2019_programme: OUT OF BAND SIGNAL PROCESSING FOR LOCALIZING DISTANT SOURCES IN A NOISY DEEP OCEAN



  • Session: 25. Signal and image processing
    Organiser(s): N/A
  • Lecture: OUT OF BAND SIGNAL PROCESSING FOR LOCALIZING DISTANT SOURCES IN A NOISY DEEP OCEAN
    Paper ID: 770
    Author(s): Geroski David, Dowling David
    Presenter: Geroski David
    Presentation type: oral
    Abstract: Frequency Difference Source Localization (FDSL) methods have proven successful for localizing sources both in the deep ocean [Geroski and Dowling, 2018] and in the shallow ocean [Worthmann and Dowling 2015, 2017] using recordings from a vertical array of hydrophones. These source localization algorithms are based upon analyzing the phase content of a product of measured field values, termed autoproducts, at frequencies outside of the recordings’ bandwidth. Similar to Matched Field Processing (MFP), the measured autoproduct is correlated to a replica that is calculated based on the user's knowledge of the acoustic environment. Unlike MFP, these out-of-band source localization methods have proven to be robust to the problem of environmental mismatch in both shallow and deep-water environments. These efforts were initially undertaken using measured data with a high signal-to-noise ratio (SNR) to demonstrate robustness to mismatch. This paper explores the performance of FDSL in the presence of noise. Specifically, simulated noise and noise measured during the PhilSea10 experiment [Worcester et. al. 2013] are added to both simulated and measured pings to determine at the SNR at which performance begins to degrade. For the scenario considered here, the source was located 210 km from a receiving array in a deep ocean environment with an active internal wave field. The source broadcast was a linear frequency sweep from 200 to 300 Hz. At high SNR (>20 dB) using single-digit-Hertz out-of-band frequencies, the source is correctly localized in all simulations, and in 98 out of 100 trials using single-ping ocean recordings. For simulated and measured forward signals distorted by simulated noise, this performance persists down to SNR's of –20 and –18 dB, respectively. For measured forward signals with elevated levels of measured noise, successful performance persists down to –8 dB SNR.
      Download the full paper
  • Corresponding author: Mr Geroski David
    Affiliation: Applied Physics Program, University of Michigan
    Country: United States
    e-mail: