Datasets on PhysioNet
Source for all datasets: LIP6 Nuage Repository
ECG & PPG Signal with Arrhythmia Episodes – 2022
Simulation model for ECG and PPG signals with arrhythmia episodes
A dedicated software tool capable of generating synthetic ECG and PPG signals containing a broad range of arrhythmic events (e.g., atrial fibrillation, bradycardia, ventricular tachycardia). Key features include:
- Normal sinus rhythm
- Atrial fibrillation (AF)
- Bradycardia
- Ventricular tachycardia (VT)
- Atrial premature beats (APB)
Realistic measurement noise can be superimposed to mimic real acquisition conditions. Users can configure signal duration, sampling frequency (PPG: 75–1000 Hz; ECG: 250–1000 Hz), abnormal episode timing, and noise type/level. This simulator facilitates generation of realistic cases to augment training datasets.
Motion Artifact Contaminated fNIRS and EEG – 2014
Simultaneous fNIRS and EEG recordings with controlled motion artifacts
fNIRS and EEG signals were recorded simultaneously in an experimental setting using two sensor groups: one deliberately exposed to motion artifacts and one kept still. A triaxial accelerometer captured the motion data.
- fNIRS: ~25 Hz at two wavelengths (690 nm and 830 nm)
- EEG: 2048 Hz
- Accelerometer: 200 Hz
CSV File Details
fNIRS/EEG signals and accelerometer data were acquired on independent systems but synchronized by trigger signals.
fNIRS data (9 experimental sessions, with sessions 5 and 8 of lower quality): two optical channels per wavelength.
EEG data (23 recordings): two frontal channels.
Trigger coding:
- fNIRS trigger: rising edge = experiment start; low level = motion‑artifact phase; high level = clean phase; final drop = experiment end.
- EEG trigger: only start (rise) and end (fall) markers; no phase segmentation.
fNIRS data structure:
| Column | Description |
|---|---|
| 1 | Sample number |
| 2 | Raw light intensity 690 nm – Channel 1 (25 Hz) |
| 3 | Raw light intensity 830 nm – Channel 1 (25 Hz) |
| 4 | Raw light intensity 690 nm – Channel 2 (25 Hz) |
| 5 | Raw light intensity 830 nm – Channel 2 (25 Hz) |
| 6 | fNIRS trigger (25 Hz) |
| 7–9 | Accelerometer 1 – X/Y/Z (200 Hz) |
| 10–12 | Accelerometer 2 – X/Y/Z (200 Hz) |
| 13 | Accelerometer trigger (200 Hz) |
EEG data structure:
| Column | Description |
|---|---|
| 1 | Sample number |
| 2 | Raw EEG – Channel 1 (2048 Hz) |
| 3 | Raw EEG – Channel 2 (2048 Hz) |
| 4 | EEG trigger (2048 Hz) |
| 5–7 | Accelerometer 1 – X/Y/Z (200 Hz) |
| 8–10 | Accelerometer 2 – X/Y/Z (200 Hz) |
| 11 | Accelerometer trigger (200 Hz) |
Note: Channel 1 is generally motion‑free, whereas Channel 2 is deliberately moved.
ScientISST MOVE – 2024
Multimodal biosignal recordings in natural living environments with annotated daily activities
Seventeen participants were monitored for ~37 minutes each during natural activities (standing, walking, running, chair displacement, greeting, etc.), with precise activity annotations.
Devices and sampling frequencies:
- ScientISST‑Chest and ScientISST‑Forearm: ECG, EMG, EDA, and finger PPG at 500 Hz.
- Empatica E4 wristband: PPG, EDA, skin temperature, and accelerometry.
| Signal type | Sampling frequency (ScientISST / E4) |
|---|---|
| ECG (gel electrodes) | 500 Hz |
| PPG | 500 Hz / 64 Hz |
| EDA | 500 Hz / 4 Hz |
| EMG | 500 Hz |
| Accelerometer (chest/wrist) | 500 Hz / 32 Hz |
| Temperature | – / 4 Hz |
BIG IDEAs – 2023
Glycemic variability and wearable-device data
Continuous glucose monitoring combined with Apple Watch or Empatica E4 recordings provides heart rate, accelerometry, blood volume pulse (PPG), electrodermal activity, and temperature.
