Neural network identification and performance assessment of a multipath time-varying satellite channel

This paper proposes a Neural network (NN) based channel estimation method for identifying the parameters of a nonlinear time varying satellite channel. A multipath time-varying Ricean-fading channel has been considered in the analysis for a down link scenario. To study the flexibility and the performance of the proposed method, the channel in question has been varied over a reasonable range of Doppler frequencies, and the estimation for each case has been made by employing 16-quadrature amplitude modulation (16-QAM) technique. Both back propagation (BP) and natural gradient (NG) algorithms have been studied for the channel identification technique. Finally, a NN maximum likelihood sequence estimator (NN-MLSE) based receiver has been studied for the addressed system. Simulation results show that the NG algorithm (NG,) can reasonably track the variation of the channel parameters when compared to BP algorithm (BPA). Moreover, the results demonstrate that the NN-MLSE receiver, trained with NGA, performs very close to that of the ideal (the channel is perfectly known) MLSE receiver in terms of symbol error rate (SER).