Learning for Detection: A Deep Learning Wireless Communication Receiver Over Rayleigh Fading Channels

Abstract

The evolution of data driven optimization has been shown advantageous in many applications. In this paper, we propose a deep learning architecture for the wireless communications receiver to optimize detection in terms of Bit Error Rate (BER), which enables reliable communication over AWGN and multipath Rayleigh fading channels with bandwidth constraints. Our approach uses a deep autoencoder to estimate the received signal along with an additional layer for symbol detection, which are jointly trained and fine-tuned. Significant improvement in terms of BER is achieved in simulation compared to the theoretical baseline.