UACE2023: OPTIMIZING A SLIDING M-OF-N TRACK INITIALIZER IN CLUTTER



  • Session: 11. Large Time-Bandwidth acoustic signals for target detection and tracking
    Organiser(s): Tesei Alessandra
  • Lecture: OPTIMIZING A SLIDING M-OF-N TRACK INITIALIZER IN CLUTTER [invited]
    Paper ID: 939
    Author(s): Abraham Douglas, Grimmett Douglas, Itschner Jonathan
    Presenter: Abraham Douglas
    Presentation type: oral
    Abstract: In active sonar applications with moving targets, objects are tracked by associating and combining single-ping measurements (clusters) over short segments of time and space. A common track initialization algorithm (TIA) requires observing M single-ping clusters across any N consecutive pings within a local search region. Design of this TIA is straightforward when the background consists of false alarms (clusters) occurring randomly in the search space, but less tractable in the more realistic scenario of clutter tracks formed on echoes of physical objects (e.g., rock outcrops, fish schools, etc.). In practice the cluster-level decision threshold (h) is chosen in the clutter-dominated scenario through trial-and-error to manage the quantity of single-ping clusters, N is chosen heuristically to balance robustness to signal fading (large N) and target maneuvers (small N), whereas M can be chosen to optimize a static target-track detection probability. The process generally does not result in a constant false-track rate and is more complicated in high duty cycle (HDC) systems where the effective ping rate (and therefore N) varies with the coherent processing interval.\n\nIn this paper, a process is described for optimizing the sliding M-of-N track initializer through use of a Markovian clutter-event model, representing clutter-cluster false-alarm probabilities with a generalized Pareto distribution (GPD), and choosing M and N to optimize target detection performance (latency or static probability of detection) while using h to maintain a desired false track rate. In situ adaptation is feasible when clutter model parameters are estimated from previously observed data. The clutter modeling and design of the TIA are illustrated using data from the CMRE Littoral Continuous Active Sonar (LCAS) 2015 Experiment.\n
      Download the full paper
  • Corresponding author: Dr Abraham Douglas
    Affiliation: CausaSci LLC
    Country: United States
    e-mail: