Rapidly Adaptive Matched
Field Processing for
Environments and
Active Sonars



James Ward
MIT Lincoln Laboratory
244 Wood Street, Room J-149J
Lexington, MA 02173-9108
tel: (781) 981-0617

Arthur B. Baggeroer
Depts. of Ocean Engineering and EECS
Massachusetts Institute of Technology
77 Massachusetts Avenue, Room 5-206A
Cambridge, MA 02139-4307

Abstract Two of the most important concerns with matched field processing are the sensitivity to mismatch and the performance in the strong interference scenarios characteristic of shipping noise, and in active systems, reverberation. Furthermore, source/receiver motion limits the time that a target remains within a single resolution cell, and therefore the number of snapshots in which the field is stationary. The snapshot problem is compounded by the presence of the target signal in the sample covariance matrix, and its steering vector may not be known exactly due to uncertain propagation parameters.

In this talk we investigate adaptive matched field processing architectures for typical shallow water environments. The objective was to provide a fair comparison of different DOF reduction strategies based upon an SVD of the sample covariance matrix. Approaches considered include the well known principal components (PC) method, the recently published cross-spectral metric (CS), and other heuristic approaches that incorporate mismatch sensitivity directly in the DOF selection process. Comparisons with the old standby (but less analytically pleasing) white noise gain constrained MVDR (i.e. diagonal loading) are also performed. The trade-off is not a simple one that results in a single universally superior approach; it is shown how the relative performance results depend on the interference DOF (which can be large with extended surface interference), the adaptive DOF, the sample support, the signal strength, and the mismatch level. The CS approach, although optimum in a known covariance sense, is shown to offer benefits only when the adaptive DOF is less than the interference DOF and there is plenty of sample support. In many practical cases, principal components and other selection metrics can offer better because they are influenced less by the noise eigenvalue fluctuations. In very low sample support situations and where mismatch is present, the performance of the white noise gain constraint approach compares quite favorably with reduced DOF approaches. We hope to have additional results for broadband MFP scenarios by the time of the workshop.

[1] Goldstein and Reed, "Theory of Partially Adaptive Radar," October 1997 AES.

[2] H. Cox, "Further Results in Dominant Mode Rejection," IEEE Underwater Acoustics Signal Processing Workshop, October 8, 1997.



Direct comments and questions to:

MIT Lincoln Laboratory. All rights reserved.