Design of an Indoor Wireless Network with Neural Prediction Model

Base stations in wireless local area networks (WLAN) need to provide good link to the backbone of communication system. Generally, problem can be reduced to a given building, where it is needed to determine the number and the positions of base stations in order to cover the building with minimum resources, i.e. proper selection of base station positions is necessary to provide adequate signal coverage and to minimize co-channel coverage overlap. Prediction of the signal strength for indoor propagation environments can be faced with the analysis methods that include effects of multipath propagation, such as signal attenuation, reflection, and diffraction. When the considered building is architecturally complex (non-parallel and/or non-smooth walls of unknown effective permittivity) this approach is extremely complicated. To avoid this computational complexity the neural network model, trained with measurements, is proposed. The same neural network propagation model is used for optimization of the base station position. The optimization process is performed with the particle swarm optimization (PSO) algorithm, genetic algorithm, Powell's conjugate direction method and simplex search method. Our intention was to verify PSO algorithm speed and results with the other optimization algorithms.