Anda di halaman 1dari 6

1999 High Frequency Postgraduate Student Colloquium

I7 September 1999, UniversiiY of Lee&

Genetic Algorithm using Real Parameters for Array


Antenna Design Optimisation
Yee Hui Lee, Andrew C. Marvin and Stuart 3 . Porter
Department of Electronics,
University of York
Email: yhllOO@ohm.york.ac.uk,acrn4@ohm.york.ac.uk,sjpI@ohm.york.ac.uk
Abstract :- The use of genetic algorithms (GAS) [l] in the design of antennas has become
increasingly popular in recent years. This is due to their versatility and ability to perform
a rigorous search in complex multi-modal search spaces. Much attention has been placed
on binary [2]-[6] genetic algorithms. However, few have looked at the use of continuous [7]
(decimal) genetic algorithms, paticularly for antenna design. In this paper, the similarities and differences between the continuous parameter and binary genetic algorithms are
examined. The use of continuous parameters in genetic algorithms for the optimisation
of antenna design is also presented. The antenna presented here is designed for near field
immunity and susceptability testing.

1. INTRODUCTION
In the design of antennas [1]-[2], much work has been done on the use of the binary genetic
algorithm [7]-[13] as a search tool. Few have looked into the use of the continuous genetic
algorithm. In this paper, an in depth look into the difference between the binary and
decimal genetic algorithm (GA) is presented. The comparison is done using a realistic
antenna design as an example. Through the example, it is shown that the decimal GA
performs relatively better than the binary genetic algorithm.
There are many advantages of using decimal GAS over binary GAS. As is well-known,
most practical problems which require optimisation consist of a number of real parameters.
Traditionally, in order to apply the genetic algorithm to these problems, the real parameters
have to be encoded into binary form. By using real parameter GAS, this process can be
eliminated. Another advantage of real parameter GAS is that, by using appropriate real
value crossover, it is shown that the problem space is searched more thoroughly.
The paper will look at the comparison between binary coding and decimal coding methods
using a practical antenna design problem. The array antenna used for optimization is for
near field electromagnetic immunity and susceptibility testing purposes. It is required that
the antenna is able to illuminate a plane situated 3m in front of the antenna with a uniform
field of 10V/m over a frequency range of 3 0 M H z to 100MHz. and also that the size of
the antenna is no larger than this plane of interest which is 3m by 5m large. Through the
example, it has been shown that the decimal GA performs better than the binary GA.

2. BINARY AND DECIMAL GAS


When applying the binary genetic algorithm to an antenna design problem, the following
steps are taken:

A . Decimal to binary coding

B. Crossover
C. Mutation
These steps are repeated over a number of iterations until the GA converges to a desired
fitness value. If the values do not converge, these steps are repeated until a maximum
number of generations is reached. In this case, the maximum number of generation is set
to 50. When applying the decimal GA, similarly, the above mentioned steps are repeated.
The following describes the differences and similarities of the binary and decimal GA.

A . Decimal to binary coding


When a GA is applied to a real parameter problem such as antenna design, the real parameters have to be coded into binary bit strings. One of the advantages of using real
parameters in GA is that it reduces the computational complexity of coding the real parameters into binary bit strings in order to form the chromosomes. In the real parameter
GA, the chromosomes are formed by placing the parameters to be optimised directly in the
form ( 1 1 x2 1 3 . .. zi).

B. Crossover
Crossover is a reproduction process in which two parent chromosomes are taken to produce
two child chromosomes. This is done by choosing one or more crossover points in a bit
string, and interchanging the bits after the randomly chosen crossover point.

As was shown in [3], a simple one point binary crossover can be seen as a perturbation
of the parents. This can be illustrated using a simple example. Taking the range of the
parameters to be optimised to be between 0 and 1, and the parameters to be encoded into
4 bit strings, two arbitrary parents, 0.5 and 0.875, are chosen. The corresponding binary
form would be 10 00 and 11 10 respectively. During crossover, a random crossover point
is taken, in this case, between bits 2 and 3. The children obtained through a single point
crossover from the parents would then be 10 10 and 11 00, which correspond to decimal
values 0.625 and 0.75. The interchange of the less significant bits 10 and 00 correspond to a
difference of 0.125. Looking back at the two child chromosomes produced from the parents,
this is simply a perturbation of &0.125 where child one is obtained from parent one through
(0.5 0.125) and child two from parent two through (0.875 - 0.125).

If the size range of the two parents, PI and


and Cz, will be given by

c,

4, is R = PZ - 4, then, the two children, Cl

= (1- F)Pl

+ F(P2)

cz = (1- F)P2 + F(P1).

(1)

(2)

where F is a factor to be chosen. This is similar to the random crossover point chosen in
the binary GA. In this particular example, the factor, F , has a value of $.

