A pilot-aided neural network for modeling and identification of nonlinear satellite mobile channels

We propose a neural network pilot symbol-aided (NN-PSA) receiver for nonlinear satellite mobile channels. The NN-PSA receiver is composed of a two-layer memory-less neural network (NN) nonlinear identifier and a pilot symbol-aided (PSA) fading estimator. In comparison with traditional techniques, the main advantage of this receiver is that it is able to identify and track both the nonlinearity and the time-varying fading simultaneously without prior knowledge of them. The natural gradient (NG) descent is used for NN training, which shows superior performance in comparison to the classical back propagation (BP) algorithm. The paper is supported with simulation results for 16-QAM modulation in terms of symbol error rate (SER) and mean square error (MSE) performance.