either to loading dielectrics which were nearly lossless or to those
with significant loss. For nearly lossless dielectrics, a resonant cavity of like transverse

geometry as the three-dielectric waveguide was used to investigate the physical

phenomenon and verify the numerical computations. For dielectrics with significant
losses, a finite waveguide arrangement through a shielded enclosure was employed. The
results of both measurement techniques compared very well with the theoretical predictions
even though a number of rough approximations had to be made to recover from a lack of
proper equipment. The agreement between the calculated values (the resonant frequencies
of the cavity and the attenuation in the firite-guide arrangement) and the laboratory
measurements gives credence to the theoretical and numerical work in this thesis. Coupling
this with additional verification provided by other published work for the two-dielectric
waveguide attests to the accuracy of the theoretical development, presented next, and the

subsequent numerical solutions to this inhomogeneous waveguide arrangement.

 6

CHAPTER 2

THEORETICAL FORMULATION

The primary concern of this chapter is the theoretical development of the infinite
inhomogeneous waveguide. The particular class of inhomogeneity considered here is a
radial variation in the constitutive parameters forming homogeneous, lossy annuli with
invariance in the 0 direction. The resulting set of equations is too complicated to solve

analytically since the expansion functions for the radial variation are Bessel functions;
therefore, numerical techniques are employed. The principal objective of this numerical
analysis is to find the axial wavenumber, k. Once this has been determined, all other

quantities such as the field components, attenuation, and power flow can be calculated.
This determination of k. is based on an eigensolution concept which will indicate situations

that can exist and not necessarily those that do exist in a physical arrangement. Further
analysis of the source coupling would be needed for this. However, since a sum of

solutions is also a solution in a linear treatment, this eigensolution method is very useful in
explaining results from measurements by using either one solution or perhaps adding a
number of solutions.
The existence of hybrid modes in this type of inhomogeneous guide is one big
factor distinguishing it from the homogeneous case. A hybrid mode is one that has both
nonzero axial electric and magnetic components. This is in contrast to the homogeneous

case for which modes of vanishing axial electric or magnetic fields can be constructed. The
method of mode nomenclature for these hybrid modes is still a fuzzy issue. Many authors
have put forth their suggestions but no method has won sweeping acceptance. The method

chosen here is somewhat of a "hybrid" in that Snitzer's method [18], discussed later, is
used ip conjunction with a mode-tracing concept outlined by Waldron [19]. The latter topic

is discussed fully in the next chapter. It is stated without proof by Waldron that all of the
modes in any waveguide have a 1:1 correspondence with those of a homogeneous, pec
 7

guide of appropriate cross section. This allows a method of nomenclature based on the

homogeneous guide as a limiting case. In any event, each mode c,, be established by its
axial wavenumber. Computation of this wavenumber is possible as a solution to a set of
equations constructed from first principles. This formulation, shown next, is an exact one
and no approximations are needed.

2.1 Scalar-Wave Function Method

The cross section of the infinite waveguide having multiple concentric circular

dielectrics is shown in Figure 2. 1. (All figures and tables appear at the end of each

chapter.) It is assumed that all dielectrics are concentric about the sheath center and that the

constitutive parameters for each dielectric are constant. Maxwell's equations for a time-

harmonic field in a homogeneous space are given by

VxE = -jcogH V-D = 0

(2.1)
VxH = juE V.B =0 .

The time dependence e-J1 is assumed throughout this work and subsequently suppressed.

In (2.1), ± is the permeability of the medium and c is the permittivity. There are a number

of ways in which these equations may be combined and the fields solved to obtain the

characteristic response in a bounded environment. One general approach is to stay with the

six field components and write solutions in terms of E z and H; [20]. Another approach is

to use the vector potential method and generate solutions to a two-dimensional scalar

problem for the transverse geometry assuming invariance in a third Cartesian direction

[211. This second rnetb,.% . -hosen here for two major reasons. One, it is elegant in

principle sincL All :he fidld co."'onents for a mode can be written in terms of one scalar
function of position. Two, some interesting properties of mode designations can be more
clearly delineated when scalar wave functions are used. Whichever method is chosen,

identical results will be obtained. The only differences will occur in the intermediate steps.
What follows is a brief review of the vector potential method to solve for the

characteristic field response of a circular cylindrical waveguide. This formulation is needed

to introduce the methods and nomenclature for extending the normal, homogeneous
waveguide methods to the more complicated situation with q concentric annuli.
Using the vector potential method then, these equations can be combined and the

field quantities solved yielding

E =-VxF - jwo4A + -V(V'A)

JoR

(2.2)
H = VxA -joEF + -V(V.F)
jO(L

with A and F being the vector magnetic and vector electric potentials, respectively. The
vector potentials satisfy the vector Helmholtz equation

V2 T + k2 T = 0 (2.3)

where T = A or F and k0.--2ge. Two characteristic field responses, namely, Transverse

Electric (TE) and Transverse Magnetic (TM) to the z direction, can be constructed with
appropriate choices of these vector potentials. Then from these two mode types, all field
patterns in the guide can be expressed as a linear superposition of them. In particular, in

the it dielectric let

Fi = 0 Ai = zv4 (2.4)
 9

where Wi' is the magnetic scalar wave potential, and (2.2) reduces to

EP= . z Hp

n
(2.5)
4

= ' t 0 2 + kA2
Ezj(a2
H,=0

with k, being the intrinsic wavenumber of the medium. Since Hzi=O, the field is TM to z.

Likewise, in the i th dielectric let

F i = zllj Ai = 0

where We is the electric scalar wave potential. Then (2.2) reduces to

I a2W

Hp. =
HPi =jW,.t pa
paz - - ET =
EPi Tpo

(2.6)

1 (a2v 2 o
H=-I- ,e

Hz =jo')Il aZ2 Ez =0

Since Ezi=O in this case, the field is TE to z. The field in any annulus within the guide can

be considered as a linear superposition of these two field types.

 10o

Also, since the yj's are chosen as Cartesian components of the vector potentials,

the 1i's necessarily satisfy the scalar Helmholtz equation

V + k1 = 0. (2.7)

In addition, the geometry inside the guide remains separable with the inclusion of these

concentric dielectrics. Therefore, it is quite easy to construct solutions to the scalar

Helmholtz equation in each dielectric. Solutions to this homogeneous partial differential

equation can be found using the separation of variables method. Accordingly then,

Wi = P(p)(O)Z(z) (2.8)

By substituting this equation into (2.7) and expressing the Laplacian in cylindrical
coordinates, (2.7) can be separated into three equations, each a scalar homogeneous

ordinary differential equation in one cylindrical coordinate only. For the circular cylindrical
geometry here, the general form of Ai will be

Vi = Bn(kpip)h(nO)h(kzz) (2.9)

where B n is some linear combination of solutions to Bessel's equation, h(no) is some linear

combination of the trigonometric functions sin(no) and cos(no), and h(kz) represents the

longitudinal variation of the propagating wave. For a wave propagating in this axially

invariant environment

h(kzz) = e- k'z (2.10)

 1

is a correct choice. Here, k. represents the longitudinal wavenumber. In a transversely

bounded environment, the Bn function will be a linear combination of two independent

solutions to Bessel's equation of the form

Bn(kp p) = [aiJn(kpjP)+biNn(kpp)] (2.11)

with Jn and Nn the Bessel functions of the first and second kinds, respectively, and a, and
bi being constants. For any source-free region containing the origin, b will be zero. For a

physical system, one would expect the fields to be single valued. This implies that n
should be chosen as an integer. Finally, for the W'i to be a valid solution to (2.7) the

separation relation

ki = (0 iEj =k1+ k; (2.12)

must be satisfied. By phase matching considerations of the tangential fields, k will be the

same throughout the cross section of the guide.

2.2 Multiple Concentric Dielectrics

To formulate the problem of q concentric dielectrics, the xj's are constructed in

view of (2.9) and the ensuing discussion. In each dielectric then,

=CtJn(kpip)cos(nO)e -jiz
I
- z
W1 C13Jn(kPP)sin(nO)e
 12

-jkz
T= [C 21J(kP2p) + C22Nn(kp2p)]cos(no)e
02= [C23(kP2P) + C24 Nn(kp2p)]sin(no)e -Jkz (2.13)

- jk
z
,t= [Ct 1Jn(kptp) + Ct2Nn(kpp)]cos(nO)e
)e - jk.z
Wt= [Ct3Jn(k ,p) + Ct4Nn(kPtp)]sin(no

- jl z
q = [CqIJn(kpqP) + Cq 2 Nn(kpqp)]cos(nO)e z

-j k "'
=
=q [Cq3 Jn(kpP) + C Nn(kpP)]sin(nO)e

The coefficients C,, are unknowns at this stage and will be solved for in the next chapter.

The index r refers to the layer and s to the coefficient of the specific radial variation in the
expansion (an integer from I to 4). C12 and C 14 are both chosen equal to zero since the
Neumann function, Nn, is infinite at the origin and there is no reason to expect singularities

in the field there. By the azimuthal symmetry of this waveguide arrangement, there is a
degeneracy in the 0 variation except for the modes having n=O [7]. For simplicity, only

one of these is shown here.

The boundary conditions for this problem (that the tangential fields are continuous
across each dielectric boundary) must be applied in order to evaluate the coefficients. At

each dielectric interface there will be four equations of continuity- one equation for each
of E., E., HO and H.. This gives a total of (4q-4) equations. The total number of

unknowns is (4q-2) due to the fact that C12 and C14 have been evaluated on physical

grounds. Two more equations are needed. These, of course, are provided by the guiding
structure which determines the major propagating characteristics of the system. Assuming
in this case a pec, then E =0 ana Ez=O at p=b. Therefore, the total number of equations

and unknowns is equal at (4q-2).

 13

As mentioned previously, the total field in each annulus will be a linear

superposition of TE and TM fields to z. To apply the boundary conditions then, the fields

must be computed according to Equations (2.5) and (2.6). Proceeding, let

= iIh m (n) {[CIJfl,(kpi p ) + Ci2Nn(kpip)Ie-JkI}cos(no)

and (2.14)
= Viehe(n) = { [Ci3Jn(kpp) + Ci4 Nn(kpp)]e-jkZIsin(no)
tiV

Using this notation,

2
E =, .(2.15)

k2

I; = j.vi (2.16)

and

'J'=
jOC.IP - +" - (2.17)

= -. "Vh(no) + kpy h-(no)

Also,

H, = 9 jp (2.18)

H, a:Lz (no)-kpi hm(no) .

 14

The above primed quantities denote differentiation with respect to p as

m= [Ci 1J'(kpp) + Ci2Nnu(kpue -k (2.19)

1 = [Ci 3Jn'(kpP) + Ci 4NnI(kpp) Iejkz

Equating tangential components of E and H at p=a gives

(i) for E. -

Ez(_,
= 7 at p=a

then,
kp(.,)

or,

k t[~-) Jn(kp
1 a) + C(t..1 ) 2 Nn(k(,,a)] (.0

-k !e([Cdlnf(kPa) + Ct2Nn(kpa)] =0

(ii) for Hz
H%_,
Hztat p=a

then,
2 2

j (1)P4 t- 1 )'4 ~1
or,

+ (t-I) (2.21)
-k;,.I(t...j)[C
3J(kPta) + Ct4Nn(k, a)] 0
 15

(iii) for E
E E0,at p=a

then, k
nkz V.tl)h-(flO) + k(_)ejh~O

- kz
flzNIhe(n) + k~e he(no)1 0

or,

nk C(t 1 )lJfl(k ( )a) + q tl )2 Nf(k a)I

+ k PI[t1) 3 Jn'(kp(,Ia) + C(t...)4 Nf'(kp(,,a)I (2.22)

nk,
-i~ , C,,(Pa + Ct2Nn(kpa)I

- k,,[C,3J,'(kp a) + C%-
4 Nn"(kpa)I 0

(Iv) for H. -
= Ot
H06-) at p=a

then,
k( ) 'V(,-1 )hm(flo) + kp(LI)v hmn-

-Lz-fh m (no) + kymhm(flo) = o

or,
k P,-)[Ct-) 1J,1 (kp(, )a) + C(,l )2 N,'(kp(-,a)]

+ nk,r4)a (tI1)3Jn(k c~a) + C(t..I4Nl(kPLa)] 2.3

 16
- kp,[CtJ,'(kpa) + CtNn'(kp a)]
-C---[ Jn(k a) + Ct4 N (kpa)] = 0
Wgt a Ptnka)=0

The remaining boundary condition to apply is that for the guiding structure itself. For a
pec, the tangential electric fields E0 and E, must equal zero at p=b. This implies from

considering Equations (2.13), (2.15) and (2.17)

CqlJn(kpq b) + Cq2Nn(kpqb) = 0

and (2.24)
Cq3Jn'(kpqb) + Cq4Nn'(kpqb)=O

Equations (2.20) through (2.24) form the primary means by which this inhomogeneous

waveguide problem will be solved. Collectively, these equations may be considered to

form a (4q-2) matrix on the "variables" Cs. Since there are no forcing terms, these

equations can be written in the matrix form

[AI[C] = 0 (2.25)

where [A] is a (4q-2) square matrix and [C] is a (4q-2) column matrix. For the nontrivial

cases, (2.25) is valid only for

det[A] = 0 (2.26)

The variable that will be solved for in order to enforce (2.26) will, of course, be kz . Once

kz is known, all other variables and fields can be computed as will be shown later.
 17

2.3 Two-Dielectric Matrix Construction

It may be instructive at this point to actually construct the matrix [A] for the case

having two dielectrics. This matrix will then be compared to other work as a verification.
Let the inner dielectric have constitutive parameters gt1 , el with radius a, and the outer p2,
E2 with radius b. The guiding structure will be assumed a pec. The matrix [A] is then

constructed using Equations (2.20) through (2.24) giving

2 2
0;E l 2E 0 22iE 0 k zlE
1 2 kE
2 4

= 2a ((2.27)
[A]
kpE' E kk, k ,E 4 ' -kzE 2 -2 E 4 (.7
2 P2 cikPE~f
2a C(09 2a
0 0 E2 E4 0 0
0 0 0 0 E2 ' E 4P

where
El = Jn(kpa) E3 = N,(kp, a)

E 2 = Jn(kP2a) E 4 = Nn (kP2a)

and

C1 3
C21
[C) C2
C23
C24 _

with
 18

This matrix is in agreement with Harrington [21]. Here, the matrix could be made smaller
in size, from (4q-2) down to (4q-4) through elimination of the last two rows by combining
the third and fourth columns and the fifth and sixth columns. However, to be more general

and allow for easier computer programming with the inclusion of other guiding structure
boundary conditions, this form is preferred.

2.4 Hybrid Modes

Examining matrix [A], when n--O, coefficients C1 1 , C21 and C 2 2 are not related to

coefficients C 13 , C23 and C 2 4 and the field separates into modes TE and TM to z [21].
From (2.14) then, the Vil 's are not related, or coupled, to the 4i m 's. This is precisely
what is meant by TE or TM "modes." That is, there exists a relationship between five of

the field components describing an allowed field pattern with the sixth field component,
either E, or Hz, equal to zero. But there is no relationship between the coefficients and
hence the fields of the two mode types. In all other cases, where nO and ks)O, the fields

are neither TE nor TM, but are hybrid modes. This means that each mode in the guide will
have, in general, both nonzero E. and H.

The main emphasis here is that the field pattern in each dielectric is considered to be

a superposition of two "basis fields": the TE and TM modes. This is done in each dielectric
and the Vi 's are constructed as per separation of variables of the scalar Helmholtz

equation. The boundary conditions are enforced on the system and the expansion
coefficients are solved for. If the expansion coefficients of one mode type are not related
to, or not coupled to, the expansion coefficients of the other mode type, then TE and TM
fields, or modes, can exist in the guide. Herein lies the main point. This means that the
field can have its pattern described by just the TE field relationships or just the TM field
 19

relationships. However, if the guide is inhomogeneous, the coefficients of the TE and TM

modes become interdependent. Then the field must be written as a superposition of TE and
TM field patterns and cannot be separated. The expansion coefficients of the two mode
types become coupled in this case.
Another way to look at this situation and investigate this coupling of coefficients

concept is to go back and consider Maxwell's equations in (2.1). Not every solution to
(2.3) (with T = E or H) will satisfy MaxweU's equations. One also needs to enforce the
divergence relations in (2.1). Expanding these out in cylindrical coordinates and
substituting the linear superposition of (2.5) and (2.6) for the field components in a source-

free homogeneous medium give in the itdielectric

V'E i= 0 =

# a2, oF
a _,( 2 ,,m
"a [ t' - ~ J + _L_ OV- ) + -D-J (2.28)
k2'

+ at vim) "

(-J]zj+ .D-
[ ]""l( o
jj+a "P"I(,m (2.29)

which is only a function of Wim . Likewise,

V"l4tH i -0 =

+ a -+
(2.30)

+
a G)
2
 20

T-p J +-' + - (2.31)

which is only a function of Vie. What has tacit' y been assumed here is that

1 W i 1 i kIi 2Vi
P or1Pa -a L-o = ~a . (2.32)

This is true only if these partial derivatives are continuous and the medium homogeneous.
All of the functions in (2.32) are discontinuous at p=O. Otherwise, they are continuous in
a homogeneous space so that Equations (2.29) and (2.31) are valid, and the coefficient sets
(Cit,Ci2) can be determined independently of the sets (Ci3,Ci 4). Hence, TE/TM modes are

possible.
However, in an inhomogeneous guide this is no longer the case. Consider from
(2.28) and (2.30)

TP'i"-P D' and -- i-.- **--- (2.33)

since in general, ei=f(p) and gi=g(p). (No variation of ei and l.

i in 0 is assumed in this

study. If this is not the case, modal separation of fields TE and TM to z would not be
possible and only hybrid modes would exit.) In order for the interchanging of the partials
with respect to p and 0 to be a valid mathematical operation, the mixed partials of yi must

be continuous. In an inhomogeneous guide this will generally not be true. The Wi's will
be discontinuous. Consider Figures 2.2 and 2.3 which are graphs of the Vi's for a 2.005
in radius pec waveguide having a0.1908 in thick Teflon tube (E1 =2.1 F/m) of inner radius

0.5008 in filled with lossy methanol (F=23.9 - j15.296 F/m) at 3 GHz. (The permittivity
 21

values are from von Hippel [22], and the 4ti's and their derivatives will be solved for in

Chapters 3 and 4.) In both cases, the Vi's and their derivatives are discontinuous, within

the discretization of the plot, across each material boundary but continuous within each
annulus. This is in accordance with the previous discussion. These Vi's are discontinuous

functions in p and interchanging partial derivatives is not a valid mathematical operation.

Therefore, the two coefficient sets are related as in (2.28) and (2.30). The manner in
which the coefficients of the two mode types are coupled is not as important as the fact that

they are just related. From this information alone it is sufficient to deduce that TE and TM
fields cannot exist in the guide, but field modes having all six components each must exist,
i.e., hybrid modes. Also evident in these graphs is at p=b

This is consistent with the boundary conditions imposed on the system for E, 3 and Ez 3 .

