Deep Learning-Based Beam Training for Extremely Large-Scale MIMO Communications

We propose a deep learning-based beam training method for extremely large-scale multiple-input multiple-output (XL-MIMO) systems. This method addresses the challenge of increased beam training overhead in near-field environments by activating specific subsets of the antenna array. Using beam gain measurements from an angular codebook, we train a deep learning model to select the optimal codeword from a hybrid codebook that incorporates near-field and far-field scenarios. Simulation results demonstrate that our approach improves normalized beam gain and data rate while significantly reducing the beam training overhead.