The fundamental and most meaningful part of coding a real paramzter problem into binary
code for genetic algorithm is the conservation of the schemata, "similarity template", in the
chromosomes. Using the relations above, it can be seen that the schemata in the parent
chromosomes can also be preserved even in a real parameter genetic algorithm. This is
done by perturbating the real parameter parents during the crossover stage. For one point
crossover, a single factor value F is chosen as the crossover factor, in place of the crossover
point.
In the decimal GA, the resultant child chromosome is very much affected by the choice
of the factor F . Therefore, this factor, similar to the randomly chosen crossover point in
the binary GA, determines how well the search space is being searched. In the binary
GA, very often the more significant bits in the chromosome are not altered due to the
lower probability of a high crossover point. However, with an appropriate value for F , the
probability of crossover at more significant bits is increased. This results in a more rigorous
search of the entire problem space.

C. Mutation
In GAS, mutation is performed to ensure that the population is not caught in a local
minimum and also to ensure a diversity in the population. In the binary GA, mutation is
performed by randomly flipping a bit from its original state to its complementary state, from
0 to 1 or 1 to 0. Similarly, in the decimal GA, mutation can be performed by randomly
altering the value of the parameter. This is done by performing a perturbation of the
original parameter.

For example, if a particular parameter (Pi)is chosen to undergo mutation, then a factor
(F,,t) of the parameter range ( R )is added or subtracted from the parameter. This results
where R is the range of the parameter, R = P,,
- Pmj,.
in a new parameter (Pi2~ F,&)
In order to maintain a fair comparison of the two genetic algorithms, the probability of
mutation for the decimal GA is lowered. For example, if the 5 parameters are represented
by a chromosome of length, 100, and the probability of mutation of each bit is 1%,then the
probability of mutation for the decimal GA will be reduced to 0.2% for each parameter.

3. ADVANTAGES OF DECIMAL GA
From the above analysis, it can be seen that the real parameter genetic algorithm can be
viewed as a binary genetic algorithm by means of perturbation. Notice that the conservation
of schemata is still preserved. There are many advantages of using real parameter genetic
algorithms in antenna optimisation:

I. The real parameter need not be converted from decimal to binary every single time
the cost function is calculated. This increases the efficiency of the code.
11. In the binary GA, the precision of the parameters are determined by the number of
bits in which the parameters are coded, the resolution. However, with the use of real
parameters, there will be increased precision as there will be no loss in resolution
during coding.

111. The problem space can be searched more thoroughly by using different crossover and
mutation schemes. As explained above, by choosing a larger range of the factor F ,
the problem space can be searched more rigorously.

10

4. THE GENETIC ALGORITHMS


Having described the decimal genetic algorithms and how it functions, the decimal and
binary genetic algorithms were compared using a practical antenna design. The GA used
here are the Elitism Roulette Wheel Decimal Algorithm and the Elitism Roulette Wheel
Binary Algorithm. In these algorihtms, the crossover and mutation schemes are as described
in the previous section. As for the selection of chromosomes for each new generation, as the
name implies, a combination of the Elitism and Roulette Wheel methods are used. After the
cost of each chromosome has been calculated, the cost are ranked according to their fitness.
The fittest top 10% of the population is taken directly and placed in the new population.
The remaining 90% of the population is filled up by the roulette wheel scheme as described
in [4]. In this scheme, all the members are allocated a section on the wheel. The proportion
of the section allocated to each chromosome depends on its fitness. The wheel is then spun
and a chromosome is selected into the new population. These algorithms are used together
with the numerical electromagnetic code (NEC) [14] during optimisation.

5. RESULTS AND ANALYSIS

Figure 1. Antenna structure to be optimised.

Fig. 1 shows an antenna which may be used for immunity and susceptability testing. The
requirement of this antenna is to radiate a uniform near electric field of 10V/m over an area
of 5m by 3m at a distance of 3m in front of the antenna. It would be desirable to achieve
this with a limited input power of 1OOW.
The structure under optimisation is as shown in Fig. 1. There are 10 parameters to be
optimised. The lengths ( F A , F B , F C , BA, B B and B C ) , the spacing between dipoles
( D F , D B and S P ) and the phase delay between the excitations (Lo).
In this optimisation, 3 objectives have to be met. Firstly, 77 uniformly spaced grid points
were taken from the plane of interest (where the uniform field is required). All 77 points
should be within 3dB range from each other. Secondly, these field points have to be as
close to 10V/m as possible. Finally, the antenna is optimised to give maximum power
transmission. The optimisation is done over a frequency range of 30MHz to lOOMHz with a
frequency stepping of 10MHz. With these criteria in mind, the structure as shown in Fig. 1

11

is optimised using both the decimal and binary GAS. For the binary GA, each parameter
is encoded into 8 bit strings, making a chromosome of length 80 bits. A population of 50 is
used for both GAS and a maximum of 50 generations are run.
From the two optimisations, it was found that the decimal GA converges to a cost value of
73.9 whereas the binary GA converges to a value of 69.5. Fig. 2 shows the distribution of
electric field strength of the 77 grid points taken from the plane of interest. All the points
are plotted for each frequency steps. From the results, it can be seen that the decimal
GA produces a more uniform field profile then the binary GA. The lines in the figure show
the 3dB band for each frequency. As can be seen, all 77 points for each frequency in the
structure produced by the decimal GA lies within the desired 3dB range. However, for the
antenna design produced by the binary GA, 4 points at IOOMHz are out of the 3dB range.

j.
;
..
.................................................
..
..
.
.
:
:
:
~

:I

$
........... ~..
.: _ - -.

