Kernel high-order approximation of time-varying parameters of harmonics (Q1298201)

This paper presents a nonparametric method for estimating the time-varying frequency, amplitude and phase of the model \(y(t)=A(t)\sin(\omega(t)t+\varphi(t))=e(t)\), \(t=\dots,-2,-1,0,1,2,\dots\), where \(e(t)\) are i.i.d. r.v. The local polynomial approximation approach with high-order nonparametric kernels is used to design estimation logarithms. The details are demonstrated for the linear in time frequency and amplitude. For instance, the covariance matrix of the estimates is obtained in the cases of symmetric and nonsymmetric rectangular windows, the asymptotic normality is proved, and the convergence orders are established. For arbitrary time-varying \(\omega(t)\) and \(A(t)\) the asymptotic bias is studied as well as the optimal window size. Various simulation results are presented.