ABSTRACT
This chapter applies neural networks (NNs) to transmission over nonlinear channels. We present applications to satellite communication channels composed of a nonlinear high-power amplifier (HPA) followed by a linear filter. The applications include adaptive predistortion, HPA modeling, maximum likelihood sequence estimation (MLSE) receiver design, and adaptive equalization. We show, in particular, that the natural gradient (NG) descent method used for NN training outperforms the classical ordinary gradient descent-based backpropagation (BP) algorithm in terms of convergence speed, mean square error (MSE), and bit error rate (BER) performance.
