Brain-Computer Interface using P300 Speller Based on Word Prediction

초록

BCI(Brain-computer interface) is a system of providing the communication to control a computer or external device using the signal from the brain activity. P300 is one of the common methods to detect the brain activity. It is relatively easy to generate the brain response. The P300 Event Relation Potential(ERP) is a positive response signal in the EEG over parietal cortex, occurring approximately 300ms after the presentation of a visual stimulus. Word speller with rows and columns of the matrix is commonly used to research P300 based BCI. Intensification of each row and column are presented in a random sequence and intensification is performed by flashing the row or column. When the row and the column having a target letter intensified, P300 is generated by user’s attention. Only flashes of groups containing the attended item should elicit a P300. In the traditional P300 BCI, items are grouped for flashing as rows and columns. However, when non-target rows and columns are adjacent to the target rows and columns, the traditional P300 method has problems that non-target rows and columns attract the user’s attention, thereby they can generate the unexpected P300. This paper proposes a new word speller method to reduce the unexpected P300 signal. After selecting the first target letter, we can identify which letter should follow and shouldn’t follow by using word prediction engine. So the word speller can disable the flash of non-target letter that are adjacent to the target and it will decrease the frequency of unexpected P300. As a result, the proposed method can reduce the noise signal in P300 and extract more accurately the feature vectors, which are composed of samples from the selected electrode. This paper uses the band pass filer, which has cut-off frequencies of 1 and 20Hz and the spatial filter for adjusting the raw recorded EEGs to the user specific subspace. P300 signals are acquired many times and averaged to reduce the signal variab