The PAAWS Dataset
We collected data across three complementary data collection protocols—a simulated free-living and laboratory (SimFL+Lab) protocol, a free-living (FL) protocol, and the Sleep protocol—to create a comprehensive multi-modal sensor dataset.
| Protocol | Participants (N =) | Hours/Nights | Duration | Measurements |
|---|---|---|---|---|
| SimFL+Lab | 252 | 717.89 hours | ~4 hours (~1h SimFL + ~3h Lab) |
• Accelerometer & IMU (21 locations) • Heart rate • Metabolic data • Phone sensors & usage • Activity annotations |
| Free Living | 236 (in progress) | 3,288.32 hours | 8-day protocol (7+ full days) |
• Accelerometer (6 locations) • Phone sensors & usage • Activity annotations |
| Sleep | 155 | 157 nights | Up to 2 nights at home |
• Full PSG (EEG, EOG, EMG, ECG) • Respiratory & SpO2 • Accelerometer (5 locations) • Expert sleep scoring |
Choose Protocol to Explore
Data Processing Progress
| Protocol | Collected (N =) | Processed | Annotated | QC Complete | Released |
|---|---|---|---|---|---|
| SimFL+Lab Protocol | 252 | 252 | 252 | 252 | 252 |
| Free Living Protocol | 236 | 236 | 137* | 22 | 20 |
| Sleep Protocol (PSG) | 157 | 155 | 157 | 15 | 15 |
*Free Living Status
To release the 137 annotated datasets:
- 16 require experts' approval
- 61 awaiting QC
- Remaining datasets are in progress
SimFL+Lab Protocol
The Simulated free-living and lab (SimFL+Lab) protocol was an approximately 4-hour supervised session: ~1 hour minimally controlled simulated free-living activities followed by ~3 hours of highly controlled laboratory exercise protocol
Accelerometer & IMU Data
Comprehensive motion capture from 21 ActiGraph GT9X Link sensors (20 on-body locations + 1 taped to back of the phone)
DATA COLLECTED
- Accelerometer: 80 Hz
- IMU (gyroscope, magnetometer, temperature): 100 Hz
SENSOR LOCATION (21 LOCATIONS)
- Wrists: Right/Left wrist top and bottom (4 sensors)
- Waist: Right/Left waist anterior, mid, posterior (6 sensors)
- Thighs: Right/Left thigh anterior (2 sensors)
- Ankles: Right/Left ankle anterior, lateral, medial, posterior (8 sensors)
- Phone: Back of the participant's phone
DEVICE
- ActiGraph GT9X Link
Heart Rate & Metabolic Data
Continuous heart rate monitoring during protocol, plus metabolic measurements during 13 specific laboratory activities
DATA COLLECTED
- Heart Rate: 1 Hz throughout entire protocol (Note: Sixteen participants have 60-second epoch data due to configuration error)
- Metabolic (13 activities): Breath-by-breath VO2, VCO2, respiratory rate, tidal volume
- Calculated MET values for steady-state activities
- Activities: lying, sitting, standing, treadmill (multiple speeds), stationary biking, household tasks
SENSOR LOCATION
- Heart Rate: Chest strap
- Metabolic: Face mask with breathing apparatus
DEVICE
- Polar H10 heart rate monitor
- Cosmed Quark RMR CPET indirect calorimeter
Built-in Phone Sensors & Usage Data
Available only for participants with compatible Android devices—comprehensive smartphone usage tracking plus built-in phone sensors
DATA COLLECTED
- External Accelerometer: 80 Hz (GT9X taped to phone back)
- Built-in Accelerometer: ~50 Hz target rate
- Environmental: Ambient pressure (hPa), light (lux), proximity (cm)
- Usage: App interactions (screen on/off, foreground/background), notification logs, hourly step counts
SENSOR LOCATION
- One ActiGraph GT9X taped to back of participant's phone
- Built-in phone sensors (internal to device)
DEVICE
- ActiGraph GT9X Link (external)
- Participant's Android smartphone (built-in sensors)
SimFL Protocol Activities
Sixteen minimally controlled real-world scenarios simulating typical free-living activities (performed in ~1 hour)
Realistic Outdoor Ambulation
- Public transit use (walking to train station, waiting for train, riding train)
- Walking variations (while on phone call, carrying groceries)
- Stair climbing (with and without groceries)
- Outdoor biking around common area
- Elevator use (up and down)
Sedentary Behaviors
- Sitting on bench (naturally, conversing with researcher, reading, using phone)
- Waiting in line at coffee shop
Realistic Free-Living Activities
- Grocery shopping simulation (selecting and returning items)
- Playing frisbee with researcher
Lab Protocol Activities
Highly controlled activities (32 activities performed in ~3 hours under precise supervision)
Activities Include
- Postures: Lying (back, sides, stomach), sitting (still, typing, reclining), standing (still, conversation)
