PSO and ACO Algorithms Applied to Location Optimization of the WLAN Base Station

Ivan Vilovi1, Nika Burum1, Zvonimir ipu2 and Robert Na2

University of Dubrovnik, Croatia 2 University of Zagreb, Croatia E-mail: ivan.vilovic@unidu.hr, niksa.burum@unidu.hr zvonimir.sipus@fer.hr, robert .nadj@fer.hr Abstract - The main goal of this work is to show the use of evolutionary computation techniques The Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO) in indoor propagation problem. These algorithms employ different strategies and computational efforts, but also they have something in common. Therefore, it is appropriate to compare their performance with the Genetic algorithm (GA). We have demonstrated their ability to optimize base station location using data from neural network model of Wireless Local Area Network (WLAN). The results show that PSO has better properties compared to ACO algorithm. The ACO algorithm needs further work to optimize the algorithm parameters, improve analysis of pheromone data and reduce computation time. However, the ant colony based approach is utilizable for solving such problems. 1. INTRODUCTION The field strength prediction in indoor environments is a difficult and complex task. The methods for field strength analysis can be empirical, semi-deterministic and deterministic. The raytracing method provides rather accurate propagation model, but it is very difficult to calculate accurate signal strength in every point of interest in indoor environments using this method. The results are highly dependable on the accuracy of the data base and require long computing time. In all these models is impossible to include time variant effects, such as movement of people. The position of reflection and diffraction points depends of complete situation in the indoor space and they are influenced by the time variant effects [1]. An alternative approach to field strength prediction is based on the neural network model [2]. We try to predict signal strength in indoor wireless communication in given environment without any detail knowledge about building geometry and construction characteristics. The selected environment (the ground floor of Dubrovnik University building) is rather difficult for ray-tracing calculation because of its irregular shape and a lot of different objects inside (different information tables, boat with sail, pots with palms). The relevant network architecture is trained using the measured field strength from two base stations at randomly distributed locations (Fig. 1). Such trained neural network is used for predicting the field strength distribution as well as for prediction of the optimum base station position. The unconstrained optimization techniques are selected according to the penalty function approach. The Particle Swarm Optimization (PSO) algorithm [3] and Ant Colony Optimization (ACO) [4] algorithms are compared with results of the genetic algorithm (GA) [5]. PSO has been presented as effective method in optimizing complex multidimensional problems. In particular, successful application of this method to antenna design has been shown [6]. Recently, the ACO has been proposed for antenna array design [7]. In the considered case, we were faced with multiple local optima. The problem is overcome by fine tuning the parameters of the each optimization algorithm. 2. DESCRIPTION OF THE MODEL The ground floor of Dubrovnik University building is chosen for simulation environment. The part of the floor under consideration is bordered by points ABCDEFGHI, Fig. 1, which area is 323 m2 and height is 3 m. The origin of the coordinate system is located in the left lower corner as it is shown in the Fig. 1. The locations of base stations, used for training neural network, are denoted by BS1 and BS2, and the height of the base stations was fixed on 2.75 m above the floor. The base stations are Cisco Aironet 1100 that supports 802.11g standard with data rates up to 54 Mbps. The walls are made of the bricks with large windows in aluminum frames. The doors of side rooms are made of wood, while the ceiling is covered by metal plasters and the floor is made of the stone blocks. Measurements of the received signal strength for the various locations of the receiver for selected base station positions (Fig.1) have been made in the first step. The each WLAN access point was operating at the 4th channel at 2.427 GHz (100mW). The signal strength measurements were made by a laptop computer with PCMCIA wireless card positioned 1.2 m above the floor.
1

H G

Up

MAIN ENTRANCE

BS2
GROUND FLOOR HALL
Up

F E
Up

BS1
Reception desk Student administration Student administration

Fig. 1. Plan of the university building, ground floor with training BS The measurements were performed for 233 receiving locations that were 1 m apart from each other. Three measurements were made for each location and mean value was used as the field strength at the considered locations. These values were used in the training and testing of the neural network. The trained neural network model that is used in the optimization procedure is the same as in [8]. The inputs to the model are coordinates of the receiving location, while the output is the field strength at that point. 3. BASE STATION OPTIMIZATION In order to find the optimal location of a single transmitter for a given distribution of receiving points, we need to develop a numerical representation for the quality of signal coverage over the given space as a function of the transmitter location. To obtain such function we need to divide given space into grid of possible receiver and transmitter locations. The density of the grid is determined by the desired accuracy. The trained neural network is used to determine the signal level at arbitrary point wherever the base station is located. According to such approach cost function is presented as sum of all weighted relative signal level predictions (in dBm) along with a penalty value that represents a violation of a maximum tolerated path loss threshold at receiver location, which in our case was the receiver threshold (72dBm for 54 Mbps). The cost function, then, can be expressed as

where N and M are the number of possible locations of base stations and receiving points respectively. Si is relative signal level (in dBm) received from base station i at location with coordinates (xj , yj,), while wj is relevant priority weight ascribed to the jth receiver location, and it makes constraint in the cost function. This constraint requires that the quality of signal coverage at each receiver location over a given space must be above a given threshold value (-72 dBm). In our case the value of weight wj is obtained as
S i ( x j , y j ) > 60dBm w =1 w j = 60 S i ( x j , y j ) 72dBm w = 10 w = 100 S i ( x j , y j ) < 72dBm

