IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES,

VOL.

MTT-33, NO. 10, OCTOBER 1985

1067

Analysis of an Arbitrarily-Shaped Planar CircuitA Time-Domain Approach

WOJCIECH (Invited K. GWAREK Paper)

Abstract we anafysis theoretical tions waveguides.

study

of an arbitrarily-shaped is presented Experiments circuits show

planar good

circuit

is reported. method between of the to

d z

theoretical

background

and theri a numerical agreement The examples

is introduced. cafcsdations stripline

and the measurements. but the method

of applica-

concern

may he afso applied

)x

I.

INTRODUCTION

Fig.

1.

A planar

circuit,

dimension

HE

GENERAL by very

CONCEPT Okoshi

of a planar with

circuit having

was one state solution idea presented wave propagation conditions. boundaries conditions. parameters II. for a complicated in the structure boundary starting problem. numerically The the in this paper is to simulate

introduced

[1] as a circuit

small in comparison

the wavelength

and an arbitrary shape in two other dimensions. The idea proved to be important from both the theoretical and the practical points of view. In microwave theory, it initiates a two-dimensional circuit theory which is an extension of the O-dimensional lumped-element theory and one-dimensional transmission-line theory. In microwave techniques, there are a wide range of applications of the theory including the use for microwave The problem for idea of solving integrated circuits. leads to a mathematical Hehnholtz equation To boundary conditions. relations of a planar V, with circuit proper

from zero initial

The wave propagating and reflecting from the sets up automatically the proper boundary When the steady state is reached, all the circuit can be obtained.
BASIC DEFINITIONS CONSIDERED AND RELATIONS FOR THE

CIRCUIT

Hsu Jui-Pang circuit beyond theory

and T. Anada of a planar

[2] have presented extending

a general

a two-dimensiond Okoshi

circuit

its definition

a potential

the circuit

of a very small height.

They have based

solve the problem, solution the potentials

and Miyoshi

[1] used Webers between This led

for cylindrical

waves to obtain

the theory on a set of E ,md H height modes. Their idea is followed in this paper but the formal description of the circuit is based on the Hertz gives mathematical precision convenient in practical use. circuit shown in Fig. 1. The is limited by two Let us consider space in which the planar potentials. This approach for the definitions and is

on different

parts of the boundary.

to an approximate solved numerically. developed obtained ticular elements Much numerically

linear matrix relation Hsu Jui-Pang and based on for a particular a circuit

which had to be Anada [2] have eigenfunctions There were of parof such prob-

a method

a set of planar

circuit.

the wave is transmitted

also some efforts [3], [4].

to characterize

elements composed

shape and to analyze

electric walls A and A situated in the planes z = O and z = d. We consider two sets of modes E. and He. Definition: electric Hertz An E. mode is a mode described by an potential IT, [5] of the form y)cos

progress has been made and many practical

lems solved; however, none of the mentioned methods seems to lead to a computer program capable of solving an arbitrarily-shaped planar circuit without an extensive theoretical and programming effort by the user. This limits their applications and justifies the search for methods more practical for engineers. All the mentioned methods, while are based on the same idea of considering only alternative

me n =a,qbg(x, where a ~ is a unit n = 0,1,2, . . . . An H. mode potential

()
to

nn ~z

kt eJ the z-axis,

(i) and Hertz

vector

parallel

being very different, a steady-

is a mode described

by a magnetic

fi: of the form (2)

Manuscript received Febnmry 5, 1985; revised June 3, 1985. The author is with the Instytut Radioelektronikci, Politechnika zawska, Warszawa, Nowowiejska 15/19 Poland.

Wars-

where

n=l,2,

... . IEEE

0018-9480/85

/1000-1067$01

.00 01985

1068

IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. MTT-33 , NO. 10, OCTOBER 1985

From we obtain

the general

properties

of the Hertz

potentials

[5],

planar changing

circuit height.

methods

may be applied circuits

to a guide

with

for an E. mode z= Et= 4C(X, Y)cos ()~z ~sin ~z () &eJt V+j(x, y)eJt

As shown, all the cases for planar equations in a standard form VV(x, vJ(x, juea, Xv#~(x, y)cos ~z () eJ (3) y)=

lead to a set of may be written

(6) and (7) or (8) and (9) which

juL,J(x,

y) y).

