ChannelGAN: Deep Learning-Based Channel Modeling and Generating

Abstract

The increasing complexity on channel modeling and the cost on collecting plenty of high-quality wireless channel data have become the main bottlenecks of developing deep learning (DL) based wireless communications. In this letter, a DL-based channel modeling and generating approach namely ChannelGAN is proposed. Specifically, the ChannelGAN is designed on a small set of 3rd generation partnerships project (3GPP) link-level multiple-input multiple-output (MIMO) channel. Moreover, two evaluation mechanisms including i) power comparison from the perspective of delay and antenna domain and ii) cross validation are implemented where the power comparison proves the consistency between the modeled fake channel and real channel, and the cross validation verifies the effectiveness and availability of the generated fake channel for supporting related DL-based channel state information (CSI) feedback.