Frequency analysis of sleep spindles by Extended Prony model: Validation by signal reconstruction

In order to classify sleep spindles from human EEG the advantages of the Extended Prony Method in spectral estimation are presented. This method assumes that the original signal is a sum of damped complex exponential sinusoids. The method has the best frequency resolution compared to the other parametric methods. Given the parameters of the Prony model, a synthetic spindle can be generated. Since these parameters are used both in spectral estimation and in the reconstructi on process, the associated Power Spectral Density (PSD) is more especially estimated as the error function between the reconstructed and the raw signal is minimized. Limitations of the method are also discussed.