Practical detection with calibrated arrays

This paper discusses some of the practical limitations of detection methods formulated in terms of the eigenvalues of the sample covariance matrix of the output of a sensor array. It presents an approach based on the principal eigenvectors and the measured array manifold that appears to be at least as sensitive, but apparently much more robust than methods such as AIC and MDL. Comparative performance results are given for simulation data with a variety of noise statistics and for data obtained from an experimental array.<>