Stephen P. Czenszak
Jeffrey L. Krolik
Dept. of Elec. Eng., Duke Univ., Box 90291, Durham, NC 27708
Broadband matched-field processing approaches have centered around incoherent averaging of narrow-band MFP ambiguity surfaces. This technique works well when environmental mismatch results in a frequency-independent shift in the source position. This appears to be true for mismatch in the bottom parameters when the sound speed is constant in the lower half of the water column. In the presence of more general sound-speed profiles, the performance of incoherent averaging techniques is significantly degraded. In this paper, coherent wideband matched-field processing is proposed to exploit statistical dependencies between the signal wavefronts at different frequencies for more general settings. The fourth-order statistics of the data are used to exploit these cross-frequency dependencies. For Gaussian data this fourth-order moment can be expressed in terms of the narrow-band second-order data cross-spectral density matrices. The coherent wideband MV-EPC processor consists of minimizing the squared output power of the beamformer subject to a set of constraints designed to provide robustness over an ensemble of random environmental realizations. In the presence of sound-speed mismatch this processor is shown to yield improved probability of correct localization performance versus incoherently averaged wideband matched-field processors. [Work supported by ONR.]