Low-Overhead Near-Field Beam Training Based on Bayesian Regression

In extremely large-scale multiple input multiple output (XL-MIMO) systems, near-field beam training is an essential way to acquire channel state information. To reduce the high training overhead brought by the additional distance dimension of the near-field codebook, some overhead-reduced near-field beam training schemes were proposed in the literature. However, existing schemes ignore the correlation between different near-field beams. In this paper, we propose a Bayesian regression-based near-field beam training scheme, which fully utilizes the correlation between near-field code-words to reduce the training overhead. Specifically, inspired by Bayesian regression, we model the received signal corresponding to different near-field codewords as a Gaussian process and determine the optimal codeword by iteratively updating the posterior distribution and designing the codeword searching order. Besides, different searching strategies are analysed and compared. The proposed scheme only requires searching for a few codewords instead of the entire codebook, which reduces the high training overhead. Simulation results verify the effectiveness of the proposed Bayesian regression-based near-field beam training scheme, which significantly reduces the training overhead while maintaining the high achievable rate performance.