Hybrid Beamforming in MIMO-OFDM Systems with Model-Driven Deep Learning

Due to the high-mobility of aeronautical communications, real-time hybrid beamforming (HBF) is indispensable. In this paper, a novel HBF scheme in the aeronautical multiple-input multiple-output orthogonal frequency division multiplexing systems is proposed with model-driven deep learning. In particular, we formulate a mean square error minimization problem with manifold constraints. As the conventional iterative optimization algorithm has high computational complexity, we propose a deep-unfolding HBF network which unfolds the iterations and introduces a set of trainable parameters. The proposed algorithm produces hybrid beamformer with several layers, which avoids cumbersome iteration procedures. Moreover, the deep-unfolding algorithm preserves the operation of the iterative optimizer, which brings in high interpretability. Numerical results show that the proposed HBF algorithm is superior to the conventional model-based counterparts in terms of reliability.