Spatio-Temporal Graph Convolutional Network-Based Beam Training and Tracking for XL-MIMO Systems

In highly dynamic environments, the performance of extremely large-scale multiple-input multiple-output (XL-MIMO) is obviously degraded due to rapid and frequent beam switching. To address this challenge, this letter proposes a spatio-temporal graph convolutional network (STGCN) based beam training and tracking scheme. Initially, the spatio-temporal graphs are constructed to build the spatial relationships between base station (BS) and user equipments (UEs). And then, graph convolutional network (GCN) learns and updates the spatial relationship of all nodes, leveraging the adjacency matrix to enhance the robustness of beam selection in dynamic environments. Finally, gated recurrent unit (GRU) outputs the best beam at the subsequent times by learning the changes of the UEs’ motion trajectory. Simulation results demonstrate that the proposed method achieves superior accuracy and robustness compared to conventional beam training approaches.