Evidenced in this example is perhaps the biggest difference between the scalar wave
potential and field approaches to solving these EM problems. In large, the NV's will be

discontinuous functions, whereas the fields will be continuous except for bounded-charge
induced discontinuities.
Another interesting case is that of k,=O. For all n, the hybrid mode separation into

modes TE and TM also form identically to the cases when n=O. As an example, the matrix
(2.27) constructed for two dielectrics is informative. Here, n and k, occur in pairs. When
kz=O the same behavior as for n--O will occur, that is, hybrid modal separation into TE and

TM modes. However, this separation is for all modes at cutoff and not just the n---O
modes. This gives rise to another modal designation scheme, namely, as the mode tends to
cutoff, the correspondence betweeri the homogeneous cases can be correlated. Although it
has been reported that this method of classification agrees with the limiting case as the
 22

guide becomes homogeneous [23], this is not always true. If no backward-wave region

(Section 3.3) is present in the dispersion graph, the above statement is correct. As will be
seen in Sections (3.5) and (4.2), with a backward-wave region present, a small amount of
loss must be introduced into the dielectrics to choose the appropriate eigenvalue and
corresponding mode designation. It was found from the results after applying this
technique that these two schemes do not always agree if cutoff is defined by the frequency
where k, is zero. Clarricoats and Taylor [14] note something similar to this but place more

restrictions on the rod insert , value which yields this effect than is really necessary.
For more than two dielectrics, when either n or k, is zero, the modes become TE

and TM in the same fashion as with two dielectrics. As each new dielectric is added, the
original coefficient sets of the two-dielectric matrix remain uncoupled and four new
columns are added in which the new coefficient sets themselves are uncoupled. These

columns are added in virtue of Equations (2.20) through (2.23). From (2.20) and (2.21) it
is seen that the z components of the fields do not provide this coupling. It is only the 0
components which contribute. With an eye on (2.22) and (2.23), when n or k. =0 the

(Ctl,Ct2 ) become uncoupled from all other coefficients, as do (Cd,Ct4). The Ct

coefficients have the same form as the C(t-l) 's so the two sets of coefficients in the-

previous four columns remain uncoupled with the addition of four new columns.
Similarly, when n and k. are not zero, the addition of four new columns for each new
dielectric interface will not have any decoupling effect or leave any other marked
impression on the matrix [A] other than increasing the size. The work on modal
designations that has been applied to two dielectrics here and in the literature can be
extended to three or more dielectrics without modification.
 23

2.5 Modal Nomenclature and Designation

Quite surprisingly, there is no universal scheme of nomenclature for modes in

dielectrically-loaded waveguides. There have been many attempts and Bruno and Bridges
[6] give an excellent history of the major ones including a new suggestion of their own.
Beam had one of the earliest methods, which was based on the relative contributions of E z

and H. to a transverse field component [24]. If H-Imade a larger contribution, the mode
was designated HE, and if Ez larger, EH. The fundamental was assigned EH1 1. This

scheme was reported to be arbitrary since it depends on which transverse component was
observed and how far the mode is from cutoff. Snitzer devised a new scheme by which
most modem day methods are based [18]. Although Snitzer's work was with a dielectric-
rod waveguide, the extension to the bounded guide is trivial. This method considers the
HE,, mode fundamental by both definition and common usage at the time (1961). All
modes that have the same sign as the fundamental for the ratio of the coefficients of Ez, and
Hz1 ,namely, C1I/C 13 , are designated HE. All other modes with different signs are

designated EH. Note that the use of HE and EH here is just opposite to that of Beam's.
From numerical work in Section 4.2, this scheme works well for modes above cutoff but
apparently does not apply to evanescent modes since no consistent relationship among the

coefficients could be found.

An efficient and unambiguous method of classification, which can be used in
conjunction with Snitzer's scheme, is to start with the homogeneous case, either a TE or
TI mode, and gradually increase (or decrease) the radii of the dielectrics within the guide
and trace kz. Thereby, the modes HE and EH can be associated with TE and TM modes.

respectively, of the homogeneous waveguide. This is the Correspondence Idea of Waldron

[19] from which a 1:1 correspondence of all modes in the dielectric-filled guide is
maintained with those of the homogeneous pec guide as all nonideal characteristics of the
 24

former approach those of the latter. For example, the dielectric annuli radii approach zero,
the Eri approach one, or the losses in the dielectrics or waveguide walls approach zero.

2.6 Nonzero Wall Losses

If in a practical situation, the metal losses were significant compared to the dielectric
losses, the formulation for this situation can be accurately approximated by considering the
metal to be infinite in extent [25]. If the skin depth is small with respect to the guide wall
thickness, this is a good approximation. Then instead of Jn and Nn functions, In and K

can be used where In is the modified Bessel function of the first kind and Kn is the
modified Bessel function of the second kind. Since K exhibits e-P asymptotic behavior
while Inhas ekp behavior, the coefficients multiplying the Infunctions must be set equal to

zero. That is

Wq = [CqlIn(kpqP) + Cq2 Kn(kpqp) ] eC- (2.34)

1q = [Cq31n(kp qP) + Cq4Kn(kpqp)Ie-*-z

where Cql=Cq3--O and there are (q-I) dielectrics in the waveguide. No other boundary

conditions need be applied to this system and the size of the [A] matrix is reduced to
(4.q-4). This example shows that the analysis used in this chapter is general enough that it

is relatively easy to modify the equations and corresponding matrices to account for other
waveguiding boundaries, not jubt pec waveguide structures.
To actually find solutions to these radially inhomogeneous waveguide problems,
the characteristic equation (2.26) is solved yielding the eigenvalues k,. This matrix [A], as

noted earlier, is composed of higher transcendental functions and in general cannot be

solved analytically. Instead, numerical techniques are employed, which is the subject of
 25

the next chapter. ft should be noted in passing that certain special cases may be solved
analytically (for example, n= 0 or 1 and two dielectrics) and using tables of Bessel
functions to obtain numerical results. However, if lossy materials are considered or higher
modes are needed, solutions are not possible using only analytical techniques and tables of
Bessel functions, even for these special cases.
 26

(a) Oblique view

(b) End view

Figure 2.1. Geometry and physical arrangement of the infinite inhomogeneous

circular waveguide with concentric layered homogeneous annuli.
 27

V
eqe

UU

sentva ;ad oynau~vx

 28

_________ ___________________ -c
C

C U U
S.-
0
U
o
a ~N

41
'J -~

S
U

2
- g
U

U a
0
I.

C @
d ~
U ~
U *~ E~
.5
U

U
1-

U
U -
U

r r i I I I
N C - N ~) '4' ~O Q ~- ~
I I I I I I
 29

CHAPTER 3
NUMERICAL IMPLEMENTATION

The characteristic equations constructed in Chapter 2 are too complicated to solve

analytically, and require a numerical treatment instead. This chapter is concerned with
computation of these solutions through implementation of a root-finding routine to search
for the roots of these characteristic equations. A group of five Fortran computer programs
was written to accomplish this task plus many others, including the calculation of k, versus

co plots, k z versus relative radius plots, transverse power flow and more. With space
limitations present, only one code, MDCW, is listed here. The other codes are basically
derivatives of this main one.
The use of these programs and interpretation of the ensuing data will be

demonstrated first for the two-dielectric case. This dielectric-rod insert guide case has been
thoroughly investigated in the past as was noted in Chapter 1. Comparison of the results
here with those in earlier published work will lend some validity to the theoretical

development and confirm the accuracy of the computer programs. In addition to this, some
extensions of this earlier work will be presented. Most notable of these is in the area of
power flow, where it will be shown that the idea of a backward wave becomes more
sharply defined for dielectrically-loaded waveguides.
Much emphasis in this chapter is placed on the two hybrid modes HEI and EH 1.
The reasons for this are that the HE,1 mode is always the fundamental mode (has the
lowest cutoff frequency) and that the EHlII mode was found to exist in some finite
waveguide experiments conducted in the laboratory. Some astonishing modal conversions

between these two modes were observed in these experiments. Hence, for a comparison of
the theoretical predictions and the experimentally measured results, these modes are

emphasized.
 30

3.1 Root Finding by Muller's Method

Solutions to (2.26) must, in general, be obtained numerically. This equation can be

thought of as a one-dimensional function in the argument k,

f(k,) =0 . (3.1)

However, if losses are present or in some fashion k. becomes complex such as

= P -ja , (3.2)

then the one dimensionality of the function is lost and solutions to (2.26) must be found in

the complex plane. Probably the most popular numerical method for solving a generic
problem such as (3.1) is the Newton-Raphson method. The difficulty in applying this
method, however, is that the derivative of the function must be computed. Since the
derivative is unknown analytically and numerical differentiation is not a stable operation,
the Newton-Raphson method will not be used. Instead, the roots of (3.1) will be found
using Muller's method. This method is a generalization of the widely known secant
method, but an inverse quadratic interpolation is used instead of a two-point linear
interpolation [26]. Muller's method can be derived by considering the quadratic function
[27]

f(x) = a(x-x 2 ) 2 + b(x-x 2 ) + c . (3.3)

The coefficients a, b, and c can be evaluated with the specification of three initial guesses,
x0 , x1 , and x2 , for the root. The computed root of this quadratic function is
 31

X3 = x2 - (x 2-xl c (3.4)

with

X2 -X 1
q-
xl -xo
a a qf(x 2) - q(l+q)f(xl) + q2f(xO)
b a (2q+1)f(x2 ) - (+q)2 f(x1 ) + q 2 f(xo )
c S (+q)f(x 2 )

A visual image of this quadratic fit and the implementation of this method are shown in
Figure 3.1. The zeros of this unique quadratic are calculated and the next point for another
quadratic fit is located using x 1, x2 , and now, x3 . This process is repeated until the zero of

the exact function is found within a specified tolerance. The sign in (3.4) is chosen to
make the magnitude of the denominator the largest. The two main advantages of this
method are that, one, complex roots may be found due to the quadratic nature of the
interpolation and, two, the root does not necessarily have to be bracketed with the initial
guesses in order to converge. In fact, just as with the secant method, the root may not
remain bracketed after the searching process begins even if it initially was. This advantage
may also be a disadvantage, since for certain functions, Muller's method may actually
diverge from the root depending upon the behavior of the exact function near the root. All-
in-all, Muller's method is very useful and accurate for most applications of this type. It has
its own shortcomings in certain instances, most of which can be rectified by simply
restarting the search routine and changing the initial guesses.
 32

3.2 MDCW Computer Program

Muller's method forms the basis of the computer program "MDCW" (Multiple

Dielectric Circular Waveguide) which is used, among other things, to solve the
characteristic equation (2.26). (A listing of the Fortran source code is given in the
Appendix along with a brief description of the program. This program is the basic
computer code used in the study of this waveguide problem. Most of the other programs

written and used in this work are derivatives in whole or in part of the philosophy and
methods employed here.) In order to search for the roots of (2.26), the determinant of the

matrix must be computed repeatedly. That is, in each iteration of this root-finding scheme,
the function, f(k,), being evaluated is the determinant of [A]. An efficient method for

finding the determinant of a matrix is by LU decomposition [26]. This is a tridiagonalizing

method whereby the determinant is computed as the product of the diagonal elements in
either an upper or lower triangular matrix. The LU decomposition can w.so be used to
solve a set of linear equations by back substitution which will be of value later on when
computing the coefficients C,. This method then is efficient in that it can perform double

duty so to speak.
The iteration process searching for the root k, that satisfies the characteristic

equation is terminated upon satisfaction of some predetermined convergence criterion. The

criterion chosen here is for two k. values to differ by less than some user specified

accuracy. Typically 10- 8 or 10-9 is used. Although these are quite small numbers,
Muller's method converges extremely fast for the functions encountered here. In addition,

specifying small tolerances such as these can actually speed up convergence in some
instances as when this program is automated for k. versus (o plots. Calling these
"tolerances" an "accuracy" can be somewhat misleading since convergence to incorrect

answers is also possible. Once a certain root has been located, other factors must be taken
into account to verify that this is a valid root and the one being sought. These factors
 33

include the value of the determinant, the values of the coefficients and kpi's, and the

resulting field patterns which can be computed once the coefficients have been solved.
After the k. value has been found, the corresponding [C] vector (cf. (2.25)) can be

calculated. Since det[A]--O is the root-finding criterion, the rows and columns of [A] are
not all linearly independent. In general, with n=(4q-2), the rank of [A] is (n-i). Since
there are n unknowns and now (n-I) linearly independent equations, one component of the
[C] vector can be chosen arbitrarily. In this case, C 1 will be chosen equal to one. (The
reason C I is chosen and not another component is important and will become evident
when power flow is discussed.) An inhomogeneous set of equations results of the form

[a 22 ...
a2 , C13 a21
•. - :(3.5)
Lan 2 .. aj][Cq4 j tan1l

where the a's are elements of the [A] matrix. This system of equations is solved giving the
Crs's using LU decomposition plus back substitution [26].
The final step in computing the fields from (2.5) and (2.6) is to substitute the Crs'S
into (2.13). This then will give all of the fields thi'oughout the guide. As has been
mentioned in Section 2.4, the fields within the guide will be a linear superposition of fields

TE and TM to z.
One principal requirement for this program was that it be able to run on a personal
computer (PC). Using Microsoft's Fortran v4.01 and double precision real and complex
numbers (64 bits of precision for each of the real and imaginary parts), one total iteration
for three dielectrics takes approximately 2 sec on a 12 MHz clocked 286 AT compatible PC
with an 8 MHz 80287 math coprocessor. On a 386 Compaq running at 16 MHz with a 16
MHz 80387 math coprocessor, the iteration time is reduced to approximately I sec,
 34

Typically, for initial guesses within 5 to 10% of the final answer, 4 to 8 iterations are
usually needed to converge. The total times needed to find a solution using a PC, as
illustrated here, are certainly reasonable.
The process of calculating k, for particular dielectric radii can be automated,

enabling kz to be plotted as a function of a continuously increasing (decreasing) radius-

something similar to a function generator. This is quite useful, among other things, in
mode identification. As a standard, the subscripts on HEn and EHnn refer to the n th order

and the mth rank [18]. In a homogeneous guide, the rank is determined by the successive

ordering of zeros for J. In the inhomogeneous case this no longer holds. The method of

identification used here, as discussed in Section 2.5, is to use the homogeneous guide as a

limiting case. Thereby, the dielectric radii within the guide are gradually made larger and

the kz versus radius plot is made. The starting point for this plot is chosen as the desired

TE or TM mode in the homogeneous guide, usually air filled. This method of tracing out

the k., has a number of advantages over successive applications of MDCW including speed

and accuracy. A speed increase is quite obvious since all of the parameters do not need to

be entered for each increment in radius. Less obvious is the greater speed realized by using

the previous last two calculations for k, along with a new calculated guess. This generally

halves the number of iterations needed for the calculation of the new root. Accuracy is also

increased since it is much easier to stay with the same mode when tracing than with

MDCW. One can accidentally jump to other modes with the same order but different rank

if the initial guesses are too far from the desired mode.

3.3 Results for Two Dielectrics

At this point it may be instructive to consider a few examples and compare the

results with published data. Two modes of particular interest are the TEII and TMII in

addition to their hybrid counterparts the HE,, and EHI modes. In the homogeneous case,
 35

the TE1t mode is fundamental and the HE, remains so for the inhomogeneous case,

hence, its importance. The importance of the EH 1 mode will become more evident later in

this chapter and in Chapter 5 when the experimental measurements on the finite guide are
discussed. Consider a 0.24 m diameter pec waveguide with a dielectric filling having
eri=10 and Er2=l F/m as in Figure 3.2. For the TE1t mode in an air-filled homogeneous
guide, k,=14.3 rad/m at 1 GHz. There are two ways in which a k, versus relative radius

plot can be constructed as the insert dielectric rod becomes larger (or smaller) in diameter.
One way is with successive applications of MDCW while gradually increasing (decreasing)
the radii. The best way is through automation of this process and allowing the computer
program to perform these calculations. (As has already been mentioned, due to space
limitations, this program is not listed in this thesis.)
Starting the tracing process for an exceedingly small radius, the plot in Figure 3.3 is
constructed. The HE,, mode is christened in the limit, as a/b-- 0 or 1 the TE1l mode is

obtained for the homogeneous guide. One way this graph can be interpreted is as a
transition diagram from one homogeneous case to another. The k, is a function of the inset

radius and is known analytically only for the homogeneous cases, i.e., at the endpoints.
This graph indicates how the axial wavenumber varies between the two homogeneous

situations. This plot compares well with one given by Harrington [21] and is, in fact, an
exact match. The transverse electric fields for the four cases labeled A, B, C and D are
shown in Figure 3.4. In each case, EP is discontinuous at the boundary by the
accumulation of bound charge at this interface. The E0 field is particularly interesting in

that the field maximum begins outside the dielectric rod and gradually moves inside as the
rod radius is increased. Related to this is the transformation of kp 2 from a purely real to a
purely imaginary quantity at k/ko= 1. The wavenumbers are listed in Table 3.1 for these

four cases. The region much below this point is characterized by electric fields primarily
concentrated in the outer dielectric, whereas the region much above k,/ko=l has its electric
fields concentrated in the inner dielectric. The magnetic fields retain most of their shape as
 36

a/b is increased with the maximum fields located in the center dielectric. The transverse

electric fields also display an increasingly exponential behavior in region 2 as a/b increases.
This is a further manifestation of the imaginary behavior of kp 2 which gives rise to
"trapped" waves in the inner dielectric. Imaginary arguments for the Bessel functions in

(2.14) give linear combinations of modified Bessel functions which vary smoothly as
opposed to the oscillatory behavior of the normal Bessel functions. Behavior of this sort is
reminiscent of surface waves and demonstrates itself here for kp2 sufficiently large and
imaginary. If there were a coating instead of an insert (meaning er1 =I and £r2>l), then the

situation would be reversed. Discussions of this arrangement are given by Chou and Lee

[71 and Lee et al. [28].

Another interesting example is an air-filled waveguide of 1 in diameter having a
dielectric insert of Erj=37.6 F/in. For the TE, mode in an air-filled guide, kz=-j 118.3
rad/m at 4 GHz serves as a starting point for the kz versus relative radius plot, with the
results of this given in Figure 3.5. This graph has a peculiar shape in that the HEl and
EHIl modes tend to form a continuous curv- This is in stark contrast to these similar
plots for the homogeneous case. Also, a backward-wave region (where the phase and
group velocities have differing signs) has formed in the k. versus a/b plot where the slope
aw.
is negative. This backward-wave assignment can be arrived at here by rewriting a/b as

with c=speed of light and a is a constant. The x axis can now be thought of as an
increment in co instead of a/b. Although this graph was generated as a 0 versus relative

radius plot, the x axis can be regarded as proportional to frequency from which backward-
wave assignments can be made by observing the slope so long as the constitutive

parameters are not frequency dependent.

Another interesting feature of this graph is the region from a/b=0.302 to 0.358
where no purely real or purely imaginary solutions for kz exist. Instead, complex solutions

are found in the form of (3.2). These solutions are the complex or "leaky" modes
 37

discussed in Chapter I and have been investigated in some detail in the literature [14], [29],

[30]. For conservation of energy reasons, these modes exist in pairs having negative

complex conjugate propagation coefficients k, and -k*. In this trace, the backward-wave

and complex mode regions exist simultaneously. Clarricoats and Taylor [14] show that

th!z is always the c-1se for the two-dielectric inhomogeneity but the converse does not

apply. That is, complex modes can exist without propagating backward-wave regions.

