Research On Extraction Methods Of Single-Trial Event-Related Potential And Implement Of USB Interface

A brain-computer interface (BCI) based on event-related potential provides a wholly novel communication and control mode between human and outside world. It is expected to improve the human living level effectively. However, the available extraction methods of single-trial event-related potential performed slower feature extraction and classification speed, as well as lower accuracy. Most of these methods not only can not meet the real-time needs of BCI system but impede BCI system in practice. Therefore, it's of great significance to research on useful methods to improve extraction speed and accuracy. The wavelet transform, with grown-up theory, which produces a good local representation of signals in both time and frequency domain, provides an important tool for extracting the single event-related potential (ERP).This paper concentrated on researching and improving the accuracy and speed of BCI system. In the "software" part, two feature extraction methods of event-related potential were proposed; 1) the feature extraction based on adaptive wavelet threshold; 2) the feature extraction based on False Discovery Rate combining with Independent Component Analysis. Besides, collection platform was designed using USB technique in the "hardware" part.Firstly, wavelet theory and some technologies of threshold filtering were analyzed and discussed. The wavelet threshold filtering mainly includes three parts: optimum wavelet basis selection, threshold function and threshold determination. The study emphasis of this paper is concentrated on these three parts and is expected to improve signal to noise ratio and mean square error. The experimental results showed, by the reformative methods, the pure signal was efficiently extracted from the noise.The problems in existing feature extraction methods of ERP and wavelet were analyzed. Then the following two feature extraction methods of ERP were presented; 1) ERP extraction based on adaptive wavelet threshold; 2) ERP extraction based on FDR and independent component. To verify the effectiveness of these methods, the data of the BCI competition 2003 were analyzed. Compared with other existing methods and the results showed that the research methods could effectively improved the extraction accuracy.Finally, this dissertation analyzed whole designation of BCI system and emphasized on implementation of Universal Serial Bus (USB) interface between data acquisition and computer. It is expected to improve data communication reliability and speed of BCI system. The core part of hardware includes PDIUSBD12 and dsPIC6010A chip, which is mainly responsible for data storage, reading and writing data. The software part uses Visual C++ and Matlab for data displaying, storage and so on.