A Deep Learning-Based Receiver for LEO Satellite Communications With Nonlinear Distortion

In order to solve the problem of nonlinear distortion of transmitted signals due to the nonlinear characteristics of high power amplifiers (HPAs) and multipath effects in low earth orbit (LEO) satellite communication systems, a deep learning-based receiver that combines a linear minimize mean squared error (LMMSE) equalizer and signal distortion mitigation network (SDMNet) demodulator, namely LMSDNet, is proposed in this letter. The LMMSE equalizer first mitigates the multipath effects of the signal, while SDMNet further addresses the nonlinear distortion. To capture the global dependencies of the nonlinear, the receiver is skillfully designed with residual blocks consisting of multiple global attention mechanism (GAMs). Numerical results over the non terrestrial network-tapped delay line (NTNTDL) channel models illustrate that the proposed LMSDNet receiver significantly outperforms both conventional LMMSE and existing deep learning-based receiver.