Rate-Adaptive Joint Source Channel Coding Using Deep Block-Based Compressed Sensing

This paper introduces a novel Rate-Adaptive Compressed Sensing-based Joint Source-Channel Coding scheme, termed RACS-JSCC, which leverages deep block-based CS to dynamically adjust the encoding rate based on available channel bandwidth and input image statistics. RACS-JSCC selects the encoding rate using both local and global statistics of the input image, alongside channel state information, prior to the feature extraction stage. This approach eliminates the transmission of redundant features and ensures the input image is encoded at an optimal rate. By training a deep learning-based JSCC encoder-decoder pair to operate across multiple rates and channel conditions, the proposed method eliminates the necessity for multiple models and enhances practical applicability in diverse communication environments. Our experimental results demonstrate that RACS-JSCC achieves superior performance in terms of image quality and robustness against varying channel conditions, making it a highly efficient solution for real-world wireless image transmission.