Three-dimensional passive source localization using a deep-water vector sensor vertical line array: A particle filtering-based track-before-detect method

In deep water, a near-bottom vector sensor vertical line array (VSVLA) enables three-dimensional (3-D) passive localization of submerged sources. For a shallow source, localization is typically performed in two steps: azimuth and range estimation using 2-D spatial spectrum estimation (2DSSE), and depth estimation by matching the frequency-domain interference period. However, in multitarget scenarios, the limited spatial resolution of the VSVLA can easily cause missed detections, while multiple threshold detections encounter a severe measurement-to-track association (MTA) problem. To address these challenges, this work adopts a track-before-detect (TBD) framework to avoid the explicit MTA process. The proposed algorithm jointly scans the intensity over the azimuth-range-depth grid using a pre-generated interference-matched kernel function and directly inputs it into TBD as the measurement, effectively mitigating missed detections caused by 2DSSE bright spot overlap. Building on the incorporation of multiple auxiliary particle filter as the tracker, adaptive weight filtering and Rao-Blackwellization strategies are further employed to improve convergence speed and reduce state dimensionality, respectively. The proposed algorithm demonstrates high robustness even under low signal-to-noise ratio conditions and environmental fluctuations. Its localization performance is validated through simulations across multiple complex deep-water scenarios, as well as a towed-source experiment.