Self organized radial basis function network for symbol detection over linear dispersive AWGN channels

Abstract

In this study, the implementation of the optimal equalizer as a radial basis function (RBF) network is considered. The RBF network has three layers; an input layer, a hidden layer and an output layer. The hidden layer can be trained using a self-organized (unsupervised) or a supervised clustering algorithm and the output layer can be trained using the classical LMS algorithm. An adaptive equalizer operates in two modes; the training mode and the decision directed mode. The advantage of the use of an unsupervised algorithm in the decision directed mode is presented. Its performance is compared with the performance of a supervised clustering algorithm through computer simulations. The performance of the former is found better at low SNR for cases in which the decision regions are very complex and highly nonlinear. On the other hand, being computationally simpler, a supervised algorithm is found to be appropriate for the training mode.