Hybrid Driven Learning Aided Beam Tracking in Air-to-Ground MIMO-OFDM Communications

A novel beam tracking approach is proposed to realize reliable air-to-ground (A2G) transmissions with reduced pilot overhead and time delay. The proposed beam tracking strategy consists of two stages, namely the model-driven channel tracking and the model-data dual driven hybrid beamforming (HBF). For the model-driven channel tracking, the angle-of-arrivals/angle-of-departures (AoAs/AoDs) are predicted by leveraging the regularity of the three-dimensional flight track and attitude, as well as the A2G geometrical information with temporal correlations. Then, the high-dimensional channel matrix estimation problem is converted to the low-dimensional multipath components parameters estimation tasks, which substantially reduces the pilot overhead. The proposed model-data dual-driven HBF module unfolds the iterative HBF algorithms and introduces a set of trainable parameters, which brings in both low complexity and high interpretability. To further improve the HBF robustness against imperfect channel state information, the denoise neural network is employed to exploit spatial-domain channel correlations for improved channel accuracy. Numerical results unveil that: 1) the proposed model-driven channel tracking scheme achieves satisfying normalized mean square error of the tracked A2G channel with significantly reduced pilot overhead; and 2) the proposed model-data dual-driven HBF algorithm is superior to the conventional counterparts in terms of reliability and robustness.