DFT-Based Near-Field Beam Alignment: Model-Based and Data-Driven Hybrid Approach

Accurate beam alignment is a critical challenge in extremely large-scale multiple-input multiple-output (XL-MIMO) systems, especially in the near-field regime, where conventional far-field assumptions no longer hold. While 2D polar-domain codebooks can capture near-field effects, they incur high time and computational complexity. To address this, we propose a novel line-of-sight (LoS) near-field beam alignment scheme that reuses the conventional discrete Fourier transform (DFT) codebook, commonly employed in far-field systems, ensuring backward compatibility without modifying the beam set. By introducing a new method to analyze the energy spread effect, we define the concept of an ϵ-approximated signal subspace, spanned by DFT vectors that exhibit significant correlation with the near-field channel vector. Building on this analysis, the proposed hybrid scheme integrates model-based principles with data-driven techniques, using DFT properties for efficient coarse alignment and a deep neural network (DNN)-aided fine alignment in a reduced search space. Extensive evaluations demonstrate that the proposed scheme attains higher alignment accuracy than existing methods, while providing substantial reductions in both computational load and neural network model size. These advantages make it well suited for practical large-scale near-field deployments.