Robust source localization using predictable mode subspace in uncertain shallow ocean environment

Conventionally used matched-field source localization methods are sensitive to environmental parameter mismatch. In this paper, a new robust localization method is proposed. It is based on a decomposition of the field into predictable and unpredictable subspaces of the acoustic normal mode representation. The method uses the predictable subspace to reconstruct the replica vector and the orthogonality between the predictable and unpredictable subspaces is used to eliminate the impact of environmental uncertainty. The performance of the method is evaluated and compared to other matched-field methods using computer simulations. Results show that, in the presence of mismatches, the proposed method has superior probability of correct localization than the robust maximum-likelihood and the Bartlett methods.