Deep Reinforcement Learning for Joint Beamwidth and Power Optimization in mmWave Systems

This letter studies the joint beamwidth and transmit power optimization problem in millimeter wave communication systems. A deep reinforcement learning based approach is proposed. Specifically, a customized deep Q network is trained offline, which is able to make real-time decisions when deployed online. Simulation results show that the proposed approach significantly outperforms conventional approaches in terms of both performance and complexity. Besides, strong generalization ability to different system parameters is also demonstrated, which further enhances the practicality of the proposed approach.