Improving Robustness of Passive Source Localization via Convex Optimization Based Mode Filtering

Model-based passive source localization is known to be quite sensitive to model mismatch. In this paper, we present a new implementation based on matched-mode processing. Specifically we formulate mode filtering as a convex optimization problem, and l(1)-regularized least squares is applied, which is robust in terms of the numerical implementation. Simulations with respective full-depth and partially-spanned apertures show that with sound speed mismatch in water column, the proposed method has a better or equal tolerance over traditional matched-field processing; with sound speed mismatch in sediment layer, combining with mode selection, where higher order modes strongly interacting with the bottom are removed, our approach has a slightly decreased bias. Overall the new method demonstrates improved robustness of passive source localization for the examples we studied.