Signal filtering preview

scipy.signal designs digital filters (lowpass, highpass, bandpass) and applies them with filtfilt for zero-phase smoothing of sensor data.

Common workflow

1. Design filter with butter, cheby1, etc.
2. Get coefficients b, a or SOS form
3. Apply sosfiltfilt for forward-backward filtering

Moving average alternative

Simple np.convolve or signal.savgol_filter for denoising without full IIR design—pick based on latency and phase requirements.

Butterworth lowpass sketch

import numpy as np
from scipy import signal

b, a = signal.butter(4, 0.1, btype='low')
x = np.random.default_rng(0).normal(size=100)
y = signal.filtfilt(b, a, x)
print('input std:', x.std(), 'filtered std:', y.std())

Important interview questions and answers

1. Q: filtfilt vs lfilter?
   A: filtfilt runs forward and backward for zero phase delay—better for offline analysis.
2. Q: Cutoff normalized?
   A: For butter, Wn is fraction of Nyquist (0–1), not Hz—scale by sample rate.

Self-check

1. What does a lowpass filter remove?
2. Why normalize cutoff to Nyquist?

Tip: filtfilt is for offline analysis; real-time streams often use causal lfilter.