Comparative analysis of time-frequency methods estimating the time-varying microstructure of sleep EEG spindles

Abstract

Parameter estimation for an assumed sleep EEG spindle model (AM-FM signal) is performed by using four time-frequency analysis methods. Results from simulated as well as from real data are presented. In simulated data, the Hilbert Transform-based method has the lowest average percentage error but produces considerable signal distortion. The Complex Demodulation and the Matching Pursuit-based methods have error rates below 10%, but the Matching Pursuit-based method produces considerable signal distortion as well. The Wavelet Transform-based method has the poorest performance. In real data, all methods produce reasonable parameter values. However, the Hilbert Transform and the Matching Pursuitbased methods may not be applicable for sleep spindles shorter than about 0.8 sec. Matching Pursuit-based curve fitting is utilized as part of the parameter estimation process.