Project Overview
StreamSense is a synchronized multi-sensor data acquisition platform designed for psychophysiological research, enabling real-time collection and analysis of biosignals from multiple sources with precise temporal alignment. ProSense is the companion analysis tool providing interactive visualization and signal processing capabilities.
Institution: Hasso Plattner Institute, University of Potsdam
Role: Lead Researcher, System Architect, Developer
Technical Specifications
Hardware Integration
- Muse EEG Headband: 4-channel frontal EEG @ 256Hz
- Empatica E4: PPG (64Hz), GSR (4Hz), accelerometer
- Synchronization: Lab Streaming Layer (LSL) protocol
Software Stack
- Backend: Python, LSL, Pandas, NumPy, SciPy
- Stimulus: PsychoPy framework
- Analysis: Matplotlib, Seaborn, MNE-Python
- Storage: HDF5 format
Key Results
The stress-yoga validation study demonstrated measurable physiological changes:
| Metric | Baseline | Stress Peak | Post-Yoga | Recovery |
|---|---|---|---|---|
| Alpha Power (EEG) | 12.3 μV² | 6.8 μV² ↓ | 14.1 μV² ↑ | +107% |
| Heart Rate | 68 bpm | 89 bpm ↑ | 62 bpm ↓ | -30% |
| HRV (RMSSD) | 42 ms | 28 ms ↓ | 56 ms ↑ | +100% |
All metrics verified through repeated-measures ANOVA (p < 0.001). Statistical methodology detailed in master's thesis.
Technical Challenges & Solutions
1. Sensor Synchronization Nightmare
Problem: Each sensor had independent clocks drifting 200-300ms over 30 minutes.
Solution: Used motion artifacts visible in both EEG and accelerometer as a "ground truth" for cross-correlation alignment. Validated <50ms sync across 48 sessions.
2. Motion Artifacts During Yoga
Problem: Strong EEG oscillations during movement were actually motion artifacts, not brain waves.
Solution: Implemented ICA-based artifact rejection validated against manual annotations. Achieved 94% EEG epoch quality (up from initial 30%).
3. Buffer Overflow During Recording
Problem: System crashed during session 12 due to single-threaded architecture overwhelmed by high-frequency streams.
Solution: Rewrote architecture with multi-threading: separate threads for acquisition, storage, and monitoring. Zero crashes in subsequent 36 sessions.
What I Learned
- Temporal precision is non-negotiable in multi-modal research. Hardware limitations require sophisticated software compensation.
- Personalized protocols can be more scientifically rigorous than standardized ones when studying subjective experiences.
- Real-time systems fail in production even if they work in testing. Design for worst-case scenarios.
- Listening to participants is as important as reading papers—real-world feedback beats literature review sometimes.