Combining AR filter and sparse Wavelet representation for P300 speller

A variety of experimental paradigms have been proposed in the field of Brain-Computer Interface(BCI). Among them, the P300 speller allows participators to input characters to a computer directly from their own brains. Estimating available features of P300 from raw electroencephalogram(EEG) is a key step of implementing P300 speller. In this paper, a novel combination of Autoregressive model and sparse Wavelet representation is proposed to estimate the P300 features in raw EEG acquired from the P300 speller experiments. Instead of superposition, the P300 features are estimated from raw EEG of single trial in this way. By introducing this method to process signals for BCI, the number of repeated trials may be reduced so that the information transfer rate of P300 speller could be remarkably improved. The proposed approach was tested in off-line data. The results show that the number of repeated trials for a wanted character could be reduced to 4 in general when the feature estimation method is used together with the linear discriminant functions.