An End-to-End and Adversarial Learning-Based Communication System for Anti-Interference and Anti-Eavesdropping
Tao Liu, Yuanxiang Chen, Jiawei Xin, Chunhui Du, Cong Hu, Shuo Wang · IEEE Transactions on Cognitive Communications and Networking · 2025
Due to the broadcast characteristics of wireless channels, wireless communication systems not only encounter interference from other legitimate users but also face the threat of eavesdropping from illegal users. This paper proposes a multi-user symmetric encryption end-to-end (E2E) communication system with dual capabilities of anti-interference and anti-eavesdropping. During the training phase, we assign a hypothetical eavesdropper to each pair of legitimate users and employ adversarial training to enable multi-user encryption. We also design a dynamically weighted iterative loss function to balance user performance. Furthermore, we propose a sub-batch power control training scheme to mitigate the significant reliability degradation caused by encryption effectively. Experimental results demonstrate that the proposed system remains robust against changes in interference conditions, ensuring secure communication with minimal performance degradation for all user pairs. Specifically, with two user pairs, a modulation order of 16, and a key length of 6, achieving a symbol error rate (SER) of 10-4incurs only a 0.71 dB signal-to-noise ratio (SNR) penalty. Moreover, system reliability gradually approaches pre-encryption levels as the key length increases. In addition, the eavesdropper's SER is as high as 93.75% when the key is unknown, equivalent to randomly guessing 16 symbols.