Q-learning based adaptive beam selection in Massive MIMO system

• 1、 School of Information and Communication Engineerging, Beijing University of Posts and Telecommunications, Beijing, 100876

Abstract：Massive MIMO is one of the key technologies of 5G and 6G in the future. Massive MIMO technology adjusts the phase and amplitude of the antenna to form a strong directional beam aimed at the user, so as to achieve the purpose of expanding signal coverage and increasing the system capacity. However, in highly dynamic scenarios, to maintain a high beam gain, new beams need to be selected frequently, which leads to large delays and thus reduces user performance. To solve this problem, reinforcement learning is introduced in this paper, and a beam selection algorithm based on Q-learning is proposed. The algorithm uses the information reported by the user to make beam selection decisions with beam coherence time as the interval, in order to balance the impact of beam quality and the delay caused by beam switching on the user performance, thus maximizing the long-term average throughput of the user in mobility. In addition, from the perspective of 3GPP protocol, this paper designs the corresponding beam selection process for the proposed algorithm. Simulation shows that the beam selection algorithm presented in this paper has advantages in reducing beam switching frequency, improving service stability and long-term average throughput of users.

Keywords： Information and communication Massive MIMO beam selection Q-learning Long-term average throughput