Equalization of non-linear perturbations by a multilayer perceptron in satellite channel transmission

Perturbations generated by the channel are among the most important limits in digital communication applications. A possibility for decreasing the intersymbol interference is via the use of equalization techniques. The satellite channel transmission is an important source of non-linear perturbations. The cancellation of this type of perturbation requires the use of a non-linear equalizer. In this context, many studies have been done using decision feedback equalizers, Volterra series, etc. Despite the improvement in the results compared to those obtained with transversal filters, these results are generally not sufficient compared to bit error ratio in the linear channel. Neural networks, thanks to their ability to create decision regions as complex as the application needs, constitute a new alternative. We show that the multilayer perceptron, with a modified backpropagation algorithm, provides, compared to transversal filters with least mean square algorithm, an interesting improvement concerning bit error rate