Semantic and technical noise modeling in semantic image communication over wireless channels

As wireless networks approach their Shannon capacity limit amid exponentially increasing data demands, semantic communication, which transmits the meaning of data rather than its exact form, has emerged as a compelling alternative. However, a key challenge lies in effectively modeling and mitigating both semantic and technical noise in such systems. We analyze a framework for semantic image transmission that independently and jointly models these two types of noise while employing Turbo codes for robust error detection and correction. We systematically examine their individual and combined impacts through seven experimental scenarios and benchmark the results against state-of-the-art image codecs. Our findings reveal that joint noise modeling not only improves semantic fidelity, but also enhances overall system robustness. Turbo coding proves effective in mitigating both noise types, enabling accurate meaning preservation even under adverse channel conditions. The simulation results demonstrate a reduction of approximately 33% in the channel bit rate when using joint modeling compared to alternative approaches, without compromising visual quality. Comparative evaluations against conventional codecs such as JPEG, JPEG2000, HEIF, and Turbo-coded image transmission systems highlight the superior performance of our semantic-aware strategy in noisy, bandwidth-constrained environments. This research underscores the value of integrating semantic intelligence with classical coding techniques, paving the way for next-generation wireless communication systems that are more efficient, resilient, and customized for emerging applications such as 6G, autonomous systems, and AI-enhanced networks.