The problem of target characterization by phased array sonar is approached as one of spectral estimation of a bandlimited function from a finite number of arbitrarily located samples. A unique solution is extracted from the resultant underdetermined system by application of singular value decomposition (SVD) techniques to derive a minimum norm, or best least-squares estimator of the spatial spectrum. The process of a single matrix decomposition is employed to derive a general basis set spanning the signal space, dependent only on the sensor locations. This can be used to reconstruct arbitrary target profiles two orders of magnitude faster than standard high-resolution methods but with significantly better resolution than a beamformer. Since only a single snapshot is used, coherency in the source is not problematic. Through a suitable regularization scheme the algorithm works well under conditions of unknown correlated noise, and produces super-resolution of features through flexible introduction of a priori knowledge in the localization and/or shape of the target.
展开▼
