Quality-Aware Dual-Path Ecg Denoising Under Non-Stationary Noise: A Reproducible Simulation Study

Electrocardiogram (ECG) recordings acquired by wearable and portable devices are affected by baseline wander, power-line interference, broadband noise, and transient motion artifacts. These disturbances can obscure waveform morphology and reduce the reliability of downstream beat analysis. This paper presents a quality-aware dual-path denoising scheme that combines a Savitzky–Golay smoothing path with a baseline-corrected low-pass path and blends their outputs using a local residual-energy estimate. A reproducible 20-second synthetic ECG experiment sampled at 360 Hz was constructed with baseline wander, 50 Hz interference, Gaussian noise, and three non-stationary noise bursts. Performance was assessed using signal-to-noise ratio (SNR), root mean square error (RMSE), and Pearson correlation against the known reference waveform. In this simulation, the proposed dual-path output achieved an SNR of 6.61 dB and an RMSE of 0.0875, while the Savitzky–Golay baseline achieved 7.82 dB and 0.0762 RMSE. The results show that adaptive fusion can improve over the noisy input but does not automatically outperform a well-tuned single filter. The main contribution is a transparent, reproducible evaluation pipeline and a quality-aware fusion design intended for subsequent validation on annotated clinical ECG data. The present results are simulation-based and should not be interpreted as clinical evidence.