Broadband MFP: coherent vs. incoherent

Matched-field processing (MFP) is now a mature technique for source localization and tracking. There are at least two aspects that emerge, by their relevance, to the success of MFP: one is the ability of a given MFP processor to accurately pinpoint the source location while rejecting sidelobes, and the other is the impact of erroneous or missing environmental information (known as model mismatch) in the final source location estimate. This study addresses the first aspect regarding sidelobe rejection while considering that the processor is working on a mismatch free situation. One well known procedure to reduce sidelobes is to use a broadband MFP processor (whenever a band of frequencies is available). There are a number of different ways to combine MFP information across frequency that can be classified in two broad groups: the conventional incoherent methods, that are based on the direct averaging of the auto-frequency inner products and the, say, less conventional methods, that perform a weighted average of the cross-frequency inner products where the weights are the frequency compensation phase-shifts. The latter are generally termed coherent broadband methods since they combine complex inner products. The coherent broadband methods proposed in the literature are either suboptimal or very computationally intensive, even for a small number of frequencies. An alternative method is presented that combines cross-frequency information with the same localization performance as the standard coherent methods and a computation load similar to that of the incoherent processor. The performance of the various broadband processors is compared in simulated data