Multi-Agent Reinforcement Learning with Common Policy for Antenna Tilt Optimization

This paper proposes a method for wireless network optimization applicable to tuning cell parameters that impact the performance of the adjusted cell and the surrounding neighboring cells. The method relies on multiple reinforcement learning agents that share a common policy and include information from neighboring cells in the state and reward. In order not to impair network performance during the first steps of learning, agents are pre-trained during an earlier phase of offline learning, in which an initial policy is obtained using feedback from a static network simulator and considering a wide variety of scenarios. Finally, agents can wisely tune the cell parameters of a test network by suggesting small incremental changes to slowly steer the network toward an optimal configuration. Agents propose optimal changes using the experience gained with the simulator in the pre-training phase, but also continue to learn from current network readings after each change. The results show how the proposed approach significantly improves the performance gains already provided by expert system-based methods when applied to remote antenna tilt optimization. Additional gains are also seen when comparing the proposed approach with a similar method in which the state and reward do not include information from neighboring cells.