Later in Section 4.2 this same occurrence will be shown to exist in the three-dielectric

inhomogeneity. Since there is no fundamental difference between three dielectrics and four

or more, it can be safely stated that for any number of dielectric layers, the existence of

propagating backward waves in a k. trace is always accompanied by a complex mode

region, but not necessarily vice versa.

Once this plot has been constructed, the transverse fields can be computed using the

k. value for the desired relative radius. Choosing a/b--0.788, then from Figure 3.5,

k,=461.7 rad/m for the HE,, mode and kz=265.1 rad/m for the EH 11 mode. The fields for

these two cases as computed by MDCW are shown in Figures 3.6 and 3.7. They are all

normalized so that the maximum field component, either E or H, has a magnitude of one.

Figures 3.6 (a) and (b) show the transverse electric and magnetic fields, respectively, for

the HEII mode. All of the fields are continuous except for the EP component which is

discontinuous by the accumulation of bound charges. Similar plots for the EH I 1 mode are

shown in Figure 3.7. These graphs compare exactly to those of Zaki [4] except for a

different normalization of the H fields. Zaki also subscribe5 to some nonstandard

nomenclature for the modal designations [3]. Graphs of these mode patterns are shown in

Figure 3.8. Figure 3.6 agrees with Figure 3.8(a) and Figure 3.7 with 3.8(b). The

excellent correspondence shown here serves as a good source of verification for the

accuracy and validity of the results from the programs for two dielectrics. For more than

two, verification will be provided by experimentation. It is not surprising that such good

agreement has been reached. The analysis used here and that of Zaki are founded on the
 38

same approach, namely, expression of the eigenmodes as a product of the expansion

functions as in Chapter 2. Figure 3.8 also contains the field patterns for the transverse E
and H fields across the waveguide cross section.

Up to this point, fairly typical topics have been discussed concerning the
multidielectric waveguide. Now, some additional discussions will be given for the two-
dielectric case. These examples, which include kz versus ca, kz versus Eri , and 3-D power
flow plots, will lend further insight into the rod-insert guide and also bridge the way for

three and more dielectrics.

3.4 Other Axial Wavenumber Plots

The first of these plots to be demonstrated is the k,--w plot for the frequencies 1 to

10 GHz. Construction of this begins by determining the starting point for the frequency
and mode of interest. This is done by tracing out the k2 versus relative radius plot at I
GHz. The results of this are shown in Figure 3.9 for b=0.5 in, cr1=3 7 .6 and £r2=l F/r.
The two modes shown are HE,, and EH1 . Again, a region of complex mode solutions
exists as for the 4 GHz case except that here it connects two evanescent hybrid mode
regions. It is interesting to observe that in this complex-mode region the distinction
between HE,, and EH 1I disappears and both modes seem to exist simultaneously. It is in
this complex-mode region where the starting point for the k,-o plot is obtained at
a/b=0.788. Since HE, and EH1 appear to exist simultaneously at this point, it suffices as
a starting point for both kz-o plots. The resulting k,--w curves are shown in Figure 3.10.
At a frequency slightly greater than 2 GHz, the modes revert from complex to hybrid with
the EH1I experiencing a small backward-wave region from approximately 2 to 2.4 GHz.
Also in Figure 3.10 are the 3-- plots above cutoff for the TEII and TMII modes
in the homogeneously filled guide of 1 in diameter with e,=37.6 F/m. At lower
frequencies, one can see a large discrepancy between the transverse and hybrid modes as
 39

expected. At higher frequencies, however, it is quite surprising that the plots all have
basically the same shape and trends. This indicates that the inhomogeneous guide could be
modelled as homogeneous at frequencies much beyond cutoff, at least for these two cases,
by selecting an effective e.
In addition to the variation of k. as a function of (o, is the important case in which
kz varies as a function of the dielectric Eri. This is useful from a number of standpoints.

One, as different loading materials are used, the cutoff frequency of the composite structure
can be observed. Two, the variation in attenuation afforded by each mode can be
scrutinized as the imaginary part of Eri (representing conduction losses) is varied. An

example of the former is shown in Figure 3.11 for the two-dielectric problem having
alb=0.788, b=0.5 in and Er2=I F/m at 4 GHz. The inner dielectric's real permittivity is
varied from I to 38 F/m for both the HE,, and EHII modes. For Er1=l F/m, both modes

revert to the homogeneous case which serves as a convenient starting point for mode
classification reasons. As the dielectric constant is increased, k increases monotonically in
these two situations. Cutoff occurs at Er1 =5 .3 and eri=13.9 F/m for the HE 1 and EH11

modes, respectively.

3.5 Power Flow

Power flow is an important topic in wavep.ide considerations and will be discussed

here for the multidielectric case. In the most general sense, the total power across some
transverse plane along the guide is given by

P =fs E x H* ds (3.6)
 40

where the * refers to complex conjugation. In general, all six components of both E and H

will exist, but since only the z directed power flow is of importance here

S = E x H*

and

SZ, = Epi" Hi* - E i-* i=1....q • (3.7)

Considering Equations (2.5) and (2.6) for the p components gives

E - L n'm~o + 2-lM (no~) (3.8)

Hp,= -[ he(no) + ihe(no) (3.9)

where the primes denote differentiation in accordance with (2.19). Substituting these two

equations plus (2.17) and (2.18) into (3.7) gives in part

E Pi iwe,
kz 2.1 ogi*p 2 W
^ 1

(3.10)
nkpiIkz 2 nkp..
+Wi'W' + "- i- )
*([hm(n3 .)

E p* -kz* I npie
2kz i

(3.11)
 41
n ~I~12 •' Ik
+ 2 + c Iih(no) '

Equation (3.7) can be used to calculate the power flow in the z direction by Equations
(3.10) and (3.11). In particular, the real part of Szi is of primary importance in this study

since it represents the time-averaged power flow in the axial direction.

An important special case is the azimuthally symmetric modes for which n=O.
Equation (3.7) then reduces to

_kz_Ik i '12
(3.12)
_ I kpi[Ci 1Jn (kpip) + C2Nn'(kpp)] 12

If the guide is homogeneous, only one coefficient will be left, namely, CI1 . To ensure

nonzero power flow, this coefficient is chosen equal to one when solving for the

coefficients [C]. This is important since it was found that in the limit as the guide becomes
homogeneous, choosing other coefficients equal to one may force the C11 coefficient to
zero in the case where n=O. Since nonzero Sz1 is to be expected, a boundary condition of

sorts is implied by this seemingly nonphysical result.

An example of the power flow through a cross section of a waveguide is shown in
Figures 3.12 and 3.13. These figures contain graphs of the three-dimensional (3-D) power
flow along with a slice through the middle of the 3-D plot in the 0=0 plane. Figure 3.12
corresponds to the point labeled A in Figure 3.3 and Figure 3.13 corresponds to point D.
These two graphs show a tremendous redistribution of the power flow from largely in the

outer dielectric to mostly in the inner dielectric. For a/b=. 1, 99.5% of the power is in the
outer dielectric while for a/b=0.4, 95% of the power concentrates in the inner. Although
the radius of the inner dielectric increases fourfold, at a/b--0.4, the inner dielectric still only
 42

comprises 16% of the cross-sectional area. A question arises as to why the power shifts in

concentration. This redistribution of power is marked by the magnitude and phase of kp2 .

As kp 2 becomes increasingly imaginary, the power tends to concentrate in the inner

dielectric as noted by these two figures and the values in Table 3. 1. Although the kp 2 --0

point at a/b=0.244 does not mark a 50/50 power distribution between inner and outer

dielectrics (this actually occurs at a/b=0.267), a large imaginary radial wavenumber does

signal a small real axial power flow in that dielectric.

Another example is shown in Figure 3.14. This plot is for the point labeled C in

Figure 3.5 having km165 rad/m in the HE,, mode. Here a/b=0.36 and b=0.5 in. One

interesting feature of this power plot is the existence of regions, in the vicinity of the 4=0

plane, where power is actually flowing backwards towards the source. In this particular

example, the total net power flow is positive with 69% of the power concentrated in the

inner dielectric and 31% in the outer dielectric and negative. However, the inner dielectric

covers only 13% of the cross-sectional area of the waveguide. This is another good
example of how the axial power flow concentrates in the dielectric with a higher real radial
wavenumber. This is a rule, in general, but the actual percentages are also a function of the
mode type and the relative dielectric radii.
The negative power flow phenomenon seen above appears to be a remnant, so to
speak, from the backward-wave region. That is, from taking a number of cases along the
03 versus relative radius plot, it was determined that the power flow in the center and outer

dielectrics is oppositely directed for all values of a/b in these two modes. The results of

this are shown in Table 3.2. Although the last entry has a positive value in the outer

region, there still is some negative flow; however, it is less than the positive flow in that

annulus. The values listed in Table 3.2 are ordered as one would trace out an ascending

path along the outside edge of the plot in Figure 3.5 beginning at point A. Examining these

percentages, a picture begins to emerge of the power flow slowly diminishing in the

backward direction within the inner dielectric and a greater percentage going forward.
 43

Further insight into this is gained through consideration of Figure 3.15 which is the power

flow plot for point A in Figure 3.5 and the first entry in Table 3.2. Here the positive power
flow is in the outer dielectric and the negative within the inner. Owing to the backward-
wave nature of the wave here, the phase fronts are toward the source and the energy

propagation is away. In the backward-wave region, the power flow is positive in the outer
dielectric, negative in the inner. The converse of this applies to the forward-wave region of
the EH 1I mode at a-b=0.5815 depicted in Figure 3.16.

This concept of a "backward wave" seems to be somewhat muddled for the multiple
dielectric cases considered here since power is actually traveling in both directions in both
the forward- and backward-wave regions! One needs to keep in mind that in the backward-
wave region, the phase and group velocities have opposite signs. To choose the correct
sign for k, as in Table 3.2, an introduction of a small amount of loss to the inner dielectric
will make the right choice apparent. Letting Erl= 3 7. 6 -jO.01 F/m for the case with

a/b=0.3646 in Figure 3.15, two roots to the characteristic equation were found using
MDCW. These are k,=68.22+jO.179 and k=-68.22-jO.179 rad/m. The first represents an

augmented, forward traveling wave while the second an augmented backward one. The

second is the correct one since the augmentation is bounded by the source and hence a
physical result. In this backward-wave region, the correct result for k is with both 13 and

ct negative. This is the reason for plotting abs(13) in Figures 3.5 and 3.10, for example.
It is also interesting to inquire as to which mode, either HE, or EH 1I, to name the

backward-wave region. By Snitzer's scheme for the sign of C1 1/C1 3 , the mode is correctly
called EH 1 1 since the sign is negative. However, if 13 were mistakenly chosen as positive,
the HE,, mode would result since the sign of the ratio then becomes positive. In the past,
the backward-wave region would indeed be labelled HE,,. In the next chapter, it will be
shown in considerable detail that this region is actually part of the EH11 mode trace.
 44

Table 3.1.

Wavenumbers of the four cases labelled in Figure 3.3 for the

HE, mode in the rod-insert guide having b=0.4, f=l GHz,
e, =10 and i 2 =1.

Case k, (rad/m) a kp (rd/m) kP2 (rad/m)

b

A 14.85 0.1 64.59 14.79

B 17.39 0.2 63.95 11.70

C 34.37 0.3 56.66 -j27.25

D 49.36 0.4 44.22 -j44.69

Table 3.2.
Axial wavenumber and percent total axial real power flow for the five
cases labelled in Figure 3.5 for both the HE, I and EH 1 modes in the
rod-insert guide having b=0.5 in, f=4 GHz, e. =37.6 and irj= 1.

Case kz(rad/m) "a flow

of total real powerue
%ne
6 in=outer

A -68 0.3646 -22.77 77.23

B -89 0.3600 -31.61 68.39

C 165 0.3600 68.76 -31.24

D 262 0.3900 94.42 -5.58

E 462 0.7880 99.78 0.22

 45

y
exact function

f(x2).

quadr ti c
f~f(XO)

initial guesses

Figure 3.1. Example of the quadratic fit procedure to locate

the zero of a function by Muller's method.

pec b

Figure 3.2. Rod-insert waveguide arrangement.

 46

.4 .

* >

CQu
0 -

04'4 c C
oi Ai C4 ej46 a
 47

-4

U C

=~ <

-J 0

00

.44
o>
o 0
CA

LL0 CU!

(M.LL.
 48

N6

-4

a
e

~
(MM/A) ~ ~
ap*joj!~j
 49

I,-

0Q

oo

0* 0

o ,
q 0 0-

fb 9 0
t I II I
- ,. o
,.4t qD

( /A opOTIOl,,';,oloOIZAS' ,,o

, ,
, , , , ,
, , , ,
 50

'o
ci'

0. 0

Wu

a.XOAUU
1-nOTO 8VOW
(1=/)
 51

Vi -

elf

W)~ Im
 52

cii
-i
o
C

-c
I-

-4
C
C

- -
- -

S
0
o
o *~ ~;KS
~
~
S ~ ~oo*
*~~3 Eoo
o ii

* > ~
h
* c
o
q -; ~u~j
o
.- ~ ~o

cii
U, g~
C cUU~
* I-c
o

I-

-o
4 S C- S I~2 e2 Cii -4 C
d ~

(u/A) *~i.u o;zasTs muz..&uuur

 53

0*
0
0

0
0

0
o
0 ,~

a 4.)
0.

C
a
o Li
* U

U
o
b'.
a
o
0
U
I-.

0*
0
0

C
K- CVI
-i -i i i .~
0* -0~6
-i 0000000000
K- CUI CI 0*-ic
ooeoooo*qo*qo*o*ooo*ooo
000000000000000000

(my) 'ptsu on.uDum SuWA*UVJ~

 54

a7

a > t

V A
*>

i*

.4.

cm >

(u/A 8POT OT-00 09JA0~a

 55

CE

0 b

C>

cuU

02 93 cis

ci ci ci o
 56

iii

790

v-a "B- -;

0
cc.
 57

iE

0 C D C
to to CD t
wC

(1=/-P~z)
x2
 58

CF

00

>0

CA

AE

7-7

t t'
 59

__________________ ___________ -o

4.) -
o

4.)

b. --
- -II

LLI
.~ *0 .~
o ~ -
I.. C~ ~

LL~ U 00
- S
C
o II
Q
Q
* I. ~
- Co
-
-h - -
U ~

.~ ~

.~ .~' 4.)
o -~
- 0 ~
4.) Co
4.)
'U
'U

4.)

- Co

I 0
o 0 0 0 0 0 0 0 0
o a C a 0 o a o
C~) N - - N
I I I

(u4dTw-aAflwDuu u~@q - ~A;~;mad) zj

 60

0.7

0.0

II
0.7

0.2

(b

Fiue31.Ailra0oe lwo h EImd a nte00paead b -)

for pon6.ntek rc o iue33hvn /3-.1 -04,f1G z

II
E,=0an
 61

e .

* 0.3
0.

3.8
b 11.4

[ .0.3

0.1

6 0*
a

-0.146 -3.1 -O.U -0.02 0.08 0.06 0.1 I.1l

aVrngttd* dtamoev (in)

(a)

(b

E,=0 n E,.

Figure 3.13. Axial real power flow of the HE1 1x mode (a) in the 4--0 plane and, (b) 3-D,
for point D in the kz trace of Figure 3.3 having a/b--0.4, b-0.4X., f=l GHz,
=10 and 2=1.
 62

0.7
0.a
0.1
0.4

* 0.3

-0.4

(b

Fiur 0.14.Ailra o e lwo h E oe()i h = ln n,()3D

fo on ntek.taeo iue35hvn /=.6 =. n
1,F,=7. f= n ,2=1
 63

0.7
* .e
0.5
0.4
S 0.3
0.

- 0.28

-0.3 -
* -0.4 -
o 0.3 -

0.6

-o.a 4~

-0.016 -0.012 -0.006 -0.004 a 0.004 0.004 0.013 0.016

(a)

(b)

Figure 3.15. Axial real power flow of the El-I I mode (a) in the 0=0 plane and. (b) 3-D,
for point A in the k trace of Figure 3.5 having a/b=0. 3646, b--O.5 in,
f=4 GHz, F, 1 =37.6 and F,=1.
 64
I
0.0
0.6

*e 0.6
0.1
. 0.4
-0.5

'. 0

o -0.1
-0.8
, -0.2..
-0.2
- -0.4
S -0.48 (

-0.87

-00

-0.016 -. 01* -0008 -0004 0 0.004 0.001 0.012l .016

wav~gbpid. ,1.inet.,n (in)

(a)

f0.6, , 376ad 2

(b)

Figure 3.16. Axial real power flow of the EH1 tmode (a) in the f-.-0 plane and, (b) 3-D,
for k=70 in the k ztrace of Figure 3.5 having alb----.58 15, b=0.5 in,
f--4 GHz, et=37.6 and 21
 65

CHAPTER 4
INHOMOGENEOUS THREE-LAYERED DIELECTRICALLY-

LOADED WAVEGUIDES

The theoretical analysis used to formulate and construct the computer programs for
the circular waveguide filled with layered, concentric dielectric media is general enough to

find solutions for any number of dielectrics having real or complex constitutive parameters.
In the preceding chapter, the cases where two dielectrics filled the waveguide were

compared with the previously published work as a verification of the computer codes in
addition to investigating the wave behavior for the rod-insert arrangement. In this chapter,
the waveguide will be assumed to be filled with three concentric, perhaps lossy, dielectrics.

The three-dielectric problem is a reasonable way in which a waveguide-below-cutoff

penetration into a shielded enclosure can be modelled when a hose filled with some material
is passed through the waveguide, into the shelter. Since the three-dielectric problem is so
important, a considerable amount of emphasis will be placed on investigating the wave
behavior in such an environment and obtaining representative samples of the key

characteristics, such as attenuation.

The amount of attenuation the wave undergoes in the waveguide is the most
important characteristic in shielding applications. Hence, the wave attenuation will be
calculated as a function of both relative dielectric radius and frequency for varying
permittivities. Only losses from the dielectrics will be considered here since the typical
waveguides are short (6-12 in) so the metal waveguide wall losses will be relatively small.
A very important result, which will be shown in Section 4.3, is that if the dielectric losses
increase in the guide, the wave attenuation is not necessarily increased and may actually be
decreased! This remarkable finding is due to a redistribution of power flow into less lossy
dielectrics, that is not possible in homogeneous waveguides.
 66

Also in this chapter, a new modal designation will be presented for propagating

modes in the backward-wave regions. This new designation will be shown to be more
consistent with other properties of the wave, such as net power flow and boundedness of
the wave. A spinoff from this discussion is an investigation into perhaps a refined
definition for the modal cutoff frequency when backward waves are present in the

dispersion graph.