- Ambulation speeds: Treadmill at multiple speeds (2-5.5 mph) with variations (hands in pockets, carrying items, phone use, conversation)
- Exercise: Stationary biking, arm curls, push-ups, crunches, machine exercises
- Household: Folding laundry, shelf loading/unloading, organizing, sweeping, vacuuming, washing dishes, chopping food
Activity Annotations
Second-by-second manual annotations from video footage (video not released, only derived annotations)
DATA COLLECTED
- Physical activity type (mutually exclusive)
- Posture (mutually exclusive)
- High-level behavior (non-mutually exclusive)
- Contextual parameters (non-mutually exclusive)
- Experimental situation (Lab vs SimFL)
- Expert manual annotation with multi-stage quality control
SENSOR LOCATION
- Primary: Third-person video camera following participant
- Backup: Egocentric front-facing camera (used in 6 cases of corruption and 35 cases of poor footage quality)
DEVICE
- C3 Cammpro camera (third-person)
- Front-facing camera (egocentric backup)
Protocol Documentation
Comprehensive notes documenting deviations, sensor issues, and participant-specific circumstances
Documentation Types
- General participant behavior notes
- Protocol deviation records (structured CSV)
- Sensor malfunction logs
- Sensor non-wear periods
- Activity modifications or skipped activities
- Equipment issues
Free Living Protocol
Eight-day naturalistic data collection providing 7+ full days of continuous sensor data, capturing complete weekly rhythms and day-to-day variations in natural settings. Participants performed normal daily activities while wearing sensors and a chest-mounted camera during waking hours
Accelerometer Data
Continuous week-long motion tracking from 6 ActiGraph GT9X Link sensors (5 on-body + 1 taped to back of the phone)
DATA COLLECTED
- Accelerometer: 80 Hz (IMU disabled for week-long battery life)
- Thigh sensor worn continuously (waterproof)
- Other sensors removed for water activities
SENSOR LOCATION (6 SENSORS)
- Right wrist (dorsal aspect)
- Left wrist (dorsal aspect)
- Right waist (anterior axillary line, rotated 90° from lab placement)
- Right thigh (anterior, waterproof tape)
- Right ankle (lateral aspect, rotated 90° from lab placement)
- Phone: Back of the participant's phone
DEVICE
- ActiGraph GT9X Link
Activity Annotations
Comprehensive second-by-second activity labels created from expert analysis of chest-mounted camera video (video not released)
DATA COLLECTED
- Physical activity type (mutually exclusive)
- Posture (mutually exclusive)
- High-level behaviors (non-mutually exclusive)
- Contextual parameters (non-mutually exclusive)
- Manual annotation from chest-mounted camera footage
- Video gaps filled with self-reported recall data
- Multi-stage quality control validation
SENSOR LOCATION
- Chest-mounted camera worn during waking hours
DEVICE
- Chest-mounted camera
Built-in Phone Sensors & Usage Data
Available only for participants with compatible Android devices—naturalistic smartphone usage patterns and built-in sensors
DATA COLLECTED
- External Accelerometer: 80 Hz (GT9X taped to phone back)
- Built-in Accelerometer: ~50 Hz target
- Environmental: Ambient pressure, light exposure (lux), proximity sensor readings
- Usage: App interactions (foreground/background), screen state changes, notification patterns, hourly step counts
SENSOR LOCATION
- One ActiGraph GT9X taped to back of participant's phone
- Built-in phone sensors (internal to device)
DEVICE
- ActiGraph GT9X Link (external)
- Participant's Android smartphone (built-in sensors)
Protocol Documentation
Comprehensive notes documenting deviations, sensor issues, and participant-specific circumstances
Documentation Types
- Daily recall sessions (structured CSV)
- Activities during camera non-wear periods
- Sleep quality ratings and duration
- Protocol deviation records
- Sensor malfunction logs
- Sensor non-wear periods
- Notable events or deviations
Sleep Protocol
Optional protocol occurring during Free Living: comprehensive at-home polysomnography collected in participants' natural sleep environment with expert sleep stage scoring following AASM standards
EEG Signals
10-lead electroencephalography following 10-20 system for comprehensive brain activity monitoring during sleep
DATA COLLECTED
- EEG Channels: C3-M2, C4-M1, F3-M2, F4-M1, O1-M2, O2-M1
- Reference: M1/A1 (left mastoid), M2/A2 (right mastoid)
- Expert manual scoring by trained sleep researchers
