Experimental Analysis of Air-Bubble-Induced Turbulent Underwater Optical Channels with Evaluation of a CNN-LSTM-Based Channel Classifier

Underwater Visible Light Communication (UVLC) is not only affected by intrinsic absorption and scattering, it is additionally challenging owing to the bubble-induced Underwater Optical Turbulence (UWOT) observed in the Underwater Optical Channels (UOC). Unlike Free Space Optical (FSO) channels, the fading process in UOCs exhibit time varying non Gaussian, non linear statistical properties. This makes it essential to experimentally determine the statistical and time domain fading characteristics of the channel and to find methods to perform efficient adaptive communication on the basis of the instantaneous channel properties. Numerous works have used simulation-based methods in UVLC channel modeling and Channel Classification (CC) associated with the fading models. However, relatively few works have dealt with the problem in an experimental context. Motivated by this gap, our study involves the practical establishment and analysis of a 530 nm wavelength LED, Intensity Modulation-based Direct Detection (IM-DD) optical link induced with various air bubble based attenuation types amounting to a total of 15 classes of attenuation. We analyze both the statistical characteristics (PDFs and scintillation metrics) and the timedomain behavior parameters (outage time and coherence time) in the experimentally observed channels. The data captured has been utilized to train a CNN-LSTM neural network based channel classifier to classify the type of bubble attenuation and correspondingly predict the channel fading characteristics. The efficacy of the Neural Network has been analyzed against the assumption of low pilot symbol (5 percent) communication.