Deep Joint Source-Channel Coding for Two-Way Relay Channels

This paper proposes a deep joint source-channel coding (DeepJSCC) scheme for two-way relay channels (TWRC) where two source nodes exchange image samples by means of a cooperative relay node. Unlike conventional DeepJSCC approaches developed for one-way relay channels (OWRC), the proposed DeepJSCC-TWRC facilitates self-interference cancellation at source nodes, thereby utilizing time resources more efficiently. We propose to share identical encoder and decoder models at source nodes so that they can observe a number of image samples during the training step. The effectiveness of the proposed method is demonstrated through numerical results.