4.1 Lossless Three-Dielectric Waveguides

In most applications of the waveguide below cutoff where a fluid is transferred into

a shielded enclosure, the waveguide can be modelled as a three-dielectric problem, that is, a
hose filled with some material inside an air-filled guide. Theoretical investigations into this

arrangement proceed almost identically to those in Chapter 3 for the two-dielectric guide.
The computer programs were modified slightly to reflect a hose as the second dielectric by
keeping the radius difference between the inner dielectric and that of the hose's outer radius
constant as the inner radius is increased in a kz versus relative radius plot. An example of

the results for the kz versus a/b tracing process in the three-dielectric problem is given in
Figure 4.1 for the lossless case. Here a 0.1908 in thick hose of e=3 F/m is increased in

diameter inside a 2.005 in radius waveguide at I GHz while the permittivity of the filling
inside the hose is varied for the HE, I and EH 1 modes. All u_ -.curves vary smoothly

until a/b-0.9 where the outer edge of the hose meets the waveguide wall. For radii greater
than this, the hose is gradually "pushed out" of the guide until a/b=l where the guide

becomes homogeneous. This a/b=l point is a convenient check for the correctness of the
mode classification as discussed in Section 3.3. These k, versus relative radius plots, in
addition to their interpretation as k,-o plots, can also be viewed as transition diagrams of

the axial wavenumber as the guide changes from an inhomogeneous guide of (q-1)
dielectrics to a homogeneous guide filled with a material having parameters (J-tri,Er).
 67

If the permittivity of the hose is increased from 3 to 10 F/m while all other

parameters are left unchanged, a new k. versus relative radius plot shown in Figure 4.2

results. There is little difference between this plot and the one in Figure 4.1 except when

Er, reaches 7 F/m where the HE,, and EHII mode traces form continuous curves. A

complex-mode region develops between these two curves from a/b=0.5 to 0.6.

Interestingly, a complex-mode region has formed but the propagating backward-wave

region usually associated with the complex mode is absent. This has been reported

elsewhere [14] and this graph provides additional verification of the phenomenon when

more than two dielectrics fill the guide.

The EH 11 modes in both figures contain regions of negative slope indicative of a

backward-wave evanescent mode. Two of these regions exist bracketed by forward-wave

evanescent mode regions. Although purely evanescent here, the introduction of losses will

impart propagating-wave behavior onto the fields and the amount of attenuation in these

modes will then become important. The HE,, modes do not exhibit this backward-wave

behavior in the evanescent region, but in Figure 4.2, for a/b greater than approximately

0.9, propagating backward waves are evident. (An exception to this is when eri= 2 F/m

and a/b>O.9, when the HEl mode does have a slight negative slope.)

When the permittivity of the innermost dielectric is made larger than approximately
10 F/m, the tracing process for the k, versus relative radius plots becomes complicated by

the interactions of modes with the same azimuthal variation but differing radially. Figures

4.3 through 4.6 form a series of graphs pictorially depicting these interactions in the same
waveguide airangement as above, but Er, ranging from 10 to 30 F/r. Both the HE and EH

modes with unity azimuthal variation are shown for the radial variations 1, 2 and 3. As Eri

increases, a few mode traces break closed contours with other modes and form smooth

traces which cover the entire a/b range as with the HEII and EH,1 modes in Figures 4.3

and 4.4. Conversely, some modes that were smooth across the entire range form closed
contours, as with the HE 1 2 and EH 12 modes. Where the slope becomes infinite in any of
 68

these traces, complex modes begin and continue with increasing a/b until either a proper
evanescent or propagating mode is reached. (For clarity, the complex modes are not

indicated in these plots and only purely propagating or evanescent modes are shown.)
A difficulty arises in Figure 4.5 where it is not obvious how the EHII mode can

begin at a/b=O, and then end as a TM12 mode at alb=l. Physically this is not possible,

since by the definition of the mode, the number of p variations is a constant integer. A
similar situation occurs for this mode in Figure 4.4 for a smaller Erl, but the transition

occurs between two infinite slope regions characteristic of the complex modes. It is the
HE12/EH 12 and HE 12/EHII coupled complex modes which provide the vehicle for smooth

mode traces across the entire range of a/b. Reasoning similar to this explains how the
mode transition from EHII at a/b=O to TM 12 at a/b=l can take place in Figure 4.5 since it is

these same complex modes which are providing the bridge for transition from one trace to
another. This reasoning is further justified by Figure 4.7 where the complex modes are

now included in the figure. The axial wavenumbers for each mode should, and can, be
traced by a smooth curve across the entire range of increasing relative dielectric radius.
However, as shown here, some complex modes must be included occasionally to allow for
these smooth transitions. Without them, seemingly nonphysical traces can result which are
difficult to explain as was seen in Figure 4.5.
By increasing En further to 30 F/m the traces in Figure 4.6 emerge. The
complexity in the interactions among the various modes increases a great deal in this graph
from the previous examples. The HE 12 /EH 12 closed contour in the previous graph has
separated and now the HE 13 and EH 12 modes form closed contours in addition to the
HEl /EH, modes. Without graphing these traces in this figure and those in the past three,

it would be difficult to properly identify some segments in the traces of this figure, such as
the segment S. However, by examining the other traces and observing the progression
with increasing En (which can be thought of as "pegging" the left-hand side of the traces

and slowly increasing the right-hand edge) this segment can be identified as an HE13 mode
 69

trace. Although the mode is evanescent in this region, for lossy materials, this mode and
those nearby in the graph will propagate with some large amount of attenuation.
Knowledge of the behavior of these lower-order modes is important even though they are
evanescent in the lossless case.

4.2 Complex and Backward-Wave Modes

The question arises as to what mode the backward-wave region in Figure 4.5
(which includes the point A) should be named. Following the past work of Clarricoats and
Waldron [9], [13], this region would be designated the HE, mode- in particular, a
"paranormal sub-mode" in keeping with Waldron's denotation [31]. This idea seems

reasonable since as the permittivity of the loading dielectrics is increased, the HE,1 mode's

13-co curve gradually changes from that of a homogeneous guide to one containing a

backward-wave region fcr the inhomogeneous case as in Figure 3.10 and t131. Waldron,
among others, expresses the belief that the whole curve belongs to the HE 11 mode, but in

the region below cutoff, in the backward-wave area, k, is double-valued for this mode, so
that the EH 1 mode apparently "disappears" for the range of values where the backward-
wave region exists. It will be worthwhile at this point to investigate whether or not this is a

correct assessment.
The wavenumbers and (C1 1 ,C1 3 ) coefficients for a number of points labelled in
Figure 4.5 of the HE11 and EH 1 modes in the propagating region are listed in Table 4.1.

In order to choose the correct kz root for these points and especially point A, a small
amount of loss is added to the dielectrics (Eri=20-jO.5, er2=3-j0. 1 F/m). From the table, it

is apparent that the sign of the real part of C13 for point A Agrees with points B, C and D,
all of which belong to the EH11 mode trace. In keeping with Snitzer's scheme (cf. Section
2.5), the backward-wave region is labelled EHII . If on the other hand, the root k, for

point A had been chosen on the other branch, which for the backward wave 's kz=9.3+j 1.5
 70

rad/m, the new C13 coefficient would be 40.4+jl.1. The sign of the real part of this
coefficient would then agree with the HE,, mode but the net power flow would be

negative. Since the net power flow is to be away from the source, the first root is the

correct one (it also has positive net power flow) and the backward-wave region is rightly
labelled an EH 11 mode and nowhere in the k. trace does the EH 11 mode disappear. This

idea was illustrated earlier in Section 3.5 for the two-dielectric guide in reference to the sign

of the net power flow. Depending on which branch kz was chosen to be on, the 3-D
transverse power flow plot appeared similar to either the HEII or EHII mode graphs from

other points on the dispersion plots. The correct choice was made by examining the type of
wave propagation present (either forward augmented or backward augmented) in addition

to the sign of the net power flowv.

An argument against this type of modal designation is that the HE 1 mode does not

appear to have a cutoff frequency which is not to be expected for a propagating mode. In
the usual lossless case, the cutoff frequency is defined by kz--0. From the separation
relation (2.12), the phase velocity, vp=-, is then infinite and the group velocity, v,=7,

is zero at cutoff. Since the group velocity can be a measure of energy velocity under the
appropriate conditions, the criterion of vg9=0 may be more fundamental than vP=** for

defining the cutoff frequency. In Figure 4.7 for example, the HE, 1 mode has v--O but
Vp'O at a/b-0.48 where the slope becomes infinite. This point could represent a cutoff

frequency of the two coupled complex modes as a/b increases and the cutin frequency of
two other modes, the forward-wave HEI1 and the backward-wave EH 1 .
The plausibility of such a notion as defining the cutoff frequency by v9--0 and not
necessarily with vp=oo can be investigated simply by considering the separation relation in

the definition of vg as

1 ( = --- Re{kz } = R (4.1)

 71

Z+Z*
Re{z} =
After some manipulation and noting that

4I~2
awo 41 k,,
, (4.2)

2(ogi(kz+kz*) - * - " Jkz

)

By defining the cutoff frequency as

lrn-o (4.3)

(4.3) is satisfied, by considering (4.2), in the two cases ,hen

(i)
IkI0 ~ 2 =-a 2

= =0 for c,3 real (4.4)

and

(ii)

UM- 1 -- 00 (4.5)

Equation (4.4) is the usual definition for the cutoff frequency, k,=0. Equation (4.5),
however, is a new relation valid only in dispersive situations, since for a homogeneous
guide, kP is a constant depending only on the "electrical diameter" of the waveguide.
Therefore, in a homogeneous guide, (4.5) will never be satisfied. The kpi becomes
 72

frequency dependent in the inhomogeneous guide because k, is now the known variable
from which kpi is calculated and not the other way around as in the homogeneous guide.
Clearly, kz is frequency dependent as any kz-c plot of the inhomogeneous guide will

attest. In an inhomogeneous guide, (4.5) will be satisfied and vg=O whenever a .

Although this was known beforehand, Equation (4.2) lends some insight into why an

alternative definition of the cutoff frequency is necessary in the inhomogeneous waveguide.

In this light then, the homogeneous guide criterion for cutoff, vg=O and vp=o, is a special
case of the more general situation where cutoff is defined by v--0 and vp*=o in the

inhomogeneous guide.
Additional confirmation for this method of naming the ba -wave modes is
provided by the graphs in Figures 4.8 through 4.11. Here the HE,, and EH11 mode plots
are given for the three-dielectric inhomogeneiry with a small amount of loss as F,' is varied

from 10 to 30 and Er "=0.5 F/m for eC1=CFh'-jer,". The physical parameters used to generate

these graphs are the same as before except for the hose, which has a new thickness of 0.2
in, that will also be used in the next section for attenuation calculations in the three-
dielectric lossy guide. This thickness is close enough to the previous one of 0.1908 in for
an accurate comparison. All these plots illustrate the splitting of the complex modes with-
the introduction of loss into the dielectrics. In Figure 4.8, the modes progress smoothly
from evanescent to complex and then either to propagating for HE,, or back to mostly
evanescent for the EHII mode. No propagating backward wave is encountered by these
two modes. In the complex-mode region, the HE, I mode is an augmented backward wave
while the EH 11 is an attenuated forward wave. The introduction of loss produces some

phase difference between the two complex waves such that nonzero net power flow is
possible.
The appearance of the backward-wave region begins in Figure 4.9 for eri=15-jO.5
F/m. The EHII mode becomes backward augmented here in contrast to the previous figure
 73

when the HE 11 mode was. Indeed, the backward-wave region in Figure 4.9 has 13
negative as previously predicted. This appears even more reasonable here since a smooth
trace for ]3, from a negative value in the complex-mode region for conservation of energy

reasons (in the limit as the losses go to zero), to the positive values for forward
propagation, is provided by negative values of P3
in the backward-wave region. Figures
4.10 and 4.11 contain plots displaying similar behavior for Eri=20-jO.5 and Eri=30j0.5

F/m, respectively.
In the Figures 4.8 through 4.11, the backward-wave propagating region only
appears when the EH uI mode changed from evanescent to propagating as a/b is varied from
zero to one. As was mentioned in Section 3.3, if anywhere in the k, trace there exists

propagating backward waves, this necessarily implies that somewhere else in the trace

complex modes exist. Considering these traces then, it is evident that the backward-wave
region is needed to preserve the continuity in the 03 trace and assure that conservation of

energy is satisfied in addition to the boundedness of the wave. In summary of this, under
certain conditions, complex-wave regions in the k. trace can exist without backward-wave

regions appearing. However, if backward-wave regions do exist, somewhere in the trace

complex modes form since there is no other way for the mode to end as evanescent, for
decreasing a/b, and still preserve continuity in the k, trace.
One very curious phenomenon apparent in the plots having a backward-wave
region is the metamorphosis of the modes, as a/b is increased, from either evanescent HE 1

or EHtI modes, to not propagating HE 11 and EHIt modes as would be expected, but
rather to propagating EHI1 or HEII modes, respectively. This behavior is not witnessed in

Figure 4.8 where there is no propagating backward-wave region. As an example of this,

starting with the HE,, mode at k,--j27 rad/m in Figure 4.9, as a/b is increased from zero
to one, the mode is first evanescent HE,,, then backward augmented in the complex-mode

region. The mode next becomes a propagating backward EH 11 mode and finally a forward
propagating and attenuated EH1 1 . The gradual change from an HEI -like to an EHI -like
 74

mode is clearly evident in this example. In this line of thought then, naming the backward-
wave region EH11 may seem up to the designator and not really dependent on any physical

reasoning. However, in addition to the material already presented in support of naming

this mode EH 11 , this trace clearly has propagating behavior exclusively associated with the
EH 11 mode, whereas the HEII mode is cut off on this trace. The HE 1 mode displays
propagating behavior only on the other trace which begins an evanescent EH 11 in Figure

4.9 and ends a propagating HE,, mode.

Another curiosity, which will be mentioned in .Assing, is the EH1 I trace in Figure
4.10. In Figures 4.9 and 4.11, the EH 1I trace is "bent" as it passes from 13 negative to
positive with the associated bulge in cc. In the lossless cases, as in Figures 4.4 and 4.6,

this bulge provides the transition from a backward propagating EHII mode to a forward
propagating EHt mode without v, becoming infinite, which would not be physical.
When this type of behavior in the EHt mode is observed, as in Figure 4.10, it has been

reported by Waldron [13] that a maximum backward-wave region has been attained. For
all other values of Eri, with i=l,..,q, the bulge will be smaller as evidenced by the plots in

Figures 4.9 and 4.11. This phenomenon has been previously associated with the
HE 1 1 /EHtt degeneracy at the point where 13=0 as in Figure 4.10 [101, [13]. From the
work that has been presented here, the mode is better named EH11 , so there is no mode

degeneracy as both the backward- and forward-wave regions are EH 1 1. This is certainly

only a modal nomenclature and in no way is meant to invalidate the results of the previous
authors. The material presented in this section and in the preparatory ones is meant simply
to report a new way to name the backward-wave region that is more consistent with other
properties of the wave and modal designations.
 75

4.3 Attenuation in Lossy Three-Dielectric Waveguides

The most important characteristic in this study of the three-dielectric waveguide is

the attenuation afforded by dielectric losses above cutoff. Once the modes have been
properly identified, the losses for a particular mode can be calculated as a function of
relative radius or frequency with the loading dielectric's Er as the parameter. The physical

dimensions of the waveguide and hose, for these theoretical calculations, were chosen as
representative of a typical application for the waveguide-below-cutoff penetration in a

shielded enclosure discussed in Chapter 1. A 4 in waveguide of inner radius 2.005 in was

chosen and a I in hose of inner radius 0.4909 in and thickness 0.2 in will enclose the
filling material in the center.
The Er of the hose was selected by consideration of the traces in Figure 4.12. Here,
a is plotted against frequency for the HE,, mode inside the 4 in waveguide with the 1 in
hose filled with a three-point dispersive methanol having values given in Section 5.4. The
parameter that is varied among these plots is the loss in the hose. A value of -.2'=2.7 F/m

was chosen from listings of values for rubber derivatives in von Hippel [22]. From the
traces in Figure 4.12, a value of E2"=2 F/m was selected since this value placed the a-o

trace in with a "family" of traces with higher ," 's.

It appears from this graph quite strange that more hose loss causes less wave
attenuation. The reason this occurs is as e2" increases, the power flow is "expelled" from

the inner two dielectrics and out to the third, where there is no loss and the wave is
attenuated less. If the guide walls were lossy, this effect would be less pronounced;
however, in this study, the guide is very short so the wall losses will be negligible

compared to the dielectric losses. In this section then, the attenuation afforded by dielectric
losses will be calculated using a value of Er- 2 .7 -j2 F/m for the hose and all the
permittivities of the dielectrics inside the guide assumed to be independent of frequency.
 76

The modes that will be considered for these attenuation plots are the three lowest-
ordered nonazimuthally symmetric modes-- the HE,,, EHII and HE21. Figures 4.13
through 4.18 contain traces of a versus a/b for these three modes with f '= 2 and 10 F/m.

These two values were chosen heuristically and yield enough disparity to discern trends in
the attenuation as various parameters are changed. For each e4" value, five 4i" values (1,

3, 5, 7 and 10 F/m) were chosen as parameters for the loss. (Although these plots give a
in Np/m, conversion to dB/m is possible by multiplying a by 8.6859=20 logl 0 (e).)
Probably the most noticeable characteristic of these plots is that a higher " value does not

necessarily iiply more loss as would be the case in a homogeneous guide. These graphs
clearly show that the relative amount of attenuation between two iossy filling materials is
not ony a function of the dielectric loss but also on how large the inner diameter of the pipe

is in relation to the sheath size. A specific example of this is in Figure 4.13 with a/b=0.6.
At point A, k,=1 1.1-j14.1 for Erl= 2 -j10, and at point B, kz=3.7-j24 rad/m for Er1=2 -jl
F/m. More attenuation is provided by the F.1=2-jl F/m dielectric than by the other. This

relation reverses itself as the guide becomes homogeneous, as it should, since in this case a
higher er" implies higher attenuation from dielectric losses [201. Although only the HE 1

mode was considered in more detail here, all of these modes in Figures 4.13 through 4.18

contain elements of the same basic behavior.

Another common characteristic among these modes is the existence of a region from
a/b--O to some point nearby where all traces have approximately the same values for a,
after which the traces separate. This region is the cutoff region where the modes are
propagating only slightly but are highly attenuated. The wave impedance in this region is
mostly reactive and the attenuation is primarily caused by the wavelength of the excitation
field being too large to create transverse standing waves in the waveguide. Beyond this
cutoff region, the attenuation is due to losses in the loading dielectrics. In a lossless guide,
this transition is abrupt and the cutoff frequency is prominent at k,=O. When losses are
 77

present, the cutoff becomes "smeared" somewhat and it becomes difficult to locate the

,,pecific point of cutoff.

