Model-Driven Deep Learning Assisted Beam Tracking for Millimeter-Wave Systems

Millimeter-wave (mmWave) systems employ directional beams to provide high-speed data communication, which demands continuous tracking of the beam direction. Currently, the technology based on Kalman filter and/or channel sparsity is considered for beam tracking, and its solutions are based on specific assumptions, which may not apply to the actual scene. In this letter, a model-driven deep learning (MD-DL) network is proposed, which combines the traditional signal processing method with the convolutional neural network for beam tracking. Specifically, the traditional signal processing is designed to drive the network for enhancing the process of feature extraction. In contrast to previous work based on deep learning, due to model-driven, fewer pilot sequences are required to track the arrival angle of the main path. The simulation results demonstrate that the proposed MD-DL network can improve the beam tracking accuracy and reduce the probability of losing tracking when vehicles travel at 72 km/h in complex urban scenarios.