Removal of Noises from an ECG Signal Using an Adaptive S-Median Thresholding Technique

The proposed work describes the process of eliminating the real-time noises that gets added during an electrocardiogram (ECG) recording. The real-time noises were taken from MIT-BIH noise stress database and added to clean ECG records, taken from MIT-BIH arrhythmia database. The noisy ECG signal is decomposed using discrete wavelet transform (DWT) algorithm. The signal decomposition is carried out by selecting the mother wavelets that looks similar in shape to an ECG signal. The Mean Square Error (MSE) between the mother wavelet and the clean ECG signal is calculated to choose the wavelet that gives the least MSE. The existing S-median threshold technique is made adaptive in this paper, by opting for threshold values that gives the lowest MSE with respect to cleaner levels. The proposed method is flexible to any decomposition levels and is applicable to both individual and composite noisy signals. The noise-free levels are retrieved back using soft thresholding. The proposed technique is evaluated on two different noise levels to verify its effectiveness. The comparison is carried out with existing thresholding techniques and the proposed method shows highest SNR improvement and lowest MSE values among others. The simulations are performed in Matlab and the time domain results are included to show the removal of unwanted noise with preservation of the important ECG features and its baseline.