The other range against which the attenuation will be plotted is frequency. Figures
4.19 through 4.24 contain plots of a versus frequency for the same physical arrangement

as in the previous Ct versus a/b plots- specifically, the I in hose of Er2=2.7-j2 F/r.. in the 4
in guide. The Eri values used as parameters are the same for this graph series as in the

previous, which are listed on the plots. The frequency band is slightly different for each
mode and was selected as a good example of attenuation near the cutin for each mode.
A common characteristic throughout these plots, and also very similar to the cc
versus a/b plots, is that the relative amount of attenuation between two F." values with

similar physical dimensions is dependent not only on the sizes of the c," 's but also on the

frequency. This is equiva.ent to the conclusion reached with the a versus a/b plots
provided Er is not a function of frequency, which it is not here. A similar statement can be
made for the relation between e4" and the amount of attenuation: a larger 4," does not

automatically imply higher attenuation for all frequencies in these inhomogeneous guides.
An interesting phenomenon displayed in Figures 4.19, 4.20 and 4.22 is that the
largest amount of attenuation is obtained by the smallest E" value. That is, for F, = 1 F/m,
much more attenuation is realized at higher frequencies than for a larger E,I". The reason
for this was discussed earlier in conjunction with Figure 4.12, where for the low., lip

value the power flow is concentrated primarily in the inner dielectric and moves primarily to
the outermost dielectric for larger Er" values where there is no loss. This redistribution of

power flow is marked by the magnitudes and phases of the radial wavenumbers in the
dielectrics. As discussed earlier in Section 3.5, the axial power flow concentrates in the
dielectric with a higher real radial wavenumber. With losses present, the kpi's are, in

general, neither purely real nor purely imaginary. Instead, the axial power flcw
concentrates in the dielectric having a kpi that is large and has a real part that is much larger
than its imaginary part.
 78

4.4 Power Flow in Lossy Three-Dielectric Waveguides

It was pointed out in the last section that the amount of loss present in the loading

dielectrics can have a very large effect on the power-flow distribution in the
inhomogeneous guide. This redistribution can greatly change the amount of attenuation the

wave will undergo when the dielectrics within the guide have largely differing
permittivities. The effect that the losses in the dielectrics have on the power flow and wave
attenuation can be quite pronounced depending on the many factors mentioned in the

previous section, including the relative radius of the inner dielectrics in relation to the

sheath diameter. A good example of this can be illustrated by considering the four cases
marked in Figure 4.13. The percentages of the total absolute real power flow in each

dielectric are listed in Table 4.2. At alb=0.6, the power flow is distributed in roughly a
similar manner for both cases A and B having eri= 2 -jlO and eri= 2 -jl F/m, respectively.
This is reaffirmed by the 3-D plots of the power flow in Figures 4.25 and 4.26 for these
two cases. However, the wave attenuation in each case is largely different with a=14.1

Np/m in case A and 24 Np/m in case B, even though the loss is ten times less in the second
case. When a/b is increased to 0.88, these distributions change drastically and display a
couple of interesting attributes. For one, in case C (eri=2-j 1 F/m, a=22.5 Np/m) nearly

90% of the total absolute real power flow is in the innermost dielectric while in case D
(Eri=2-jl0 F/m, a=33.4 Np/m) there is only 25%. Secondly, in case D, nearly half the

power is flowing in the pipe when it coN .rs only 19% of the cross-sectional area, with the
remainder roughly distributed between the two other dielectrics. The higher dielectric loss
in region one has caused the power flow to concentrate more in the pipe (region two) and

also in the air-filled region (region three). This tremendous change in the distribation of the
power flow is illustrated in the 3-D plots for cases C and D in Figures 4.27 and 4.28,

respectively.
 79

The attenuation characteristics have reversed from those at a/b=0.6, such that now,

for alb=0.88, the waveguide arrangement having the larger dielectric loss also has the

larger wave attenuation. That is, case C has ai=22.5 Np/m while case D, which has more

dielectric loss in region one, has ct=33.4 Np/m. By increasing the losses in region one,

less power flows in this region, in comparison to a similar guide of lesser dielectric loss in

region one (as cases A to B and D to C), but the amount of attenuation is dependent on the

relative sizes of the dielectric diameters to the sheath size.

The effects that the relative sizes of the dielectric constants have on the po,.ker-flow

distribution and attenuation can also be quite pronounced. The four cases marked in Figure
4.14 correspond to the four cases in Figure 4.13, except , ' has been increased from 2 to

10 F/r. Between cases B and F at a/b=0.6, the higher er, value unexpectedly gives rise to

a smaller attenuation- from 24 to 16.9 Np/rn. When the radius is increased to a/b=0.88 in

cases D and H, the wave is attenuated almost equally in both instances v,ith cx= 33.4 and

31.5 Np/m, respectively. Furthermore, cases C and G show the greatest amount of

difference in the power-flow distributions, but surprisingly have nearly the same
attenuation. The reason for this is that both points have ,"=l F/m, which is near that of

C1"=2 F/m producing roughly the same amount of loss in both dielectrics and, in addition,

nearly the samepercentage of power flows in the inner two dielectrics in both cases C and

G. These facts in combination produce the effect of attenuation which varies only

moderately when the dielectric constant in region one is increased by a factor of five.

These three comparisons demonstrate that the differences in the dielectric constants among

the regions in the waveguide can have a large impact on the power-flow distribution and

wave attenuation. More importantly, these comparisons also show that there is no simple

relation between the attenuation and power-flow distributions to the relative dielectric

diameters and dielectric constants for the inhomogeneous waveguide. The relationships are
very complicated and must be obtained for each waveguide arrangement individually as
 80

was done for these examples- by tracing out the k, versus a/b plots for many cases while

varying the physical parameters.

Furthermore, a large redistribution in power flow by itself does not necessarily
imply a significant change in the attenuation. For between cases B and C there is a huge
change in the power-flow distribution with only a small change in attenuation, while in

cases A and D there is a large redistribution of power flow and a large change in
attenuation. A large variation in the losses among the dielectrics is an additional factor
which must also be considered when making attenuation evaluations for varying physical

parameters such as the dielectric radii. Again it should be emphasized that the power-flow
distributions and the wave attenuation inside the inhomogeneous, layered waveguide are
very complicated functions of the dielectric radii, constitutive parameters, sheath size and
frequency. No simple rules-of-thumb are applicable to these waveguides and the correct
relationships must be obtained numerically by tracing out the axial wavenumber and
computing the desired fielded characteristic.
 Table 4. 1.
Axial. wavenumbers and field expansion coefficients for the seven
cases labelled in Figure 4.5 of the HE, and EH modes in the
iossy three-dielectric guide having F,,20-,0.5, e,.=3-j0. 1 and F, I1.

