Adaptive Beam Steering and Divergence Control for Underwater Optical Wireless Communication Using Reinforcement Learning

Underwater optical wireless communication (UOWC) is a promising technology enabling high-speed, low-latency communication for beyond 5G/6G systems. However, UOWC faces significant challenges due to the complex nature of the underwater channel, including absorption, scattering, turbulence, and dynamic sea wave conditions, which complicate static analysis. To address these challenges, we propose a neural network-based beam adaptation technique for UOWC systems, combining deep Q-networks (DQN) and long short-term memory (LSTM) models. These models dynamically optimize beam divergence and steering angles based on the properties of sea waves. Our approach offers a robust solution to maintaining communication quality in diverse and challenging underwater environments. In this work, the turbulence is modeled using the exponential generalized gamma (EGG) distribution, which provides an excellent fit for various types of turbulence. Simulation results show that the proposed LSTM-DQN-based approach consistently outperforms fixed-beam, DQN-only, and heuristic methods in a range of underwater environments. The system successfully compensates for random vessel movements and turbulence-induced intensity fluctuations, ensuring reliable communication. These results highlight the effectiveness of the LSTM-DQN-based method in optimizing beam alignment under various water conditions. Furthermore, a comparison with other machine learning (ML) methods revealed that similar performance can be achieved with those techniques. However, the proposed method demonstrated superior stability. By accounting for variations in vessel size and the movement of the transmitter, we have shown that the proposed method is effective under different environmental conditions.