ELEE Doctor of Philosophy Dissertation Defense by Abner C. Barros - ECE
Topic: Physics-Based Hierarchical Bayesian Inference for Localization in Active Underwater Acoustic and Passive Seismoacoustic Sensing
Abstract:
Spatiotemporal inference of objects from sensed acoustic and seismic fields is challenging due to refractive wave propagation, environmental uncertainties, and the dynamics of the sensed body. Nevertheless, a-priori information is often available regarding the media, the object’s composition, and its state of motion. Such information provides soft constraints to aid the knowledge building task. This dissertation advances the breadth of the hierarchical Bayesian philosophy to incorporate such disparate streams of physical and situational information into the inference process. To advance the utility of the framework, computationally efficient methods are developed to judiciously align and manage resources. Two inverse problems are considered: localization of a mobile submerged scattering body using high–frequency active sonar with a small-aperture array in an uncertain ocean waveguide, and terrestrial seismic source localization using a fiber–optic distributed acoustic sensing (DAS) system in which the broadband source disturbance is not known.
For the active sonar case, localization of a scattering body in a dynamic ocean waveguide is complicated by boundary interactions, uncertainty in the medium’s refractive profile, and the short coherence time associated with mobile platforms and dynamic environments. The problem is exacerbated by aperture constraints that limit focus and resolution of the scattered field below the Rayleigh bound. A computational Bayesian framework is developed to quantify intrinsic uncertainty associated with the scattered wavefield by decomposing its phase-front representation in the Doppler-angular domain. A Gibbs-like scheme efficiently constructs the joint posterior probability density (PPD) of the scattered field as a continuous mixture based on conditional densities in the presence of numerous nuisance parameters. Tractable conditional densities of eigenray amplitudes and ambient acoustic noise power are in the Gaussian–Inverse–Gamma family enabling computational resources to be concentrated on the conditionals of the frequency-angle dependent wavevectors, which are constructed using two–dimensional quantile sampling. The resulting PPD of the scattered field is then mapped to the PPD of the scattering body’s range, depth, and speed under an uncertain sound speed profile (SSP). The nonlinear relation between the SSP and the scattered field is numerically expanded over the posterior mixture representation to quantify eigenray spreading via Laplace marginalization. Performance analysis and case studies using HYCOM SSPs lend credence to the approach.
For terrestrial inference, DAS systems are increasingly employed for passive detection and localization in geophysical monitoring. Inference is challenging due to DAS strain-based measurements and the associated directional and sensitivity characteristics of the fiber-optic sensor. A localization approach is developed for an impulsive seismic source on the Earth’s surface in the near field of a DAS array, extending DAS performance beyond conventional proximity-based detection. Marginalization over the weakly known source disturbance yields a PPD of the source location. The forward propagation model incorporates a near–field Green’s function with surface–wave dispersion appropriate for layered media while accounting for the angular response of fiber-optic sensors to surface waves. Modal information is extracted via frequency–wavenumber analysis, and empirical dispersion curves are fitted using a two–term exponential model. Application to DAS observations show promising localization accuracy, underscoring the potential to expand the capabilities of DAS across a wide range of geophysical applications. [This work is supported by the NUWC Division Newport's Doctoral Fellowship and ILIR programs, and the Office of Naval Research]
Advisor: Dr. Paul J. Gendron, Associate Professor, Department of Electrical & Computer Engineering, UMass Dartmouth
Committee members:
- Dr. David A. Brown, Professor, Department of Electrical & Computer Engineering, UMass Dartmouth
- Dr. Dayalan P. Kasilingam, Professor & Chairperson, Department of Electrical & Computer Engineering, UMass Dartmouth
- Dr. Zoi-Heleni Michalopoulou, Distinguished Professor, Mathematical Sciences, New Jersey Institute of Technology (NJIT)
- Dr. Tod Luginbuhl, Head, S & T Division, Sonar and Sensors Department, Naval Undersea Warfare Center Division Newport (NUWC)
Note: All ECE Graduate Students are encouraged to attend. All interested parties are invited to attend. Open to the public. *For further information, please contact Dr. Paul J. Gendron.
Lester W. Cory Conference Room, Science & Engineering Building (SENG), Room 213A
: Zoom Link: https://umassd.zoom.us/j/91905521054 Meeting ID: 919 0552 1054 Passcode: 209887
Paul J. Gendron
508.999.8510
pgendron@umassd.edu
https://umassd.zoom.us/j/91905521054