In shallow-water environments, the sound generated by low-frequency sources interacts significantly with the sea surface and seabed, inducing a strong dependence on source range in the received pressure field due to modal dispersion. As a result, these signals are ideal for range-based localization. However, in complex environments that are range dependent or that exhibit stratified water column and/or seabed properties, inverting the received signal for range in real time is nontrivial. Here, we develop a deep-learning approach to detect and range received signals that accounts for the distributed nature of the deployed sensor network and the stratified nature of the environment which varies with spatial position. The model is trained using simulated data generated based on the parameter distributions observed in the target environment. The detection and ranging method is validated using both simulated and experimental marine data. Work supported by NDSEG and ONR.
Goldwater et al. (Tue,) studied this question.