An Improved SSVEP Based BCI System Using Frequency Domain Feature Classification

Brain Computer Interfacing systems provide a new communication channel for disabled people. Among the many different types of the BCI systems, the Steady State Visual Evoked Potential (SSVEP) based ones has attracted more attention due to its ease of use and signal processing. SSVEPs are usually recorded from the occipital lobe of the brain when the subject is looking at a twinkling light source. Following our previous report[10], a novel set of features along with a new high-speed classifier are introduced and used in this work. These used for SSVEP classification elicited by LED light sources separated by D = 4, 14, 24, 44 and 64 cm from each other while the LEDs’ plane was located 60 cm away from the subject's eyes. Using various SSVEP sweep lengths, the results show that the LDA and SVM classifiers outperform the other method used when applied to 0.5 and 1-second sweep lengths and to 2 and 3-second sweep lengths respectively. The Max classifier needs, however, longer sweep lengths but with a comparable Information Transfer Rate (ITR). In addition, for D=44 cm and D=64 cm the algorithm could produce the highest accuracy rate of 90% and 92% respectively compared to the other distances. Also, the performance of the proposed algorithm for D=4 cm is not acceptable (p-value<0.001). Finally, it was showed that the sweep length of 0.5 second could provide a more practical online ITR.