UC San Diego

Passive acoustic monitoring (PAM) enables the study of marine mammals and anthropogenic sources without active sound transmission, offering a non-intrusive means to observe underwater environments. Yet, in practical deployments, PAM systems must operate under resource constraints—sparse hydrophone geometries, limited power, and noisy, non-cooperative sources—requiring robust signal processing methods to extract spatial information.This dissertation presents two distinct contributions to passive source localization. The first addresses the challenge of localizing and tracking odontocetes using their echolocation clicks, which are impulsive, broadband signals (Chs. 1 and 2). Through simulation and at-sea experiments with sparse volumetric arrays, we demonstrate that time-difference-of-arrival (TDOA) methods outperform beamforming approaches—including frequency-difference beamforming—in terms of accuracy and robustness. Building on this finding, we develop a particle filter-based multi-target tracking framework that localizes multiple animals in 3-D from the TDOA measurements while accounting for multiple targets, false alarms, and intermittent detections. The method is validated using acoustic recordings of echolocation click signals of from goose-beaked whale (Ziphius cavirostris), yielding automated movement tracks of deep-diving whales.The second contribution focuses on passive range estimation of large commercial ships in shallow water, where acoustic propagate as modes. Using single-hydrophone recordings, we apply waveguide invariant (WI) theory to analyze striation patterns as a result of modal interference in the low frequency band. A statistical model is developed to characterize the received tonal and broadband acoustic signature of ship noise, enabling maximum likelihood-based range estimation. Applied to real data from the Seabed Characterization Experiment 2017 (SBCEX17), this method yields accurate range estimates out to 45 km. Building on this result, Ch. 4 examines the validity and limits of the WI approach when only a small number of modes propagate in a low frequency band. Through simulation and observational analysis, we show that range estimation remains effective when interference is dominated by a single mode pair. This finding highlights the practical viability of long-range passive ranging in environments based on WI.Collectively, these results advance the use of PAM in both ecological and operational contexts by enabling precise source localization in scenarios where sensor density, acoustic conditions, or signal structure pose major challenges. The techniques developed here contribute to scalable monitoring of marine mammals and improved acoustic situational awareness.