<title>Adaptive multitaper time-frequency spectrum estimation</title>

Abstract

ABSTRACT In earlier work,1 Thomson's adaptive multitaper spectrum estimation method2 was extended to the nonstationarycase. This paper reviews the time—frequency multitaper method and the adaptive procedure, and explores someproperties of the eigenvalues and eigenvectors. The variance of the adaptive estimator is used to construct an adaptive smoother, which is used to form a high resolution estimate. An F-test for detecting and removing sinusoidalcomponents in the time—frequency spectrum is also given.Keywords: Time—Frequency Analysis, Time—Frequency Resolution, Multitaper Methods I. INTRODUCTION Time—frequency analysis is the process of estimating the time—varying spectral content of nonstationary signals, which would not be completely described by stationary spectral analysis. A general approach to time—frequency analysis is given by Cohen's bilinear class,3 which provides a general formulation for time—frequency representations. This general class includes well—known analysis methods such as the spectrogram and Wigner distribution, as well as