Reinforcement Learning-Based Link Switching in Underwater Optical Wireless Communication Systems Using Intelligent Reflecting Surfaces

Intelligent reflecting surfaces (IRS) are emerging as a promising technology for establishing adaptable links and addressing the limitations of direct line-of-sight (LOS) in underwater optical wireless communication (UOWC) systems. In challenging underwater environments where direct LOS is often disrupted, the IRS can provide alternative paths for stable and reliable communication. In this paper, we consider an IRS-assisted UOWC system with an autonomous surface vehicle (ASV) and autonomous underwater vehicles (AUVs) operating in realistic, dynamic conditions. To enable adaptive link switching between LOS and IRS-based paths, we apply a reinforcement learning approach using a deep $\mathbf{Q}$ network (DQN). We conduct simulations over gamma-gamma channels, taking into account key degrading factors, including underwater attenuation, pointing errors, turbulence, and obstruction. System performance is evaluated in terms of both average reward and cumulative reward. Based on the obtained results, we demonstrate that the proposed method effectively maintains transmission reliability in the presence of obstacles, improving performance across various scenarios, particularly under obstruction, low attenuation, and minimal pointing error conditions.