Diversity combining for digital mobile radio using radial basis function networks

This paper presents the results of an analytical and computer simulation-based study into the application of radial basis function (RBF) neural networks to diversity combining in digital mobile radio. By exploiting the finite state nature of the digitally modulated received signals the RBF diversity combiner is able to provide a significantly better bit error rate (BER) than a linear combiner in a given dispersive channel. Computer simulations have been used to quantify this improvement in channels with varying levels of time dispersion. Finite precision effects in a digital hardware realisation of an RBF diversity combiner are also considered and it is shown how these effects will limit the achievable BER. The performance of the optimal RBF diversity combiner is compared with that of a finite precision minimum distance classifier. (C) 1997 Elsevier Science B.V.