Multitaper Covariance Estimation and Spectral Denoising

A seamless and smooth transition from nonparametric multitaper (MT) spectral estimation (SE) to autoregressive (AR) parametrization is made possible via autocorrelation estimates from multitapered data. We show that AR spectral estimates obtained from MT autocorrelation estimates outperform the conventional method and the Capon-AR estimates. We derive a direct method of autocorrelation estimation from multitapered data. We show that choice of tapers yields low-bias and consistent autocorrelation and AR spectral estimates. We also show that MTAR spectra provide spectral smoothing comparable to denoising by wavelet thresholding. We show examples on simulated AR and ARMA processes and tests on aero-acoustical data.