(2)

f i = S i (x j , y j )w(S i (x j , y j )),
N M i =1 j =1

The cost function as a function of two variables (x, y), that represent location of base station, is calculated according the equations (1) and (2) where the needed signal levels are obtained from neural network trained model. The coverage is not a differentiable function of the base station locations, so small changes in the base station location can cause great changes in received signal strength, which is caused by completely different pattern of reflected, transmitted and diffracted rays. We may expect a lot of such rapid changes in real indoor environment. The mentioned arguments make such cost function extremely limited in accuracy when it is evaluated at limited number of grid points. As in our method the cost function is calculated from neural network propagation model, there are no limits in the number of grid points.

(1)

3.1 Particle Swarm Optimization Algorithm The PSO, although originally invented for research on simulating the movement of the swarm in 2-dimensional space, can be applied as an optimization method in n-dimensional space [6]. The particles are defined with its own position x, velocity v, and personal best result so far (pbest). The key element of the entire optimization is the changing of particle's velocity [3]. For the k+1 particle movement, the j-th coordinate component of velocity i-th particle, we can write for the particle velocity

vij

k +1

= c0 vij + c1 rand1 pbest ij xij c 2 rand 2 gbest ij xij .

)+

carefully. Considering different values for c0, c1, c2, we have selected 0.4 for the time step value. We have carefully selected population size among large populations with a lot cost function evaluations and longer computation time, and smaller populations that give the result much faster. It was determined by many parametric studies [6] that relatively small populations can sufficiently explore the space under consideration, so population of 30 particles is used in our algorithm. Among the suggested boundary conditions, introduced by various authors, we have selected socalled "reflecting walls" to avoid moving the particle out from the given space [6]. 3.2 Ant Colony Optimization Algorithm The ACO algorithm is originated from ant behavior in the food searching. When an ant travels through paths, from nest to food location, it drops pheromone. According to the pheromone concentration the other ants choose appropriate path. The paths with the greatest pheromone concentration are the shortest ways to the food. The optimization algorithm can be developed from such ant behavior. The first ACO algorithm was the Ant System [4], and after then, other implementations of the algorithm have been developed [7]. Our approach requires some modifications of the algorithm proposed for Traveling Salesman problem solution [9]. In our case the pheromone matrix is generated with matrix elements that represent a location for ant movement, and in the same time it is possible receiver location. The ant population is randomly generated (30 ants) and each ant is associated to one matrix location (node). Each ant can move to any of its eight adjacent locations (matrix elements). The next nod (location) is selected according to the probability with which ant k will choose to go from current location i to next location j

(3)

where i= 1,2,.m, where m is the size of the swarm; j = 1,2,.n, where n is dimension of the space; c0, c1, and c2 are positive constants that scale the old velocity and increase new velocity toward pbest (local best result) or gbest (global best result), respectively. rand1 and rand2 represent random numbers that are uniformly distributed in the interval [0,1]. The parameter c0 is called "inertial weight" and it determines if the particle will stay on its current trajectory or if it will be strongly pulled toward pbest or gbest. Its value is between 0 and 1. The new particle location is given by

xij

k +1

= xij + tvij

k +1

(4)

The new velocity is applied after some time-step t, which is usually one. In other words, particles exchange information about results they obtained, so they know the best of all results so far. According to this information they accelerate in the direction of the global best result (gbest) and at the same time toward its own best result (pbest), so their trajectory is altering between these two goals depending which direction prevails. A proper selection of parameter values is very important to obtain qualitative result. Various authors have proposed different inertial weights and other constants. After running PSO algorithm with different parameters we got the best result when inertial weight c0 was changed linearly from 0.9 to 0.2 during the run of algorithm. In this way, particles in the beginning are less pulled toward pbest and gbest, but after a number of iterations they are more rapidly pulled toward these values. Higher value of c0 means faster move toward gbest, faster convergence, but less accuracy. For the constants c1 and c2, value of 2 is used, since in our case where very little change in coordinates may result in great change in cost function value, the time step needs to be chosen

pij =

lN i k

[ ] [ ]
il il

[ ] [ ]

ij ij

(5)

where tijk is the pheromone content of the path from location i to location j, Nik is the neighborhood locations for ant k when it is at location i. The neighborhood includes only locations that have not been visited by ant k. ij is the desirability of location j, and it depends of optimization goal so it can be our cost function. The influence of the pheromone concentration to the probability value is presented by the constant , while constant do the same for the desirability. These constants are determined empirically and our values are =1,

=10. The ants deposit pheromone on the locations they visited according to the relation

j new = j current + j k ,

(6)

that PSO shows better performance than two others, which is manifesting in shorter CPU time, and the lowest value of cost function. Note that all three methods satisfy the coverage requirements (i.e. that field strength is larger than -72dBm).

where tjk is the amount of pheromone that ant k exudes to the location j when it is going from location i to location j. This additional amount of pheromone is defined by

9850

9800 Objective function (gbestf)

j =
k

1 , fj

(7)

9750

9700

where fj is the cost function of the location j (because the goal of the algorithm is to find the minimum of the cost function). The pheromone evaporates during time and diminishes if there are no new additions. The pheromone evaporation is applied to all locations as follows

9650

9600

9550

10

20

30

40

50 60 Iterations

70

80

90

100

new

= (1 ) j ,

0 < 1.