Case k.,,(rad/m) C11 C13

A -9.3-jl1.5 1.0 -40.4-j1. 1

B 4.9-jO.7 1.0 -54.6+j5.5

C 23.0-jl1.0 1.0 -58.0+j2.6

D 28.8-jl1.2 1.0 -47.5+j2.0

E 42.3-j2.9 1.0 33.3+jO.5

F 62. 1-jl1.8 1.0 40.0+jO.4

G 78. 1-j 1.4 1.0 66.389+j1. 1

 82

r- tn 00
\C 0 0

00~0 00 00(

-o-
00
t; 0% 0000
wl-
00 000
~ 0- N

(7 C

00 't. C4 e

r 00 %AI(~

~~~iC -- __ _ __4 __ _

UL
 83

oDd
to P-c to

=w~~-~ va *m4Ia& od :
 84

4J~

E I

E~ ca

CSu

0~~. 0 0 C -D0

W5 ~ ~~~~ C1
M0 iaC 0 - to o

(uqdlm-oLn*Seu~
v,* -ANud ~
 85

._

.4
-II
78

.. ,,

...

CDC

r -

M (

I- - (4I I-
-fu

o0 0 0 0 0
v 10 0o : 0
C 'p C C 0 O 'p C cmC C~

- - 4 C~ Cn
 86

V I. .

uj I u

-d -d
 87

C u-

L) -IL

E op

LLI uj

=C
wvj j - .;4~ i ,-a ~ q - .;4w d :
 88

*-II

A- c*

E.
oe

72

0 u

LT .-

aI *k
N~

II
 9

L -i
LUU

wa -

CEU

0.
cm cc
m 0
 90

U-5

c6-

U -ju

z 0>

7 - 1 i 11 I I I
M CQ-4
m Cl) P U
 91

UU

~~0

LL

3 cm -O N 1

Z31I
 92

LUU - wI

x0

II
 93

ci

6I-

'41

ot

cQo, 6
>4I

EC

x~o
 94

11
Up1
C C'

d ~ l

LL ~

4.3I

(uz/dN) vqdlv
 95

7=-

III

II L " H'

C0

- -,

Hl T " -I..

I II. I I1-

(u
l /d ) H
 96

0 311

l-L

cc >

to t 0 InJ
lccQ
m li

(am/df4 miS.-
 97

.u

0 Cur

d L_
a.

do4.)

'I, 4 JIn

~~ u

(Umldm0 wqt
 98

-- pa

ang

IU f

cia CQ

- ~ C ~ cncv

(um/dp) vqda
 99

- I-
C
-
H

*0~
z II

*
o
U
U
-o
U
U
I-

*~ ~II

*
d

* ~L

.~
g ~U*~

*~ 9
~. ~0

6
~ E

V
0 I~*

L
a
U.- -
~ >

4
U
I-

I 0
o 0 0 0 0 0 0 0 0 0 0 0
- C~i c~ ' * r- * ~ 0 -~ N
-
I I I

(uzdN) uqd~w
 100

K
I
h

-a4.) .- 9

~CN
~J lI~4

U-
~II
C
C

*
- C
U
S
~
U
* NV

~ 2
~v-~ ~

C
a ~

I C
o 0 0 0 0 0 0 0 0 0 0 0 0
- N ~ IL~ C t- C * 0 N
- -~ -4
I I

(uz/dN) ut~dju
 101

e-4

t .3
mI

TC

U'i c co C

I II II

(iu/dN) uqdtm
 102

cr~

IAl

vlqKli
cq MIm
 103

C4

C'4

u4 U

C4I

to- 10
W
(ma/dN wq--
 104

V 04

0c/N dt --
 105

"U

CF

-S

cc 0 cc

ccv I
(ua/(N uIIl
 106

u 7L

I- I

ii 4 (tU/db) tqdiu K
 107

I I0,

* 0.7
0.5

\ 0.3
* 0.7
0.2
S

i. .4

0.3
+

-0.O8 -0.02 0 0.02 0.0 0.04

(a)

(b)

Figure 4.25. Axial real poxer flow of the HE,, mode (a) in the 0=0 plane and, (b) 3-D,
for point A in the k, trace of Figure 4.13 having a/b=0.6, b=2.005 in,
f=I GHz, e, =2-j 10, er2=2.7-j2 and 3= 1.
 108

* 0.9
0.3

S. 0.7

0.5

* 0.0

S.• 0.3 -
\

6 0, I I waqad dlmb ( i n)

-0.00 -0.04 -0.02 0 0.01 0.04 0.06

(b)

Figure 4.26. Axial real power flow of the HEII mode (a) in the 0--0 plane and, (b) 3-D,
for point B in the k. trace of Figure 4.13 having a/b=0.6, b=2.005 in,
f=l GHz, er =2-jl, e. 2=2.7-j2 and e,=1.
 109
0.6

0.5

U 0.4
\

0.2

a 0.1

0 ,

-6.06 -6.64 -6.1 6 0.08 0.64 0.0

'arguitde diameter (m)

(a)

other half of power-

flow plot symmetric

(b)

Figure 4.27. Axial real power flow of the HE11 mode (a) in the 0=0 plane and, (b) 3-D,
for point C in the k. trace of Figure 4.13 having a/b--0.88, b=2.005 in,
f=1 GHz, .T=2-jI, e 2 =2.7-j2 and E,3;.
 110

0.43

S* 0.3

waeud iaee m

(a

ote af fpwr
flo pltym eti

* (.)

Figue42.Ailra.oe lwo h E, oe()i h ~ i nk)3D

0.n13nte fo .taeo Fgr .3haigab088 =.0 n
0,e,2jO f= ,=.-2ad 1
 111

CHAPTER 5
EXPERIMENTAL INVESTIGATIONS

The subject matter up to this point has been primarily directed toward the theoretical

and numerical aspects of the inhomogeneous circular waveguide and developing an

understanding of the principles involved. The emphasis now will be shifted toward the
more physical concerns of this problem and the experimental investigations. The purpose

of this laboratory work is to see how well the theoretical models and methods used in this
development match the physical phenomenon. This also provides a verification for the
accuracy of the computer programs developed and written for this project in addition to

teir utility.
Two very different experimental techniques were devised and used in these

measurements. One is a resonant cavity technique which proved very useful for
investigations with near lossless materials. The second technique is a finite open-ended
waveguide method which is much harder to model theoretically but is useful for lossy and
very lossy materials. A major problem encountered in analyzing these experiments was not
having accurate values for the permittivities of the loading materials. This made analytical
verification difficult but not impossible as additional boundary conditions which appeal to

logical argument made predicted permittivities for a select number of materials reasonable.
The finite open-ended waveguide measurements have an additional appeal in that
they are also used in the shielding measurements discussed in Chapter 1. In Section 5.4,
the results of the attenuation measurements for this arrangement will be seen well modelled
by infinite waveguides of identical cross section. This is encouraging for further
justification of the use of these programs. However, the use of these codes is limited to
explaining measured results. Their use as predictive tools is highly questionable.
 112

5.1 Resonant Cavity Method Justification

One technique for measuring the properties of the dielectrically-loaded infinite

waveguide is with a resonant cavity [11 ]. At first glance, the finite, bounded geometry of
the cavity may seem far from that of the infinite waveguide. However, the inclusion of end
plates on the waveguide does not affect the separability of the geometry inside the cavity.
That is, the interior of the cavity remains a separable geometry and the interior fields can be
found by using separation of variables as in Chapter 2. Since one would now expect the
waves to be traveling back and forth in the cavity and not propagating in only one direction,

the choice of e- j k z for the z variation must be replaced. The choice of

h(kz) = sin(kzz) or cos(kzz) (5.1)

allows for all the transverse electric fields to meet the zero field boundary condition at both
pec end plates. Therefore, for zero tangential fields (Ep and E$) at z= 0 and L, the length of

the cavity, from Equations (2.5) and (2.6)

sin(k.) =0 =

p7t
k', U- p = 0,± 1,±:2.... (5.2)

In order to satisfy this boundedness criterion in the z direction, another free parameter
(either o, g or e) must be fixed. Using (2.12), assuming gt and e are known, the resonant

frequencies can be fixed for each p, thus enforcing (5.2). This is in contrast to the
waveguide problem where the frequency is given and the kz determined. This implies that
 113

solutions to the resonant cavity problem must be found in a different manner than for a
waveguide of similar cross section.
With the geometry separable, the dispersion relation given in Equation (2.12)

remains valid for the interior of the cavity. Plots of the dispersion relations for both the
homogeneous and inhomogeneous waveguides considered earlier are contained in Figure
3.10. The methods for determining these plots in those two cases are dramatically
different, but, nonetheless, they are both graphical depictions of the dispersion relations. If
metal end caps are placed perpendicular to the guide ax;z at any two points along z, a
resonant cavity is formed for which the fields still obey the same dispersion relation as the

guide fields. The additional boundary condition in (5.2) forces a discretization of these
dispersion graphs so k. assumes only discrete values spaced uniformly by ir/L starting at
kz=0. Using the same dispersion relation for the cavity as for the infinite guide, the
resonant frequency for a particular mode can be found through this functional relation of k.
and co. Therefore, the formulation of the infinite waveguide can be experimentally tested

and verified in the laboratory with a resonant cavity! The major benefit of a resonant cavity
is that a bounded and controlled environment is provided for more accurate laboratory

measurements.

5.2 Resonant Cavity Measurements

The resonant cavity used in the laboratory measurements was fabricated from

aluminum with the interior dimensions given in Figure 5.1. With the excitation and
measuring probes located in the centers of the circular end caps, the TMo modes are
excited [321. Although the TM0I mode is not fundamental to the guide, these azimuthally
symmetric modes are easy to excite and prove very stable in the sense that small radial
variations in the probe dimensions do not exc .e other spurious modes. The inside of the

cavity is approximately 3.7 in in diameter and 5.2 in in length. The measurements

 114

performed in the laboratory were for the three-dielectric case: a hose filled with some
material. The hose was suspended concentrically along the cavity axis with two styrofoam
rings of approximately 1/4 to 1/2 in thickness. It was determined by measuring a number
of resonant frequencies for the cavity, both empty and with the styrofoam, that the rings
provided less than a 0.1% change in the center frequency of the resonance, f', implying the

styrofoam's dielectric constant is near that of air. The loss due to the rings was also small,
since Af3d B increased only slightly. The measurements of the cavity Q (fJAf 3,d) were a

source of great difficulty. The primary reason for this was the lack of repeatability in the
Af3dB measurements from one setup of the cavity to the next and from one day to the next.
Since with this cavity method the f. measurements were very consistent, the dielectric

losses, which can be derived from the Q, were measured by a different, more repeatable
method to be discussed in Section 5.4.

The excitation and measuring probes in the cavity were straight, 32 gauge wires
protruding into the cavity 0.13 in from the end cap. From trial-and-error measurements,

both the diameter and length of wire proved to be critical in two respects. One, too large a
diameter or length will perturb the geometry of the system excessively and large deviations
from the simple end-plate boundar conditions 1Ep=E=O) become evident in the smoothed

resonant peaks. Two, for a probe too short, the coupling to the fields becomes too small

and the resonances are lost in the noise.

The source of excitation for the cavity was a Wiltron 6637A signal sweep
generatr. To measure the resonances, a Wiltron 560A scalar network analyzer was
connected to the cavity output through type-N connectors. Figure 5.2 shows the output
from the network analyzer for the arrangement of an empty 0.46 in inner diameter cpvc
hose of thickness 0.0825 in with the frequency swept from 2 to 10 GHz. The resonant
peaks are quite prominent above the noise level near -65 dB. These resonances are all
associated with azimuthally symmetric TM modes. The first peak, the TM,01 resonance
near 2.3 GHz, corresponds to the cutoff frequency of the TM0I mode in an infinite
 115

waveguide of similar cross section. The loaded Q's for these lower-order resonances range
from a low of approximately 300 to a high near 1100. The next higher mode, the TM02 in
the infinite guide, is first seen in the stepped region near 5 GHz, and the TM0 3 resonances

begin near 8 GHz in the next stepped region. Justification for these designations will
become evident shortly when theoretically predicted resonances are compared with these

results.
When nose is filled with a silicone sealant (Dow Coming 732), the
resonances in FIgL.re 5.3 result. All the peaks experience a significant downward shift in
frequency, but the higher-order TM0 2 and TM03 resonances undergo the greatest amount.

A comparison of the resonant peaks for the above two cases pius those of the empty cavity

is shown in Figure 5.4. Although the permittivities of the cpvc and silicone sealant are
relatively small (near 3.4 and 2.4 F/m, respectively) there is a large downward shift in the

peaks as shown in the figure. Figures 5.5 and 5.6 show the resonant peaks of the same
cavity arrangement but having a 3/4 in cpvc hose of inner diameter 0.697 in and 0.09 in
thickness, both empty and filled with the sealant, respectively. The TMOt resonances of

the empty 3/4 in hose case occur lower in frequency than do the corresponding ones of the
1/2 in hose case. Curiously, for both the TM0 2 and TM03 modes, the peaks are higher in
frequency for the 3/4 in hose than for the 1/2 in one. When the hose is filled with the
sealant, however, all the resonances of the 3/4 in hose, including the higher-order modes,
occur at much lower frequencies than the corresponding resonances of the 1/2 in hose. All
graphs of the resonant responses shown here confirm that the density of the resonant peaks
increases with increasing frequency, making it quite difficult to distinguish the individual

peaks of the higher-order resonances.

 116

5.3 Theoretically Predicted Resonances

Comparison of the measured cavity resonant frequencies with the theoretically

calculated ones is a straightforward process. As discussed in Section 5.1, the kz-w plot

for the infinite waveguide is constructed, then the appropriate resonant frequencies of the
cavity are obtained by the discretization of the axial wavenumber into its allowed
eigenvalues. This process assumes known constitutive parameters. In reality, the 's

were unknown for the materials used in the laboratory. These parameters can be measured
of course, but due to a lack of proper equipment, this was not possible. One way to get
around this stumbling block and find an approximate e,is to assume a value, and if many
calculated resonant frequencies can be made to match laboratory measured values, then one
may assume, with some degree of certainty, that the F, isindeed the correct parameter. One

difficulty in this is that if the materials are dispersive, as most are in the microwave region,
the Er would need to be continuously varied in many small frequency bands to accurately

model the behavior. Since these bands are unknown to begin with, it is not possible to
accurately model dispersive materials using this method. Another difficulty is that the
materials must be relatively loss-free since this peak-matching method is unique, or nearly
so, only in the loss-free case. Therefore, it will be assumed here that the materials are
lossless and nondispersive.
For the cpvc hose, Fink and Christiansen [331 give a value of 3.4 F/m for the
dielectric constant at 1 GHz and tan 8=0.006. Using this value for the dielectric constant

across the entire 1 to 10 GHz band, the prcdicted resonant frequencies were calculated from
the l-.to plots for both the empty 1/2 and 3/4 in lossless cpvc hoses. Excellent agreement
was realized considering for 32 peaks which clearly matched the theoretical and
experimental results, there was an average error of less than 1% and a maximum error of
2. 1%. For the filled hose, no data for the permittivity were available on the silicone sealant
although Dow Coming has stated that E.=2.8 F/m with tan 8=0.0015 at 100 kHz [34).
 117

Using the heuristic peak matching method, a value of 2.4 F/m for the dielectric constant
was arrived at for the sealant. The TM01 , TM02 and TM03 kz--o plots for the 1/2 and 3/4
in cpvc pipes filled with a material of 4=2.4 F/m are given in Figu-e 5.7. For 29 peaks

which matched, there was again an average error of less than 1% and a maximum error of

2.1% with the measured values. A selected sample of values for these two sets of
comparisons is listed in Table 5.1. More than likely, the permitti-Aties are nu, exactly 2.4
F/m for the sealant and 3.4 F/m for the cpvc since it was required they be lossless and

nondispersive. However, it is probable that the permittivities are very near those given
above since this additional boundary condition based on the logical argument of matching

many resonances was met exceptionally well.

Finally, the resonant cavity technique inherently assumes a low-loss material(s)
within the cavity. Too high a loss will not allow the cavity to resonate since by the time the
wave traverses the length of the cavity twice, the signal is too weak to contribute to the
resonance effect. Measurements on methanol, tap water and glycerin (all very lossy) gave
the same results- the measured output was completely devoid of resonant peaks. (A

typical example of this is shown in Figure 5.8 for the 1/2 in cpvc hose filled with methanol
in the cavity.) Other techniques are needed to measure the loss effects. One method that

gave surprisingly good results was the finite waveguide excited by horn antennas to be
discussed next.

5.4 Finite Open-Ended Waveguide Measurements

To overcome the inherent shortcomings of the cavity-resonator technique's inability

to measure attenuation for very lossy dielectrics, a finite-waveguide technique was
employed. A small section of guide was attached to a panel on a shielded enclosure, as
depicted in Figure 5.9, and illuminated by a horn antenna. (A more detailed description of
this type of general experimental arrangement is given in [35].) The wave attenuation
 118

provided by the guide can be measured with regard to a set of reference measurements

performed with the two antennas in free space. This is a total attenuation from all sources
including diffraction, reflection and wall and dielectric losses. The antenna gains and
system losses are assumed removed by the reference subtraction. To approximate the
losses due only to the loading dielectrics, a new reference level can be chosen, namely, the
total guide setup with the empty hose. Then upon subtraction of this reference level from

the guide measurements with the filled hose, the diffraction and reflection losses will be
subtracted in addition to the system and antenna losses. This is an approximation since the
fields, in addition to reflection from the inner filling, will be redistributed in the guide from

the empty to filled hose cases. Hence, the losses from other sources will be modified.
However, if these other sources of loss are considered negligible and the filling material
inside the hose the only sourcL of loss, then only this material can contribute to the
attenuation once diffraction and reflection are subtracted. To reduce the reflection losses
from the inner material (which are not accounted for in this referencing subtraction
method), conical-shaped styrofoam plugs were used to help reflect the incident energy
down the guide. All these seemingly incredulous assumptions prove quite reasonable for
these finite-guide measurements since for one, the guide is only 6 in long so the wall losses
are small and two, a Teflon hose (Er=2.1-j0.000315 F/m at 3 GHz [221) is used to contain

the material and is quite lossless. Two separate examples of these finite guide
measurements will be discussed, both of which use methanol (which is very lossy) as the

filling inside a Teflon tube.

The first example is for a 3/4 in Teflon hose of inner radius 0.3733 in and thickness
0.1908 in filled with methanol inside a 1.5 in diameter aluminum waveguide of inner radius
0.7425 in and length 6 in. The results of this comparison are shown in Figure 5.10. The
measured values given are the results of four measurement cycles to verify the repeatability
of the experiment and identify any experimental anomalies such as humidity effects. The
calculated attenuation shown in the same figure is for the HEI mode of a perfect guide of
 119

similar cross section and length. As discussed earlier, this eigensolution method, used for
the calculated values, indicates only those modes that can exist and not those that actually

do in a physical arrangement. However, these two plots are remarkably similar which
strongly suggests the existence of this mode in the guide. The TM0) mode also has an

attenuation versus frequency graph similar to this one, but azimuthally symmetric modes

are not likely to be excited by the incident fields from the pyramidal horn antenna used

here. Most likely, there are many modes excited in addition to the HE, , but here it

appears that the major behavior of the attenuation is dictated by the HE 1 1 modal properties

alone.

As was the case with other materials, the permittivity values fcr the methanol are

primarily unknown; however, von Hippel [22] lists these three values

er=30.9-j2.472 F/m at 0.3 GHz

rr=23.9-j 15.296 F/m at 3.0 GHz
eT=8.9-j7.209 F/m at 10.0 GHz

Linear interpolation was used between these known ,!alue" and is shown pictorially in

Figure 5.11. The use of interpolation here is a bit presumptive since the real and imaginary

parts of the permittivity cannot be chosen independently. These quantities are related

through the Kramers-Kronig equations which specify their unique behavior for proper

satisfaction of the causality principle [36], [37], [38].

Even with this series of rough assumptions, the agreement between the theoretically

calculated attenuation for the methanol and the measured values is remarkable. Both predict

