Signal Denoising Task

Overview

The objective of denoising is to remove noise and artifacts from physiological signals while preserving the underlying signal information. The dominant noise sources include baseline wander (BW), muscle noise (EMG), electrode movement artifacts (EM), and powerline interference (PLI). For physiological signals such as ECG and PPG, removing the artifacts is difficult due to the non-stationary nature of the noise and overlapping frequency bands with the signal. While traditional signal processing techniques such as filtering and wavelet denoising have been used to remove noise, deep learning models have shown great promise in enhanced denoising.

---

Noise Characteristics

The following table summarizes the characteristics of common noise sources in ECG signals:

Type	Causes	Spectrum	Effects
Baseline Wander (BW)	Respiration, posture changes	0-1.0 Hz	Distorts ST segment and other LF components
Powerline Interference (PLI)	Electrical equipment	50-60 Hz	Distorts P and T waves
Muscle Noise (EMG)	Muscle activity	0-100 Hz	Distorts local waves
Electrode Movement (EM)	Electrode motion, skinimpedance	0-100 Hz	Distorts local waves

The following table summarizes the characteristics of common noise sources in PPG signals:

Type	Causes	Spectrum	Effects
Motion Artifacts	Movement, pressure	0-10 Hz	Distorts signal
Ambient Light	Sunlight, artificial light	0-100 Hz	Distorts signal
Blood Pressure	Blood flow, pressure	0-10 Hz	Distorts signal

---

Dataloaders

Dataloaders are available for the following datasets:

• LUDB
• PTB-XL
• ECG Synthetic
• PPG Synthetic

---

Pre-trained Models

The following table provides the latest performance and accuracy results of denoising models. Additional result details can be found in Model Zoo → Denoise.

NAME	DATASET	FS	DURATION	MODEL	PARAMS	FLOPS	METRIC
DEN-TCN-SM	Synthetic, PTB-XL	100Hz	2.5s	TCN	3.3K	1.0M	18.1 SNR
DEN-TCN-LG	Synthetic, PTB-XL	100Hz	2.5s	TCN	6.3K	1.8M	19.5 SNR
DEN-PPG-TCN-SM	Synthetic	100Hz	2.5s	TCN	3.5K	1.1M	92.1% COS

---

References

• DeScoD-ECG: Deep Score-Based Diffusion Model for ECG Baseline Wander and Noise Removal
• DEEPFILTER: AN ECG BASELINE WANDER REMOVAL FILTER USING DEEP LEARNING TECHNIQUES