ON A NOVEL SUBSPACE-BASED APPROACH TO PARAMETER-ESTIMATION

A novel method for the estimation of temporal or spatial frequencies from measurements corrupted by additive white noise has been recently proposed in this journal. In the present paper we rederive that method in a more direct way than in the original work. Our simple derivation makes it possible to relate the subject method to the estimation-of-signal-parameters-via-rotational-invariance-techniques (ESPRIT). We show that ESPRIT and the new method under discussion are identical in finite samples, for a sensible choice of the user's variable in the latter technique. We also discuss the use of centrohermitian sample covariances with the previous two methods or, in fact, any other eigenanalysis-based frequency estimator. (C) 1995 Academic Press, Inc.