Michael D. Collins
Laurie T. Fialkowski
W. A. Kuperman
John S. Perkins
Naval Res. Lab., Washington, DC 20375
The MUSIC method [R. O. Schmidt, Ph.D. dissertation, Stanford Univ. (1981)] and other beamformers based on covariance matrix eigenvectors suppress sidelobes and provide high resolution. These methods exploit the fact that energy arriving from sources located in different directions tends to partition into different eigenvectors. Eigenvector processors have been less successful for matched-field applications because of the prevalence of environmental uncertainty and the sidelobe structure of matched-field ambiguity surfaces. The multivalued Bartlett processor, which involves applying the Bartlett processor to the eigenvectors corresponding to the largest eigenvectors, is less sensitive to mismatch than high-resolution eigenvector processors. The well-defined estimates of the source locations correspond to the locations of the peaks in the ambiguity surfaces for the different eigenvectors. Since the sidelobes in matched-field ambiguity surfaces are distributed, energy partitioning is not necessarily favorable for sources in different locations. The multivalued Bartlett processor is effective for moving sources, however, because the partitioning is likely to be favorable for randomly selected locations. As the sources move, the partitioning may be favorable frequently enough for the source paths to be come evident. [sup a)]Present address: Scripps Institution of Oceanography, La Jolla, CA 92093.