Inversion for sound speed profile in shallow water based on long short-term memory networks and ray theory

To address the problem of underwater sound speed profile (SSP) inversion in underwater acoustic multipath channels, this paper combines deep learning and ray theory to propose an inversion method using a long short-term memory network (LSTM). Based on the equidistant characteristics of the horizontal line array, the proposed method takes the perceptual matrix composed of multi-modal data, such as time difference of arrival and angle of arrival, as input, and utilizes the ability of LSTM network to process time-series data to mine the correlations between spatially ordered receiving array elements for sound speed profile inversion. On this basis, a time delay estimation method based on hard threshold estimation method and cross-correlation function is proposed to reduce the measurement errors of the perceptual matrix and improve the anti-multipath performance. The feasibility and accuracy of the proposed method are verified through numerical simulations. Compared with the traditional optimization algorithm, the proposed algorithm better captures the nonlinear characteristics of SSP, with higher inversion accuracy and stronger noise resistance.