- Five stages: Wake, N1, N2, N3, REM
- 30-second epoch scoring following AASM standards
- Arousal and sleep/wake transition identification
SENSOR LOCATION
- Central: C3, C4 (scalp)
- Frontal: F3, F4 (scalp)
- Occipital: O1, O2 (scalp)
- Reference: M1, M2 (mastoids behind ears)
DEVICE
- Nox A1 PSG system
Eye Movement & Muscle Activity
Electrooculography for REM detection and electromyography for muscle tone assessment
DATA COLLECTED
- EOG: E1-M2, E2-M2 for REM sleep identification and sleep/wake transitions
- EMG: Submental (chin) EMG derivations 1-2, 1-F, 2-F
- EMG: Bilateral anterior tibialis (legs)
- Muscle tone assessment for sleep staging
- Periodic limb movement detection
SENSOR LOCATION
- EOG: E1 (1 cm lateral and below left outer canthus), E2 (1 cm lateral and above right outer canthus)
- EMG Chin: Submental (below chin)
- EMG Legs: Bilateral anterior tibialis muscles
DEVICE
- Nox A1 PSG system
Respiratory Monitoring
Multi-modal respiratory assessment including airflow, effort, and pattern analysis for sleep apnea detection
DATA COLLECTED
- Airflow: Nasal cannula pressure and flow
- Derived: Flow limitation signal, RIP sum, RIP flow
- Effort: Thoracic (chest), abdominal movement via RIP belts
- Snoring: Audio volume in dB (no audio recording)
- Apnea and hypopnea identification
- Central vs obstructive event classification
SENSOR LOCATION
- Nasal Cannula: Below nostrils
- Thoracic Belt: Around chest
- Abdominal Belt: Around abdomen
DEVICE
- Nox A1 PSG system with nasal cannula and RIP belts
Cardiovascular & Oximetry
Single-lead ECG for heart rhythm and pulse oximetry for oxygen saturation throughout sleep
DATA COLLECTED
- ECG: Single-channel ECG (Lead II configuration)
- Derived: Heart rate, bradycardia (BPM <40 for ≥20s), tachycardia (bpm>100 for ≥20s)
- SpO2: Oxygen saturation, PPG waveform, pulse rate
- Desaturation event detection (≥3% drop)
- Signal quality assessment and artifact detection
SENSOR LOCATION
- ECG: Chest (Lead II placement)
- Pulse Oximeter: Index finger
DEVICE
- Nox A1 PSG system with pulse oximeter
Sleep Events & Position
Comprehensive detection and expert scoring of sleep disorders and body position throughout the night
DATA COLLECTED
- Breathing Events: Central, obstructive, and mixed apneas (≥10s duration)
- Hypopneas: Central and obstructive (≥10s, ≥30% amplitude decrease, with ≥3% desaturation or arousal)
- Movement Events: Periodic limb movements, limb movement series, major body movements
- Arousals: EEG arousals (≥3s frequency shift)
- Snoring: Individual snores and snore trains (≥4 consecutive snores)
- Body Position: Supine, prone, left, right, upright (auto-detected from gravity sensors)
SENSOR LOCATION
- Derived from combined PSG sensors (respiratory, EEG, EMG, gravity)
DEVICE
- Nox A1 PSG system
Concurrent Accelerometer Data
Participants continued wearing their 5 Free Living accelerometers throughout sleep for sensor fusion opportunities
DATA COLLECTED
- Accelerometer: 80 Hz
- Synchronized with PSG data
- Enables comparison of wrist actigraphy with gold-standard PSG
SENSOR LOCATION (5 SENSORS)
- Left wrist (dorsal)
- Right wrist (dorsal)
- Right waist (anterior)
- Right thigh (anterior, waterproof tape)
- Right ankle (lateral)
DEVICE
- ActiGraph GT9X Link
At-Home Natural Sleep
PSG data collected in participants' own homes during natural sleep, maximizing ecological validity
Data Collection Approach
- Nox A1 PSG system configured for home use
- Sensors applied in lab, then participant goes home
- Self-applied nasal cannula and pulse oximeter at bedtime
- Natural sleep in familiar environment
- Bio-calibration before sleep
Quality & Standards
- Following AAST Technical Guideline for Standard Polysomnography (AASM)
- Expert manual scoring by trained sleep researchers
- Signal quality monitoring and artifact detection
- Impedance checks for EEG/EOG electrode quality
Dataset Releases
Release 1 (October 2025): Initial Dataset Release
Complete, validated data from 252 participants with expert annotations. All sensor data synchronized to sub-second precision with multi-stage quality validation and comprehensive documentation.
Release 1 Dataset Composition
| Protocol | Participants (N =) | Hours / Nights |
|---|---|---|
| SimFL+Lab | 252 | 717.89 hours |
| Free Living | 20 | 3,288.32 hours |
| Sleep | 15 | 29 nights |
Download Dataset
Access the complete PAAWS Release 1 dataset with all sensor data, expert annotations, and comprehensive documentation.
Download PAAWS Release 1