On Large Language Model Based Joint Source Channel Coding for Semantic Communication

Deep learning is extensively proposed in Semantic Communication (SemCom) for semantic extraction and optimization, but often requires complex, application-specific neural networks and struggles with generalization to new environments. Building on the success of large language models (LLMs), this paper explores the adaptability of LLMs to a more efficient and generalizable SemCom design. We introduce SemComLLM and demonstrate its viability by integrating the Llama-2 model with Joint Source-Channel Coding (JSCC), enhancing transmission efficiency and semantic accuracy. The empirical results show an improved contextual understanding. We discuss key challenges and develop future research directions for further refinement.