Stable nonlinear methods for sensor array processing;

Most nonlinear high-resolution bearing estimators are unstable in the presence of correlated noise, system phase errors, and multipath arrivals because they inadvertently place too much emphasis on unstable eigenvectors of the cross-sensor correlation matrix. For moderately correlated noise there will be sufficiently many stable eigenvectors to resolve and localize discrete sources. A method is given (the "stable nonlinear method" or SNLM) whereby reweighting of the eigenvectors is achieved implicitly, without actual calculation of the eigenvectors. This SNLM is compared with Capon's maximum likelihood method (MLM) in simulations of correlated noise, partially correlated signals, and phase errors, and is shown to provide good stability in the cases considered.