You are on page 1of 4

83.

Genetic Algorithms and Method of Moments (GA/MoM):


A Novel Integration for Antenna Design
J. Michael Johnso; and Yahya Rahmat-Samii
Department of Electrical Engineering
University of Califomia, Los Angeles
405 Hilgard Ave
Los Angeles, CA 90095-1594

INTRODUCTION
Recently, several investigators have reported encouraging results coupling a relatively
new form of optimization technique known as the Genetic Algorithm (GA) with method
of moments (MOM) [1,2]. In particular [l] reported the design of compact microwave
filters using GA coupled with a 2.5D method of moments analysis technique and [2]
reported some interesting results of applying the well known NEC code and GA to the
design of circularly polarized wire antennas with broad angular coverage.
The problem that is often encountered in linking GA to MOM is that GA can require a
large number of trials (MOM analysis operations) before reaching an optimum. When
dealing with the optimizationof an electromagneticstructure composed of many wires and
metal plates, these large number of trials can become prohibitively expensive.
This paper introduces a novel technique for efficiently integrating Genetic Algorithms
with Method of Moments (GA/MoM) for antenna design and related electromagnetic
problems through the application of direct Z-matrix manipulation (DMM). In the DMM
approach to GA/MoM a mother impedance or Zmatrix is filled only once at the
beginning of the optimization process, and then the GA optimizer operates on subsets of
the original mother %Matrix to perform the optimization. Application of DMM for
GA/MoM significantly reduces the total optimization time by eliminating multiple Zmatrix fill operations. Although beyond the scope of this paper, DMM also facilitates the
use of matrix partitioning and pre-solving to further reduce the optimization time in many
practical cases.

GENETICALGORITHMS
Genetic Algorithms (GAS) are a class of robust optimization methods modeled on the
concepts of natural selection and evolution. As optimizers, GAS are adept at handling
complex, multi-modal optimization problems, especially those that are naturally
combinatorial. GASdiffer from more conventionaltechniques in that 1) they operate on a
group (or population) of trial solutions in parallel, 2) they normally operate on a coding of
the function parameters (chromosome) rather than on the parameters themselves, and 3)
they use simple, stochastic operators (selection, crossover, and mutation) to explore the
solution domain in search of an optimal solution.
In a simple GA optimizer, a set of trial solutions is caused to evolve toward an optimal
solution under the selective pressure of the fitness function. The trial solutions are
represented by a string of parameters, generally encoded in some form such as binary.
Operators called selection, crossover and mutation act on the population of trial solutions
to produce a new generation from the current generation. The optimization objectives are
used to influence the constitution of the new generation through evaluation of the fitness
function that assigns a numerical value to each individual. A block diagram of a simple
0-7803-4178-3/97/$10.00

0 1997 IEEE
1664

GA optimizer is illustrated in Figure 1. Details of the implementation of a Genetic


Algorithm optimizer, and the various operators involved are covered in [3]. EM
applications of GA along with many relevant references are presented in [4].

GA/MoM METHODOLOGY
The application of the Electric Field Integral Equation (Em)Method of Moments
technique ultimately results in the representation of the physical problem by a set of linear
equations of the form of equation 2
7 = V ,where Z is the so called 2-matrix or impedance
matrix that specifies the electrical relationship between the various surface patches and
wire segments in the problem, V is a vector accounting for voltage sources, and Z is a
vector containing the unknown currents that are to be found. Once a 2-matrix is
constructed, the desired solution in terms of surface currents is obtained by solving the
linear equatnon [5].
Now, coinsider a new structure that is similar to the original structure except for the
absence of various metal regions or elements. The Zmatrix of the new structure can be
obtained by simply removing rows and columns from the original Zmatrix corresponding
to the absent metal elements.
We can, therefore, propose an optimization methodology as depicted in
Figure 2. In this methodology, a mother Z-matrix is constructed that includes the
presence of metal everywherethat we might allow metal to be. All possible structures that
can be constructed from this mother configuration become subsets of this mother
configuration. A solution methodology is then apparent that allows for the exploration of
a large number of potential configurations without the need for costly, multiple 2-matrix
filling.
Upon coupling this approach to GA, we have that the chromosome is simply a string of
ones and zeros, one bit for each metal element in the mother structure that may be
removed. A zero in the chromosome represents the absence of metal while a one
represents the presence of metal. The fitness function involves solving the ZZ = V
equation for a sub-matrix derived from the original, mother Zmatrix and then extracting
the appropriate parameters from the results to compare with the optimization goals.
Refemng to Figure 2, a sub-matrix, 2, is created from the mother 2-matrix by removing
the rows and columns corresponding to a mapping of the zeros in the chromosome to
metal regions in the model. The ZI = equation is then solved and the resulting current
vector is used to extract the desired parameters such as input impedance or antenna
pattern.

