Accurate knowledge of the ocean environment is a crucial factor in establishing national security, being intimately tied to accurat'e threat detection, localization, and identi cation. The goal of this work is to improve our knowledge of a changing ocean medium by solving inverse problems that use physics of sound propagation and statistical signal processing. Speci cally, using sequential ltering to track multiple paths from one receiver to the next, we plan to estimate the arrival times of those; the arrival times are related to source location and properties of the propagation medium. The ltering process will provide complete probability density functions of arrivals rather than point estimates which are typically computed by conventional methods. Once these densities are available, a forward model will be developed for inversion. This will determine path propagation along the direct path, the surface and bottom reflections, and a reflection from the first/second sediment interface. For efficiency as well as accuracy, a linearized model based on ray-tracing will be devised. This is a challenge because it requires partial derivative computation of arrival times with respect to the unknown parameters, including frequency and depth dependent sediment properties. Using this linearized model and the probability densities of arrival times will facilitate the calculation of probability density functions for source location, sound speed in the water column, bottom depth, sediment thickness, and sediment sound speed. To further the research, we will consider data received from sound emitted by moving sources. The sequential ltering method just described tracks arrivals in space from one receiver to the next. This process will be combined with a new tracker, tracking arrivals in range as well, using prior information from step to step. The combination is expected to result in efficient inversion for a changing environment. In a di erent direction, data science methods will be implemented for geoacoustic inversion and seafloor classifi cation. The process will use training sets consisting of features extracted from received time-series. The features characterize the structure of the eld as it relates to sediment parameters and serve as propagation medium signatures. Advanced regression techniques as well as probabilistic neural networks will be applied to the problem. Two more methods will be investigated: a coherent matched-field scheme and a direct approach. The methods to be developed within this project will allow the fast identi cation of environmental properties and can be integrated by' the US Navy in already existing techniques for accurate source detection and localization.
|Effective start/end date||3/1/20 → …|
- U.S. Navy: $450,588.00