Underwater Wireless Optical Communication Channel Characterization Using Machine Learning Techniques

Recently Underwater Optical Wireless Communication (UOWC) has attracted major attention due to its high transmission rate, low link delay, high communication security, and low implementation cost. However, optical signals suffer from severe attenuation loss due to absorption and scattering effects, which impedes the establishment of an effective and reliable UWOC system. Hence, it is important to identify the characteristic of the underwater channel in order to overcome the mentioned challenges. In literature, the combination of the Exponential and the Generalized Gamma Distribution (EGG) has been shown to model the underwater channel environment with great accuracy. EGG is a comprehensive channel model incorporating the effect of temperature-induced turbulence in the presence of air bubbles, in both fresh and salty aqueous environments. In this work, we built a Machine Learning (ML) based system that utilizes Convolutional Neural Network (CNN) to estimate the parameters of the EGG channel model from the received signal. Furthermore, we take one more step and train a separate deep network to predict bubble level and temperature gradient in the UWOC channel using the estimated parameters. The two networks together form a pipeline enabling us to estimate the channel state from the received signal. The results confirm well with the experimental data from the literature.