WEIF-12

SYNTHESIS OF GENERAL TOPOLOGY MULTIPLE

COUPLED RESONATOR FILTERS BY OPTIMIZA,TION

Walid A. Atia, Kawthar A. Zaki, and Ali E. Atiat

University of Maryland, Electrical Engineering Department
College Park, Maryland 20742
+OrbitalSciences Corporation, 2030 1 Century Blvd
Germantown, Maryland 20874

, ABSTRACT The second category [6-91 creates, from the

filter’s transfer function, a coupling matrix with
A synthesis procedure, using optimization, arbitrary topology. Successive planer (2-
for multiple coupled resonator filters having dimensional) similarity transformations are then
general topology and general response is applied to the coupling matrix to reduce certain
described. The error function for the prescribed set of its elements to zero, in order to
optimization is based on the values of the achieve a desired network topology. In many
characteristic function at its zeros and poles. cases, the reduction process using these
The optimization is performed directly on the successive 2-dimensional rotations does not
element values of the coupling matrix. converge. Combinations of the two approaches
Convergence of the optimization is extremely have also been used [4], where the classical
fast and nearly independent of the starting synthesis is first applield, and then 2-dimensional
coupling matrix. Examples of the design of rotations are used to transform the coupling
practical filters with symmetric or asymmetric matrix to a desired topology.
responses and topology are presented. In many practical cases it is desirable to
define the topology of the filter to conform to
INTRODUCTION certain mechanical and packaging constraints.
Also, realization of asymmetric structures as
Synthesis procedures for multiple coupled well as asymmetric filter responses using the
resonator filters of general response have minimum number of resonators is desired. In
previously been developed [1-81. These these cases it is often not possible to use the
procedures generally belong to one of two above synthesis techniques to achieve the
categories. The first category is the classical desired designs due to failure of convergence.
synthesis procedure [l-51, in which a synthesis Direct network optimization has been used with
cycle is established to extract an elementary commercial software packages, where the
section from the filter’s transfer function, coupling matrix elements are the optimization
leaving a lower degree realizable transfer variables and the difference between the transfer
function. The same synthesis cycle is then function frequency response of the filter and the
repeated until the degree of the remainder frequency specification mask is the basis for the
transfer function becomes zero. The resulting objective error function. However this approach
coupling matrix from the classical synthesis has is usually inefficient and often results in non-
a given topology that corresponds to the optimum (local minimum) solutions.
generalized cascade of the extracted sections.
82 1
0-7803-4471-5/98/$10.00 0 1998 IEEE 1998 IEEE MTT-S Digest
 , " " "
+ ! + i

"([; VI
- 1
~

Figure 1: General multiple coupled resonator filter Fig. 2: Filter as a lossless two port network.

This paper presents a different synthesis where h = (@BW)fl/f, '-Jf is the normalized
approach based on using optimization to find the frequency variable, E is a scale factor related to
coupling matrix of a prescribed topology. The the pass band ripple, and
error function is based on evaluating the N

reflection coefficient and transmission

coefficient at zeros and poles of the desired
characteristic function from the trial coupling
matrix. The coupling matrices resulting from j=l
minimization of the error functions are is the characteristic function, the problem is to
realizations that approximate the polynomials of find the coupling matrix M with a given
the numerators and denominators of the desired topology T that realizes the desired insertion
filter characteristic function. loss ratio.
The filter network is a lossless two port
PROBLEM STATEMENT AND CIRCUIT network, driven by a source of internal
ANALYSIS resistance RI and terminated in a load R2, as
shown in Fig. 2. Analysis of this network leads
Consider the equivalent circuit of the to the following expressions for the insertion
multiple coupled resonator filters shown in Fig. loss ratio S21 and the reflection coefficient SI1 :
1. The circuit consists of N series resonators
with fiequency independent couplings Mg
between cavities i and j , driven from a source of
internal resistance R1 and terminated in a
resistive load R2. Cavity i has resonant D-PAR, +&,RI -P,,R,)
SI, =
frequency f o i = fo + Ai where fo is the center D+ --j(41Rl + P,,R,)
frequency of the filter. This circuit model is where P11,Pl2, P22, Po, and D are polynomials
valid for narrow band filters (with bandwidth < in h related to the coupling matrix M by:
20%). The topology of the filter is defined by a 4, 42
topology matrix T, which is an N x N matrix [U - M]-Il1= - [AI
- =-
D D
with Tu = 1 if Mu is non-zero, and Tu = 0 if My is 42
zero. Given an insertion loss ratio: POD= 4; - cIP22[AI- M]-'12 =
D
~

1 All the required polynomials can be readily

s21 = 1+ E 2 Q 2 ( A ) obtained using the Souriau-Framealgorithm [9]
for the inversion of the matrix [AI-MI.

