Prewhitening of background brain activity via autoregressive modeling

The detection of steady-state visual evoked potentials (SSVEP) is important in some clinical audiometry and ophthalmology applications. The SSVEPs are usually concealed in the ongoing background electroencephalogram (EEG) generated in the brain. The EEG is highly colored with unknown covariance matrix. In this paper we model the background noise using an autoregressive (AR) model whose parameters are estimated on line, on a block by block basis. The problem of estimating the AR parameters in the presence of the signal and its effect on the bias of the parameter estimates is addressed. We show that in the case of a low level sinusoidal signal, the parameters of the AR model are only slightly perturbed and an accurate estimate of the parameters can be found using the Yule-Walker equations.