Dynamic Link Selection Algorithm for RIS Enhanced VLC under Power Consumption Metric

In visible light communications (VLC), reconfigurable intelligent surfaces (RIS) are adopted to improve the system bit error rates (BER) performance in case of direct line-of-sight (LoS) link blockage. Traditional RIS-VLC design parallelly enables both LoS and RIS links, leading to higher power consumption if LoS link is blocked by obstacles. A critical issue in RIS-VLC systems is whether the LoS link should be always enabled for good system BER performance, even if the blockage is occasional. In this paper, we propose a dynamic link selection algorithm for RIS-VLC, named DLS-VLC, by making full use of the evolvement of link blockage probability. Specifically, DLS-VLC defines two transmission modes, M-0 where both RIS and the LoS links are enabled, and M-1 where only RIS link is enabled, respectively. We formulate the dynamic link selection problem as a Markov decision process (MDP) and study three representative link selection policies. As a benchmark, the first one is the optimal policy that can be derived from dynamic programming, but it has high computational complexity. The second one is the greedy policy, which is easy to implement but sub-optimal because of considering myopic performance only. We have proven that both the optimal policy and the greedy policy have a dual-threshold structure. To maintain good performance while keeping low computational costs, we propose a deep reinforcement learning (DRL) policy, where the actions are generated using a pre-trained deep neural network. Simulation results show that all three policies can outperform traditional RIS-VLC. Specifically, our proposed DLR policy can achieve similar performance of optimal policy, and can reduce the power consumption by 22% compared with conventional RIS-VLC. Our findings indicate that dynamic link selections in RIS enhanced VLC systems can greatly reduce the system power conservation, while maintaining the same BER performance.