Adaptive equalization for digital channels RBF neural network

The paper investigates adaptive equalization of linear and non-linear digital communication channel. An architecture of equalization based on a radial basis function neural networks is proposed. Then we describe the hybrid training technique developed: the complex LMS training for the supervised part, and the complex rival penalized competitive learning method for clustering. The basic idea is that for each input not only the winner input is modified to adapt to the input, but also its rival is learned by a smaller learning rate. Performances obtained are better than LTE and MLP equalizers, because the RBF network has a similar structure to the optimal Bayesian symbol-decision equalizer