Beam Management in 5G NR using Geolocation Side Information

Abstract

A machine learning solution leveraging geolocation side information is proposed for enhancing beam management in 5G NR millimeter wave (mmWave) wireless systems. An important building block of our solution are the support vector machines (SVMs), which are used to model the mapping between the user equipments’ (UEs) geolocations and their serving beams/cells in a multiuser, multi-cell environment. Building upon the SVM models mentioned above, we introduce a multiuser scheduling algorithm that uses local beam assignment information from the cells adjacent to the users to reduce the amount of required real time channel state information (CSI) feedback. Simulations carried out using a realistic antenna array radiation pattern, as well as, data from a ray tracing channel model in a dense urban mmWave deployment show that the proposed multiuser scheduler has remarkably good performance, while its algorithmic complexity is kept low. We further quantify the improvements that our SVM-based beam management methods enable by comparison against the conventional exhaustive beam sweeping approach typically employed by 5G NR mmWave implementations. In this case, we show that our proposal enables the network to achieve a 50% reduction in initial access latency at a fixed signaling overhead, or 34% reduction of signaling overhead at a fixed latency requirement.