One of factors that influences brain computer interface (BCI’s) with high information transfer rates is selection of appropriate signal processing methods to extract the user's messages and commands from electroencephalograph (EEG).EEG is highly colored and non-gauss signal

the frequency of EEG is irregular and was the same as the event-related potential(ERP).According to the theory that decay of wavelet local modulus maxima of signal and white noise is different

we develop a new method based on Autoregressive Model and wavelet multiresolution analysis.Results from a series of actual experiments show that

this method can extract ERP components from single trial exactly.