Statistical modelling, matched-field processing and matched-field inversion at high frequencies

Matched-field processing (MFP) is a generalized form of beamforming which uses knowledge of the acoustic environment to predict the pressure field at the receiver array, from which a weight vector is computed. This is used to weight and sum the measured field components to product an output analogous to beamforming. It has been shown that MFP is sensitive to mismatch between the modelled and actual environments. In this paper it is shown that the sensitivity of MFP to mismatch increases with frequency, because prediction of the field detail becomes more difficult with increasing frequency. It is shown that certain statistical properties of the acoustic field can be predicted at high frequencies, and these can be used to infer environmental parameters and hence improve the performance of MFP. An algorithm is proposed which uses statistical properties of environmental uncertainty and focalization to give improved MFP performance at high frequency