Q-Learning Based Two-Timescale Power Allocation for Multi-Homing Hybrid RF/VLC Networks

This letter investigates hybrid networks composed of a radio frequency (RF) access point (AP) and multiple visible light communication (VLC) APs. We consider mobile multi-homing users that can aggregate resources from both RF and VLC APs. In hybrid RF/VLC networks, RF channel gains vary faster than VLC channels due to small scale fading. By leveraging multi-agent Q-learning to interact with the dynamics of wireless environments, we develop an online two-timescale power allocation strategy that optimizes the transmit powers at the RF and VLC APs to ensure quality-of-service satisfaction. Simulation results demonstrate the effectiveness of the proposed Q-learning based strategy.