Accurate posterior probability estimates for channel equalization using gaussian processes for classification

Abstract

In this paper we propose to use Gaussian processes for classification (GPC) for solving the channel equalization problem. GPC provides not only accurate decisions as other nonlinear machine learning tools do, i.e. support vector machines or neural networks, but it also assigns posterior probabilities to each one of its output. This is a significant advantage of GPC with respect to other machine learning tools for channel equalization, because the channel decoder benefits from its soft outputs to provide significantly better error correcting capabilities for the entire communication system. As, for previous schemes, the channel decoder had to rely on the hard decisions given by the equalizer, because the output of these methods cannot be transformed into posterior probabilities. We show that the GCP equalizer is able to estimate posterior probabilities accurately in a variety of real digital communications channel.