Learned Compression and Restoration of Beam Measurements for Coordinated Hybrid Beamforming in Multi-Cell mmWave Systems

The accurate acquisition of downlink channel state information (CSI) is essential for designing coordinated hybrid beamforming (HYB) in multi-cell (MC) millimeter-wave massive multiple-input multiple-output systems. In this paper, we present a novel beam management (BM)-based MC-HYB design framework with limited feedback. Different from the traditional BM procedures, our approach feedbacks all the beam measurements to the base station via beam reporting to facilitate HYB design without the estimation of explicit CSI. Furthermore, to handle the noisy beam measurements feedback problem with reduced overhead, we propose a swin transformer-based beam compression and restoration method featuring its ability to extract the relationships among different beams. Finally, we develop a low-complexity coordinated HYB algorithm in the proposed framework. The method involves analog domain beam selection and the design of two-layer digital precoders for inter-cell and intra-cell interference cancellation, respectively. Extensive simulations demonstrate that the proposed beam restoration method efficiently removes measurement noise and reduces feedback overhead and the proposed coordinated HYB scheme reveals a 90% achievable sum-rate of the ideal scheme, highlighting the effectiveness in interference elimination.