Denoising of mixed noises in ECG with separate noise estimators based on discrete wavelet transform

ECG may contain different types of noise. For example, it may be corrupted simultaneously by baseline wander and some kinds of high-frequency noise. A technique capable of removing mixed noises using separate noise estimators based on discrete wavelet transform (DWT) is proposed in this paper. By applying a multilevel DWT to the noisy ECG a set of detail and approximation coefficients was used to estimate different types of noise. The approximation coefficients are used to reconstruct a waveform which is in turn used as the estimate of the low-frequency noise, while parts of the detail coefficients are used to reconstruct a waveform which is in turn used as the estimate of the high-frequency noise. This is useful for the case in which the noise spans over a wide range of frequencies. The number of levels used for each noise estimator is dependent on the noise. Because only parts of detail coefficients are used in the high-frequency noise estimator, significant reduction of computation is obtained compared to previous methods using adaptive thresholding techniques. Simulation results show significant improvement compared to traditional approaches using empirical mode decomposition.