Adaptive Beam Management in 5G-NR: A Machine Learning Perspective

Abstract: In this rapid era of technology, speed of communication has become an important factor. To overcome this challenge, 5G/NR technology is being explored. One of the ways of doing this is by using the so called millimeter wave spectrum. In order to make usage of the mmWave spectrum, due to the high path loss it introduces, the beamforming technique is adopted. Beamforming is a method used in high frequency signal transmission for focusing or directing the beam towards the user in such a way that it will improve the signal strength and throughput and reduce the signal wastage, resulting in better experience. This can be achieved by constructively adding the phase shift from each antenna element towards the desired direction. Beam refinement or tracking, which are used in 5G/NR, are the mechanisms that selects the active beam used for transmission between the base station and User Equipment. Beam management is based on the continuous measurement reports from the UE on the reference signals related to the available beams. With higher frequencies and more narrow beams for coverage enhancements, beam management is becoming increasingly important. However, it comes with a collective cost due to the frequent measurements that are needed to monitor the positions of all connected UEs. The purpose of this Master’s thesis work is to investigate if machine learning methods could be applied to optimize the beam measurements and find the most efficient way to use these results to select the best beam for multiple UEs. This will allow the UEs and Base Station (BS) to have a strong signal for better transmission. We further compare the results obtained, with 3GPP baseline algorithm to gain insights about its performance.