Power Allocation based on Q-Learning for NOMA Visible Light Communication Networks

Non-orthogonal multiple access (NOMA) has been proposed to enhance system capacity for visible light communication (VLC) systems. However, the effective power allocation strategy is one of critical problems that needs to be solved in NOMA. In this paper, a new method for multi-user downlink power allocation in VLC NOMA based on reinforcement learning is proposed. This method utilizes distributed multi-agent Q-learning algorithm with low complexity to maximize sum throughput of the multiuser VLC downlink system which is subject to both user fairness and quality of service (QoS). The numerical results show that a large sum logarithmic user rate can be obtained with higher probability compared with other conventional power allocation algorithms.