Neural network equalization for frequency selective nonlinear MIMO channels

Abstract

In order to provide high data rate over wireless channels and improve the system capacity, Multiple-Input Multiple-Output (MIMO) wireless communication systems exploit spatial diversity by using multiple transmit and receive antennas. Moreover, to achieve high date rate and fulfill the power, MIMO systems are equipped with High Power Amplifiers (HPAs). However, HPAs cause nonlinear distortions and affect the receiver's performance. In this paper, we investigate the joint effects of HPA nonlinearity and frequency selective channel on the performance of MIMO receiver. Then, we propose two equalization schemes to compensate simultaneously nonlinear distortions and frequency selective channel effects. The first one is based on a feedforward Neural Network (NN) named (NN-MIMO-Receiver) and the second uses NN technique and LMS equalizer (LMS-NN-MIMO). The Levenberg-Marquardt algorithm (LM) is used for neural network training, which has proven [1] to exhibit a very good performance with lower computation complexity and faster convergence than other algorithms used in literature. These proposed methods are compared in term of Symbol Error Rate (SER) running under nonlinear frequency selective channel.