Frequency Recognition of SSVEP with Adaptive Reference Signals Using CCA

Background: The brain signal obtained by electroencephalography (EEG) is widely used as low cost arrangement for BCI implementation. Steady-state visual evoked potential (SSVEP) is an EEG obtained by illustrating flickering type of visual stimuli. In the SSVEP based BCI, the main concern is to recognize the flickering (stimulus) frequency of test SSVEP signal. Canonical correlation analysis (CCA) has been widely used in the detection of one of the most SSVEP based BCI systems. The performance and usability in our daily life of SSVEP-based BCI system depends on the higher recognition accuracy with shorter processing time. MultisetCCA (MsetCCA) is one of the most popular SSVEP frequency recognition methods, whereas, it still has significant limitation due to the noise effects in the reference signals. Materials and Methods: In this study, a novel method is introduced to eliminate the noise contamination of the reference set yielding the improvement of performance in stimulus frequency recognition. The required number of training trials and the average over all those trials are added in the training set. The joint spatial filter with MsetCCA is applied to the obtained set to optimize the artificially generated sin-cosine reference signals set by the set of training data and its average. The maximally correlated canonical variates and their corresponding weights are used to construct the reference set. Then the standard CCA is used to determine the multivariate correlation between test data and the obtained reference set. The frequency corresponding to the maximum canonical correlation is selected as the stimulus frequency of the test SSVEP signal. Results: The experiments are conducted for twelve different stimulus frequencies for ten individual subjects. The performance of the proposed method is compared with standard CCA and MsetCCA in terms of recognition accuracy as well as information transfer rate (ITR). The results show that this method performs better than that of CCA and recently developed MsetCCA. Conclusion: An effective method for frequency recognition of SSVEP signals is introduced here using adaptive reference signal. This is a training based approach to SSVEP based BCI implementation. To reduce the noise effects from the training signals the average over all the training trials is included in the training set. This study performs a quantitative comparison of the proposed method with standard CCA and MsetCCA. The proposed method outperforms the existing methods.