Intelligent Extraction of Salient Feature From Electroencephalogram Using Redundant Discrete Wavelet Transform

Area of Science:

• Signal Processing
• Biomedical Engineering
• Neuroscience

Background:

• Electroencephalogram (EEG) signals are crucial for medical diagnosis and research.
• Processing EEG signals is essential to mitigate interference and noise.
• Wavelet Transform (WT) is widely used for EEG feature analysis due to its time-frequency representation.

Purpose of the Study:

• To investigate translation variability (TV) in Discrete Wavelet Transform (DWT) and its impact on EEG signal analysis.
• To demonstrate that TV, caused by downsampling in DWT, leads to degraded time-frequency localization and feature distortion.
• To introduce Redundant Discrete Wavelet Transform (RDWT) as a translation-invariant alternative.

Main Methods:

• Numerical simulations were performed to verify the cause of TV in DWT.
• The discrete delta impulse function was used to test the time-frequency response of DWT and RDWT.
• Actual EEG signals were decomposed using both DWT and RDWT to compare performance.

Main Results:

• DWT exhibits significant sensitivity to the translation of the delta impulse function, causing feature distortions.
• RDWT, by eliminating downsampling, demonstrates translation invariance, maintaining consistent decomposition results.
• These findings were validated through the decomposition of real EEG data.

Conclusions:

• Translation variability in DWT can lead to significant distortions in EEG signal features.
• RDWT provides a more stable and reliable method for EEG signal decomposition due to its translation invariance.
• RDWT is recommended for Brain-Computer Interface (BCI) research and clinical applications for improved performance.