Tracking tomographic arrivals.

Abstract

Ocean acoustic tomography records an acoustic pattern of peaks that is interpreted in terms of ray arrivals. Before peaks can be tracked in the time-evolving environment, they must be associated with the correct ray path. Scattering by range-dependent ocean structure is a complicating factor. Fine scale structure limits the time and bandwidth coherence of an arrival and thus a single peak per ray path becomes multiple fading peaks. Associating the peak data with the ray model is often done by hand, but this has two drawbacks. First, a human is required to look at the data precluding processing the data in situ. Second, there is no guarantee that the analysis is repeatable. To address these concerns, three popular methods from radar data processing, the nearest neighbor standard filter, the probabalistic data association filter, and Viterbi data association, are proposed to solve the automatic data association problem. The peak arrival time, angle, and amplitude are clues used to construct a scoring function based on the likelihood ratio. A Kalman filter is used to restrict the expected arrival pattern to shifts and stretching between transmissions. A comparison of these methods with actual tomographic data will be presented.