Near-Field Beam Training for Cell-Free Networks Leveraging Distributed Angle Observations

Near-field communications significantly increase spatial degrees of freedom by exploiting both angular and distance information. However, near-field beam training remains challenging due to the high-dimensional joint angle–distance search and inherent distance ambiguity. To address these issues, we propose an iterative near-field beam training algorithm leveraging cell-free network architectures. Specifically, in our proposed scheme, each access point (AP) in the cell-free network first utilizes a far-field codebook to obtain a coarse estimate of the user’s angle of arrival. The central processing unit then applies a least-squares based localization method to fuse these multi-AP angular estimates and derive the user’s position. Leveraging this position information, the APs iteratively update their position-adaptive near-field codebooks and refine the angular estimation. The process is repeated until convergence, which enables stable and accurate beam training without explicit distance-domain search. Simulation results demonstrate that the proposed cell-free near-field beam training design achieves higher positioning accuracy than existing state-of-the-art schemes.