Deep Q-Network-Based Near-Field Beam Tracking Under Hardware Impairments

초록

This letter proposes a deep Q-network (DQN)-based near-field (NF) beam tracking method under hardware impairments (HWIs). In extremely large aperture array deployments, conventional NF tracking suffers from hardware-induced distortions and model mismatch. The proposed method sequentially updates the position estimate directly from received-signal measurements, without explicit parametric HWI compensation. The DQN agent learns, over a discrete angle-of-arrival–distance action space, a policy that maximizes the received-signal power under intra-subarray mutual coupling and residual local-oscillator-induced phase distortions. Simulation results against model-based and learning-based baselines confirm reliable tracking and low online inference latency under HWIs.

키워드