A Neural Network Approach to Estimate Modal Propagation in Coastal Oceanic Waveguides

Understanding underwater acoustic propagation is crucial for applications like target detection, environmental monitoring, and underwater communication. In shallow coastal waters (D < 500 m) and low frequencies (f < 1000 Hz), sound propagation is modeled using normal mode theory. However, repeated simulations using normal mode codes can be computationally expensive. To address this, two conditionned neural networks (NNs) are used to approximate normal mode propagation, significantly reducing computation time once trained. These NNs predict modal parameters, such as the modal wavenumber k<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">rm</inf> and the modal depth function Ψ<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</inf> which represent the physics of the waveguide. Their predictions are tested on unseen environments and used to reconstruct transmission loss. This method, is applied to Pekeris waveguides with different physical property and operates on frequencies between 50 and 300 Hz