Tracking and identification of ray acoustic arrivals by means of neural networks

Multipath arrivals tracking and identification of SOFAR propagation are central points in ray ocean acoustic tomography data processing because one must know the ray geometry to solve the tomography inverse problem. Tracking resolved peaks all along an experiment allows to create temporal series of acoustic arrivals. Identification is achieved by comparing peak time series with outputs of a ray tracing model. The authors present a method based on neural networks to perform the tracking step. Two types of neural architecture (multi-layer perceptron and Hopfield) are used. As an intermediate result of the tracking, a global arrival peaks pattern is computed which used in conjunction with the predicted one in an iterative constrained minimisation allows one to perform the identification phase in a few iterations. Results obtained with a data set acquired during a four-month experiment illustrate the paper