Joint Source-Channel Coding for Wireless Image Transmission: A Neural Architecture Search Approach

In this paper, we propose a joint source-channel coding (JSCC) scheme for wireless image transmission, where the encoder and decoder of the transmission system are implemented by convolutional neural networks. A neural architecture search (NAS) method is introduced to find promising network structures aiming at minimizing the distortion of the target image. Experimental results demonstrate that our proposed scheme, referred to as NAS-JSCC, significantly outperforms the conventional manually designed ones in terms of peak-signal-to-noise ratio in nearly all tested signal-to-noise ratio regions.