High performance SVD-like procedure for spectral estimation using Rayleigh function estimates

The author describes how Rayleigh estimates can be viewed as a method which performs singular-value decomposition (SVD) procedure without doing it. As a short cut to get principal-component reduction, Rayleigh quotients allow the resolution of frequency detectors yet preserve the asymptotic behavior of the actual power spectral density. In a filtering framework the estimate is extended to adaptive schemes and 2-D spectral estimation. The resulting estimate provides the means for adaptive processing with low computational complexity. It avoids also the crucial decision between signal subspace and noise subspace which promotes undesired distortion and false peaks in spectral estimation applications.>