Recurrent radial basis function networks for optimal blind equalization

A recurrent version of a radial basis function (RBF) network can compute optimal symbol-by-symbol decisions for equalizing Gaussian channels in digital communication systems, but the (linear or not) channel response and the noise variance must be known. Starting from theoretical considerations, a novel technique for learning the channel parameters in a non-supervised, non-decision directed way is proposed. This technique provides a simple and fast algorithm that can be used for tracking in time variant environments or for blind equalization purposes