Deep Ensemble Learning: A Communications Receiver Over Wireless Fading Channels

Abstract

Deep learning algorithms have proven themselves powerful in different applications because of their ability of generalization. In this paper, we introduce a deep learning wireless communications receiver network that enables reliable data transmission over wireless multipath Rayleigh fading channels. Significant improvements in terms of Bit Error Rate (BER) are achieved in simulation for various channel models compared to the traditional receiver. We also present a training procedure that leads to better generalization of the network by dividing it into jointly trained subnetworks as a Deep Ensemble learner while leveraging the regularization by combining information from different subnetworks.