Supported Signal Types¶
compressionKIT is designed for physiological signal compression — signals that originate from the human body and are captured by wearable or clinical sensors. These signals share common properties that make them good candidates for learned compression:
- Quasi-periodic — heartbeat-driven repetitive structure
- Low bandwidth — typically 0.5–40 Hz of useful content
- Low sampling rates — 64–500 Hz (compared to audio at 44 kHz)
- High redundancy — temporal and morphological redundancy across beats
Currently Supported¶
| Signal | Module | Sampling Rate | Frame Size | Status |
|---|---|---|---|---|
| PPG | compressionkit.datasets.ppg |
64 Hz | 320 (5 s) | RVQ published; SPIHT/hybrid local comparisons — models |
| ECG | compressionkit.datasets.ptbxl |
256 Hz | 512 (2 s) | RVQ published; SPIHT/hybrid local comparisons — models |
PPG Section¶
These pages cover the PPG signal path:
- PPG for signal context and preprocessing details.
- PPG Workflow for the end-to-end supported task.
- PPG Models (v1.0) for the five golden reference operating points.
- PPG Codec Demo for the browser and hardware evaluation experience.
ECG Section¶
These pages cover the ECG signal path:
- ECG for signal context and preprocessing details.
- ECG Models (v1.0) for the six golden reference operating points.
Signal Properties Comparison¶
| Property | PPG | ECG |
|---|---|---|
| Typical bandwidth | 0.5–8 Hz | 0.5–40 Hz |
| Morphology | Smooth, pulse-shaped | Sharp QRS complexes |
| Compression challenge | Preserve pulse shape | Preserve QRS timing & amplitude |
| Key clinical metrics | HR, SpO2, HRV | HR, HRV, ST-segment, arrhythmia |
| Typical sensor | Optical (wrist/finger) | Electrode (chest/wrist) |