Lightweight Deep Joint Source-Channel Coding for Semantic Communications Over Fading Channels

Deep joint source-channel coding (DeepJSCC) emerges as a novel technology in semantic communication, coin-ciding with the rising demand for edge devices in the Internet of Things (IoT). Consequently, the deployment of DeepJSCC on edge devices becomes a pivotal research direction in semantic communication. However, DeepJSCC confronts issues related to the fading of complex channels. Besides, implementing DeepJSCC on edge devices also poses challenges due to the constrained computing resources. In this paper, we propose a method named DeepJSCC with Ensemble learning (DeepJSCC-ES) to enhance its ability in resisting fading in the Rician channel. Furthermore, we present a solution by introducing a signal-to-noise ratio (SNR)-adaptive pruning algorithm called the DeepJSCC SNR-Adaptive Pruning method (DJSAP) to make the DeepJSCC network lightweight, reducing computational complexity and enhancing its suitability for edge devices. Our simulations show that the DeepJSCC-ES system significantly outperforms the baseline DeepJSCC, particularly excelling in low SNR conditions. Further, the parameter size of the pruned model using DJSAP is compressed by 93.37%, while the average structural similarity index metric (SSIM) for the images only decreases by 0.92% compared with the baseline DeepJSCC.