an almost linear increase in attenuation, on the log scale, as the frequency is increased. For

frequency values less than 4.2 GHz, the air-filled reference guide is in cutoff and no

comparison can be made

 120

The second example is with a 4 in diameter guide of radius 2.005 in filled with a 1

in diameter Teflon tube of inner radius 0.5008 in and thickness 0.1908 in. The same
measurement referencing method is used here as in the previous case. Excellent agreement
is evident again, with the results given in Figure 5.12 for methanol in the Teflon tube.
Surprisingly, as the frequency is increased here, the attenuation peaks near 6 GHz and
quickly decreases to very small values near 10 GHz. This behavior could not be modeUed
with only one mode, but when the effects of the EH11 mode were included with those of
the HEII, the results were very close to the measured values. This inclusion of the EH 1

mode indicates a strong modal conversion between the HE,1 and EH 1 modes with the
EH 1I dominating at higher frequencies. In the midband region where neither mode

dominates, the correspondence of measured to predicted values is marginal. This is to be

expected since the theoretically calculated values assume a dominance of one mode over all

others; clearly, this is not satisfied in the midband region. Also, above 8 GHz the

attenuation becomes negative which is not to be expected for this passive system. This

behavior is attributed to the out-of-band responses of the pyramidal horn antennas which

have single-mode operation only up to 8 GHz.

A number of observations can be made concerning these finite-guide

measurements. One, the EH 1 mode was needed in the 4 in guide measurements to

improve correspondence with measured values but not needed for the 1.5 in guide.
Examination of this mode in the small guide showed that the minimum attenuation afforded
was near 200 dB in the 1 to 10 GHz band. Clearly, this mode would not noticeably affect
the attenuation measurements. When it does dominate as in Figure 5.12, the attenuation
trend changes considerably. This introduces the next observation: the attenuation has a
negative slope and approaches a small value in the 4-in guide case at higher frequencies.
This type of behavior is curious from the standpoint that as the frequency increases, the
attenuation from the dielectric losses is decreasing. A question arises as to whether this is

reasonable and if it is unique to this physical arrangement. The attenuation as a function of

 121

frequency for the homogeneous waveguide of varying radius filled with methanol was
examined using the linearly interpolated r, values for methanol as shown in Figure 5.11. It
was observed that attenuation relationships with frequency for the TE11 and TMI1 modes

assume similar shapes as in the inhomogeneous guide by appropriate choices of the radius
(not necessarily corresponding to either guide), but at much higher attenuation (400-600
dB). However, the negative slope behavior in the high band of the inhomogeneous guide
could not be reproduced with the homogeneous guide- the attenuation is always
increasing in this band or asymptotically so. Although the conduction losses are decreasing
in the methanol with increasing frequency, in the homogeneous guide the losses remain on
the rise, while in the inhomogeneous 4-in guide they decrease. This decreasing attenuation
with increasing frequency appears to be unique to this inhomogeneous waveguide
arrangement.
 122

CU 0"",- . . 0. " ,... .00 -"

N w

- I

-~0 00 ce00 !

E i.-0 "" E'---

- i ]
-i",, ..
00v r t-: C4I
= w
- 0
C4 v: rz
-

, 0 e t ,' 0 e
C,,.-, = = = -
-=C 1 . ...
CA C

C C
CU
E
U 4 "'0='
CUI
 123

14 5.199 in

stymfoam
n~rngs

probe capped
with styrofoam

type-N cpvc pipe

connector

(a) Side view

3.732
in

(b) End view

Figure 5.1. Geomtc-y and physical arrangement of the resonant cavity

filled with the cpvc hose.
 124

00

tow

Z 100

I V .-

olovq I

(gp)
po,4oC.&ndlim
 125

__
__ _ __ __ _ __ __ _ __ _ __ __ _ __ __ _ __

I EOKL

OEOW.U

ZZOW.L c-

IZOVL-

ozow_

ZN IO

I I oasu
OIOW,-
 126

u qT

E=

lob OOO
CA

~ 0c

to~

(gp)~1. tT&n -0400

ICOKL127

ZIOVU

I 10ju
olow-

(Up)Indno
a-4o-4a
 128

ocO-o

tu~

74

izolu oc-S

ZIOI-

olowo

(13p) i4ndv4no ps-4o*1.p

 129

, >

0i
I "- "

1.. , ,.

Q
.

N N - I --

I irI

I inl n ili mm nlmll

n nlllali I
 130

______________________________________________ C

V I-
U
U
E

I-
: ~
~ 1.,
N

ZHD 9Z6S
'J

61 Q-~
o* .~ ~
ZHS 9O[~ U
C

~
00

K
oc
If~1
U
I-
Co

C C
 131

pyramidal horn waveguide under

antenna test

to tx to rX

6 in
120 dB shielded
room

Figure 5.9. Experimental arrangement for the finite-length waveguide

measurements on the hose-filled 6 in guide.
 132

E. A m

400

c(sp)
 133

4-.

NA
 134

<<

I-

-- 0

ej a

CD cc r- 0 in, N -
 135

CHAPTER 6

CONCLUSIONS AND SUGGESTIONS

FOR FURTHER STUDY

In this thesis, solutions to the radially inhomogeneous infinite circular waveguide

have been obtained numerically from an exact theoretical development beginning with first

principles. These inhomogeneous waveguides are layered with circular, concentric annuli
having, perhaps, complex constitutive parameters. The modes that were found to exist in

these waveguides are hybrid, meaning that they have both axial E- and H-fields. In
general, no TE and TM modes are possible in the inhomogeneous guide except for modes
with n=O and all modes for which kz-0. The modal derignation scheme and modal

nomenclature for these hybrid modes have not yet been standardized in the literature. The
modal designation technique used in this work takes the homogeneous guide as a limiting

case such that as the constitutive parameters of the loading dielectrics approach those of free
space or the dielectric radii are increased or decreased until the guide becomes
homogeneous, the hybrid modes approach those of-the homogeneous guide as

HEnm"+TEnm, EH.-4TMnJI

By Waldron's Correspondence Idea [19], these hybrid modes have a 1:1 correlation with
those of the homogeneous pec guide. Using Waldron's method in conjunction with
Snitzer's scheme of modal nomenclature (the sign of C 1 1/C 13 ), a unique method of

defining the proper modal designation is obtained which is both useful and necessary in the
tracing process for kz. This tracing process can be complicated by such occurrences as

mode traces intersecting other mode traces of similar azimuthal variation and by the
development of backward-wave regions in the traces when the permittivities become
 136

sufficiently large in the loading dielectrics relative to the radius of the dielectrics and the
waveguide wall.
It was discovered by applying this method of modal designation and nomenclature
that a new scheme of naming the backward-wave modes was needed since older methods
led to nonphysical results such as unboundedness in the waves and net power flow directed
toward the source instead of away in the backward-wave regions, as examples. This new
scheme becomes further justified after a small amount of loss is introduced into the loading
dielectrics. When this is done, the complex modes are found to be coupled forward and
backward waves, as expected from the conservation of energy arguments. After increasing
a/b (or some other physical parameter) to a sufficiently large value, these complex modes
decouple into forward and backward waves. As to whether either of the forward and/or
backward waves propagate, additional factors must be considered, such as the relative sizes
of the e values and relative dielectric radii. This idea of introducing a small amount of loss

to separate the roots is a very useful tool. Here, however, it complicates matters
considerably since the complex determinant of a matrix filled with Bessel functions of
complex arguments must be computed.
In addition to using dielectric losses as tools for separating roots, these complex -

constitutive parameters also provide attenuation of the wave which is especially important
in shielding applications. In this study, only the attenuation provided by dielectric losses
was examined since this was anticipated to be the major source of attenuation in the
inhomogeneous guide above cutoff. There was a heavy concentration on the attenuation
provided by the lossy three-dielectric guide since this is a good model for the hose through
the waveguide-below-cutoff penetration. The results of this investigation showed that
contrary to homogeneous guides, there are no simple rules-of-thumb for computing the
amount of attenuation the wave suffers in relation to either the amount of dielectric losses or
the relative sizes of the dielectric radii. It was concluded that each case must be analyzed
separately by tracing out the axial wavenumber as a function of relative dielectric radius or
 137

frequency, as examples. It was also shown that the redistribution of power flow in the
dielectrics can vary the attenuation in counterintuitive ways. For example, in one case,
even though the tosses were inireased considerably in the inner region of a three-region
waveguide, the attenuation was shown to decrease appreciably.
As a verification of the theoretical formulation and the numerical solutions, two
types of experimental measurements were performed. These measurements were separated
into those for near lossless dielectrics where a resonant cavity technique was employed and

those for lossy materials where the finite guide arrangement for the waveguide-below-
cutoff penetration was used. Both gave excellent agreement with the theoretically predicted
values. In the finite guide arrangement, a number of speculative assumptions were made
concerning the reference measurement subtraction from the complete, loaded waveguide

arrangement so that only the losses from the loading material would be observed. This was
necessary, since in this study, only the characteristic properties of the waveguide itself

were scrutinized and the source properties were not considered.

For future study into this inhomogeneous waveguide topic, an important
characteristic of this penetration which deserves considerable attention is this source
coupling to the waveguide. By considering the particular source of EM excitation along

with the waveguide geometry, the total penetration arrangement could be analyzed such that
a determi~tion of the actual wave attenuation, including diffraction and reflection effects,
could be calculated in a deterministic fashion rather than identifying excited modes and the
corresponding attenuations which have occurred. This would allow for a predictive tool in
the analysis of the shielding effectiveness for waveguide penetrations.

As additional study into this phenomenon, actual measured values for the
permittivities of the loading dielectrics could be obtained, and using these new values as
input to the tracing programs, more accurate comparisons could be performed with the
measured field quantities. The additional boundary condition of matching many resonant
frequencies applied in this study is logically consistent; however, using actual measured
 138

permittivity values is more appealing since the losses present in the loading dielectrics will

have some effect on the resonant frequencies. In addition, materials with dispersive
constitutive parameters could be used in both measurement arrangements with a higher
degree of accuracy.
Finally, a new precision resonant cavity fabricated from brass, rather that
aluminum, would allow for more accurate resonant frequency measurements in addition to

repeatable, accurate Q measurements. From the Q, another measure of the dielectric losses
could be correlated with the theoretical calculations.
 139

APPENDIX

MDCW COMPUTER PROGRAM LISTING

The program MDCW is a user interactive computer program for finding the kz roots,

the resulting field coefficient vector and the transverse fields for a circular waveguide filled
with q concentric, layered dielectrics. Once all the data for each annulus has been entered
and three guesses near a root given, the program fills the matrix [A] in subroutine
MTRXVAL. Here, by virtue of Equations (2.20) through (2.23), values are added row-
wise to the matrix. These equations contain the Jn and N n Bessel functions which are not

trivial to compute, especially for complex arguments. Recursion schemes to calculate these
Bessel functions suffer from excessive error corruption if upward recursion is used for
more than a few orders. Fortunately, Donald Amos recently published a package of higher
transcendental function subroutines which give incredibly accurate and fast results without
recursion [391. Instead, ti. uses a number of series approximations, in appropriate ranges
of the argument, with syster hardware dependent coefficients to maximize the available
precision. This package will compute Jn and N n for positive n and complex arguments in

addition to other. special functions.

The computation of these Bessel functions and their derivatives is carried out in

subroutine TOTBES. Evaluation of the determinant of the filled matrix is the last operation
performed in subroutine MTRXVAL. Here, as discussed earlier in Chapter 3, LU
decomposition is used through subroutines DETI and LUDCMP. It is the value of the
determinant which is returned to the main program. Once the determinant becomes small
enough, as directed by the root finder, signalling the location of a root, the last section of
the main body of MDCW computes the transverse fields and the scalar potentials as a
function of the waveguide diameter for this k. root.
 140

program mdcw 10

* Program -- Multiple Dielectric Circular Waveguide (mdcw) *

* Written by Keith W. Whites *

* This version 10 dated March 16, 1988 *

* This program will calculate the axial wavenumber for a circular waveguide *
* containing q concentric dielectrics having complex constutive .ir3rineters. *
* Data needed by the program for these calculations are frequency, rumber of *
* dielectrics, number of azimuthal variations, and the constitutive
* parameters of the dielectrics and their dimensions. The numerical root *
* finding technique used is Muller's method which configures three points for *
* a quadratic fit and uses a root of that quadratic fit to move towards *
* the zero of the expression to be evaluated - in this case the det[A]. *

Output Files: fldmag.wvd - transverse E and H fields (magnitudes, phases, *

* real and imaginary parts) as a function of *
* the guide radius. *
* psie(m).wvd - scalar electric and magnetic potentials as a *
* function of guide radius. *
* pwrin.wvd - input file for "pwrflw3" code which calculates *
* the total transverse, % distribution of real, *
* and 3-D plot of power flow. *
* zkztra.wvd - trace of kz (axial wavenumber) in complex *
* plane as the root finder searches for the *
* correct root. *

* Variables: arg - argument of the Bessel functions *

* besj 1 - Bessel function of the first kind (J) of the *
* inner dielectric of the two under evaluation *
* besj2 - J of the outer dielectric *
* besn 1 - Bessel function of the second kind (N) of the *
inner dielectric of the two under evaluation *
* besn2 - N of the outer dielectric *
* bes**p 1 - Bessel function of order plus one *
* coeff - coefficients [C] for the field expansions *
* dbes** - derivative of the Bessel function '**' *
* dc - array of matrix elements for which kz is to be *
* determined from *
* freq - radian frequency *
* ipart - number of divisions the radius is divided into for
* the field and scalar wave functions plots
* ix I - matrix row pointer *
* ix2 - matrix column pointer *
* nord - order of mode, number of azimuthal variations *
* numdi - number of dielectrics *
 141
* psie(m) - scalar electric and magnetic potential functions *
* radii - radius of dielectrics and the conductor (m) *
* sacc - accuracy sought for kz value *
* ump l,ump2,ump3 - determinant values for three guesses
* to be used with Muller's method *
* uma,umb,umc - updating information for Muller's method *
* zeps - permittivity *
* zkro - radial wavenumber *
* zkz - axial wavenumber *
* zmu - permeability *

* Subroutines: detI - calculates the determinant of a double precision *

* complex matrix using lu decomposition *
* lubksb - called by deti to perform the backsubstitution
* after the LU decomposition in order to calculate *
* the coefficients [C]
* mtrxval - calculates the elements of the matrix from *
* the boundary conditions for the multiple *
* concentric dielectric circular waveguide *
* totbes - calculates the Bessel functions and their *
* derivatives for both dielectrics bordering *
* an interface *
zbesj(y) - called by totbes to calculate the Bessel *
* functions of the first and second kinds with *
* complex arguments

implicit complex* 16(a-h,o-r,t-z)

implicit real*8(s)
parameter (pi=(3.14159265359dO,O.OdO),maxdie=5,maxcon=maxdie*4-2)
parameter (zepsO=(8.854d- 12,0.OdO),zmu0=(1 2.56637062d-7,0.OdO))
parameter (ipart= 100,ibesnum--maxdie)
complex* 16 zkro(maxdie),radii(maxdie),zmu(maxdie),
+ zeps(maxdie),dc(maxcon,maxcon),coeff(maxcon),
+ efra(ipart),efphi(ipart),hfra(ipart),hfphi(ipart),
+ psie(ipart),psieder(ipart),psim(ipart),psimder(ipart)
complex* 16 worksp(maxcon),indx(maxcon)
real*8 siunk 1(ibesnum),sjurk2(ibesnum)
real*8 sbjr(ibesnum),sbji(ibesnum),sbyr(ibesnum),sbyi(ibesnum)
integer*4 ipivot(maxcon)
common/combes/nord,zkro,radii
common/mtrx/freq,zrnu,zeps,numdi,sacc,dc

open(unit= I ,file='a:zkztra.wvd',status='unknown') :set up

open(unit=23,file='a:fldmag.wvd',status-'unknown') :output fies
open(unit=26,file='a:plwrin.wvd',status='unknown')
open(unit=29,file='a:psie.wvd',status='unknown')
open(unit=3 I,file='a:psim. wvd',status='unknown')
 142
itcont=O
mstp=0
iflagd= I
sguinc=O.Od0
write(*,*) 'enter frequency(Hz) and the number of dielectrics'
reaa(*,*) sfreq,numdi
freq--dcmpx(sfreq)*(2.OdO,O.OdO)*pi
write(*,*) 'enter order'
read(*,*) nord
write(*,*) 'enter 3 kz guesses'
write(*,*)' (middle guessjlar.gest,smallest:recommended)Y
read(*,*~) zkzl1
szkzre=dreai(zkz 1)
szkzim=dixnag(zkz 1)
write(1 1,12) szkzre,szkzim
read(*,*) zkz
szkzre-=dral(zkz2)
szkzim=di-ag(zkz2)
write(I 1,12) szkzre,szkzi--n
read(*,*) zk3
szkzre=dreal(zkz3)
szkzixn-dimag(zkz3)
write(] 1,12) szkzre,szkzim
write(*,*) 'desired acc-aracy?'
read(*,*) sacc

do 10 i=l,numdi
write(*,*) 'enter radius, mu, and epsilon'

radii(i)=dc-mplx(sradii)
zrnu(i)=zmu(i)*zmuO
zeps(i)=zeps(i)*zeps0
10 continue
write(*,*) 'Calculate Field Magnitudes? (1/0)'
read(*,*) magcal
" with 3 itz guesses, call subroutine mtrxval to calculate matrix elements
"and the value of the resulting determinant

call mrrxval(zkzl ,det,iflagd)

umplI=det
call mtrxval(zkz2,detiflagd)
ump2=det
call mrrxval(zkz3,det,iflagd)
ump3=det
*the 100 goto loop utilizes Muller's method to determine the kz value

100 itcont--itconts-
umnq=(zkzl -zkz2)/(zkz-u3)
uma=umq*ump I-umq*(( 1.OdO,O.OdO)+umq)*ump2+umq*umq*ump3
umb=((2.OdO,0.OdO)*umq+( 1.OdO.0.OdO))*ump 1+-(1 .OdO,O.OdO)+umq)
+ *('( IOdO,O.OdO)+umq)*ump2+umq*umq*ump3
 143

umc=(( I.OdO,O.OdO)+umq)*ump I
uradi=sqrt(umbub-(4.OdO,.OdO) *upa*umc)
if (abs(umb+uradi).gt.abs(umb-uradi)) then
uradi=umb+uradi
else
uradi=umb-uradi
endif
um3up
ump3=ump2
ump2=ump 1
zkz3=zkz2

*calculate a new point and update last two points for quadratic fit

zkzl=zkz2-(zkz2-zkz3)*(2.0dO,0.OdO)*umc/uradi
szkzze=dreal(zkz 1)
szkzim-dimag(zkz 1)
write(l 1,12) szkzre,szkzi.m
call mtrxval(zkzl ,det~ifagd)
umpl=det
*if kz varies less than the specified value, then end

if ((cdabs(zkzl-zkz2).lr.sacc)) then
write(*,*) 'finished'
write(*, *) *kz-',zkzl
write(*,*) 'total number of iterations = ',itcont
write(*,*) 'determinant of matrix = ',det
do 200 i=1,numdi
if(abs(zkro(i)).lt. I .OdO) then
write(*,*) 'potential error: zkro small'
write(*,*) 'zkro(',i,')=',zkro(i)
endif
write*,*) 'zkro(',i,')=',zkro(i)
200 continue
write(*,*) 'finished'
rnstp=lI
endif
if(mstp) 100,100,300
300 continue
if(magcal)999,999,305
305 continue
maxnum=4*numdi-2
 144

call mtrxval(zkzl,det,0)

do 310 i=1,(maxnum-1) :shift matrix up

coeff(i)=-dc(i+1,1) :and to the left
310 continue :one element

do 315 i=l,maxnum
do 317 j=1l,maxnum
317 cniueciljl
315 continue
31* oniu

*calculate field expansio,- coefficients

call ludc-np(dc,maxnum-l,zraxcon,indx,d)
call lubksb(dc,maxnum- I ,maxcon,indx,coeff)
write(*,*) 'the coefficients are:'
do 320 i=maxnum.2,-1
cocff(i)=coeff(i- 1)
write(*, *) 'coeff(',i,')',coeff(i)
320 continue
coeff( I)=( I .OdO,0.OdC) :arbitrarily define
write(*,*) 'coeff( 1)',Coeff(1)
guidsp=madii(nunxli)/dcmplx(ipax )
*calculate transverse E and H fields knowing the kz and coefficients

do 400 j=1Ilpari
sguinc=sguinc-tdreal(guidsp)
if(sguinc.lt~drealI(radii(I))) then
arg=zkro(l1)*dcmplx(sguinc)