(11) (12)

H==O H,= where # The surface current equal to J.= Let us define a= XH,l==d= jum~~(x, y)eJ. (4) III.
The proposed

y) = j~C,V(x,

We shall now concentrate on the E. mode. It does not mean that the formulas presented in the next section are invalid for the higher modes. They are valid, but their physical interpretations are more complicated and will not be considered For unitary square here. Eo, the value of the circuit, C, is the capacitance and the value of L. of a is the the mode

= ~2pc _ ()

which flows in the upper plane A is

inductance

of an arbitrary
THE PROPOSED method [6], [7]. of

square of the circuit.

METHOD analysis Let us OF ANALYSIS is based consider (11) on the and finite(12) in

a potential V,= d+~(x, y)~~e~. into (5) consideration equation, we

difference the

method

Using that obtain

the above the potential the relations

relations between
v)7=-j%~J e

and taking J, and V=

time-dependent

form

(1) satisfies the Hehnholtz

vV(x,

y,t)=

L,

dJ(x,

y, t) ~1 v,t)

(13)

till d

(6) (7) the formulas dual

vJ(x,

y,t)=

C,

dv(x,,

at

(14)

v.

Je=j%v

The xy-plane (the domain of the-functions V and J) is divided into a set of square meshes of the size a. The coordinates lth column (13) and frequency of the middle of a mesh in the k th row and the are denoted by xi and y~. Let us assume that

In the case of an H. mode, we obtain to (3), current

(4), and (5). These formulas include a magnetic Jk and a magnetic potential Vfi, which are related

by the equations (8)

(14) describe propagation of a wave of the u and the wavelength A. If we assume that 2V a << ~ and At << , we may replace the differentials in

(13) and (14) by fin~te differences At and a. After algebraic transformations, we obtain the relations for V and two components J, x,+; of the current ,yk, to++ .JXand JY inside the circuit x[+:, y~, toz At ) y,, to)V(xl, y,5t0))&
s

vJ~=

j~V~.

(9)

There is a following correspondence between the defined modes and the popular transmission lines. 1) Stripline may be considered as a composition of two planar structures with E. modes. The fringing fields may be included in the calculations by extending the real dimensions of the circuit in the xy-plane by a value of [1] S = 2d(log 2) Microstrip mode, but the makes its analysis of stripline. is basically complicated ,2)/7r. structure with (lo) an E. fields

)(

=JX

-(

V(xl+a,

(15)

JY xl, y~+~,
(

to+;

=JY )( y,+a,

x[, y~+:,

to$ )

a planar nature

-(v(x{,

to)-v(x[,

y,, fo))~
s

(16)

of its fringing

a more difficult

problem

than in the case

v(X/,

yk,

to+

At)=

V(X/,

yk,

to)+

((

JX

X[+&,

to+~ )

3) HOI rectangular waveguide situated with the E field parallel to the z-axis of Fig. 1 may be considered as a planar circuit of E. mode with the short-circuit boundary conditions circuit width. planar tions on the sides of the guide. In this case, the planar methods solve the problem of a guide with changing When the guide is turned by r/2, it becomes a circuit for of an HI mode with open-boundary condithe Vh potential on the sides. In this case, the

Jy (

X,;

,yk, to++ ) At )

+JY ( J, (

x[,y~+j,to+~

Xl, yk;

>to++ )1

g. 3

(17)

GWARSK:

ARSITRARILY-SHAPED

PLANAR

CIRCUIT

1069

ulars dropped be written

on it from

the nodal form

points.

Since (14) may

in an integral

$ J.a~dl=
L. Fig. 2. Lumped circuit equivalent to a segment of a planar circuit,

C,

J v

ds = C&-#

dV (18)

~m at

where S~ is the surface of the m th mesh, an is the unit vector perpendicular to L~, L~-border line of the m th mesh, and the capacitances Cl, C2, and Cd in the equivalent circuit in Fig. 4 are proportional meshes. L2 and L3 are the standard inducmeshes Lz = L3 = L,. Lq, we note that to the surfaces of the corresponding

l%.

