Wavelet denoising is a signal processing technique that decomposes a CSI signal into multi-resolution frequency-time components using wavelet transforms, then suppresses noise by thresholding or discarding coefficients that fall below a significance level before reconstructing the cleaned signal. In WiFi CSI sensing, it is particularly valuable because raw CSI measurements are contaminated by environmental multipath interference, hardware noise, and minor irrelevant movements, all of which can obscure subtle features tied to human presence, count, or activity. Key variants applied in this domain include soft and hard thresholding strategies applied to discrete wavelet transform (DWT) decompositions, as well as multi-level wavelet packet decomposition, which provides finer frequency band resolution and is especially useful when noise and signal energy overlap spectrally across the 30 OFDM subcarrier streams typical of commodity WiFi MIMO systems.
Dictionary term