Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Genetic Algorithm Optimization

and its Application to Antenna Design


J. Michael Johnson(*) and Yahya Rahmat-samii
Department of Electrical Engineering
University of California, Los Angeles
Los Angeles, CA 90024-1594

I. Introduction
Synthesis of antenna pattems employing iterative optimization techniques has been
studied by many authors. However, successful application of these approaches to
pattern synthesis has usually been limited to relatively simple arrays or has required
careful, intelligent selection of the optimization starting points dictated by the nature
of the optimization techniques used and the functions being optimized. This is
because conventional functional optimization techniques are either based on greedy,
local optimization methods such as gradient methods or consist of random walk
solution space searches. In either case, these conventional techniques are often
poorly suited to the task of arbitrary pattern synthesis in 1D and 2D antenna arrays
due to the high dimensional, multimodal functional domains involved. In addition,
traditional optimization techniques usually require the object function to be, at the
veiy least, continuous and, in many cases to be differentiable, placing severe
limitations on the form and content of the object function. This paper presents a
radically different and relatively new functional optimization methodology known as
Genetic Algorithm (GA) optimiLation that overcomes the above-mentioned
problems of the traditional techniques and discusses how C A optimiiation is applied
to 1D and 2D antenna design.
Genetic Algorithm (GA) optimizers are robust, stochastic search methods modeled
on the concepts of natural selection and evolution. GA optimizers efficiently search
for and locate global maxima in high dimension, multimodal function domains in a
near optimal manner. CAS differ from traditional techniques in that they operate on
a group, or population, of trial solutions in parallel, they operate on a coding of the
function parameters rather than the parameters directly and they use simple,
stochastic operators to explore the solution domain. CAS are blind to the intemal
operation of the object function placing few constraints on it and GAS possess
implicit parallelism enabling the extensive search of multiple hyperplanes in the
function domain without directly testing each hyperplane value. Functional
optimizations using GAS are able to handle discrete, continuos, discontinuous, and
mixed functional domains and GAS are readily implemented in panllel processing
environments. GAS population versus single point operation and the lack of an
object function differentiability requirement permits GAS to function effectively in
noisy and convoluted function domains

0-7803-2009-3/94/$4.00 0 1994 IEEE.

326

11. A Simple Genetic Algorithm Optimizer


The concept of Genetic Algorithms, first formalized by Holland (1975) and
extended to functional optimization by De Jong (1975). involves the use of
optimization search strategies pattemed after the Darwinian notion of natural
selection and evolution. During a GA optimization, a set of trial solutions is chosen
and "evolves" toward an optimal solution under the "selective pressure" of the
object function.

Populations and Chromosomes: in GA based optimizations a set of trial solutions


is assembled as a population. The parameter set representing each trial solution or
individual is coded to form a string or chromosome and each individual is assigned a
fitness value by evaluation of the objective function. The objective function is to
only link between the GA optimizer and the physical problem.
Parents: following this initialization process, pairs of individuals are selected
(with replacement) from the population in a probabilistic manner weighted by their
relative fitness and designated as parents. A typical selection scheme, a weighted
roulette wheel, is depicted in Figure 1. Each individual in the population is assigned
space on the roulette wheel proportional to the individual's relative fitness. The
wheel is spun each time a parent is required. Individuals with the largest spaces on
the wheel have the greatest chance of being selected and, therefore, the greatest
probability of passing on their characteristics to the next generation.
Children: a pair of offspring, or children, are then generated from the selected pair
of parents by the application of simple stochastic operators. The principle operators
are crossover and mutation. Crossover occurs with a probability of pcross
(typ. 0.6-0.8) and involves the random selection of a crossover site and the
combining of the two parent's genetic information. Specifically as shown in
Figure 2, child 1 receives the chromosomal substring that precedes the cross-site in
parent 1 and the substring following the cross-site in parent 2. Child 2 gets the
remaining genetic information not given to child 1. The two children produced
share the characteristics of the parents as a result of this recombination operators.
Other recombination operators are sometimes used but crossover is the most
important. Recombination (e.g., crossover) and selection are the principle way that
evolution occurs in a GA optimization.

