Applications of neural networks to the prediction of propagation path loss for mobile communications systems

Summary form only given, as follows. The purpose of this paper is to present the results of radio coverage studies using neural networks. The problem of finding an exact or approximation model for propagation path loss occurs frequently in planning mobile communication systems. Two strategies for propagation path loss prediction are in use: one is to derive an empirical formula for propagation path loss from measurement data and the other is a deterministic method that is based on the theory of diffraction. While the deterministic methods suffer from excessive computation time and the need for very detailed databases, the empirical methods have difficulties in making efficient use of all available data. An empirical formula based on Okumura's results has been developed by Hata in order to make the propagation loss prediction easy to apply. The advantage of using neural networks for field strength prediction is given by the possibility of deriving training patterns directly from measurements. This allows the system to become very flexible and to adapt to an arbitrary environment. The applications of neural networks discussed in this paper can be viewed as a function approximation problem consisting of a nonlinear mapping from a set of input variables containing information about potential receiver locations (i.e. distance to the transmitters, terrain, frequency) onto a single output variable representing predicted path loss. Our intention is to train a neural network with measurement data for the purpose of field strength prediction. Applications to Hata's formula and knife-edge diffraction are included to demonstrate the effectiveness of the neural network approach.