Preprocessing noisy EEG data using time-frequency peak filtering

This paper considers the problem of spectral parameter estimation for electroencephalogram (EEG) data in the presence of white Gaussian noise. A comparison of three known spectral estimation techniques is first presented. The methods work well at high signalto-noise ratio (SNR). However a t low SNR, which often characterises the EEG, these methods fail to produce good spectral estimates. Here, we discuss a new preprocessing technique for filtering noisy data. This technique is based on time-frequency peak filtering. Experimental results show that this method results in improved spectral estimates.