PROPAGATION PREDICTION FOR INDOOR WIRELESS COMMUNICATION BASED ON NEURAL NETWORKS

The installation of indoor radio systems requires rather detailed propagation characteristics for any arbitrary configuration, so appropriate wave propagation model must be established. In spite of a number proposed solutions for prediction of the propagation characteristics in WLAN environment, it is difficult to say that we have completely satisfied solution. A neural network propagation model that was trained for particular environment was developed. The network architecture is based on the multilayer perceptron. The neural network results are additionally compared with the numerical results obtained by the deterministic 3-D ray tracing model. The ray tracing model includes three reflected rays from the walls and other obstacles what was enough accurate for the given environment. The neural network is used to absorb the knowledge about given environment through training with three access points. Using such obtained knowledge the network is used to predict signal strength at any spot of space under consideration. The various training algorithms were applied to the network to achieve the best convergence results and best possible network model behavior. The network model was trained by Scaled Conjugate Gradient (SCG), Conjugate Gradient of Fletcher-Reeves (CGF), Quasi-Newton (QN), and Levenberg-Marquardt (LM) algorithms. The comparison of the obtained results is presented.