Beam Weights Optimization Based on Reinforcement Learning for 6G

With the continuous evolution of wireless radio access network architecture, artificial intelligence (AI) plays an increasingly significant role in network optimization. Native-AI will realize fast and accurate optimization, real-time policy adjustment and trend prediction in future 6G systems. To cope with the increasing user density and complex wireless environment, beam weights optimization becomes one of the key factors to promote the communication capacity, coverage and anti-interference ability. In this paper, we build an intelligent network model based on the reinforcement learning (RL) algorithm and optimize the beam pattern in a real typical scenario that contains both roads and buildings. A field test is conducted to investigate the potential impacts of the intelligent algorithm on network optimization. The data used in the field test are real measurement reports from UEs, which imply the location information and the uneven distribution of UEs. The coverage performance and received signal quality are significantly improved after optimization. We also analyze the limitations of AI in the current 5G systems and discuss the future work on native-AI and RL.