Automatic Detection and Localization of an Unknown Number of Acoustic Sources Using a Network of Unsynchronized Hydrophones in a Dispersive Waveguide

Le résumé fourni par la source

Low-frequency impulsive signals in shallow-water environments can be decomposed into dispersive modal components with time–frequency positions that are strongly range-dependent, making the signals suitable for range-based localization. Provided a collection of signals from multiple unsynchronized sensors and corresponding source–receiver range estimates, it is either laborious or impractical to manually determine the number of unique sources and which range measurements originate from each. In this article, we propose a flexible method to automatically perform data association, localization, and reject outlier measurements. The approach here considers unique combinations of range measurements from collections of$k$sensors. For every range measurement combination, if the residual between the location estimate generated using each subcombination of$k-1$measurements and the remaining measurement is less than a chosen threshold, the whole collection is labeled as group-$k$consistent. All such groups are represented as neighboring nodes in a$k$-uniform hypergraph, and dense subgraphs obtained via clustering are used to calculate the final source location estimates. The range measurements are made by a temporal convolutional network (TCN) that processes spectrograms from individual sensors in a network of unsynchronized hydrophones to simultaneously detect dispersive signals and estimates source–receiver ranges. The TCN is trained on signals simulated using normal mode theory over different source and environment configurations. Testing the full automatic pipeline on simulated and experimental data based on the 2022 Seabed Characterization Experiment, the localization results are comparable to when data association and outlier measurement identification are done manually, and multiple present sources are consistently identified.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.