Multitaper estimation over multiresolution subspaces

We apply the multitaper spectral estimation method over multiresolution subspaces to analyze nonstationary data. We define a measure of stationarity continuous between 0 and 1 and use it to show that multiresolution subspace projections of a process rank closer to being stationary than the original process. Our measure is based on the ratio of the energy of the power spectral density S(f, upsi) in the neighborhood of the f = - upsiline to the total energy as a measure to "stationarize" a process. We show that the increase in stationarity is a function of the wavelet transformation, or restriction of the process to a multiresolution subspace, rather than the process itself. We show results of the application to acoustic signals generated by an airplane in the landing process