Channel Decoding Based on Complex-valued Convolutional Neural Networks

Abstract

Inspired by the recent outcomes in deep learning, we propose a novel decoding architecture which concatenates a complex-valued convolutional neural network (CCNN) with a belief propagation (BP) decoder for combating correlated noise in the channel. The CCNN can exploit the complex noise correlation and yield a more accurate estimation of the channel noise. Depressing the influence of channel noise via the proposed architecture, the BP decoder can obtain better decoding performances. Furthermore, extensive experiments are carried out to analyze and verify performances of the proposed framework.