Enhancing ECG Signal Quality in Noisy Environments: Comparative Evaluation of Signal Processing, Deep Learning, and Embedded Implementation

Electrocardiogram (ECG) signals are highly susceptible to various noise sources, including baseline wander, power-line interference, muscle activity, and motion artifacts, which hinder accurate clinical interpretation. Numerous denoising techniques, ranging from classical filters to deep learning-based approaches, have been proposed to address this challenge. In this study, we comparatively evaluate several denoising methods—including wavelet transform, empirical mode decomposition, nonlinear filtering, and a convolutional autoencoder—using performance metrics such as SNR, RMSE, and correlation coefficient. Beyond algorithmic comparison, the novelty of this work lies in its implementation-oriented perspective: the denoised output from deep learning is further refined through polynomial regression correction, and its feasibility for real-time deployment is assessed on a resource-constrained microcontroller RP2040 with an ADS117 ADC. Benchmarking results demonstrate that while floating-point evaluation can handle sampling rates up to ~5 kSPS, fixed-point Chebyshev regression enables operation beyond 10 kSPS with significantly reduced latency. These findings highlight not only the effectiveness of hybrid denoising strategies but also their practicality for embedded biomedical devices, bridging the gap between algorithm development and real-world application.