Mar. 2005 Journal of Electronic Science and Technology of China Vol.3 No.

Parameter Optimization Based on GA and HFSS *

SUN Shu-hui, WANG Bing-zhong
(School of Physical Electronics, UESTC Chengdu 610054 China)

Abstract A new project based on genetic algorithm (GA) and high frequency simulation software
(HFSS) is proposed to optimize microwave passive devices effectively. This project is realized with a
general program named as optimization program. The program is compiled by Matlab and the macro
language of HFSS which is a fast and effective way to accomplish tasks. In the paper, two examples are
used to show the project’s feasibility.
Key words GA; HFSS; Matlab; macro language

There are a lot of optimization and self-adaptive the optimization of microwave passive devices by
problems in the field of computational combining the merits of GA with HFSS. The project
electromagnetics. It is difficult to solve these problems includes two processes: one is that HFSS simulates
in large and complex systems except for some simple each structure whose parameters are produced and
applications. optimized by GA and returns the results of simulation
Genetic algorithm (GA) is essentially a search to GA, and another is that GA optimizes the structure’s
algorithm based on the mechanics of nature (e.g. parameters continually according to the data returned
natural selection, survival of the fittest) and natural from HFSS. The two processes are being run
genetics. It combines solution evaluation with alternately until a satisfying optimization is obtained.
randomized and structural exchanges of information A program named as optimization program has
among solutions to obtain optimization. GA is been compiled to realize the project. We complete the
considered to be a robust method because no work with two softwares, Matlab and HFSS. Our
restrictions on the solution space are made during the optimization program shown in Fig.1 includes two
process. The power of this algorithm comes from its parts: genetic algorithm module and fitness module.
ability to exploit historical information structures from
previous solution guesses in an attempt to increase Optimization program
performance of future solution structures[1].
High frequency simulation software (HFSS) is an
interactive software package for calculating Exchange data
each other
electromagnetic behavior of a structure. The software Genetic algorithm module Fitness module

also includes post-processing commands for analyzing

Fig.1 The framework of the optimization program
the electromagnetic behavior of a structure in more
detail. Using HFSS, we can compute basic
electromagnetic field quantities, radiated near and far The genetic algorithm module shown in Fig.2 is
fields for open boundary problems, port characteristic compiled with Matlab. It is not only the main program
impedances and propagation constants, general but also a general optimization program. The genetic
S-parameters and S-parameters renormalized to algorithm module is based on simple genetic algorithm
specific port impedances, and the eigenmodes or (SGA) and some new genetic techniques proposed in
resonances of a structure. recent years[2,3]. It has three functions. First, it sends
By combining the merits of GA and HFSS, the each structure’s parameters produced by itself to the
optimization can be realized for high frequency’s fitness module. Second, it receives each structure’s
structures, especially for large and complex systems. fitness from the fitness module. Third, it optimizes
structure’s parameters continually according to the
1 Related Work fitnesses returned from the fitness module until a
We propose a new project which is suitable for satisfying optimization is obtained.

Received 2004-03-08
* Supported by the Specialized Research Fund for the Doctored Program of Higher Education (No.20010614003) and the Key Project of Chinese Ministry of
Education (No.104166)
 46 Journal of Electronic Science and Technology of China Vol.3

2 Examples
Initialization
of parameters 2.1 Impedance Converter of Rectangular
Waveguides
Encode and generate
The size of a section of impedance convertor of
an initial population
Output one the rectangular waveguides shown in Fig.4 has been
individual’s
parameters optimized to reduce S11 at 10 GHz. The parameters of
Evaluate fitness Fitness module optimization include the length c, the width a, and the
Return the
individual’s height b of the middle section of the rectangular
fitness
waveguides. Tab.1 shows the ranges for parameter
Whether the terminate-
condition is satisfied
Stop the running optimization. Without the middle section, S11 is equal
Y and show the result
to 0.24076 at 10 GHz. After inserting the middle
N
section and optimizing its parameters, S11 is equal to
Generate new population
0.004 93 at 10 GHz and the optimized results are also
by genetic operation
shown in Tab.1. The S parameter curves are shown in
Fig.2 The framework of the genetic algorithm module Fig.5 respectively.
20
The fitness module shown in Fig.3 is made up of 9
several subprograms compiled by Matlab and the a 15
b
macro language of HFSS[4]. It has two functions as
c
follows. 16 36 15
1) It receives each structure’s parameters from the
genetic algorithm module and then sends them to Fig.4 The rectangular waveguides
HFSS.
2) It calls HFSS to simulate the structure after Tab.1 The ranges for parameter optimization and
the optimized results
sending a structure’s parameters to HFSS and returns
the results of simulation to the genetic algorithm Parameters a/mm b/mm c/mm
module as the structure’s fitness after simulation. Minimum 21 10 1
Maximum 35 15 30
Input one Optimized results 34.835 294 12.333 333 22.266 667
individual’s
parameters
Genetic algorithm
Call subprogram
module 0
Return the
Call individual’s
ANSOFT fitness −10
HFSS
−20
Magnitude of S11/dB

