Deep Learning-Based Hybrid Precoder and Combiner Approach for MIMO-OFDM Systems

As one of the important technologies for the forthcoming 6G millimeter-wave massive multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) communication systems, hybrid precoding/combining (HPC) can realize the tradeoff between high spectral efficiency (SE) and computation efficiency. In this article, an innovative spectral-efficient end-to-end (E2E) HPC algorithm is proposed which jointly optimizes the pilot transmission, the channel state information (CSI) feedback, and HPC by three sub-networks. Specially, to achieve the accurate reconstruction of implicit channel, the pilot transmission sub-network (PTN) and the CSI feedback sub-network (CFN) are used to accurately and rapidly get phase information of the pilot and CSI feedback from the channel matrix, respectively. On this basis, the frequency division duplex HPC sub-network is developed to predict HPC matrices with the reconstructed channel matrix. Finally, via a newly defined loss function of SE, the presented approach jointly optimizes three sub-networks to achieve HPC rapidly and effectively. Simulations indicate the presented E2E HPC approach realizes better compromise in terms of SE and computation efficiency than other related approaches.