A mode separation method based on sparse Bayesian learning for explosive sources in a shallow water waveguide

Abstract

This paper presents a mode separation method based on sparse Bayesian learning (SBL) for explosive sources in a shallow water waveguide. In previous work [Niu et al., JASA, 2021, 4366], the SBL dictionary was constructed by assuming a large number of horizontal wavenumbers and utilized an approximate mode-frequency dispersion relation for low frequencies. Then, modes were separated in the frequency domain by estimating the coefficients of the dictionary atoms. However, challenges inherent to explosive sources, such as bandwidth expansion and the bubble-pulse effect, result in a mismatch in the dictionary matrix built using the approximate mode-frequency dispersion relation for low frequencies, leading to unsuccessful mode separation. To address these issues, this paper builds the SBL dictionary matrix by utilizing the acoustic model (e.g., Kraken) to derive horizontal wavenumbers under various environmental hypotheses and combines it with the secondary bubble-pulse model. By estimating the coefficients of the dictionary atoms, both environmental parameter estimation and mode separation can be achieved simultaneously. Simulation and experimental data results demonstrate the validation of the proposed method.