DRL-Based Beam Split Alleviation for Movable Antenna-Enabled Near-Field Wideband Communications

Near-field communication is regarded as a key enabling technology for future 6G wireless systems. However, when operating over wide bandwidths, the beam split effect arising from frequency-independent analog phase shifters leads to significant beamforming gain degradation. Different from existing works that address this issue through true-time-delay hardware, this paper exploits the emerging movable antenna technology for beam split alleviation. Specifically, we consider a movable antenna-enabled near-field wideband uplink system with an analog beamforming architecture. Under this setup, we jointly optimize the analog phase shifts and antenna positions to maximize the minimum beamforming gain across all subcarriers. The formulated problem is highly non-convex due to the constant-modulus constraint on the analog combiner and the nonlinear dependence of the near-field channel on antenna positions, which makes conventional optimization methods difficult to apply. To this end, we develop a deep reinforcement learning framework based on the soft actor–critic algorithm that operates in a continuous action space and effectively handles the non-smooth max-min objective. Simulation results show that the proposed approach alleviates the beam split effect and achieves a higher minimum beamforming gain than conventional schemes.