~~

'

'

.:

_.4_.

6-

30 40

60 70 80
Frequencyin MHz

90

50

100

30 40 50 60 70 80
Frequency in MHz

(4

90 100

(b)

Figure 2. Electric field strength distribution for (a) decimal GA, (b) binary GA.

Another one of the objectives is to keep the electric field strength as close to 10V/lm as
possible. From Fig. 2, it can be seen that the electric field strength is quite well distributed
and varies very little from the desired 10V/m. However, the antenna optimised by the
binary GA shows much more variation from the desired field strength.

,I:

..

..

..

10 15 20 25 30 35 40 45 50

Number of Generations

Figure 3. Tkend of convergence for decimal and binary GA.

Fig. 3 shows the trend of convergence for the 50 generations during the optimisation process.
It is interesting to note that the decimal GA converges later than the binary GA. This is
due to a greater variation in the population maintained throughout the iterations for the

12

decimal algorithm as it performs a rigorous search of the problem space. As can be seen &om
Fig. 3, the binary GA converges after about 20 generations and no further improvement
to the cost function can be obtained, whereas the decimal GA converges after about 40
generations. Results obtained hom this antenna design has shown that the decimal GA
does in fact perform better then that of the binary GA. Therefore, it has been shown that
for array antenna design, decimal GAS can perform better then binary GAS.

6. CONCLUSION
In this paper, the link between decimal genetic algorithm and binary genetic algorithms
has been studied. The use of the decimal genetic algorithm for array antenna design has
been examined. This is done using a realistic antenna design problem. It has been found
that the decimal GA converges to a higher cost value although the convergency is slower.
Various aspects such as the speed of convergence and the trend of convergence has also been
analysed.
Finally, two antenna designs for immunity and susceptability testings were arrived at; one
obtained from the decimal GA and the other from the binary GA. Results show that the
antenna design from the decimal GA performs much better than that from the binary GA.
This again cofirms that the decimal GA performs better than the binary GA for array
antenna design.

References
J. D. Klaus, Antennas, McGraw-Hill, 1988.
T. Gilligan, Modem Antenna Design, McGraw-Hill, 1985.
A. H. Wright, Genetic Algorithms for Real Parameter Optimization, Foundation of Genetic Algorithms, San Mateo, CA: Morgan Kaufmann, 1991.
D. E. Goldberg, Genetic Algorithms in search, optimization B Machine Learning, Reading, M A
Addison-Wesley, 1989.
J. M. Johnson and Y. R. Sam ii, Genetic Algorithms in Engineering Electromagnetics, IEEE Antennas Propag. Mag., vo1.39, No. 4, pp.7-25, Bug. 1997.
R. L. Haupt, An Introduction t o Genetic Algorithms for Electromagnetics, IEEE Antennas Propag.
Mag., vo1.37, No. 2, pp.7-15, April 1995.
E. E. Altshuler and D.S. Linden, Wire-Antenna Designs Using Genetic Algorithms, IEEE Antennas
Prooao. Mao.. vo1.39. No. 2 . ~ ~ . 3 3 - 4 Anril
3.
1997.
jones&d W. $. J o i n 4 Desi& ofYagi-UdaAntennas Using Genetic Algorithms, IEEE %ns.
E. i.
Antennas Propag., vo1.45, No. 9, pp.1386-1392, Sep. 1997.
R. L. Haupt, Thinned Arrays Using Genetic Algoritbms, IEEE RanS. Antennas Propag., vo1.42, No.
7, pp.993-999, July 1994.
1 A. Boag, E. Michielssen and R. Mittra, Design of Electrically Loaded Wire Antennas Using Genetic
Algorithms, IEEE %ns. Antennas Propag., vo1.44, No. 5 , pp.687-695, May 1996.
[ll] E. E. Altshuler and D. S. Linden, Design of a Loaded Monopoly Having Hemispherical Coverage
Using a Genetic Algorithm, IEEE Bans. Antennas Propag., vo1.45, No. 1, pp.1-4, January 1997.
[ l Z l Z. Altman, R. Mittra and A. Boag, New Design of Ultra Wide-Band Communication Antennas Using
Genetic Algorithm, IEEE %ns. Antennas Propag., vo1.45, No. 10, pp.1494-1501, Oct. 1997.
[13] B. Kemp, S. J. Porter and J. F. Dawson, Optimization of Wire Antennas Using Genetic Algorithms
and Simulated Annealing, ACES Symposium Proceedings, ~01.11,pp.1350-1357, March 1997.
[14] G. J. Burke and A. J. Poggio, Numerical Electromagnetic Code (NEC) - Method of Moments, Lawrence
Livermore National Laboratory Rept. UCID-18834, January 1981.

13

Anda mungkin juga menyukai