Robust and adaptive semantic noise for complex secure communication networks

Abstract

Privacy protection and robust transmission play critical roles in secure semantic communication, especially in open wireless environments with the presence of eavesdropping and for sensitive information such as self-management health information of patients with chronic diseases. Traditional physical layer technologies lack adaptability and efficiency in semantic-level threats due to sole attention to symbol protection. To address these challenges, this paper introduces a novel semantic security approach that integrates semantic noise into the transmitted semantic information to enhance the semantic security and improve the transmission robustness, even with the public semantic coding method and codebook. Specifically, we propose a vector quantized variational semantic security autoencoder (VQ-VSSAE), which can generate the semantic symbols by extracting essential features from original data and encoding them into a structured security codebook. By leveraging the semantic context and real-time channel state information (CSI), VQ-VSSAE dynamically produces the semantic noise to confuse the eavesdropper and mitigate the background interference through the noise-enabled VQ security design with semantic ambiguity, thus avoiding potential semantic leakage. Extensive simulations are conducted to evaluate the performance of the proposed system. The results demonstrate that VQ-VSSAE outperforms existing benchmark schemes in terms of semantic security and transmission robustness.