Reduced complexity implementation of the Bayesian equalizer using local RBF network for channel equalization problem

This paper examines a method to reduce the implementation complexity of the RBF Bayesian equalizer by model selection. The selection process is based on nding a subset model to approximate the response of the full RBF model for the current input vector, and not for the entire input space. By such a scheme, when the channel equalization problem is non-stationary, the requirement to update all the center locations is removed, and the implementation complexity is reduced. Using computer simulations, we show that the number of centers can be greatly reduced without compromising classi cation performance. Indexing Terms : Bayesian equalizer, RBF network.