Digital Image Semantic Communications with Joint Source-Channel Coding and Constellation Optimization

Linzi Shen, Nan Ma, Jianqiao Chen, Mengshu Song, Chen Dong

Abstract: Recently developed semantic communication facilitates an intelligent and minimalist approach to meet the growing demands for future sixth-generation (6G) communication sys-tems. However, since the constraints of backpropagation during neural network training, the predominant approach of most extant studies are based on differentiable channel inputs, which is difficult to match with existing digital communication systems effectively. Meanwhile, the research on modulation techniques considering constellation optimization for digital semantic communication is still in the early stage. To deal with these issues, we propose a novel deep joint source-channel coding (DJSCC)-based digital semantic communication with considering constellation optimization, which is referred to as DJSCC-C. Firstly, the digital semantic communication system is introduced. Secondly, the DJSCC-C system is proposed for image transmission, which is trained with a newly-constructed loss function, aiming to improve the transmission rate in the digital semantic communication system as well as optimize the joint source-channel coding capability. Finally, simulation results are provided to verify the superiority of the proposed scheme for future design of digital semantic communication systems.