- Glucose measurement every 5 min
- PPG sampling at 64 Hz (sufficient for waveform analysis)
- 16 participants including controlled food intake
Food_Log_xxx.csv files detail nutritional intake (type, time, amount, calories, carbohydrates, proteins, lipids, fibers, etc.). The dataset supports time‑aligned PPG–glucose analysis.
Labeled Raw Accelerometry Data – 2021
Annotated raw accelerometry during walking, stair climbing/descending, and driving
Thirty‑two healthy adults (13 males, 19 females) each wore four ActiGraph GT3X+ accelerometers (left wrist, left hip, left ankle, right ankle) sampling at 100 Hz.
Activities: walking ~1 km, ascending/descending stairs six times, and driving ~12.8 miles. Each trial began and ended with a hand clap for synchronization.
Each subject file contains:
| Parameter | Meaning |
|---|---|
| activity | Activity code |
| time_s | Elapsed time (s) |
| lw_x,y,z | Left wrist axes |
| lh_x,y,z | Left hip axes |
| la_x,y,z | Left ankle axes |
| ra_x,y,z | Right ankle axes |
Activity codes: 1=walk, 2=downstairs, 3=upstairs, 4=drive, 77=clap, 99=off‑protocol.
Stress and Structured Exercise Sessions – 2025
Physiological recordings during induced stress and structured exercise
Recordings with Empatica E4 wristbands under three protocols:
- Acute Stress (STRESS) – alternating mental arithmetic and emotional stimulation with rest; subjective stress levels logged (two CSVs per subject).
- Aerobic Exercise (AEROBIC) – moderate, rhythmic cycling.
- Anaerobic Exercise (ANAEROBIC) – short, high‑intensity cycling.
Sample sizes: 36 (STRESS), 30 (AEROBIC), 31 (ANAEROBIC).
BVP.csv provides 64 Hz PPG for heart‑rate, HRV, and waveform‑quality analysis.
| Accelerometer | PPG | EDA | Heart rate | IBI | Events | Temperature |
|---|---|---|---|---|---|---|
| ACC.csv | BVP.csv | EDA.csv | HR.csv | IBI.csv | tags.csv | TEMP.csv |
BigIdeasLab_STEP – 2021
Skin‑tone effects on optical heart‑rate sensing in smartwatches
Assesses how skin tone (Fitzpatrick types 1–6), activity type, and device model influence optical heart‑rate accuracy.
Fifty‑three participants (32 females, 21 males; ages 18–54) balanced across six skin‑tone categories completed a standardized protocol repeated three times:
- Seated rest – 4 min
- Paced breathing – 1 min
- Brisk walking (≈50 % HRmax) – 5 min
- Seated rest – 2 min
- Keyboard typing – 1 min
Reference ECG was recorded with a Bittium Faros patch at ~1000 Hz. Only heart‑rate values (BPM) from multiple devices are provided—no raw PPG.
| Parameter | Description |
|---|---|
| ECG | Reference (Bittium Faros 180) |
| Apple Watch | Apple Watch 4 |
| Empatica | Empatica E4 |
| Fitbit | Fitbit Charge 2 |
| Garmin | Garmin Vivosmart 3 |
| Miband | Xiaomi Miband 3 |
| Biovotion | Biovotion Everion |
| Skin tone | Fitzpatrick type (1–6) |
| ID | Participant identifier |
| Activity | Rest, exercise, breathing, typing |
Conclusion
Most datasets cited here employed the Empatica E4 to acquire PPG, EDA, and temperature signals.
However, the E4 has been superseded by EmbracePlus, which offers lower energy consumption and supports up to four concurrent measurement channels.