Intelligent near-field physical layer security: machine learning-driven approaches

With the emerging of extremely large-scale antenna array (ELAA) and high-frequency carrier, the wireless communications will mainly occur in the radiating near-field region. The unique spherical wavefront and distance-angle coupling characteristics of near-field communications provide new opportunities for physical layer security (PLS), which can significantly enhance the channel difference between legitimate users and eavesdroppers. Nevertheless, issues such as the prohibitive complexity, beam squint effect, and dynamic environment have seriously constrained the deployment of near-field PLS. To overcome these challenges, this article explores the machine learning (ML) based solutions tailored for the near-field PLS. Concretely, we begin with an overview of PLS and discuss its unique characteristics in near-field communications. Then, we outline the ML techniques and discuss the key roles of ML in the near-field PLS with the emphasis on the Mamba model. On this basis, we investigate a wideband near-field system. A penalty-based algorithm is first designed as the label generation algorithm and optimal baseline. Then, we propose the Mamba-based supervised learning to safeguard the secure transmission due to the high complexity. Through the case study, the results demonstrate that the proposed scheme can maintain over 95% of the optimal secrecy rate while guaranteeing the user quality of service and sensing accuracy. Meanwhile, the proposed scheme dramatically reduces the runtime, presenting its potential for the near-field PLS. Finally, some challenges and opportunities are pointed for future research in this direction.