Robust processing for multipath in acoustic navigation, tomography, and source tracking

Because sound travels along many refracted paths through the ocean, passing through many horizontal sound speed layers, it is often challenging to unravel the received multipath to convert arrival times to ranges (e.g., by assigning them to eigenrays calculated by a ray tracer), particularly when it must be automated. This problem is exacerbated when the source and/or receivers are in motion. Assigning the wrong eigenray to an arrival results in the arrival time being modeled for the wrong acoustic path and the wrong range assigned to that arrival, resulting in degraded navigation, tomography, or localization. We will show how to take advantage of redundant measurements to ensure that we avoid spurious eigenray assignments. We use an initial “robust” optimization process whose cost function is less sensitive to spurious measurements or outliers than processes using least squared errors. Such processes both enable outliers to be identified and produce better estimates (than if spurious measurements were admitted). Better estimates allow the assignment of eigenrays to arrivals to be better constrained, often leading to corrected assignments. This virtuous cycle results in fewer spurious assignments and improved forward modeling for applications like navigation, tomography, and source tracking.