DSP Overview
Generic and modality-specific digital signal processing for biosignals.
AnyBio's DSP layer turns raw biosignal streams into clean, time-aligned features your application can consume. It runs server-side, so your client doesn't need to bundle filter coefficients or peak detectors - you query feature series the same way you query observations.
The DSP layer has two parts:
- Generic primitives - Butterworth filters, peak detection, FFT, spectral analysis. Building blocks, not application logic.
- Modality extractors - Per-biosignal pipelines that compose primitives into validated, production-grade features (e.g., ECG to HR via Pan-Tompkins; PPG to HR via pulse-foot detection).
What you get out of the box
For each waveform biosignal, the platform produces a set of derived feature series at lower cadences. Your application consumes the feature series; you don't need to run DSP on the device or in the browser.
| Source | Features | Cadence | Algorithm |
|---|---|---|---|
| ECG | hr_ecg (heart rate) | 1 Hz | Pan-Tompkins R-peak detection |
| ECG | hrv_rmssd (HRV, 5s rolling) | 0.2 Hz | Successive R-R differences |
| ECG | hrv_sdnn_5min (HRV, 5min rolling) | 0.2 Hz | Standard deviation of R-R intervals |
| PPG | hr_ppg (heart rate) | 1 Hz | Pulse-foot detection |
| PPG | prv_rmssd (PRV, 5s rolling) | 0.2 Hz | Successive pulse-to-pulse differences |
| PPG | prv_sdnn_5min (PRV, 5min rolling) | 0.2 Hz | Standard deviation of pulse intervals |
Discrete biosignals (HR from a smartwatch, glucose from a CGM, steps from an activity monitor, etc.) are emitted as observations directly without further DSP. The same query API covers both derived features and device-reported values.
Where to go next
Getting Started
Consume feature series from the API and interpret them.
Signal Processing Pipeline
How a raw stream becomes time-aligned features - architecture, primitives, and modality extractors.
Custom Algorithms
When and how a custom DSP algorithm runs against your stream.
Testing Algorithms
QA masks, synthetic streams, and validating output against reference data.