A Deep Neural Network Codebook Approach for Near-Field Nulling Control Beam Focusing

This paper proposes a deep neural network (DNN) codebook approach for multi-user interference (MUI) mitigation in extremely large multiple-input multiple-output (XL-MIMO) systems operating in the near-field region. Unlike existing DNN-based nulling control beamforming (NCBF) methods that face scalability and complexity challenges, the proposed framework partitions the Fresnel region using correlation-based sampling and assigns a lightweight fully connected DNN model to each subsection. Each model is trained on beamforming weights generated using the linearly constrained minimum variance (LCMV) method, enabling accurate prediction of nulling control beam-focusing weights that simultaneously optimize the desired signal strength and suppress potential interference for both collinear and non-collinear user configurations. Simulation results show that the trained models achieve average phase and magnitude prediction errors of 0.085 radians and 0.52 dB, respectively, across 75 sample subsections. Full-wave simulations in Ansys HFSS further demonstrate that the proposed DNN codebook achieves interference suppression better than 31.64 dB, with a performance gap within 2 dB of the LCMV method, thereby validating its effectiveness in mitigating MUI while reducing computational complexity.