The estimation of frequency in the multichannel sinusoidal model (Q2293395): Difference between revisions

The purpose is to estimate the frequency parameter \(\omega\) in the multichannel sinusoidal model \[ x_t=\mu+\alpha\cos(\omega t)+\beta\sin(\omega t)+\epsilon_t \] where \(\omega\in(0,\pi)\), \(\epsilon_t\) is a \(d\)-dimensional stationary centered stochastic process and \(\mu,\alpha,\beta,x_t\) are of dimension \(d\times 1\). Similar to the univariate case one starts by estimating parameter for a fixed \(\omega\) and substitutes them in the log-likelihood \(l(\theta,\Sigma,\omega)\), \(\theta=[\mu\alpha\beta]\), \(\Sigma=E(\epsilon_t\epsilon_t^T)\) which this way becomes a function of \(\omega\) only. The maximum likelihood estimator for \(\omega\) and the asymptotic properties of the minimizer are shown. The central limit theorem is proved. A method to approximate the spectral density parametrically by fitting long-order autoregression is developed. The algorithm is presented and simulation results are discussed. Proofs are delivered in the appendix section.