---

NA B v, hl<c b ______ JS a J$ V,
J,

The
D ---

inductances

tances between
--

--J2 V2 - ~

To calculate

V, --

and that the current

Fig.

passing between the meshes 3 and 4 is

3.

Example

of an open boundary

approximation.

equal to ll-J1(a+ b). (19)

-r

w
L, l-5

Thus VA V3 dIA Ld ~= or Lb=

considered in Fig. 3

c2H. .
Fig. 4. Lumped circuit equivalent to the circuit

+%

(V,-K) L,

(a+b) a

L,&.

(20)

To calculate Equations esting set of lumped voltage. calculated The (15)(17) describing The circuit the circuit have an interas a of the may be values L, as preinterpretation. may be represented V has the meaning in the inductances The elements

L5, we put dJ5 7i-~-vl _ L,afi

capacitors current

C and inductances

sented in Fig. 2. The potential flowing

Since near the open boundary component of the current 15==

there exist only a tangential

as IX = JXa and IY = .@.

are C = C,a2, L = L,. A. Boundary Conditions border line passes through or one row, by assuming the centers which gives V3 VI L5

J5h (21) Cos ~a () 4

If a short-circuited conditions

of the meshes in one column are satisfied If an open-circuited border

the boundary

d15 ~= c&) ;;

V= O at those points. a side line of a row and

line follows

or a column of meshes, the boundary conditions are satisfied by putting J. = O, where J. is the current density component perpendicular to the border line. Those simple rules are sufficient to calculate a circuit of

a rectangular shape, but if we want to analyze an arbitrarily-shaped circuit without significant loss of accuracy, a procedure conforming the boundary shape is needed. That kind of procedure for an open boundary will be introduced using an example of a segment of a circuit shown in Fig. 3, and its equivalent LC circuit shown in Fig, 4, The border line was approximated by a broken line ABCD. Four meshes close to the border are considered and the potentials in the centers of them are denoted by VI .0 V//. To match the border line, the meshes 1, 3, and 4 are deformed as shown in Fig. 3. In the neighborhood of the border, they are limited by the line ABCD or by perpendic-

B.

Lumped Direct

Elements

in the Circuit between the algorithm used for

correspondence

calculation

and the equivalent

LC circuit

makes it easy to

consider lumped elements branched into the circuit. These elements may be passive (R, L, C) or active (voltage or current sources), and, since a time-domain method is used, they may be nonlinear. It should be noted, however, that (15)-(17) describe the voltages at the points of time and space different from those where and when the currents are described. The differences are At/2 in time and a/2 in

1070 space and they should impedance calculations: C. Algorithm A computer important introduced In propagates Stability algorithm be considered

IEEE TRANSACTIONS ON MZCROWAVE THEORY AND TECHNIQUES, VOL. MTT-33 , NO. 10, OCTOBER 1985

to avoid

errors

in

Isl,l
0.7

1,! ~

.QINPuT .,
4 1

based on (15)(17), This

with

excitawill be
0.4.

tion by a set of sinusoidal condition through with

sources of a frequency condition interpretation.

O, has an

of stability. its physical

the circuit

considered,

any perturbation

of

V or J

E&l

the velocity of
0.3.

(23)
0.2--

where and

A is the wavelength

of the wave in the planar

circuit,
0.1, .

*=*
(k!

In

the

algorithm

after

a time

At,

the

perturbation the
Fig. 5. Comparison

0.02s

0.05

0075

0.1

0.125

Y+

reaches the adjacent perturbation from Thus, the velocity smaller than

mesh. It requires

2A t to transmit

of the results of calculation

~=w N.

for different

mesh size

the mesh k, 1 to the mesh k +1,1+1. of propagation in the algorithm is not a2 ~ a= 2At of the algorithm

(24) we must

Development gives

of the functions

V and

JX in Taylor

series

To assure the proper have

functioning

13Jx =

83JX Xl+ (
at

;,

Yk,

to

~t2

24+ ,yk, to

(28)

or

-&>+,
The same result may be obtained from the finite-difference methods (see [6, sec. 26.2]). D. Accuracy There of the Calculations principal causes of errors theory

(25) of From tional rapidly in the pro-

133V 8V=

Xl+; ax 3

&+

... .

