DRL-Based Physical-Layer Security Optimization in Near-Field MIMO Systems

The advent of extremely large antenna arrays (ELAAs) is crucial for meeting the performance demands of future sixth-generation (6G) wireless networks. However, ELAA introduces significant near-field communication (NFC) effects, characterized by spherical wavefront propagation, in contrast to the conventional planar waves observed in far-field models (FFMs). As NFC facilitates precise beamfocusing and spatial multiplexing, it inherently increases the risk of eavesdropping, making physical-layer security (PLS) a critical challenge for safeguarding confidential communication. This article investigates a multiple-input-multiple-output (MIMO) system in the near-field regime, utilizing NFC properties to enhance secrecy performance. Unlike FFMs that rely on the angular domain, our approach leverages both angular and distance domains to achieve robust PLS, even when an eavesdropper shares the same angular direction as a legitimate user. We propose a deep reinforcement learning (DRL)-based solution to optimize beamforming, power allocation, and antenna selection. By minimizing antenna use while maximizing secrecy rates, the approach avoids resource wastage and ensures superior security. Numerical simulations demonstrate significant secrecy rate improvements.