Millimetre-Wave Beam Tracking: An Intelligent Machine Learning and Kalman Filter Fusion Technology

Millimeter-wave (mmWave) based wireless communication with beamforming is a promising technology to meet the ever-increasing demand for communication bandwidth. However, due to the narrow beam, beam tracking is a challenge since consistent accuracy tracking is required. To solve the above challenges, we propose an intelligent machine learning (ML) and Kalman filter (KF) fusion technology in this work. The proposed ML-KF fusion method provides consistent and robust high tracking accuracy with low tracking overhead. By introducing the variational Bayesian inference, the acquisition of the state transition function in the KF algorithm is converted to data-driven neural network training. Therefore, the fusion-based KF beam tracking learns the state transition matrix, which matches the practical scenario. Furthermore, a long-short-term memory (LSTM) is employed to predict the beam angle, which functions as a generative model to supply more data for the NN training. The proposed ML-KF fusion scheme improves the prediction accuracy while inheriting the KF robustness. According to the simulated results on the realistic campus scenarios by ray-tracing software, the proposed algorithm outperforms the existing beam tracking methods in bit error rate (BER) and normalized mean square error (NMSE) of angle prediction with less overhead.