Integrating Adaptive Beamforming and Classification for Passive Sonar Detection in Complex Environments
Abstract
Abstract Approved for Public ReleaseThe goal of this project is to integrate adaptive beamforming and classification stages to improve passive sonar performance in complex multipath underwater channels when both signal and interfering source wavefronts are uncertain. Conventional passive sonar signal processing consists of separately-designed stages for beamforming, short-time Fourier-analysis, and classification. Spatial processing is based on physical modeling of source wavefronts, Fourier analysis is driven by expectedsource temporal spectra, and classification relies heavily on spectral features, as well as source dynamics estimated using a tracker. The presence of multiple directional acoustic sources makes passive target detection and classification a process of manually eliminating non-target interfering sources from broadband spatial field-directionality maps and narrowband temporal spectra versus time (i.e., BTRs and LOFARgrams, respectively). Minimum variance distortionlessresponse (MVDR) adaptive beamforming has been a staple in sonar beamforming for suppressing interference to discriminate sources of interest. Unfortunately, however, when the target and interference have similar spatial wavefronts, mismatch between the received and modeled target wavefront, e.g., due to multipath propagation or array miscalibration, can result in suppression of signals which would otherwise be detectable. This problem arises because in the MVDR beamformer the modeled signal wavefront effectively defines both the target and interference spatial characteristics. In the proposed work, we avoid the deleterious effects of signal mismatch by forming a signal-free noise covariance matrix estimate using a neural-network-driven classifier to identify the spatial-temporal characteristics of interfering sources while not requiringspecification of the signal wavefront. The tight integration of spatial processing and classification yields an adaptive beamformerwhich more closelyapproximates the optimal detector when the signal and interference wavefronts are both uncertain. The adaptive beamformer-classifiers (ABCs) proposed here are designed to make use of recorded acoustic data to learn the characteristics of surface and biologic interference, including their channel-induced frequency and time-domain fading features, as well as acoustic data identified by human operators as signals of interest. ABCs will enhance passive sonar operations by classifying source types in acoustic beam outputs, thus allowing operators to focus their attention on beams containing signals of interest.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 12, 2023
- Source ID
- N000142312135
Entities
People
- Jeffrey Krolik
Organizations
- Duke University
- Office of Naval Research
- United States Navy