Studies Of Key Techniques For High-speed Steady-state Visual Evoked Potential-based Brain-computer Interface

Recently, steady-state visual evoked potential(SSVEP)-based brain-computer interface(BCI) has attracted much attention for its advantages such as litter user training, ease of use, and high information transfer rate(ITR). However, the current level of BCI performance limits its practical applications. To address these issues, this dissertation performs an in-depth study of SSVEP-BCI, mainly focusing on two aspects of information modulation methods of visual stimulation and pattern classification algorithms. These works improve the performance of this kind of BCI and then promote its practical applications.For information modulation, when using a computer screen in combination with the standard frame-based ?on/off? stimulation, frequencies are limited by the screen refresh rate. To overcome this restriction, this dissertation investigates the feasibility of using intermodulation frequencies and sampled sinusoidal stimulation method to increase the number of coded targets, and designs two SSVEP-BCI systems. The first system is SSVEP-BCI based on intermodulation frequencies. The proposed BCI system realizes eight targets merely using three flickering frequencies. The online results obtained from 15 subjects(14 healthy and 1 with stroke) reveal that an average classification accuracy of 93.83% were achieved using our propsed SSVEP-BCI system. The second system is SSVEP-BCI based on sampled sinusoidal stimulation which realizes forty-five targets on an LCD screen and obtain an average ITR of 104.65 bits/min. In addition, this dissertation proposes a new joint frequency-phase modulation(JFPM) method to enhance the discriminability between SSVEPs with very close frequencies, the most easily misclassified conditions in frequency coding.For pattern classification algorithms, filter bank analysis is introduced into canonical correlation analysis(CCA) to solve the optimization of harmonic SSVEP components for SSVEP recognition. A filter bank canonical correlation analysis(FBCCA) is proposed. And then we design a 40-target BCI speller based on frequency coding. Offline and online results reveal that the FBCCA method significantly outperformed the standard CCA method. At a spelling rate of ~33.3 characters per minute, the online BCI apeller obtains an average ITR of 151.18 bits/min. Considering individual difference of visual latency in target idenfication, individual SSVEP calibration data and a filter bank analysis method are introduced into CCA. An extended CCA method is proposed. On the basis of these studies, we build a 40-taget BCI speller using the extended CCA method and JFPM method. To our knowledge, the average ITR of 267.16 bits/min represents the highest ITR reported in BCI speller.Finally, this dissertation investigates the feasibility of hybrid frequency and phase coding methods in multi-target SSVEP-BCIs and compares JFPM method with the existing mixed frequency and phase coding method. These results suggest that the hybrid frequency and phase coding methods are highly efficient for multi-target coding in SSVEP-BCIs, providing a practical solution to implement a high-speed BCI speller.