scientific article; zbMATH DE number 3999073

From MaRDI portal
Publication:4725568





The problem considered in this paper is that of estimating the parameters of the following sinusoidal signals-in-noise model \[ (1)\quad y_ t=\mu +\sum^{n}_{k=1}(a_ k\cos \omega_ kt+b_ k\sin \omega_ kt)+e_ t=:x_ t+e_ t \] where \(e_ t\) is assumed to be a stationary linear process. The unknown parameters in (1), which are the mean level \(\mu\), the amplitudes \(\{a_ k,b_ k\}\) and the frequencies \(\{\omega_ k\}\) are estimated by the (nonlinear) least-squares method: \[ (2)\quad {\hat \theta}:=[{\hat \mu},\{\hat a_ k\},\{\hat b_ k\},\{{\hat \omega}_ k\}]=\arg \min_{\theta}\sum_{t}[y_ t-x_ t]^ 2. \] The time index t in (2) runs from 1 to N or from -N to N. Either of these assumptions on the range of t implicitly introduces some assumption on the time origin of the sinusoidal signal. However, for the estimation and properties of \(\{\omega_ k\}\) (which are the parameters of interest in many applications) this assumption does not introduce any restriction. The main results of the paper concern the asymptotic distribution of \({\hat \theta}\). It is proven that \({\hat \theta}\) is asymptotically Gaussian distributed and an explicit expression for the covariance matrix of the limiting distribution is given. Similar and, to some extent, more general results have been derived independently by the reviewer and \textit{A. Nehorai}, Statistical analysis of two nonlinear least-squares estimators of sine wave parameters in the colored noise case. Center for Systems Science, Yale University, Report No.8709 (1987).




Cited in
(26)








This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4725568)