Deep Learning Decoder for MIMO Communications with Impulsive Noise

Abstract

In this paper we consider signal detection in multiple-input-multiple-output (MIMO) systems with an impulsive noise channel. The existing, near optimal, sphere decoder (SD) achieves good performance, however, the computational complexity is directly related to the number of nodes visited during the tree search and the signal-to-noise ratio (SNR). Using neural network techniques, a Deep Learning Detector (DLD) is proposed. The DLD method detects signals transmitted in an impulsive noise channel, after an off-line training phase. The detection process of DLD has lower complexity than the average SD complexity, while exhibiting good performance. What is even more interesting is that the computational complexity of DLD is constant across SNR, in contrast to the SD detectors, which have an exponential complexity across the SNR. This constant complexity could be very helpful when implementing a detector in practice because it could allow for better optimization of resources. To evaluate the performance of our proposed method we have used a low level simulator that generates a fairly accurate model of a MIMO system with an impulsive noise channel. The complexity analysis and simulation results validate the arguments presented in this paper.