Abstract
•A clean ECG data is modeled as a combination of different components.•The components are disjoint in the time domain, and they have overlapped spectral coefficients.•A segmentation procedure is developed to segment the measured data.•Ideal filters are designed to filter each segment.•The proposed algorithm outperforms many of the existing techniques.
A new algorithm for denoising ECG data contaminated by wideband noise is proposed in this paper. In the proposed algorithm, a clean ECG data is modeled as a combination of different components. The components have the characteristics that they are disjoint in the time domain, their spectral coefficients overlap in the frequency domain, and they have different bandwidths. Based on this model, a successive local filtering approach is suggested in this paper to remove wideband noise from a recorded ECG data. In the proposed algorithm, a segmentation procedure is first developed to segment the recorded ECG data such that each segment approximately contains one dominant component. The denoised ECG signal is then constructed by successively denoising the constructed segments using ideal filters. The ideal filters are designed in the frequency domain by minimizing a penalized least-squares objective function, where the weighted ℓ0-norm is utilized as the penalty term to encourage the on-off group-sparsity of the ideal filters. In the proposed algorithm, the BW of each ideal filter is automatically adjusted to the BW of the dominant component in the analyzed segment. Simulation results on simulated and real ECG data show that the proposed algorithm can be successfully utilized to denoise ECG data contaminated by wideband noise. In addition, the proposed algorithm is also shown to produce significantly improved results compared to some existing ECG denoising techniques.