Mutation: mutation, an example of which is depicted in Figure 3 is a mechanism for


introducing new, unexplored points into the GA optimizer's search domain.
Genetically, mutation randomly changes the genetic makeup of the population.
Mutation is much less important than recombination and occurs with a probability
pmuzation (typ. 0.05) which is much less than pcross.

New Generation: reproduction consisting of selection and recombinatiordmutation,


continues until a new generation is created to replace the original generation.
Highly fit individuals, or more precisely, highly fit characteristics. produce more
copies of themselves in subsequent generation resulting in a general drift of the
population as a whole towards an optimal solution point. The process can be

327

terminated in several ways: threshold on the best individual (Le., the process stops
when an individual has an error less than some amount E), number of generations
exceeds a preselected value, or other some other appropriate criteria.
A simple Genetic Algorithm must be able to perform five basic tasks: encode the
solution parameters in the form of chromosomes, initialize a starting point
population, evaluate and assign fitness values to individuals in the population,
perform reproduction through the fitness weighted selection of individuals from the
population, and perform recombination and mutation to produce members of the
next generation.
111. Object Function

The object function used to assign a fitness value to each of the individuals is the
only link between the physical problem being optimized and the general GA
machinery. The only constraint on the form and content of the object function
imposed by the GA is that the fitness value returned by the object function is in
some manner proportional to the "desirability" of a given trial solution represented
as a set of input parameters to the object function. A shaped pattern antenna array
object function includes a means for calculating the pattern generated by a set of
input parameters (often the amplitude and phase of the array elements) and a means
for comparing the calculated pattern to a desired pattern. Calculation and
comparison can be done at any number of space domain locations and frequencies
with the fitness value returned by the object function representing a weighted sum of
the individual comparisons.
IV. GA Optimization in Several Typical Array Design Problems

The simple CA optimizer developed for antenna design optimization is depicted in


Figure 4. This GA optimizer has been applied to several typical and not so typical
optimization tasks associated with 1D and 2D antenna arrays to better understand
the strengths and weaknesses of the approach. In the first set of CA optimization
trials, a relatively large 1D array with equally spaced, identical elements is designed
having a two level, flat top, scanned beam. Two designs are produced, an
unconstrained design (i.e., both amplitude and phase are allowed to vary) and a
constant amplitude design. In the second round of tests, a shaped, 2D antenna array
pattern has been synthesized.
Since CA optimization can operate on mixed type functions, a design of a 2D array
with non-identical antenna elements has been carried out. In this case the object
function included the ability to calculate pattems resulting from arrays with multiple
element types where the types were selected from a data base of typical elements
and the parameter set is expanded to include the extra parameters identifying
antenna type. Finally, the uniform spacing assumption is removed and a 1D array
with variable element spacing was designed using GA optimization. Again the
object function is modified to accommodate the non-uniform spacing and the input
parameter set was expanded to include element location.

328

References:
Holland, John H. (1975) Ann Arbor: The University of Michigan Press.
De Jong. K. A. (1975) (Doctoral dissertation, University of Michigan).
Dissertation Abstracts International 36(10), 5 140B (University Microfilms
NO. 76-9381).

Figures:

Figure I : The weighted roulette wheel


method of selection.

I,,

iiill

irL
a

=,7
-.

I 2 3 4 5 6 7 E 9101112

I 2 3 4 5 6 7 8 9101112

I I.
,

,2; i

Reproduction

101100101100

0010t1100011

Desired

1 2 3 4 5 6 7 8 9101112

1 2 3 4 5 6 7 8 9101112

Antenna

Figure 2: An example of cross-over


with a length n=lO chromosome.

10

I1

12

m m
1

io

11

12

aleria is not met

Figure 4: A block diagram of the


GA optimizer as applied to antenna
pattern synthesis.

Figure 3: An example of the action of


the mutation operator.

329

You might also like