Definition: Pre-whitening is a technique used to eliminate autocorrelations from residuals in preparation for analysis with the General Linear Model (GLM), which is essential for fNIRS signals characterized by
noise—where noise power inversely correlates with frequency (
). This process involves applying a whitening filter W to make the noise distribution uniform, thereby minimizing errors in statistical models like GLM and reducing the False Discovery Rate. The whitening matrix W is derived iteratively by fitting the GLM, analyzing residuals, and estimating W through an autoregressive (AR) model fitted to these residuals. This model adjusts for noise structure by accounting for autocorrelation in the data, and its coefficients are used to construct W. The efficacy of W in reducing autocorrelation depends on the order of the AR model, which specifies the extent of autocorrelation it can address.
Category: Analysis
Related terms: noise, pre-colouring, General Linear Model
Abbreviation or symbol:
Synonym:
References: https://doi.org/10.1117/1.jbo.22.5.055002
https://doi.org/10.1002/1097-0193(200102)12:2%3C61::AID-HBM1004%3E3.0.CO;2-W
DOI:10.1364/BOE.4.001366
Alternative definition:
Related terms (alternative definition):
References (alternative definition):
English