Deep Joint Source and Channel Coding
Haotian Wu, Chenghong Bian, Shao Yulin, Denız Gündüz · 2025
This chapter delves into the transformative potential of deep joint source-channel coding (DeepJSCC) in the realm of semantic communications, challenging the conventional wisdom established by Shannon's separation theorem. DeepJSCC, emerging as a pioneering approach that integrates source compression and error correction into a unified process through deep learning, stands out by allowing end-to-end optimization of the communication system and promises significant improvements in bandwidth efficiency and reliability over traditional coding methods. Exploring the application of DeepJSCC across three canonical channel models – multiple-input and multiple-output (MIMO), relay, and feedback channels – this chapter aims to showcase DeepJSCC's versatility, adaptability, efficiency, and resilience.