Deep Joint Source-Channel Coding Based on Feedback-Driven Codebook Optimization

The rapid growth of wireless communication technologies has made semantic communication an increasingly important approach for efficient data transmission. In this paper, a novel deep joint source-channel coding (DeepJSCC) framework based on Feedback-Driven Codebook Optimization (FDCO) for wireless image transmission is proposed. The framework dynamically optimizes the codebook based on channel feedback to improve image reconstruction performance under varying channel conditions. Specifically, a FDCO network is introduced to adjust the balance between common and individual information in the codebook based on the input signal-to-noise ratio (SNR). In addition, the residual between the original and quantized images is encoded to obtain semantic details, which are transmitted to reduce semantic quantization loss. Experimental results demonstrate that the proposed framework improves image quality and compression efficiency, especially under low SNR, and validates FDCO's dynamic adjustment of the codebook's information balance, leading to enhanced performance.