Modeling the time-varying microstructure of simulated sleep EEG spindles using time-frequency analysis methods

Area of Science:

• Neuroscience
• Signal Processing

Background:

• Sleep spindles are crucial for cognitive functions and their microstructure holds clinical significance.
• Quantifying and modeling sleep spindle time-varying characteristics is essential for understanding their role.

Purpose of the Study:

• To evaluate and compare four distinct signal processing techniques for estimating sleep spindle parameters.
• To determine the accuracy and distortion introduced by each method in modeling sleep spindles as AM-FM signals.

Main Methods:

• Sleep spindles were modeled as Amplitude-Modulation/Frequency-Modulation (AM-FM) signals with six parameters.
• Instantaneous Envelope (IE) and Instantaneous Frequency (IF) waveforms were estimated using Hilbert Transform (HT), Complex Demodulation (CD), Wavelet Transform (WT), and Matching Pursuit (MP).
• Model parameters were derived from IE and IF waveforms, and errors were calculated for each method.

Main Results:

• Hilbert Transform (HT) yielded the lowest average error across model parameters.
• Complex Demodulation (CD) and Matching Pursuit (MP) demonstrated errors below 10%.
• Wavelet Transform (WT) exhibited the highest error (over 10%), while HT introduced the most significant distortion (0.4 sec).

Conclusions:

• Complex Demodulation (CD) and Matching Pursuit (MP) are recommended for analyzing real sleep spindles due to their balance of accuracy and distortion.
• Hilbert Transform (HT) can be considered for sleep spindles exceeding 1 second in duration, provided its distortion is acceptable.