Small-scale fading prediction using an artificial neural network

This paper proposes and evaluates an artificial neural network used for prediction of the Ricean K-factor. The model is trained with measurement data obtained by utilising the IS-95 pilot signal of a commercial CDMA mobile network in rural Australia. The neural network inputs are chosen to be distance to base station, parameters easily obtained from terrain path profiles and a clutter parameter extracted from a vegetation density data base. The Ricean K-factor indicates the small-scale fading margin required in a link budget calculation scenario, where pessimistic modelling, assuming Rayleigh fading, would lead to unnecessary high base station transmitter power and possible interference problems. The statistical analysis shows that the artificial neural network can be applied to accurately predict variations in the small-scale fading characteristics due to different terrain and vegetation.