Neural network based channel estimation and performance evaluation of time varying multipath satellite channel

Abstract

Neural network (NN) based channel estimation method has been proposed 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 performance of the proposed method, the channel 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. Back propagation (BP) and natural gradient (NG) algorithms have been studied for the channel identification technique. Based on different learning rates and normalized Doppler frequencies, a comparative analysis between the algorithms has been provided. Finally, a NN maximum likelihood sequence estimator (NN-MLSE) based receiver has been studied for the addressed system. Simulation results show that the NN-MLSE receiver performs close to that of the ideal MLSE receiver in terms of symbol error rate (SER).