Broadband modal phase speed estimation and geoacoustic inversion with sparse Bayesian learning using multi-range dataa).

A broadband modal phase speed estimation method based on sparse Bayesian learning (SBL) using multi-range data (multi-range SBL) is proposed for geoacoustic inversion in shallow water. Multi-range SBL estimates local modal depth functions and horizontal wavenumbers from multi-range signals received by a vertical line array, without a priori knowledge of geoacoustic parameters or sound source locations. It eliminates narrowband limitations caused by the approximate dispersion relation in block SBL, which utilizes multiple frequencies, allowing horizontal wavenumber estimation over a broad bandwidth. This enables geoacoustic inversion by fitting and matching the curves of phase speed changing with frequency across a broad bandwidth. Multi-range SBL decouples seabed attenuation coefficient from other parameters, allowing for their separate estimation and reducing the mutual interference. The proposed method involves local modes, which are only related to local parameters, unaffected by propagation path and bathymetry mismatch during inversion. A Bayesian optimization algorithm is used to improve search efficiency during the estimation of multi-dimensional parameters. The inversion results are used to construct a sound field model for matched-field processing localization. The feasibility and accuracy of the method are validated through simulations and experimental data, showing better results in range and depth estimation compared to conventional matched-field inversion.