(29)

(28) and (29), it is seen that the errors are proporto At 2 or a 2. This with means that the errors decrease a decrease in the mesh size and a decrease in by the condition of the function

are three

the size of a step in time. Since the values of At and a are bound (25), the error caused by the approximation

posed numerical

method

of planar-circuit equations

analysis. (13) and (14) by

1) replacing
2) 3)

the differential

(15), (16), and (17), approximation of the circuit border line, and assumption of a steady state in the circuit after finite period of time. the first cause of errors, let us consider, it from (13), we have replaced by finite differences. exists, we may write ,yk,
1

To analyze example, differentials replacement

for the

of time has a minor effect. The value of 8 ~ and similar values of errors found in the analysis of (16) and (17) depend on the field distribution in the circuit. Typically, the functions V(x, y) and J(x, y) change slowly in space following a sinusoidal-type distribution over a wavelength. Thus, the third derivatives of these functions are not very big and the errors are small even for relatively of a. An example of this is shown results are displayed for different high vahzes in Fig. 5, where the sizes of the meshes

(15). To obtain

Since some error of this

aJx X,+;

to

assumed for calculations. Nevertheless, if the boundary conditions force rapid changes of V(x, y) and J(x, y), the third derivatives of these functions may cause some increase in the errors. It is difficult to obtain mathematical approximations for the other two causes of errois. We may formulate only some general remarks. The second cause of errors obviously depends on the possibility of reasonable approximation of the boundary line by a straight line within one mesh. The experiments have confirmed that if approximation is possible, the error introduced +8 ~. (27) proximating dominating such an by ap-

JX x[+f, .

y~,to+~

JX )( At

xl+f,y~,to~

+ 6JX (26)

13v Xl+; (

,yk,

to

ax
v(x,

+a,y~, to) Jqx[,Yk>to)

a

the shape of the boundary does not have a role in the total error of calculations.

GWAREK:

AR331T2L4RILY-SH.4PED

PLANAR

CIRCUIT

1071

o ~dd -10
-20

is no irregular

visible

loss

of

accuracy

of

the

calculations

due

to

an

boundary.

F.

Computer The method

Memoiy

and Time Needed for Calculation on a wide range of

-30 )

may be implemented

Fig.

6.

&I

of a rectangular stnpline circuit calculated the 14-GHz band.

and measured

in

computers, including small personal computers. 64 kB of RAM should be sufficient to calculate circuits with 50 x 50 rectangular elements. The time depends strongly on the number of meshes tid also on the period of time required to stabilize the oscillations in the circuit. For example, analyzing the circuit of Fig. 7 for a fixed frequency takes about 1 min in Fortran on a CDC computer of the Cyber personal 70 series or about computer. IV.
The method presented CONCLUSION in this paper shows many ad-

80 ruin in Pascal on a standard

Fig.

7.

lS211 of

stnplim circuit
measured

of irregular shape c~c~ated and in the 14-GHz band.

vantages. 1) It is very versatile. A universal program may be worked out for an analysis of a planar circuit of any shape

defined The time needed to stabilize dimensions and its loaded the oscillations Another in the circuit the circuit is to be set of possibility may be assumed by the user taking into account Q-factor. to leave the time of terminating up automatically of the V(t) in the algorithm monitored function the calculations

by the user. of calculations can be made sufficient the theory to between and the of

2) The accuracy experiment. 3) There

assure a very good agreement are possibilities

for the further

development

on the basis of the shape in several chosen points

the method. For example, the circuit losses may be easily included in the calculations. It is also possible to analyze a circuit with lumped branched into it. 4) The method elements (both linear and nonlinear)

the circuit. It is always possible to reduce the error caused by a transient response to a negligible value, but it is very time consuming for the circuits of a high loaded Q-factor.

simulates

a natural

process of the wave

propagation, and the actual field distribution may be displayed during the calculations. This may help in debugging E, Examples of Application of the calculated of the frequency and in the program. It may also give some valuable designer trying to understand It may of his circuit. microwaves. information to the circuit properties of better the physical

Figs. 6 and 7 show the comparison measured 14-GHz values of IS21I as a function band for two stripline

circuits.

