Research Article
Fast Beam Switching Based on Machine Learning for MmWave Massive MIMO Systems
@INPROCEEDINGS{10.1007/978-3-030-90196-7_2, author={Kean Chen and Danpu Liu and Xingwen He}, title={Fast Beam Switching Based on Machine Learning for MmWave Massive MIMO Systems}, proceedings={Artificial Intelligence for Communications and Networks. Third EAI International Conference, AICON 2021, Xining, China, October 23--24, 2021, Proceedings, Part I}, proceedings_a={AICON}, year={2021}, month={11}, keywords={5G mobile communication Beam switching Machine learning Random forest}, doi={10.1007/978-3-030-90196-7_2} }- Kean Chen
Danpu Liu
Xingwen He
Year: 2021
Fast Beam Switching Based on Machine Learning for MmWave Massive MIMO Systems
AICON
Springer
DOI: 10.1007/978-3-030-90196-7_2
Abstract
Millimeter wave (mmWave) and massive multiple-input-multiple-output (MIMO) systems are two key technologies for 5G. Beamforming based on massive MIMO can produce high directional beams with array gain, and thus effectively compensate for the high path loss of mmWave. As the number of antennas increases, the beams become increasingly narrow, resulting in large overhead and high latency in the initial access and handover of the beams. For high-speed mobile scenarios, beam switching becomes more challenging since the traversal search among a large number of beams cannot be completed in a short period of time. To address this problem, this paper proposes a method based on machine learning to predict the optimal Base Station (BS) and the optimal beam pair at the successive instant for the User Equipment (UE) in motion. More specifically, a Random Forest (RF) classification model is trained to learn the channel’s features in a multi-cell scenario, and complete the nonlinear modeling of the propagation environment. Furthermore, this model is used to predict the future optimal BS and the optimal beam pair for a moving UE based on the present UE’s location, BS beam index and RSRP value. The simulation results show that the prediction accuracy is greater than 90% in most situations, thus the latency and the consumption of signaling resources for beam switch is reduced significantly while the loss in spectral efficiency is little.
