Machine Learning Based Classification and Modelling of Underwater Acoustic Communication

The acoustic medium in ocean has high complications due to its non-homogenous property. The speed of sound in the medium plays a significant role in acoustic computations and is more related to the density and compressibility of the propagation medium. Several acoustic propagation modelling methods that are described by wave equations are proposed for different underwater applications. The mathematical propagation models that are used widely are the empirical method (Thorp’s model), ray theory (Bellhop model), normal mode method (Kraken), wavenumber integration (Scooter), and parabolic equations (RAMGeo). The propagation models compute several parameters that include transmission loss, impulse response, arrival time, etc. with the input of the sound velocity profile and the transmission environment. The error rate of the propagation models varies with respect to the frequency, range of transmission and other parameters as well. In this paper, a classification dataset for shallow water propagation is generated with the threshold limits of range and frequency of each propagation model. Since, the limits of the propagation model are non-linear, machine learning based algorithms are proposed and validated with the data generated. Finally, an GUI is created that classifies the required propagation model and simulate the model with the inputs of range, frequency and sound velocity profiles.