WIDEBAND
PATCHANTENNA
DESIGNEXAMPLE
Consider the design of a broadband patch antenna. The original, mother, patch design
consisted of a 0.48 x 0.48 cm patch suspended 0.048 cm above an infinite ground plane.
As can be seen in Figure 3, the initial impedance match to 50 ohms, as indicated by a
dotted line, had a bandwidth of approximately 6%. The feed consists of a wire with a
voltage source at the ground plane/wire interface and located 0.24 x 0.12 cm from the
comer of the patch. The original patch structure including the triangularization used is
shown in Figure 4.
GA/MOM was applied to this mother structure. The population size was 100 individuals
and 100 generations were produced. Probability of crossover was set at 0.7 while the
probability of mutation was equal to 0.02. Elitism and roulette wheel selection [3,4] were
used. The goal was to minimize the maximum S11 magnitude at three frequencies,
2.7 GHz. 3 GHzand 3.3 GHz.

1665

The results of the best individual from the GA/MoM optimization in terms of magnitude
S 11 are shown in Figure 3 as a solid line. As can be seen, an acceptable 2 1 match has
been achieved over the entire 2.7-3.3 GHz band. The resulting structure is shown in
Figure 5. While not an intuitive result, the design is readily realizable by standard printed
circuit board techniques. A test of whether the GA optimized result still represents a
converged solution, was conducted by doubling the discretization density. The calculated
S 11 values for both the single and double density discretization are plotted in Figure 6.
Very little change is observed in either the original square patch performance or that of the
optimized patch so we can conclude that the result in the optimized case is representative
of the true performance of the design within the limitations of MOM.

CONCLUSION
A method of direct matrix manipulation was described that yields significant
reductions in analysis cost by eliminating the multiple re-filling of the Z-matrix. In
addition, computational cost savings from matrix partitioning and pre-solving of portions
of the problem was described. As was true with earlier investigators, the current work
found that GA coupled with MOM can yield some surprising designs for antennas. The
use of GA for the design of compact antennas combining metal patches and wires has
been demonstrated.
Acknowledgments- This work was supported in part by DARPA Contract number
JFBI94-222/J4C942220

REFERENCES
[ 11 John, A and R. H. Jansen, EvolutionaryGenerationof (M)MIC Component Shapes using
2.5D EM Simulation and Discrete Genetic Optimization,1996MTT-S Digest, pp, 745-748.
[2] Linden, D. S. and E. E. Altshuler, AutomatingWire Antenna Design using Genetic
Algonthms, Microwave Journal, vol. 39, March 1996,pp. 74-86.
[3] Goldberg, D. E., Genetic Algorithms in Search, Optimization and Machine Learning.
Addison-Wesley, 1989.
[4] Johnson, J. M. and Y. Rahmat-Samii,Genetic Algorithmsin EngineeringElectromagnetics,
submitted for publication in the AP Magazine.
[5] Hodges, R.E., and Y. Rahmat-Samii,An Iterative Current-BasedHybrid Method for
Complex Structures,IEEE Trans. Antennas and Propagat., vol. 45, no. 2, February 1997.

Figure 1: Block diagram of a simple genetic


algorithm optimization

1666

You might also like