Statistical Clustering Applied to Adaptive Matched Field Processing

Cluster analysis provides a tool for mapping out regions of ambiguous response in sparse array beamforming problems. This paper discusses clustering and its application to matched field processing (MFP) problems in ocean acoustics. The map of the ambiguity volume provided by clustering can be used for improved interpretation and postprocessing. By peak-picking in cluster space, rather than in spatial dimensions, we are able to identify and discard spatial peaks that result from the presence of a strong source. As shown in a data example, clustering can help collapse 3-D MFP output to bearings-only while preserving signal gains obtained by accounting for multipath. Clustering can also provide computational gains by allowing elimination of highly redundant replicas. 1. OVERVIEW OF MATCHED FIELD Matched field processing has been extensively explored for use in detecting and localizing underwater sources. MFP uses ocean propagation models to account for multipath when generating replica vectors. If accurate environmental information is available, sources can be localized in range, depth, and bearing. Accurately modeling multipath can also yield mismatch reduction and detection gains as compared to direct-path beamformers. A significant problem for MFP is that the beamformer output often displays high ambiguities in range and depth, with a single source leading to multiple spatial peaks. This is particularly true for arrays without significant vertical aperture because of their limited ability to resolve multipath. MFP for these arrays can be thought of as a form of sparse array processing with resultant high ambiguities. Many of the ambiguous peaks may be within a few dB of the overall peak, so they are not easily suppressed with adaptive beamforming. This work was sponsored by DARPA-ATO under Air Force Contract F19628-00-C-0002. Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the United States Government. MFP replicas are generated for a set of look directions in range, depth, and azimuth. If we define Θ = (r, φ, z) as the look direction, the MFP output for an N element array at frequency f0, time t0, and direction Θ is given by