Encrypted Semantic Communication Using Adversarial Training for Privacy Preserving

作者：Xinlai Luo, Zhiyong Chen, Meixia Tao, Feng Yang · 发表于：IEEE Communications Letters · 年份：2023 · DOI：10.1109/lcomm.2023.3269768 · 被引用次数：90 · 研究领域：Wireless Signal Modulation Classification、Internet Traffic Analysis and Secure E-voting、Adversarial Robustness in Machine Learning

Semantic communication is implemented based on shared background knowledge, but the sharing mechanism risks privacy leakage. In this letter, we propose an encrypted semantic communication system (ESCS) for privacy preserving, which combines universality and confidentiality. The universality is reflected in the fact that the structures of all network modules of the proposed ESCS are public, and the training database is shared, which is suitable for large-scale deployment in practical scenarios. Meanwhile, confidentiality is achieved through key encryption. Based on the adversarial training, we design an adversarial encryption training scheme to guarantee the accuracy of semantic communication in both encrypted and unencrypted modes. Experiment results show that the proposed ESCS with the adversarial encryption training scheme can perform well regardless of whether the semantic information is encrypted. It is difficult for the attacker to reconstruct the original semantic information from the eavesdropped message.