Hamming Window-Based Discrete Wavelet Transform Technique for Enhancement of Electroencephalographic (EEG) Signal Processing

Electroencephalographic (EEG) signal is a biomedical signal generated by the electrical activity of the brain. Its waveform provides information about physiological state of the brain and can help medical experts make appropriate decisions on the condition and health of the brain. For the information in electroencephalographic signal to be extracted, it has to be analysed. This analysis can only be reliable if the EEG is devoid of noise. Incidentally, EEG signal is naturally mixed with, in addition to other biomedical signals, powerline interference. It therefore becomes necessary that the EEG which is signal of interest must be extracted from these other signals which constitute noise to it so that health care practitioners can make a reliable interpretation of the signal. In this work, a system for the reliable removal of powerline interference from Electroencephalographic signal based on discrete wavelet transform technique for the signal decomposition was modelled. A new shrinkage function called hamming window-based shrinkage soft thresholding (Ham-WSST) function was developed. An EEG signal was practically captured by measurement from Federal Medical Centre, Owerri and contaminated with 50Hz powerline noise. The sampling frequency is 1000Hz. Based on the new shrinkage function, decomposition level of 7 and daubechies 7(db7) mother wavelet, the denoising of the powerline noise was extensively performed. Seven different decomposition levels including levels 3, 4, 5, 6,7, 8 and 9 were used in sequence and the optimum decomposition level was found to be 7. The performance of the system was evaluated using power spectral density (PSD), signal to noise (SNR), mean square error MSE) and maximum absolute error (MAE). The power spectral density based on the optimum decomposition level of 7 at 0.1 radian normalised frequency was 35.89 dB and signal to noise ratio was 42.26 dB. Also, mean square error was 0.00147 while the maximum absolute error was 0.1147. The developed shrinkage function was compared with two existing shrinkage functions of soft thresholding and hard thresholding functions and the developed function out performed them as can be seen from the fact that the PSD, SNR, MSE, and MAE for soft thresholding shrinkage function are 37.93dB, 40.16 dB, 0.0127 and 0.17462 respectively while for the hard thresholding shrinkage function, the PSD, SNR, MSE and MAE are 39.58 dB, 38.46 dB, 0.00147 and 0.1147 respectively. Less PSD means more noise attenuation at the considered frequency instant. The contributions to knowledge in this work is to make window thresholding shrinkage function based on Hamming window a reliable alternative to other well-known wavelet coefficient shrinkage functions in using wavelet-based systems for practical purposes. The recommendation is that the system is very suitable for hospitals that handle neurological cases and should be deployed to such hospitals and EEG analysis centres.