Description
The wavelet transform decomposes a signal into time-localised components at multiple scales, giving better time-frequency localisation than the STFT for non-stationary signals. In CSI sensing it appears as a pre-processing step for noisy amplitude time-series — denoising, motion-burst extraction, and multi-scale activity features.
When it's used
- Wavelet denoising of raw CSI amplitude
- Multi-scale activity-burst features for HAR
- Hand-crafted features fed into classical classifiers
Limitations
- Mother-wavelet choice is empirical
- Less popular than STFT in modern deep pipelines
- Computationally heavier than FFT