Extracting complexity waveforms from one-dimensional signals.

Area of Science:

• Nonlinear dynamics
• Signal processing
• Biophysics

Background:

• Nonlinear methods estimate signal complexity using fractal dimension (FD).
• Traditional methods like Higuchi's are unsuitable for very short epochs (N < 100 samples) due to low-pass filtering effects.
• Extracting running fractal dimension FD(t) requires analyzing short signal epochs.

Purpose of the Study:

• To develop a novel, simple method for estimating FD in short signal epochs (N < 100).
• To improve the accuracy of extracting time-varying fractal dimension (FD(t)).

Main Methods:

• Introduced 'normalized length density' (NLD) for FD estimation in short epochs.
• Constructed a monotonic calibration curve (FD = f(NLD)) using Weierstrass functions with known theoretical FD.
• Compared the NLD method with Higuchi's and consecutive differences methods on synthetic and human EEG signals.

Main Results:

• The NLD method demonstrated significantly lower scattering of FD values for short epochs (N < 30) compared to existing methods.
• Achieved more accurate reconstruction of FD waveforms in signals with abrupt changes in complexity.
• NLD-derived FD values for human EEG signals remained within theoretical limits, unlike Higuchi's estimations.

Conclusions:

• The NLD approach offers superior accuracy for FD(t) extraction in very short epochs.
• This method is suitable for analyzing physiological signals with abrupt changes, such as transient artifacts or phasic phenomena.
• The NLD method has potential applications across various scientific fields requiring analysis of complex, dynamic signals.