(8)

Fig. 2 PSO convergence of the cost function versus the number of iterations

The value of is selected empirically, what is in our case = 0.15. 4. EXPERIMENTAL RESULTS
Max desirability

1.05 1.04 1.03 1.02 1.01 1 0.99 0.98 0.97 0.96

x 10

The developed algorithms (PSO, ACO) are used to optimize base station location for the case from chapter 2, and these results are compared with results obtained by genetic algorithm. The suitable computer programs are developed and run for deferent optimization parameters. The convergence of PSO algorithm is illustrated in Fig. 2. The best cost function values are plotted for every iteration. The PSO algorithm converges very fast, in less then 80 iterations. The convergence in the case of ACO algorithm wasn't so fast; it needs more then 300 iterations. In other words ACO algorithm needs more computer time to converge for this type of optimization problem. The Fig. 3 shows this result for ACO algorithm. The genetic algorithm is used as reference and Fig. 4 shows cost function convergence versus the number of generations. The algorithm is run with population of 30 individuals, crossover rate of 0.8 and mutation rate of 0.0225. Very fast convergence with short computation time is characteristic of this algorithm. The complete results are summarized in the Table 1. The colon Result shows optimized locations of base station (coordinates), while colon f includes the values of cost function at optimum locations. CPU time is given in seconds. All three methods give similar results. It can be emphasized

50

100

150

200 Iterations

250

300

350

400

Fig. 3. ACO convergence of cost function value versus the number of iterations 5. CONCLUSION In this paper the optimization of the base station location is studied based on the neural network, as propagation model, and two optimization algorithms PSO and ACO were used for determining the base station location. The PSO have already been used in the optimization of various electromagnetic problems while ACO algorithm hasn't been used in the propagation problem yet. The best performance we got with the PSO algorithm, where results show the lowest cost function and consuming time. The comparison with genetic algorithm shows equality. The ACO algorithm doesn't converges so fast and its result is little worse then of other two algorithms. It must be

emphasized that the accuracy of final results depends on accuracy of the field strength estimation obtained by described neural model.

10200

10100

10000 Cost function

9900

9800

9700

9600

9500

10

20

30

40

50 60 Generations

70

80

90

100

Fig. 4. Convergence of genetic algorithm versus the number of generations Table 1. Optimization results obtained by three methods ACO CPU time Result f CPU time Result
184.47 (5.2, 8.8) 9.62103 814.44 (5.62, 8.32)

Result
(5.52, 8.2)

9.573103

PSO f

9.574103

GA f

CPU time
189.25

REFERENCES [1] T.S. Rappaport, Wireless Communications Principles and Practice, Prentice Hall, USA, 2002 [2] G. Wolfle, F. M. Landstorfer, "Field Strength Prediction with Neural Networks for Indoor IEEE Mobile Communication, 47th International Conference on Vehicular Technology, pp. 82-86. May 1997. [3] R.C.Eberhart, Y. Shi, "Particle swarm optimization: developments, applications, and resources", Proc. 2001 Congr. Evolutionary Computation, vol. 1, 2001. [4] M. Dorigo, V. Maniezzo, and A. Colorni, "Ant system: Optimization by a colony of cooperating agents", IEEE Trans. on System, MAN, and Cybernetics-Part B, vol. 26, pp. 2941, February 1996. [5] Goldberg, D. E. Genetic Algorithms in Search, Optimization, and Machine Learning, Reading MA, Addison-Wesley, 1989. [6] J. Robinson, and Y. Rahmat-Samii, "Particle Swarm Optimization in Electromagnetics", IEEE Trans. on Antennas and Propagation, vol. 52, No. 2, February 2004.

[7] E. Rajo-Iglesias, O. Quevedo-Teruel, "Linear Array Synthesis using an Ant Colony Optimization based Algorithm", IEEE Trans. on Antennas and Propagation, vol. xx. 2005. [8] I. Vilovic, N. Burum and Z. Sipus, "Indoor Field Strength Prediction Based on Neural Network Model and Particle Swarm Optimization", Proc. of 23rd International Review of Progress in Applied Computational Electromagnetics, Verona, 2007. [9] Camelia-Mihaela Pintea, D. Dumitrescu, "Improving ant systems using a local updating rule", Proc. of the 7th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing, 2005.