The Massive Multiple-Input Multiple-Output (M-MIMO) is considered as one of the key technologies in 5G, and future 6G networks. From the perspective of, e.g., channel estimation, especially for high-speed users it is easier to implement an M-MIMO network exploiting a static set of beams, i.e., Grid of Beams (GoB). While considering GoB it is important to properly assign users to the beams, i.e., to perform Beam Management (BM). BM can be enhanced by taking into account historical knowledge about the radio environment, e.g., to avoid radio link failures. The aim of this paper is to propose such a BM algorithm, that utilizes location-dependent data stored in a Radio Environment Map (REM). It utilizes received power maps, and user mobility patterns to optimize the BM process in terms of Reinforcement Learning (RL) by using the Policy Iteration method under different goal functions, e.g., maximization of received power or minimization of beam reselections while avoiding radio link failures. The proposed solution is compliant with the Open Radio Access Network (O-RAN) architecture, enabling its practical implementation. Simulation studies have shown that the proposed BM algorithm can significantly reduce the number of beam reselections or radio link failures compared to the baseline algorithm.

翻译：大规模多输入多输出（M-MIMO）被视为5G及未来6G网络的关键技术之一。从信道估计的角度来看，特别是针对高速移动用户，利用静态波束集合（即波束网格，GoB）部署M-MIMO网络更容易实现。在采用GoB方案时，如何将用户正确分配给波束（即执行波束管理，BM）至关重要。通过引入无线电环境的历史知识（例如避免无线链路失效）可增强BM性能。本文旨在提出一种利用无线电环境地图（REM）中位置相关数据的BM算法。该算法通过策略迭代方法，结合接收功率地图与用户移动模式，在不同目标函数（如最大化接收功率、最小化波束重选次数及避免无线链路失效）下，基于强化学习（RL）优化BM过程。所提方案符合开放式无线接入网络（O-RAN）架构，具备实际部署可行性。仿真研究表明，与基线算法相比，所提出的BM算法能够显著减少波束重选次数或无线链路失效次数。