Maximum likelihood geoacoustic inversion from surface generated noise

Abstract

Noise generated at the surface from wind and breaking waves is incident upon the seabed at a wide range of angles and frequencies. The reflectivity of the seabed thus plays an important role in determining the vertical directionality of the noise field. It is therefore not surprising that from an information theory perspective, the noise field encodes a significant amount of information about the geoacoustic properties of the seabed that govern reflectivity, such as layering structure, sound speed, density, and attenuation. Using a physics-based ambient noise model, the Cramer-Rao lower bound (CRLB) can be computed, which specifies the lowest possible variance that an unbiased estimator can have. Existing estimators have primarily involved vertically beamforming the noise field to recover the vertical noise directionality and bottom loss, which is then processed to estimate geoacoustic seabed properties. The method described here takes a different approach by estimating the properties directly from the measured data (i.e., without beamforming). This presentation will show that maximizing the proposed likelihood function can perform better than beamforming-based techniques, and approaches the CRLB. Further, it remains unbiased in low signal-to-noise ratio conditions, is tolerant to array tilt, and can operate beyond the nominal array design frequency.Noise generated at the surface from wind and breaking waves is incident upon the seabed at a wide range of angles and frequencies. The reflectivity of the seabed thus plays an important role in determining the vertical directionality of the noise field. It is therefore not surprising that from an information theory perspective, the noise field encodes a significant amount of information about the geoacoustic properties of the seabed that govern reflectivity, such as layering structure, sound speed, density, and attenuation. Using a physics-based ambient noise model, the Cramer-Rao lower bound (CRLB) can be computed, which specifies the lowest possible variance that an unbiased estimator can have. Existing estimators have primarily involved vertically beamforming the noise field to recover the vertical noise directionality and bottom loss, which is then processed to estimate geoacoustic seabed properties. The method described here takes a different approach by estimating the properties directly from the me...