No label defined (Q4779735)

The subject of the paper is an already classic problem related to a harmonic set of signals corrupted by additive noise. The parameters to be found are: the order \(k\) of the model (the number of sinusoids) and the variance of the noise. Many solutions have been proposed, each of them possessing merits and deficiencies. The present paper provides an excellent overview of the recents methods and offers a new solution to the problem. In fact, the present method consists in an iterative deletion and estimation algorithm. After each iteration, when one (or more) tone is detected a deflation is performed (the contribution of already detected tone is deleted by an orthogonal projection). Then after a set of iterated deflations, the initial periodogram becomes a notch periodogram, where the lower contributions stem from the noise. By averaging, the noise estimation is obtained. Some numerical examples validate the proposed method. We think the present paper represents a basic contribution in the field of harmonic estimation. For the entire collection see [Zbl 0990.00077].