Confirm the concrete configuration

−30
Call ANSOFT HFSS
−40
Evaluate the configuration
−50
---The S parameter curve without the middle section
Return the —The S parameter curve with optimized middle section
results −60

Fig.3 The framework of the fitness module −70

8.0 8.5 9.0 9.5 10.0 10.5 11.0 11.5 12.0
f/GHz
The optimization program requires users to do Fig.5 The S parameter curve
parameter preassignment by using the macro language
of HFSS to compile program sentences. In order to In the genetic algorithm module, each optimized
make the optimization program be used conveniently, parameter is expressed by 8 binary bits, and the
the preassignment is done through the interface of population size and the number of generations are
HFSS according to the datum-saving character of equal to 10 and 80, respectively. The curve of each
HFSS. generation’s best value of fitness is shown in Fig.6.
 No.1 SUN Shu-hui et al: Parameter Optimization Based on GA and HFSS 47

0
0
−0.01
−0.05
−0.02
−0.03 −0.10
Best value of fitness

Best value of fitness

−0.04 −0.15
−0.05 −0.20
−0.06
−0.25
−0.07
−0.08 −0.30
−0.09 −0.35
0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80
Number of generations Number of generations
Fig.6 The curve of each generation’s best value of fitness Fig.9 The curve of each generation’s best value of fitness

2.2 Microstrip Antenna 3 Conclusions

The size of a microstrip antenna shown in Fig.7 Two examples have been optimized, and the
has been optimized to reduce S11 at 2.45 GHz. The results of optimization show the project is effective and
parameters to be optimized include the thickness h of can offer beneficial reference for the device design. In
the substrate and the length a of the foursquare patch. the later work, we will make the program be used more
Tab.2 shows the ranges for parameter optimization and conveniently. At the same time, new genetic techniques
will be introduced into the program continuously to
the optimized results. The optimized S parameter curve
improve the convergence velocity and search ability of
is shown in Fig.8.
its optimization algorithm.

h a References
[1] David C W, Gerald B S. Genetic algorithm solution of
Fig.7 A microstrip antenna economic dispatch with valve point loading[J]. IEEE Trans.
Power Systems, 1993, 8(3): 1 325-1 332
Tab.2 The ranges for parameter optimization and [2] Wang B Z. Computational Electromagnetics[M]. Beijing:
the optimized results
Science Press, 2002. (in Chinese)
Parameters a/mm h/mm [3] Zhou M, Sun S D. Genetic Algorithms: Theory and
Minimum 25 2 Applications[M]. Beijing: National Defense Industry Press,
Maximum 40 8 1999. (in Chinese)
Optimized results 29.823 529 7.811 765 [4] Introduction to the Ansoft Macro Language[Z]. Ansoft
Corporation, 2001
0

−5 Brief Introduction to Author(s)

Magnitude of S11/dB

−10 SUN Shu-hui (孙树辉) was born in Hainan, China, in

−15 1979. He received the B.S. degree from School of Electrical
Engineering in UESTC in 2002. He is now pursuing the M.S.
−20
degree in Radio Physics. His research interests include genetic
−25 algorithm and simulation of passive microwave circuits. E-mail:
−30 [email protected]
−35 WANG Bing-zhong (王秉中) was born in Chengdu, China,
2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0
in 1962. He received the Ph.D. degree from Institute of Applied
f/GHz
Physics in UESTC in 1988. He is now the Head of Institute of
Fig.8 The optimized S parameter curve Applied Physics, UESTC. His current research interests include:
computational electromagnetics, numerical modeling and
In the genetic algorithm module, each optimized simulation of the electromagnetic behavior, EMC analysis,
parameter is expressed by 8 binary bits, and the electromagnetic modeling by artificial neural networks,
population size and the number of generations are computer-aided design for passive microwave and millimeter
equal to 10 and 80, respectively. The curve of each wave integrated circuits, and antenna design. E-mail:
generation’s best value of fitness is shown in Fig.9. [email protected]