Neural network applications to satellite nonlinear channel processing

We investigate applications of neural networks for nonlinear processing of satellite channels. On-board a satellite, a high power amplifier HPA: a travelling wave tube TWT or a solid state power amplifier SSPA, is implemented. In order to get the maximum power efficiency of on-board power sources, the amplifier is designed to operate near the saturation. So it shows a nonlinear performance. Due to the nonlinear amplification, the amplified signal suffers from a complex distortion. A good model of this amplifier is a nonlinear complex gain which can be split to an amplitude nonlinearity, or AM/AM conversion, and a phase nonlinearity, or AM/PM conversion. We present a NN based predistortion scheme, in the baseband, and two methods for updating weights and bias of the predistorter.