Nonlinear filtering for narrow-band time delay estimation

Abstract

This thesis presents a method for improving passive acoustic tracking. A large family of acoustic tracking systems combine estimates of the time difference of arrival (TDoA) between pairs of spatially separated sensors - this work improves those estimates by independently tracking each TDoA using a Bayesian filter. This tracking is particularly useful for overcoming spatial aliasing, which results from tracking narrowband, high frequency sources. I develop a theoretical model for the evolution of each TDoA from a bound placed on the velocity of the target being tracked. This model enables an efficient form of exact marginalization. I then present simulation and experimental results demonstrating improved performance over a simpler nonlinear preprocessor and Kalman filtering, so long as this bound is chosen appropriately.