Denoising semi-simulated EEG signal contaminated ocular noise using various wavelet filters

Abstract

Denoising is crucial in electroencephalography (EEG) processing to remove undesired components contaminated in a signal. Wavelet filters are a powerful and robust denoising approach to eliminate the noises in EEG. However, a broad number of wavelet families and decomposition levels confused the selection of the optimal and most appropriate wavelet filter. Therefore, this study aims to determine the optimal wavelet filter based on the signal-to-noise ratio (SNR) for EEG denoising. This work used the semi-simulated EEG signal contaminated with ocular noise as the observed signal. The wavelet filter with various wavelet families that is Haar, Daubechies (db), Symlets (sym), coiflets (coif), Discrete Meyer (dmey), Fejer-Korovkin (fk), biorthogonal (bior), and Reverse Biorthogonal (rbior) from decomposition level 1 to 8 were applied. A MATLAB wavelet toolbox with a soft thresholding method was used to denoise the desired signal. The result showed that the highest SNR value was 63.0172 dB. The highest SNR indicated that the filter had a high ability to remove the noises in EEG signals. Therefore, this work suggested that the haar, db1, bior1.1, and rbior1.1 of the mother wavelet at decomposition level 8 were the most efficient for removing the ocular noise.