Adaptive DRL for IRS Mirror Orientation in Dynamic OWC Networks

TL;DR

This paper introduces a deep reinforcement learning approach to dynamically optimize mirror orientations in IRS-assisted optical wireless networks, significantly improving data rates in mobile, blockage-prone indoor VLC environments.

Contribution

It develops a DRL-based method for real-time IRS mirror orientation control in dynamic VLC networks, outperforming traditional algorithms.

Findings

DRL outperforms deep Q-learning in mirror orientation tasks.

Significant sum rate improvements over random IRS configurations.

Effective adaptation to user mobility and blockages.

Abstract

Intelligent reflecting surfaces (IRSs) have emerged as a promising solution to mitigate line-of-sight (LoS) blockages and enhance signal coverage in optical wireless communication (OWC) systems with minimal additional power. In this work, we consider a mirror-based IRS to assist a dynamic indoor visible light communication (VLC) environment. We formulate an optimization problem that aims to maximize the sum rate by adjusting the orientation of the IRS mirrors. To enable real-time adaptability, the problem is modelled as a Markov decision process (MDP), and a deep reinforcement learning (DRL) algorithm is developed based on the deterministic policy gradient for real-time mirror-based IRS optimization in dynamic VLC networks. The proposed DRL is employed to optimize mirror orientation toward mobile users under blockage and mobility constraints. Simulation results demonstrate that our proposed DRL algorithm outperforms the conventional deep Q- learning (DQL) algorithm and achieves substantial improvements in sum rate compared to random-orientation IRS configurations