call zbesj (dreal(arg),dirag(arg),dble(nord), 1,2.sbjr,sbji,
+ nz,ierr)
if(ierr.ne.0) write(*,*) Bessel Routine error #',ierr
call zbesy(dreal(arg),dimag(arg),dble(nord), 1,2,sbyr,sbyi,
+ nz,sjunk I ,sjunk2,ierr)
if(ierr.ne.0) write(*,*) 'Bessel Routine error #',ierr
bj=dcmplx(sbjr(l1))+(0.OdO, I .OdO)*dcmplx(sbji( 1))
bjp I=dcmplx(sbjr(2))+(0.OdO, 1.OdO)*dcrnplx(sbji(2))
dbj=dcmplx(nord)*bj/arg-bjp I
efra(j)=-zkzl/(freq*zeps(l))*coeff(l)*zkro(l)*dbj
+ -dcmpix(nord)/dcmplx(sguinc)*coeff(2)*bj
hfraoj)=-zkz /(freq*zrnu(l1))*coeff(2)*zkro(lI)*dbj
+ -dcmplx(nord)/dcmplx(sguinc) *coeff(l1)*bj
efpluij)=zkz I *dcmpIx(nord)/(freq*zps()*~
+ dcinpLx(sguirc))*coeff(l1)*bj+zkro(lI)*coeff(2)*dbj
hfphioj)=-zkz 1*dcmplx(nord)/(freq*zmu(l1)*
+ dcmplx(sguinc))*coeff(2)*bj-zkro(lI)*coeff( 1)*dbj
 145
psimrnO)=coeff(I)*bj .calculate electric
psimnder j)=coeff( 1)*dbj :and magnetic
psieoj)=coeff(2)*bj :scalar potentials
psieder(j)=coeff(2)*dbj

else
do 410 i=2,nuxndi
if(sguinc.ge.dreal(radii(j- 1)).and.sguinc.le.
+ dreal(radii(i))) then
irng--i
endif
410 continue
arg=zkro~irng)*dcmplx(sguinc)
call zbesi(dreal(arg),dimag(arg),dble(nord), 1 2,sbjr,sbji,
+ nz~ierr)
if(ierr.ne.0) write(*,*) 'Bessel Routine error #',ierr
call zbesy(dreal(arg),dimag(arg),dble(nord),1I,2,sbyr,sbyi,
+ nz,sjunkl ,sjunk2,ierr)
if(ierr-ne.0) write(*,*) 'Bessel Routine error #',ierr
oj~dcmplx(sbjr(l1))+(0.OdO, 1.OdO)*dcmix(sbji( 1))
by=dcmplx(sbyr( 1))+(0.OdO, 1.OdO)*dcmplx(sbyi( 1))
bjp 1=dcinplx(sbjr(2))+(.OdO, 1.OdO)*dcrnplx(sbji(2))
byp 1-dcmpLx(sbyr(2))+(0.0dO, 1.OdO)*dcmrplx(sbyi(2))
dbj=dcniplx(nord)*bj/arg-bjp 1
dby=dcmplx(nord)*by/arg-byp I
ii=4*irng-5
efraj)-zkz 1/(freq*zeps(irng))*(coeff(ii)*
+ zkro(irng)*dbj+coeff(ji+l1)*zk(ing)*dby)
efrao)=efrao)-dcmplx(nord)/dcmplx(sguinc)*(coeff(ii+2) *
+ bj+coeff(ii+3)*by)
hfra(,)=-zkz 1/(freq*zxnu(irng))*(coeff(ii+2)*
+ zkro(irng)*dbj+coff(i+3)*ro(rng)*dby)
hfraoj)=hfraoj)-dcmplx(nord)/dcmplx(sguinc)*(coeff(ii)*
+ bj+coeff(ii+l)*by)
efphioj)=zkz 1*dcmplx(nord)/(freq*zeps(irng)*
+ dcmplx(sguinc)) *(coeff(ii)*bj+coeff(iji I)*by)
efphioj)=efphioj)+zkro(irng)*(coeff(jj+2) *dbj+
+ coeff(ii+3)*dby)
hfphij)=-zkz1I *dcmplx(nord)/(freq*zmu(img)*
+ dcmptx(sguinc))*(coeff(ii+2)* bj+coeff(ii+3)*by)
hfphij)=hfphi)-zkro(imrg)*(coeff(ii)*dbj+
+ coeff(fi+ 1)*dby)
psm)*eff)b~of~i )b
psimd(j)=coeff()*dbj+coeff(+ 1)*by
psixe)=coeff(ii2)*bj+oeff(ii+ )*by
psied(j)=coeff(ii+2)*bj+coeff(ii+3 )*by
endif
 146

400 continue
sfdmax=0.OdO
*write out the Eand Hfields

do 420 i=1,ipart
sfdxnax--max(abs(efra(i)),abs(hfra(i)),abs(efphi(i)),
+ abs(hfphi(i)),sfdmax)
420 continue
sguinc=0.OdO
write(23,*)' radius E ro (inag) H ro (maag) E phi
+ (inag) H phi (mag)'

do 430 i=l,ipart
sguinc~sguinc+drea1(guidsp)
write(23,24) sguinc,abs(efra(i))Isfdmax,abs(hfra(i))Isfdmax,
+ abs(efphi(i))/sfdmax,abs(hfphi(i))/sfdmax
430 continue
sguinc=O.OdO
write(23,4. ) radius E rc (real) H ro (real) E phi
+ (real) H phi(reai)'

do 431 i=1 ipart

sguinc=sguinc+dreal(guidsp)
write(23,24) sguinc,dreal(efra(i))/sfdxnax,dreal(hfra(i))/
+ sfdmax,dreaI(efphi(i) )/sfdmax,dreal(hfphi(i))/
+ sfdniax
431 continue
sguinc=O.OdO
wiite(23,4 . )' radius E ro (imag) H ro (imag) E phi
+ (irnag) H phi(iniag)'

do 432 i=1,ipart
sguinc=sguinc+dreal(guidsp)
write(23,24) sguinc,dimag'kefra(i))/sfdinax,dimag(hfra(i))/
+ sfdmax,dimag(efphi(i))/sfdmax,dimag(hfphi(i))/
+ sfdniax
432 continue
write(26,27) dreal(zkzl1),dimag(zkzl1),sfreq,nord,numdi
do 440 i=1,numdi
write(26,28) dreal(zeps(i)/zeps0),dreal(zxnu(i)/zinu0),
+ dreal(radii(i))
write(26,28) dimag(zeps(i)/zeps0),dimag(zznu(i)/zmu0)
440 continue
do 450 i=1,maxnum
write(26,28) dreal(coeff(i)) ,dimag(coeff(i))
 147

450 continue
sguinc=O.OdO
* write out the electric and magnetic scalar potentials
do 460 i=1,ipart

sguinc=sguinc+dreal(guidsp)
write(29,24) sguinc,dreal(psie(i)),dimag(psie(i)),
+ dreal(psieder(i)),dimag(psieder(i))
write(31,24) sguinc,dreal(psim(i)),dimag(psim(i)),
+ dreal(psimder(i)),dimag(psimder(i))

460 continue
12 format(e 12.5,5x,e 12.5)
24 format(e 12.5,4x,e 12.5,4x,e 12.5,4x,e 12.5,4x,e 12.5)
27 format(e 12.5,5x,e 12.5,5x,e 12.5,5xi4,5x,i4)
28 format(e 12.5,5x,e 12.5,5x,e 12.5)
999 close(31)
close(29)
close(26)
close(23)
close(11)
end

---- -------- -------------------------------------------------------

* Subroutine mtrxval

* mtrxval calculates the matrix elements for the multiple dielectric

* circular waveguide main program.
,
* input: zkz - axial wavenumber guess
* iflag - = 1 for determinant calculation
* = 0 for no determinant
* output: det - determinant of matrix for k-guess

*---------------------------------------------------------------------------------------------------------

subroutine mtrxval(zkz,detiflagd)
implicit complex* 16(a-h,o-rt-z)
implicit real*8(s)
parameter (spi=3.14159265359,maxdie=5,maxcon--maxdie*4-2)
parameter (zepsO--(8.854d- 12,0.OdO),zmuO=-(12.56637062d-7,0.OdO))
parameter (ibesnum=4)
complex* 16 zkro(maxdie),radii(maxdie),zmu(maxdie),
+ zeps(maxdie),dc(maxcon,maxcon)
 148

complex* 16 worksp(maxcon)
real*8 sjunkl (ibesnum),sjunk2(ibesnum)
real*8 sbjr(ibesnum),sbji(ibesnurn),sbyr(ibesnum),sbyi(ibesnum)
integer*4 ipivot(maxcon)
comnion/combes/nord,zkro,radii
commonlmtrx/freq,zmu,zeps,numdi,sacc ,dc
do 1020 i=1,maxcon
do 1030 j=1,maxcon
dc(ij)=(.OdO,0.OdO)
1030 continue
1020 continue
zkro(l1)=sqr(frtq*frcq*zmu(l1)*zeps(1 )-zkz*zkz)
ixl=1
ix2= I
do 1000 i=2,nuzndi
zkro(i)=sqrt(freqfq*znu(i)*zeps()-zkz*zkz) :principal value
arg--zkro(i)*radii(i-1)
*for the given zkz, the Bessel's functions and their derivatives are found

call totbes(i,arg,besjl1,besj2,besnl1,besn2,dbesjl1,dbesj2,
+ dbesnl,dbesn2)
" calculate matrix elements. inner dielectric is a special case and is source
" of the if statements. ix 1-row pointer, ix2-colurnn pointer, i=outer dielectric
" of the two under observation.

dc(ix l,ix2)=zkro(i- 1)*zkro(i- 1)*zeps(i)*besj1I

if(ix Lne. 1) then
dc(ix 1 ix2+1 )=zia'o(i- 1)*zkro(i- 1)*zeps(i)*besn 1
ix.2=ix.2+2
endif
*ci ~x+)-koi*zr~)zp~-1*ej
dc(ix 1,ix2+2)=-zkro(i)*zkro(i)*zeps(i- 1)*besj2
ixi =ix 1+1
iffixl.ne.2) then
ix2=ix2-2
dc(ix 1,ix2+2)=zkro(i- 1)*zkro(i- 1)*zmu(i)*besj 1
dc(ix 1,ix2+3)=zkro(i- 1)*zkro(i. 1)*zu(i)*besn 1
dc(ix l,ix2+6)=-zkro(i)*zkro(i'*zmu(i- 1)*besj2
dc(ix 1,ix2+7)=-zkro(i)*zkyo(i)*zrnu(i- I )*besn2
else
dc(ix l,ix2+1)=zkro(i- 1)*zkro(i- 1)*zmu(i)*besj I
dc(ix 1,ix2+4)=-zkro(i)*zkro(i)*Izmu(i- 1)*besj2
dc(ix I,ix2+5)=-zkro(i) Ozkro(i)*zmu(i- 1)*besn2
endif
ixl1=ix 1+1
 149

dc(ix 1,ix)=z*dcml(nnd)/(freq*zeps(i. I)*radi(i 1 )*

+ besjlI

if(ixl.eq.3) then
ix2--ix2- I
else
dc(ix 1,ix2+ 1)=zk*dcmpx(nord)/(fr-q*zeps(i. 1)
+ *rij(j..1))*besnl
endif
dc(ix 1,ix2+2)=zkro(i- 1)*dbesj 1
if(ixl.eq.3) then
ix2=ix2-1
else
dc(ix 1,ix2+3)=zkro(i- 1)*dbesn 1
endif
dc(ixr I,ix2+4)=..zkz*dcmplx(nord)/(freq*zetps(i)*radii(i. 1))*
+ besj2
dc(ix 1,ix2+5)=-zkz*dcxnplx(nord)/(freq*zeps(i)*radii(i- 1))*
+ besn2
dc(ix 1,ix2+6)=-zkro(i)*dbesj2
dc(ixl1,ix2+7)=-zkro(i)*dbesn2
if(ixl.eq.3) then
ix2--ix2+2
endif

ix1 =ix I+ I
dc(ix 1,ix2)--zkro(i- I )*dbesj 1
if(ixI1.eq.4) then
ix2=ix2- 1
else
dc(ix 1,ix2+I )=zkro(i- I )*dbesn I
endif
dc(ix I ,ix2+2)=zkz*dcmplx(nord)/(fieq*znmu(i- I )*radii(i- I ))*'
+ besjlI
iffix 1.eq.4) then
ix2--ix2-l
else
dc(ixl ,ix2+3)=zkz*dcniplx(nord)/(freq*znu(i- 1)
+ *radij(i-1))*besn1
endif
*x+)zr~)dej
dci
dc(ix I ,ix2+4)=-zkro(i)*dbesj2
dc(ix 1,ix2+6)-zk*dnlpx(nord)/(freq*zmu()*radij(.. I))*
+ besJ2
dc(ix 1,ix2+7)=-zkz*dcniplx(nord)/(freq*zinu(i)*radii(i- 1))*
+ besn2
 150

if (ixl.eq.4) then
ix2=ix2+2
endif
ixl--Ixl-'l

if~ixl.eq.5) then
ix2--ix2+2
else
ix.2=ix244
endif

1000 continue
arg--zkro(numdi)*radii(numdi)

" for last two rows calculate elements. zbesj(y) is Bessel's routine with no
" derviatives. last two rows due to PEG guiding structure.

call zbesj(dreal(arg),dimag(arg),dble(flord), 1,2,sbjr,sbji,

+ nz,ierr)
if(ierr.ne.0) write(*. ,*) 'Bessel Routine error #',err
call zbesy(dreal(arg),dimag(arg),dble(nord), 1,2,sbyr,sbyi,
+ nz,sjunkl1,sjunk2,ierr)
iftierrnne.O) write(*,*) 'Bessel Routine error #',ierr
besj2=dcmplx(sbjr(l1))+(0.OdO, 1.OdO)*dcmplx(sbji( 1))
besn2--dcniplx(sbyr(l1))+(0.0d0, 1.OdO)*dcmplx(sbyi( 1))
besj2p I=dcrnplx(sbjr(2))+(0.OdO, 1.OdO) 4.dcrnplx(sbji(2))
besn2pl1=dcmplx(sbyr(2))+(0.OdO, 1.0d0) 4.dcmplx(sbyi(2))
dbesj2=~dcmplx(nord)*besi2/arg-besj2p 1
dbesn2=dcmplx(nord) 4. besn2/arg- besn2p I
maxnum=numdi*4-2
dc(ixl1,maxnum-3)=besj2
dc(ix 1 maxnum-2)=besn2
ixl=ixl+1
dc(ix 1,maxnum- 1)=dbesj2
dc(ixl1,maxnum)=dbcsn2
nc=nuzndi*4-2
*with matrix calculated, now call determinant routine.

if (iflagd) 2020,2020,2030
2030 continue
call det I(dc ,nc,det)

2020 continue
return
end
 151

* Subroutine totbes
* totbes calculates the Bessel functions of the first and second kinds
* and the corresponding derivatives for complex arguments.

* input: i - loop counter (outer dielectric of the two under scrutiny)

* arg - argument of Bessel functions
* output: besj 1,besj2 - Bessel functions of the first kind
* 1-inner dielectric of the boundary
* 2-outer dielectric of the boundary
* besn l,besn2 - Bessel functions of the second kind
* dbesj 1,dbesj2 - derivative of the Bessel function of the
* first kind
* dbesn 1,dbesn2 - derivative of the Bessel function of the
* second kind

-------- --- -- ----------------------------------

-------------------------

subroutine totbes(i,arg,besj 1,besj2,besn 1,besn2,dbesj 1,

+ dbesj2,dbesn 1,dbesn2)

implicit complex* 16(a-h,o-r,t-z)

implicit real*8(s)
parameter (pi=(3.14159265359dO,O.OdO),maxdie=5,maxcon--maxdie*4-2)
parameter (ibesnum=4)
complex* 16 zkro(maxdie),radii(maxdie)
real*8 sjunkl(ibesnum),sjunk2(ibesnum)
real*8 sbjr(ibesnum),sbji(ibesnum),sbyr(ibesnum),sbyi(ibesnum)
common/combes/nord,zkro,radii
* with the argument and the order, call the Bessel function subroutine

call zbesj(dreal(arg),dimag(arg),dble(nord), 1,2,sbjr,sbji,

+ nz,ierr)
if(ierr.ne.0) write(*,*) 'Bessel Routine error #',ierr
call zbesy(dreal(arg),dimag(arg),dble(nord), 1,2,sbyr,sbyi,
+ nz,sjunk 1,sjunk2,ierr)
if(ierr.ne.0) write(*,*) 'Bessel Routine error #',ierr
besj2=dcmplx(sbjr(1))+(O.OdO,I.OdO)*dcmplx(sbji(1))
besn2=dcmplx(sbyr(1))+(O.OdO,1.OdO)*dcmplx(sbyi(1))
besj2p 1=dcmplx(sbjr(2))+(O.OdO,I.OdO)*dcmplx(sbji(2))
besn2pl =dcmplx(sbyr(2))+(O.OdO, 1.OdO)*dcmplx(sbyi(2))
dbesj2=dcmplx(nord)*besj2/arg-besj2p 1
dbesn2=dcmplx(nord)*besn2/arg-besn2p 1
 152

* calculate Bessel functions and derivatives for the inner dielectric

* of the boundary under consideration
arg=zkro(i- 1)*radii(i- 1)
call zbesj(dreal(arg),dimag(arg),dble(nord), 1,2,sbjr,sbji,
+ nz,ierr)
if(ierr.ne.0) write(*,*) 'Bessel Routine error #',ierr
call zbesy(dreal(arg),dimag(arg),dble(nord), 1,2,sbyr,sbyi,
+ nz,sjunkl,sjunk2,ierr)
if(ierr.ne.0) write(*,*) 'Bessel Routine error #',ierr
besjl =dcmplx(sbjr( 1))+(O.OdO,1.OdO)*dcmplx(sbj i( 1))
besnl=dcmplx(sbyr(1))+(O.OdO,1.OdO)*dcmplx(sbyi(1))
besj ip 1=dcmplx(sbjr(2))+(O.OdO,1.OdO)*dcmplx(sbji(2))
besn ipi =dcmplx(sbyr(2))+(O.OdC, 1.OdO)*dcmplx(sbyi(2))
dbesj l--dcmplx(nord)*besj I/arg-besjlpl
dbesnl=dcmplx(nord)*besn1/arg-besnlp 1
return
end

*--------------------------------------------------- ------------------------------------------------------

* Subroutine detl
,
* detl calculates the determinant of a double precision complex
* matrix using lu decomposition. reference "Numerical Recipes."

* input: a - square matrix under evaluation

* n - size of square matrix
* output: d - determinant (complex, double precision)

------------------- -----------------------------------------------------------------

subroutine detl(an,d)
implicit complex* 16(a-h,o-z)
pararneter(maxdie=5,maxcon--maxdie*4-2,np--maxcon)
complex* 16 a(npnp),indx(np)
call ludcmp(a,n,np,indx,d)

do 11 i=l,n
d=d*a(ii)
11 continue
return
end
 153

* Subroutine ludcmp
* ludcmp performs the LU decomposition of the matrix [dc] to be used in
" conjunction with the root-finding routine in searching for kz, or
* in conjunction with lubksb to calculate coefficients [C].
* reference "Numerical Recipes."

* input: a - square matrix under evaluation

*i n - size of square matrix
* np - dimension of matrix
* output: indx - row permutations for partial pivoting
d - even or odd number of row interchanges

*----------------------------- -------- --------------------------------------------------------------------

subroutine !udcmp(an,np,indx,d)
implicit complex* 16(a-h,o-z)
parameter(nmax=100)
, 16 a(np,np),indx(np),w(nmax)
tiny=(1.Od-40, 1.Od-40)
d=( 1.OdO,O.OdO)

do 12 i=l,n
aamax=(O.OdO,O.OdO)

do 11 j=l,n
if (cdabs(a(ij)).gtcdabs(aamax)) aamax=abs(a(ij))
11 continue
if (aamax.eq.0.) pause 'singular matrix'
vv(i)=(1.OdO,O.OdO)/aamax
12 continue
do 19 j= l,n

do 14 i=lj-l
sum=a(ij)
do 13 k=li-1
sum=sum-a(i,k)*a(kj)
13 continue

a(ij)=sum
14 contnue
 154

aamax=(O.OdO,O.OdO)
do 16 i=j,n
surn=a(i,j)
do 15 k=lj-1
sum=sum-a(i,k) , a(kj)
15 continue

a(ij)=surn
dum-v(i)4, abs(sum)
if(cdabs(dum).ge.cdabs(aamax)) then
imax=i
aamax=dumn
endif
16 continue
if (jauedmax) then

do 17 k=1,n
dum=a(irnax,k)
a(imax,k)=aoj,k)
aoj,k)=durn
17 continue
d=-d
vv(iinax)=vvwj)
cndif
indxoj)-inax
if(aOj,j).eq.O.) aojj)=tiny
ifoj.nexn) then
dum=(l1.OdO,O.OdO)/aojj)
do 18 i=j+1,n
a(i,j)=a(i,j)4, durn
18 continue
endif
19 continue

retur
end

4Subroutine lubksb
*lubksb calculates the backsubstitution in conjuction with ludcnp,
*to calculate the coefficients [C]. reference "Numerical Recipes."
*input: a - square matrix under evaluation15
nn-size of squarecmatrix
*np - dimension of matrix

*output: indx - row permutations for partial pivoting

* b - right-hand side of the matrix equation Ax=b
*--------------------------------------------------------- -------------------------

subroutine lubksb(a,n,np,indx,b)
implicit complex* 16(a-h,o- z)
complex* 16 a(np,np),indx(np),b(np)
ii=0

do 600 i=l,n
U=indx(i)
sum--b(l)
b(ll)=b(i)
if (ii.ne.O) then

do 610 j-ii,i-lI
sum=sum-a(ij)*boj)
610 continue
else if (sumn.ne.0.) then
endif
b(i)=sum
600 continue
do 620 i=n,l1,-l1
suni=b(i)
if (i.lt.n) then
do 630 j=i+1I,n
sum=sum-a(ij)*boj)
630 continue
endif
b(i)=surn/a(iji)
620 continue

return
end
 156

REFERENCES

[I] L. Pincherle, "Electromagnetic waves in metal tubes filled longitudinally with two
dielectrics," Phys. Rev., vol. 66, pp. 118-130, Sept. 1944.
[2 P. J. B. Clarricoats, "Propagation along unbounded and bounded dielectric rods,"
Parts I and 2, Proc. Inst. Elec. Eng., Mon. 409E and 410E, pp. 170-186, Oct.
1960.
[3] K. A. Zaki and A. E. Atia, "Modes in dielectric-loaded waveguides and
resonators," IEEE Trans. Microwave Theory Tech., vol. MTT-31, pp. 1039-1045,
Dec. 1983.
[4] K. A. Zaki and C. Chen, "Intensity and distribution of hybrid-mode fields in
dielectric-loaded waveguide," IEEE Trans. Microwave Theory Tech., vol. M'TT-
33, pp. 1442-1447, Dec. 1985.
[5] K. A. Zaki and C. Chen, "Coupling of non-axially symmetric hybrid modes in
dielectric resonators," IEEE Trans. Microwave Theory Tech., vol. MTT-35, pp.
1136-1142, Dec. 1987.
[6] W. M. Bruno and W. B. Bridges, "Flexible dielectric waveguides with powder
cores," IEEE Trans. Microwave Theory Tech., vol. MTT-36, pp. 882-890, May
1988.
[7] R. C. Chou and S. W. Lee, "Modal attenuation in multilayered coated
waveguides," IEEE Trans. Microwave Theory Tech., vol. MTT-36, pp. 1167-
1176, July 1988.
[8] P. J. B. Clarricoats and R. A. Waldron, "Non-periodic slow-wave and backward-
wave structures," J. Electron. Contr., vol. 8, pp. 455-458, June 1960.

[9] P. J. B. Clarricoats, "Backward waves in waveguides containing dielectric," Proc.

Inst. Elec. Eng., Mon. 451E, vol. 108C, pp. 496-501, June 1961.
[10] P. J. B. Clarricoats, "Circular-waveguide backward-wave structures," Proc. Inst.
Elec. Eng., vol. 110, pp. 261-270, Feb. 1963.
[111 P. J. B. Clarricoats and A. B. Birtles, "Circular waveguide backward-wave
experiments," J.Electron. Conir., vol. 15, pp. 325-330, Oct. 1963.
[12) P. J. B. Clarricoats and K. R. Slinn, "Experimental observation of travelling
backward waves ia dielectric-loaded circular waveguide," Proc. Inst. Elec. Eng.,
vol. 111, pp. 109Ij-1092, June 1964.
[13) R. A. Waldron, "Theory and potential applications of backward waves in
nonperiodic inhomogeneous waveguides," Proc. Inst. Elec. Eng., vol. 111, pp.
1659-1667, Oct. 1964.
 157

[14] P. J.B. Clarricoats and B. C. Taylor, "Evanescent and propagating modes of

dielectric-loaded circular waveguide," Proc. Inst. Elec. Eng., vol. 111, pp. 1951-
1956, Dec. 1964.
[15] P. Chorney, "Power and energy relations in bidirectional waveguides," Tech. Rep.
396, Research Lab. of Electronics, Mass. Inst. Tech., Cambridge, MA, Sept.
1961.
[16] S. R. Laxpati and R. Mittra, "Energy considerations in open and closed
waveguides," IEEE Trans. Antennas Propagat., vol. AP-13, pp. 883-890, Nov.
1965.
[17] V. A. Kalmyk, S. B. Rayevskiy and V. P. Ygryumov, "An experimental
verification of existence of complex waves in a two-layer, circular, shielded
waveguide," Radio Eng. Electron. Phys., vol. 23, pp. 17-19, Apr. 1978.
[18] E. Snitzer, "Cylindrical dielectric waveguide modes," J. Opt. Soc. Amer., vol. 51,
pp. 491-498, May 1961.
[19] R. A. Waldron, Theory of Wave Guides and Cavities. London: Maclaren and Sons
Ltd., 1967.
[20] S.Ramo, J. R. Whinnery and T. Van Duzer, Fields and Waves in Communication
Electronics. New York: Wiley, 1984.
[21] R. F. Harrington, Time-Harmonic Electromagnetic Fields. New York: McGraw-
Hill, 1961.
[22] A. R. von Hippel, DielectricMaterialsand Applications. New York: Wiley, 1958.
[23] E. J. Rothwell and L. L. Frasch, "Propagation characteristics of dielectric-rod-
loaded waveguides," IEEE Trans. Microwave Theory Tech., vol. M'IT -36, pp.
594-600, Mar. 1988.
[24] R. E. Beam, M. M. Astrahan, W. C. Jakes, H. M. Wachowski and W. L.
Firestone, "Dielectric tube waveguides," Tech. Rep. ATI 94929, ch. 5,
Northwestern University, Evanston, IL, 1949.
[25] R. A. Waldron, Theory of Guided ElectromagneticWaves. London: Van Nostrand
Reinhold Company, 1969.
[26] W. H. Press, B. P. Flannery, S. A. Teukolsky and W. T. Vetterling, Numerical
Recipes. Cambridge: Cambridge University Press, 1986.
[27] R. L. Burden and J. D. Faires, Numerical Analysis. Boston: Prindle, Weber and
Schmidt, 3rd ed., 1985.
[28] C. S. Lee, S. W. Lee and S. L. Chuang, "Normal modes in an overmoded circular
waveguide coated with a lossy material," IEEE Trans. Microwave Theory Tech.,
vol. MTT-34, pp. 773-785, July 1986.
[29] P. J. B. Clarricoats and K. R. Slinn, "Complex modes of propagation in dielectric-
loaded circular waveguide," Electron. Let., vol. 1, pp. 145-146, July 1965.
 158

[30] A. S. Omar and K. F. Schunemann, "Complex and backward-wave modes in

inhomogeneously and anisotropically filled waveguides," IEEE Trans. Microwave
Theory Tech., vol. MTT-35, pp. 268-275, Mar. 1987.

[311 R. A. Waidron, "Properties of ferrite-loaded cylindrical waveguides in the

neighbourhood of cut-off," Proc. Inst. Elec. Eng., vol. 109B, suppl. 21, pp. 90-
94, Jan. 1962.

[32] E. C. Jordan, ed., Reference Data for Engineers. Indianapolis, IN: Howard W.
Sams and Company, Inc., 1985.

[33] D. G. Fink and D. Christiansen, eds., Electronic Engineers' Handbook. New

York: McGraw-Hill, 1982.
[34] Information About Silicone Elastomers: Dow Corning 732 Multi-Purpose Sealant.
Midland, MI: Dow Corning Corporation, 1986.

[35] P. F. Williams, K. F. Heyen and R. G. McCormack, "Low cost electromagnetic

shielding using drywall composites: results of RFI testing of shielding
effectiveness," Tech. Rep. No. M-88/02, US Army Corps of Engineers
Construction Engineering Research Laboratory, Champaign, IL, Oct. 1987.

[36] J. A. Kong, Electromagnetic Wave Theory. New York: John Wile: -,d Sons,
Inc., 1986.
[377 .. D. Jackson, ClassicalElectrodynamics. New York: John Wiley and Sons, Inc.,
1975.

[38] J. T. Verdeyen, Loser Electronics. Englewood Cliffs, NJ: Prentice-Hall, 1981.

[39] D. E. Amos, "A portable package for bessel functions of a complex argument and
nonnegative order," ACM Trans. Math. Software, vol. 12, pp. 265-273, Sept.
1986.
 DISTRIBUTION

HQUSACE
ATTN: CEEC-E (2)
ATTN: CERD-L
ATTN: CEIM-SL (2)

USAEHSC, ATTN: Library 22060

ATTN: DET III 79906

US Army Engineer Districts

ATTN: Library (41)

US Army Engineer Division

ATTN: Library (14)
Huntsville 35807
ATTN: Chief, HNDED-SR

USA Natick, R&D Ctr 01760

Naval Air Systems Command

ATTN: Library

Federal Emergency Mgmt Agency

ATTN: EMP Program Mgr 20472

Hanson AFB, MA 01731

ATTN: HQ AFSC
ATTN: ESDIOCR-E

Naval Civil Engr Lab

ATTN: Library

Harry Diamond Labs 20783

ATTN: SLCHD-NW
ATTN: SLCIID-NW-E
ATTN: SLCHD-NW-EP (10)
ATTN: SLCiID-NW-EH
ATTN: SLCIID-NW-ES
ATTN: SLClHD-CS

Derense Nuclear Agency 20305

ATTN: DNA-RAAE

Defense Technical Info. Center 22314

ATTN: DDA (2)

U.S. Gov't Printing Office 22304

Receiving Sect/Depository (2)

89
9/89

m ~ mmm m l lUU IM~lwllll xHI