Deep Learning-Enhanced Channel Prediction for XL-MIMO Systems

With the dramatically increasing number of antennas, wireless channel becomes sophisticated and the pilot overhead becomes intolerable. In this paper, we construct the last-bounce cluster (LBC) based channel model to capture the spatial-temporal characteristics in extra-large multiple-input-multiple-output (XL-MIMO) systems. Then, we formulate the channel prediction problem to conserve the pilots. Leveraging the advantages of deep learning (DL), we propose convolutional neural network (CNN)-Transformer based channel prediction method (CTCP) to enhance the spatial-temporal information extraction of the channel. Simulation results validate that CTCP effectively extracts the spatial-temporal correlations of the channel, and enhances the trade-off between accuracy and comlexity in channel prediction for XL-MIMO systems.