Date of Completion

5-9-2013

Embargo Period

5-9-2013

Keywords

ML-PMHT, ML-PDA, multitarget tracking, multistatic sonar, maximum likelihood, low observable, maneuver model, extreme value, ROC, trackability, thresholds, Anscombe transform

Major Advisor

Peter Willett

Co-Major Advisor

Yaakov Bar-Shalom

Associate Advisor

Tod Luginbuhl

Associate Advisor

NA

Field of Study

Electrical Engineering

Degree

Doctor of Philosophy

Open Access

Open Access

Abstract

Multistatic sonar tracking is a difficult proposition. The ocean environment typically features very complex propagation conditions, causing low target probabilities of detection and high clutter levels. Additionally, most sonar targets are relatively low speed, which makes it difficult to use Doppler (if available) to separate target returns from clutter returns.

The Maximum Likelihood Probabilistic Data Association Tracker (ML-PDA) and the Maximum Likelihood Probabilistic Multi-Hypothesis Tracker (ML-PMHT) --- a similar algorithm to ML-PDA --- can be implemented as effective multistatic trackers. This dissertation will develop a tracking framework for these algorithms.

This framework will focus mainly on ML-PMHT, which has an inherent advantage in that its log-likelihood ratio (LLR) has a simple multitarget formulation, which allows it to be implemented as a true multitarget tracker. First, this multitarget LLR will be implemented for ML-PMHT, which will give it superior performance over ML-PDA for instances where multiple targets are closely spaced with similar motion dynamics. Next, the performance of ML-PMHT will be compared when it is applied in Cartesian measurement space and in delay-bearing measurement space, where the measurement covariance is more accurately represented.

Following this, a maneuver-model parameterization will be introduced that will allow ML-PDA and ML-PMHT to follow sharply maneuvering targets; their previous straight-line parameterization only allowed them to follow moderately maneuvering targets.

Finally, a novel method of determining a tracking threshold for ML-PMHT will be developed by applying extreme value theory to the probabilistic properties of the clutter. This will also be done with target measurements, which will allow the issue of trackability for ML-PMHT to be explored.

Probabilistic expressions for the maximum values of the LLR surface caused by both clutter and the target will be developed, which will allow for the determination of target trackability in any given scenario.

COinS