English
كوردى
عربي
A Wavelet-Based Architecture for Efficient ECG Signal Denoising
Abstract
An electrocardiogram (ECG) is one of the most important biomedical signals for the detection and diagnosisof heart arrhythmias. As an interpretable biomedical signal, the ECG is subject to various interferences and noise sources,such as baseline wander, 50 Hz power-line interference, and additive white Gaussian noise, all of which may obscure vitaldiagnostic information and distort clinically relevant features. This study aims to develop an efficient, lightweight, andmorphology-preserving denoising method for ECG signals that can suppress multiple noise sources simultaneously.This paper proposes a unified Discrete Wavelet Transform (DWT)-based architecture that combines frequency-selectivesubband filtering with adaptive soft thresholding and multilevel baseline wander removal. The method was evaluated usingboth synthetic ECG signals and real recordings from the MIT-BIH Arrhythmia Database. Performance was assessed usingSNR improvement (SNRimp), correlation coefficient (CC), mean square error (MSE), and percentage root mean squaredifference (PRD). Using the bior6.8 wavelet with soft thresholding, the proposed method achieved SNR improvement up to19.67± 0.37 dB at SNRi =−5 dB and 15.29± 0.33 dB at SNRi = 0 dB, with corresponding CC values of 0.9828± 0.0019and 0.9851± 0.0017, respectively, demonstrating strong noise suppression while preserving ECG morphology. Across alltested SNR levels (−5 to 10 dB), the method consistently maintained CC ≥ 0.98 and PRD below 18% for synthetic data.On challenging real MIT-BIH arrhythmia records (104, 105, 108, 114, 208, 228), the method achieved output SNR upto 20.16± 0.58 dB and CC in the range 0.98–0.995, while preserving key diagnostic features such as the QRS complexand ST-segment. Performance degraded only in record 114 due to severe motion artifacts, which is consistent with priorstudies. Compared with representative traditional and recent deep-learning denoising approaches, the proposed DWT-basedarchitecture achieved superior or competitive performance while remaining computationally efficient and training-free,making it suitable for real-time and wearable ECG applications. Overall, the results confirm that the proposed methodprovides a robust, unified, and clinically reliable solution for multi-noise ECG denoising.