A Study On Asynchronous P300-Based Brain-computer Interface Speller

Brain Computer Interface(BCI) is a kind of system using electroencephalogram(EEG)to establish communication with computer, which can control external devices or exchange information with the outside world. BCI has broad application prospects in the medical and rehabilitation fields. According to the different working modes, BCI system can be divided into synchronous BCI and asynchronous BCI. By setting the fixed control period,synchronous BCI system only needs to detect the subject’s EEG signals in the control period.While in asynchronous BCI system, the subject can choose when to start or to end the control period, the system needs to detect the subject’s EEG signals real-timely and distinguish between control and non-control state. Synchronous BCI is easy to analyze and implement, at present, most of the BCI systems are using synchronous mode. However, in order to make the BCI system not only used in the laboratory, but also used in the daily life, the development of asynchronous BCI system is particularly important. In this paper, the following three aspects of the research work are done about asynchronous P300-based BCI speller.1) An algorithm using only one threshold value to achieve asynchronous P300 character inputting is proposed. Traditional asynchronous BCI systems need to first finish the control state recognition and then the target character recognition, which needs two different classification thresholds for each step. In this paper, the control state P300 signal is categorized as a class, the rest control state non-P300 signal and non-control state signal are categorized as another class. Through the detection of the signal, if the output value is larger than a predetermined threshold, it is determined the system is in control state, and outputting the corresponding character; otherwise it is non-control state without character output. The method simplifies the process of signal detection with ensuring high accuracy,and realizes the detection of the control state and the non-control state.2) For the detection of asynchronous P300 signals, this paper proposes two detection algorithms based on Bayesian linear discriminant analysis(BLDA) and convolutional neural network(CNN). BLDA algorithm is a classic pattern recognition method, while CNN is a kind of artificial neural network algorithm developed rapidly in recent years.This paper attempts to use the CNN method in the detection of asynchronous P300 signals,and compared with the classic BLDA method. The experiment results show that the two detection methods achieve similar outcome, and both achieve high detection accuracy.The accuracy rate is more than 80% with the classification response score superimposed and averaged in 5 rounds; the accuracy rate is more than 90% with the classification response score superimposed and averaged in 7 rounds.3) An asynchronous P300-based BCI speller online experimental system is development.Through online experiments, the feasibility of BLDA algorithm and CNN algorithm under the single threshold method for asynchronous P300 signal detection is verified, with high accuracy rate, allowing the subjects to freely switch between control state and non control state.