Sound speed profile inversion based on distributed networked underwater sensors system and graph attention networks.

Variations in the sound speed profile (SSP) cause sound ray bending, which affects the accuracy of underwater communication and localization. Ocean acoustic tomography (OAT) is a convenient method for estimating the SSP. However, traditional SSP inversion methods based on OAT are often limited by the use of linear arrays and challenging to expand to distributed networks. To address this problem, a SSP inversion scheme within distributed networked underwater sensors (DNUS) systems is proposed. The scheme combines multimodal inputs, such as the time difference of arrival, the angle of arrival, node positions, and boundary sound speed, into a sensing matrix as input features. The graph attention network model is used to establish the mapping relationship between these features and the SSP, enabling SSP inversion under DNUS. Through numerical simulations and a shallow-water validation experiment, this study validates the effectiveness and accuracy of the proposed inversion method.