Lecture 3.2.5: Signal Preprocessing ECG, PPG + Feature Extraction, Windowing & HRV Spectral Features
Why It Matters
Robust preprocessing transforms noisy biosignals into trustworthy HRV metrics, directly influencing patient monitoring, wearable health analytics, and performance‑optimization decisions.
Key Takeaways
- •Raw ECG/PPG signals require multi-stage preprocessing to extract insights
- •Notch, high‑pass, low‑pass filters remove power line, baseline, EMG noise
- •Band‑pass (0.5‑40 Hz) and zero‑phase filtering preserve R‑peak timing
- •Sliding windows with overlap enable continuous HRV analysis across long recordings
- •RMSSD metric reflects autonomic balance; high values indicate recovery, low stress
Summary
The lecture walks through converting raw ECG and PPG voltages into actionable physiological metrics, focusing on preprocessing, feature extraction, and heart‑rate‑variability (HRV) spectral analysis.
Aksha outlines three dominant noise sources—power‑line interference, baseline wander, and EMG artifacts—and recommends a “Goldilocks” filter chain: notch for 50/60 Hz spikes, high‑pass for low‑frequency drift, low‑pass for high‑frequency muscle noise, and a 0.5‑40 Hz band‑pass with zero‑phase forward‑backward filtering to preserve R‑peak timing.
Using a bread‑slicing analogy, the talk describes windowing long recordings into 30‑second segments with 5‑second overlaps to avoid edge loss. The Pan‑Tompkins algorithm (differentiation, squaring, integration) is demonstrated in Python with NumPy, Matplotlib and SciPy on a synthetic ECG contaminated by 50 Hz line noise and baseline wander.
Accurate preprocessing enables reliable extraction of R‑peak intervals and HRV indices such as RMSSD, which differentiate sympathetic stress from parasympathetic recovery. These metrics underpin wearable health monitors, ICU patient tracking, and sports‑science training programs, making robust signal pipelines a prerequisite for clinical and commercial applications.
Comments
Want to join the conversation?
Loading comments...