Eigenvector peeling approach to coherent multiple source location problem

Abstract

We propose a novel preprocessing scheme, referred to as vector peeling, as an alternate to the conventional spatial smoothing for solving the multiple source location problem involving coherent sources or a rank deficient source covariance matrix. The essence of the technique is to preprocess the signal sub-space eigenvectors rather than the covariance matrix as in spatial smoothing. It is shown by analysis and computer simulations that these two approaches are related, and that vector peeling slightly outperforms spatial smoothing when employed with the MUSIC-type DOA estimators. In certain instances, vector peeling offers advantages in terms of computational simplicity and flexibility. The latter is especially true with eigenstructure DOA estimators in adaptive estimation problems, i.e., when the signal subspace eigenvectors are updated using fast adaptive algorithms.