The application of support vector machines with Gaussian kernels for overcoming co-channel interference

Abstract

Investigates the application of support vector machines (SVMs) for the equalization of communication systems corrupted with additive white Gaussian noise, intersymbol and co-channel interference. Performance obtained with SVMs for this task is compared to the one obtained with linear and radial basis function (RBF) equalizers. The centers and the weights of the RBF networks are determined by the k-means and LMS algorithms, respectively. Experimental results shown that the SVM equalizer outperforms both linear and RBF equalizers, particularly for small training set. In case of time-varying channels, it is envisaged that the length of the training sequence which needs to be periodically transmitted would be reduced by SVM equalizers.