822
 SYNTHESIS BY OPTIMIZATION AND
NUMERICAL EXAMPLES
Zeros)
The synthesis procedure starts by an initial
guess for the coupling matrix. Unlike many
other optimization methods, the choice of the
Symmetric
IieGmnse 1
3-2, 19690 2.52x10-’ 1.75 Fig. 3(b)
initid guess appears to have no measurable Asymmetric
effect on the final result or the computation time
in most cases, which indicates that the Asymmetric
con vergence is to a global minimum. A simple Response
initial guess for the coupling matrix is the
topology matrix, i.e. M=T. An alternative Table 1 Examples of ithe Synthesis by Optimization
choice of the starting coupling matrix guess is results.
the Tchebycheff coupling matrix for a filter of
the same order. Regardless of the initial guess PC. Typical filter responses included in Table 1
the optimization always converges to a desirable and their optimized coupling matrices are shown
solution. The error function is defined as in Fig. 3. Fig. 3(a) is the symmetric response of
follows: a 6-pole filter, having one finite transmission
zero, on either side of the pass band. Fig. 3(b) is
the response of an asymmetric 6-pole filter with
one finite transmission zero in the lower stop
, and E^ are evaluated
where the functions S I I Szl, band and two finite transmission zeros in the
from the current trial matrix M, and E is the upper stop band. Fig. 3(c) is the response of a
desired value of the scale factor related to the 5-pole filter with two finite transmission zeros
pass band ripple as defined in the insertion loss in the lower stop band. In all cases the
ratio. A standard gradient unconstrained search synthesized coupling matrix reproduced the
minimization algorithm is used to minimize the reflection and transmiission zeros at the specified
error function. Convergence of the locations, thus demonstrating the validity of the
minimization is very fast and in all cases tested synthesis procedure.
is independent of the initial coupling matrix
guess. In contrast, optimization using an error CONCLUSIONS
function based on the difference between the
mask and the response was slow, often did not A synthesis procedure for multiple coupled
converge to any acceptable solution, and in all resonator filters using optimization is
cases required an initial coupling matrix guess introduced. The procedure yields a coupling
whose response was close to the desired matrix with a given topology that produces the
response in order to converge. zeros and poles of the specified characteristic
A computer program was developed to function. The procedure is insensitive to the
perform the synthesis by optimization described starting values of the coupling matrix, and
above. Numerous examples were run to verify converges very fast. Typical examples of
and test the program. Some of these examples practical filters are given which show the
are summarized in Table 1. In all cases, the effectiveness of the procedure.
initial coupling matrix guess was chosen as the
topology matrix. The CPU times indicated in
the table is for a 200-MHz Pentium processor

823
 0
m
REFERENCES
E -10
VI

3 20 [i] Smith, B. R. and Temes, G. C. “An Iterative

E Approximation Procedure for Automatic Filter
339
U
Synthesis”, IEEE Trans. Circuit The0y , vol. CT-12,

c:2 ”
pp.107-112, Mar. 1965.
[2] R. J. Cameron, “General Prototype Network Synthesis
-
C
-50

-60 I ~
I’
!
’ !
lj
,
, !
, , ! !
Methods for Microwave Filters”, ESA Journal, vol. 6, pp.
193-206, 1982.
087 088 089 09 091 092 093 094 095 0% ( 37
[3] D. Chambers and J. D. Rhodes, “A Low Pass
Prototype Network Allowing the Placing of Integrated
Poles at Real Frequencies”, IEEE Trans. On Microwave
Theory and Techniques, vol. 3 1, pp.40-45, Jan. 1983.
[4] H. Clark Bell, Jr., “Canonical Asymmetric Coupled-
Resonator Filters”, IEEE Trans. On Microwave Theory
and Techniques, vol. 30, pp.1335-1340, Sept. 1982.
[5] R. Levy, “Direct Synthesis of Cascaded-Quadruplet
(CQ) Filters” IEEE Trans. On Microwave Theory and
Techniques, vol. 43, pp. 2940-2944, Dec. 1995.
[6] Ali E. Atia and Albert E. Williams, “Narrow-Bandpass
Waveguide Filters”, IEEE Trans. On Microwave Theory
and Techniques, vol. 20, pp.258-265, April 1972.
086 067 068 0 8 9 09 091 092 093 094 095 096
F w (GW
[7] Ali E. Atia and Albert E. Williams, “New Types of
(4 Waveguide Bandpass Filters for Satellite Transponders”,
COMSAT Technical Review, vol. 1, No. 1, Fall 1971.
[SI Ali E. Atia, Albert E. Williams and R. W Newcomb,
“Narrow-Band Multiple-Coupled Cavity Synthesis”, ZEEE
Trans. Circuit and Systems, vol. CAS-2 1, pp.649-655,
Sept. 1974.
[9] Leon 0. Chua and Pen-Min Lin, “Computer-Aided
Analysis of Electronic Circuits”, Printice-Hall, Inc,
Englewood Cliffs, New Jersey, 1975, Sec. 9-5, pp. 377-
384.

0842 0843 0844 0845 0846 0847 0848 0849 085 0851 0852
Freq. ( G k )

(4

Figure 3: Examples of optimized filter responses:

(a) 6-Pole symmetric response 1, (b) 6-Pole
asymmetric response, (c) 5-pole asymmetric
response.

824