Application of a reconstruction technique in detection of dominant SSVEP frequency

Steady-state visual evoked potentials (SSVEPs) have a number of specific properties such as an oscillating feature when compared to event-related potentials (ERPs). Based on this oscillating property, a short electroencephalogram (EEG) segment containing an SSVEP can be used to reconstruct a series of EEG segments where the SSVEP has the same initial phase but the phase of the background EEG varies randomly. When these reconstructed EEG segments are averaged, the background EEG is weakened, whereas the SSVEP is strengthened. Therefore, the signal-to-noise ratio (SNR) of the SSVEP in the averaged signal is significantly improved. This reconstruction technique is first proposed in this work, and the method of extracting SSVEPs based on this technique is referred to as the reconstruction extraction (RE) method. The RE, power spectrum (PS), and canonical correlation analysis (CCA) methods were applied to EEG segments that were 1s in length in order to detect SSVEPs. The results show that the detection accuracy of the RE method is similar to that of the CCA method, although it is higher than that of the PS method in situations where the SSVEP has low strength. However, in contrast to the CCA, the RE can be applied using only one signal electrode. This suggests that the RE method can be adopted in a real-time SSVEP-based brain-computer interface (BCI) system.