A robust approach to denoise ECG signals based on fractional Stockwell transform

•Method based on a time-frequency analysis tool known as fractional Stockwell transform is presented for ECG denoising.•Performance of the proposed method is dependent on three variable parameters a, p and q respectively, making process more flexible.•The proposed method has showed robust performance against different levels of AWGN of 5 dB, 10 dB and 15 dB levels.•Performance of the proposed method is also validated on noises and artifacts taken from MIT-BIH Noise Stress Test Database. Electrocardiogram (ECG) is a non-invasive technique, used by physicians for prognosis of underlying heart diseases. ECG being the time-varying electric signal evolving from heart, is susceptible to various kinds of low and high-frequency noises. Therefore, this paper presents a novel method for denoising ECG signals, in order to assist physicians to monitor cardiovascular disorders with good accuracy. The proposed denoising method is based on the time-frequency analysis tool known as fractional Stockwell transform (FrST). Method exploits the characteristics of FrST to perform operations in a transformed domain where required signal information is highly concentrated. This aids in suppressing noise from ECG and hence making its analysis effective under noisy environment. Performance of the proposed denoising method is tested on two databases, namely; MIT-BIH Arrhythmia Database and European ST-T Database. The proposed method is used to remove background noise corrupting ECG signal from the time of its acquisition in the form of additive noise. Performance of the proposed method is evaluated by artificially corrupting these ECG signals with additive white Gaussian noise at 5 dB, 10 dB, and 15 dB levels and with real-time noises like baseline wander and motion artifact. Simulation results prove superiority of the proposed method over existing denoising methods in terms of Root-Mean-Square Error, Percent Root Mean Square Difference, and improved Signal-to-Noise Ratio values.