The shape of the

also be useful

in the teaching

inner strip of each circuit is also presented. Both circuits were built on a stripline of the height 2d = 6.94 mm filled with a dielectric of c,= 2.5. The dimensions of the circuit 1 = 46.5 mm, included according and in the calculations to (10) in Fig. 6 are w = 5.2 mm, fields were these dimensions 1 = 49.6

h = 54.8 mm. The fringing by increasing w= to the values

A disadvantage of the method is the long computing time. This is especially important for the circuits of high loaded Q-factors. We may assume, however, that this is the price paid for the mentioned advantages, which may have crucial significance in many applications. assume that due to constant improvement computers, price. the future We may also in the speed of to pay that

8.27 mm,

mm, and h = 57.9 mm. This circuit was designed to be matched exactly by a set of squares of the size a = ~w. The results of the calculations are in good agreement with those of the measurements. The difference between them visible for higher frequencies may be attributed to higher value of the ratio a/A. For example, let us assume the frequency .f = 4 GHz (a/h = 1/17) and calculate the error described by (29) assuming a sinusoidal distribution of V(x, y) along the x-axis. We obtain the error at the level of about 45 dB, and, since several similar of about Fig. errors accumulate, similar the error level 30 dB N what we may expect. to those of Fig. 6 but

user will be more willing

IU3FE~NCES

[1]

[2]

[3]

T. Okoshi and T. Miyoshi, The planar circuiAn approach to IEEE Trans. A41crowave Theory microwave integrated circuitry, pp. 245252,Apr. 1972. Tech., vol. MTT-20, Hsu Jui-Pang and T. Anada, Planar circuit equation and its practical application to planar-type transmission-line circuit, in IEEE MTT Int. Symp. Dig. (Boston) 1983, pp. 574-576. R. Chadha and K. C. Gupta, Greens functions for triangular segments in planar microwave circuits, IEEE Trans. A4icrowiwe Theory Tech., vol. MTT-28 pp. 1139-1143, Oct. 1980. T. Okoshi, Y. Uehara, and T. Takeuchi, The segmentation method An approach to the analysis of microwave planar circuitsj IEEE Trans. Microwave Theory Tech., vol. MTT-24, pp. 662-668, Oct. 1976. R. E. Collin, Field Theoty of Guided Waues. McGraw-Hill, 1960, chs. 1 and 5.

[4]

7 shows the results

obtained for a circuit of a shape drawn at random (which is also shown in Fig. 7). The mesh size assumed for calculations is the same as in the previous example. There

[5]

1072

IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. MTT-33, NO. 10, OCTOBER 1985

[6] [7]

G. E. Forsythe and W. R. Wasow, Finite-Difference Methods for Partial Differential Equations. New York: Wiley, 1960. G. D. Smith, Numerical Solution of Partial Differential Equations. London: Oxford University Press, 1975.

electronic engineering from the Technical University of Warsaw, Warsaw, Poland, in 1970 and 1977, respectively. Since 1970, he has been with the Institute of Radioelectronics, Technical University of Warsaw, where he is currently a Deputy Director. While on leave during the academic year 19731974, he completed the postgraduate study and received the M. SC. degree from the Massachusetts Institute of Technology, Cambridge, MA, in 1974. His research interests are in the

Wojciech September

K. Gwarek was born in SkarZysko-Kamienna, Poland, on 13, 1947. He received the M. SC. (Honors) and Ph.D. degrees in

areas of electromagnetic computer-aided analysis

field theory, microwave of microwave circuits.

measurements,

and