Canonical Correlation Analysis of Task Related Components as a noise-resistant method in Brain-Computer Interface Speller Systems based on Steady-State Visual Evoked Potential

Abstract

ObjectiveBrain-Computer Interface Speller systems based on Steady-State Visual Evoked Potentials (SSVEPs) can help people write text without moving their hands. This study’s primary goal is to reduce the noise effect in the signal, which has been recorded without electromagnetic shields. For this purpose, an online available database called BETA has been used. This database has been recorded outside the laboratory conditions; Thus, ambient noise is more prevalent in this database. MethodsCanonical Correlation Analysis of Task-Related Components (CCAoTRC) method has been proposed in this research. In the structure of this method, a spatial filter called the TRC filter has been used, which can reduce the effect of noise and increase the Signal-to-Noise Ratio (SNR) in the data. In order to compare the results with previous methods, the Canonical Correlation Analysis (CCA), the Filter Bank Canonical Correlation Analysis (FBCCA), the Task-Related Components Analysis (TRCA) and the Extended CCA with Training data (ExCCATrain) methods were also implemented. ResultsThe results showed that the accuracy (70.94 %) and Information Transfer Rate (61.93 bpm) of the CCAoTRC method is significantly higher than the traditional CCA (54.06 % and 45.41 bpm). Also, the Wide-band SNR of the signal has significantly increased after applying the TRC filter (p-value < 0.05). ConclusionsThe results show that the CCAoTRC method has been able to increase the SNR using the TRC filter and eliminate the shortcomings of the CCA method. Therefore, the proposed approach seems to be suitable for real-world applications.