Matched Field Processing: Status at 2000 and Where To Now?

Matched processing (MFP) was introduced by Bucher in 1977 based upon earlier work on modal beamforming. It was first demonstrated experimentally by Yang using vertical array data taken at the 1984 FRAM IV ice station. Since then numerous algorithms have been developed and many experiments have attempted to exploit MFP. The essential concept of MFP is to incorporate the propagation physics into the array processing (beamforming) for vertical and/or large horizontal arrays. This leads to both range and depth localization and avoids signal gain degradation if a plane wave model is used. Alternatively, the emphasis can be refocused towards matched field tomography (MFT) wherein one uses the same techniques to estimate environmental parameters.

MFP (and MFT) are now almost twenty years old, so it is fair to ask what it has and has not accomplished. Certainly, there have been some spectacular demonstrations of localization by MFP. Some of the very early experiments localized sources well in both range and depth at megameter ranges; however, the experience in shallow water has been less successful for a number of reasons. Virtually all the experiments have emphasized localization (and tomography) at high SNR's while detection at low SNR, the most important problem for passive sonar, and the issues of minimum detectable levels has received far less attention. Narrowband tones have been the signal of choice for most; while there are broadband algorithms, most are ad hoc and do not include the consequences of broadband signal models such as stationarity. Finally, MFP makes certain assumptions about the spatial coherence among the rays/modes which represent the signal and ones ability to compute replica vectors for steering. At frequencies where scattering and coupling effects are small, MFP has performed very well; yet, there have been many analyses where MFP has been used in a most uncritical manner. In this lecture, we will critique the status of MFP at 2000 and suggest some significant problems it must yet overcome for it to used beyond the research community.