Problem
Respiratory rate is clinically useful and annoying to instrument. A dedicated belt is extra hardware. A single-lead ECG is often already on the patient.
Approach
A DSP pipeline that estimates respiratory rate from that ECG:
- CEEMDAN decomposition to isolate respiratory modulation of the cardiac signal.
- Discrete wavelet transforms and peak detection to recover breath timing.
- Evaluation on the BIDMC dataset against reference respiratory recordings.
Result: 2.5 bpm mean absolute error. Built during a biomedical signal processing internship at Vigo, alongside accelerometer-based posture classifiers for the same wearable monitoring product.
No deep network. The constraint was a signal you already have, and an error you can explain.