You are on page 1of 4

GSM NETWORK OPTIMIZATION BY GENETIC ALGORITHM

Luděk ZÁVODNÝ
Institute of Radio Electronic, Faculty of Electrical Engineering and Communication, Brno
University of Technology,
Purkyňova 118, 61200 Brno, Czech Republic
zavodny@feec.vutbr.cz

Abstract: This paper deals with the utilization of the genetic algorithm for optimization of
GSM transmitters parameters such as their radio channel and emitted power. Signal to noise
ratio is the main criterion of optimization.

1 INTRODUCTION

Expansive progress of mobile communications, like GSM system, requires the use of
design system for simulation of radiowave propagation. Modeling and simulations of radio
wave propagation carry a lot of advantages. Making the design of radio-network faster is the
great positive of this modelling. It can bring the saving money beside the saving time.
Optimization of the location, used radios channel and transmitted power of radio
transmitters (base transient station in GMS system) is finally the most attractive utilization of
radio channel modelling. Genetic algorithms are the one of the latest methods. Their best
advantage is the ability to optimization the tasks which have a lot of local minimums.

2 OVERVIEW OF GENETIC ALGORITHMS

Genetic algorithms proceed from principles of biologic systems evaluation. Ever


solution of solved objective can be described by the bit sequence - chromosome.
Chromosome represents values of parameters of the system in discrete form.
Genetic algorithm works with group of several chromosomes. This file is called
population and its every element is an individual. Process of optimization starts with the
starting population, with is randomly generated by the computer. Number of individuals in
population is usually constant but individuals are changed to improving quality of population.
Every new (changed) population creates new next generation.
Quality of population is improved by several principles. Value of quality function of
ever individual is calculated at first. Selection, which follows in next step, is based on value
of quality function. Purpose of selection is to eliminate worse individual from next population
and to prefer better ones. Tree basic selection methods exist: the decimation, the proportionate
selection and the tournament selection. Next generation is created by crossover and mutation
of selected individuals. Elitist strategy is used to guarantee of best individual survive.

3 SOLVED OBJECTIVE

There are six base transceiver station (BTS) on the area, witch location is fixed. The
area is divided into 15 times 10 sub areas by the calculating net (see on Fig. 1). Each of those
six BTS can radiates by 8 level of emitted power (the smallest value is zero - BTS is not
transmitted, highest value is 50W) on one of two radio cannels.

BTS 6 BTS 5 BTS 4

y
BTS 1 BTS 2 BTS 3

Fig. 1: The computational net of solved task.

Radio signal propagation is modeled by one ray approach. Geometric spreading is the
only factor of attenuation the power and it follows an inverse square law with distance (1) [1].
Pe D
Pr = (1)
(2kL )2 W
Where Pr and Pe are received and emitted power, k is the wave number, L is distance
from the source, D is directivity factor of source antenna and W is attenuation factor.
4bits define each BTS. One of them carries information about radio channel and
remaining 3 bits carry information about emitted power. It means that 4*6=24 bits can
describe every state of solved problem. These 24 bits create the chromosome. Total number of
states is 2 24 = 16.777,216 .
The criterion of optimization is the highest signal to noise ratio on either first or second
radio channel (see on error function (2)).
1 X Y
Q= ∑∑ [max(SNR 1 (x, y ), SNR 2 (x, y ))]
X ⋅ Y x =1 y =1
(2)

SNR1 and SNR2 represent signal to noise ratio on first or second radio channel in the
location (x, y).
signal n ,k
SNR n = (3)
 6 
 ∑ signal n ,i  − signal n ,k + noise
 i =1 
Signaln,k is the strongest signal. Signals from other sources cause interference.
It was choused 12 individuals in population; number of generation was fixed at 80. The
proportionate selection was used. The probability of crossover was set up to 95%, while the
probability of mutation to 5%. Elitist strategy was used to decreasing of quality function
prevent. There are displayed interesting solutions and their value of quality function on figure
2. Figure 3 displays evolution of best individual quality at five attempts.
Fig. 2: Signal to noise ratio of the samples of individual. Radio channel 1 is on the left.
Radio channel 2 is on the right. Quality of these individuals (from the top): 1st: 8.2772dB,
2nd: 7.1600dB, 3rd: 3.3187dB, 4th: 8.6765dB.

Fig. 3: Evolution of quality of the best individual from population and average quality of
the all population.
Fig. 4: Signal to noise ratio of the best solution.

On the bottom there is showed average value of quality function of all generation at the
last attempt.

4 CONCLUSION

Part of radio network with contains six base transceiver stations was optimized to
maximal signal to noise ratio. Only 2 of 5 attempts were converged to best solution in 80th
generation at 12 individuals in population. Increasing number of individuals or generations
can improve this statistic, but high complexity of computing was the greatest limitation of
optimization. One run of optimization took a lot of time, although there were only 15x10
calculating points in this simply case. Minimization of computing algorithm will be necessary
for solving the more complex objectives, which will contain lot of calculating points.

References

[1] Siwiak, K.: Radiowave Propagation and Antennas for Personal Communications, London,
Artech House Boston 1995, ISBN 0-89006-755-4.
[2] Černohorský, D., Nováček, Z. aj.: Elektromagnetické vlny a vedení, Brno, FEI VUT
Brno, 1999.
[3] Černohorský, D., Raida Z., Škvor Z., Nováček, Z.: Analýza a optimalizace mikrovlných
struktur, Brno, FEI VUT Brno, 1999, ISBN 80-214-1512-6.

You might also like