Resampling based techniques for source detection in array processing

Abstract

The source detection problem in array processing can be considered as a test for the equality of eigenvalues. This approach is implemented through a multiple hypothesis procedure which compares all pairwise differences between eigenvalues. A resampling procedure is used to estimate the null distributions of the test statistics, an advantage for small sample sizes or non-Gaussian signals since traditional techniques such as the minimum description length (MDL) assume Gaussianity. Simulations show the increased performance of the test compared to the MDL for small samples or non-Gaussian signals, with a noticeable improvement over the more accurate sphericity test.