Description
Choose a sample rate and antialiasing filter, then remove drift, mains hum and out-of-band noise from biosignals with digital filters chosen for the roll-off versus phase trade-off.
Every physiological measure in UX research (EEG, ECG and heart rate, EDA, EMG) passes through the same chain: transducer, amplifier, analogue filters, an antialiasing filter, then the analogue-to-digital converter. Frequencies above half the sample rate must be removed before digitising or they reappear as false aliases, and sampling at about five times the highest frequency of interest is common. After recording, digital filters remove DC offset and slow drift, isolate the band of interest, and strip mains interference at 50 or 60 Hz. Filters are characterised by pass, stop and transition bands, Bode plots and impulse responses, and implemented as IIR or FIR designs: Butterworth is the usual compromise, Bessel preserves waveform shape, and Chebyshev and elliptic give steeper roll-off at the cost of ripple. IIR filters shift phase, so forward-backward (zero-phase) filtering is used when timing matters. Poor choices here silently distort every downstream metric.
Attributes
- Discipline
- Neuro, Biometrics & Affect
- Loss type
- Siloed
- Stage
- Evaluation
- Data
- Quantitative
- Effort
- Days
Steps
- Decide the highest frequency of interest and sample well above twice it, commonly about five times.
- Apply an analogue antialiasing filter below the Nyquist frequency before digitisation.
- Remove DC offset and slow drift by demeaning or high-pass filtering.
- Apply a band-pass filter matched to the signal of interest (for example 15-45 Hz for QRS detection in ECG).
- Remove mains interference with a band-stop (notch) filter around 50 or 60 Hz, without relying on it to rescue a poor recording.
- Choose the filter type by the trade-off between roll-off and phase response, using forward-backward filtering when event timing matters.
- Inspect the Bode plot and impulse response before applying a filter, and document all settings.
Sources
- Signal Processing for Neuroscientists (van Drongelen, 2018)
Also known as
- Antialiasing and sampling-rate choice
- Band-pass and notch filtering of biosignals
- Zero-phase filtering
- Mains hum removal
- Filter type selection (Butterworth, Bessel, Chebyshev, elliptic)
- Biosignal preprocessing
Categories
- Analytics & Behaviour
- Neuro, Biometrics & Affect