Research On The Spatial Filtering Algorithm Of Steady-State Visual Evoked Potential-Based Brain-Computer Interface

Brain-computer interface(BCI)refers to a direct communication system established between brain and computer or other external device through the analysis of brain activity without relying on peripheral nerves or muscle tissue.In this paper,we mainly focus on the spatial filtering algorithm of steady-state visual evoked potential(SSVEP)based BCI.Brain-computer interface(BCI)refers to a direct communication system established between the brain and a computer or other external devices that does not depend on peripheral nerve or muscle tissue.Canonical correlation analysis(CCA)is widely used to spatially filter multi-channel SSVEP signals.In the existing CCA approaches,training data are used for constructing templates of stimulus targets and the spatial filters are created between the template signals and a single-trial testing signal.However,compared with task-related component analysis(TRCA)algorithm,the classification performance of CCA is relatively poor,for the spatial filter relying on the test signal.Research suggested that the spatial filters derived from CCA have the equivalent performance with TRCA,when the CCA based spatial filters are optimized by the training data only.It has been theoretically proved that the two spatial filters yielded by CCA were equivalent to each other.A classification experiment based on three CCA approaches was performed using the benchmark SSVEP dataset from 35 subjects.The results suggested that the newly proposed CCA-based method performs best.Further research indicated that the performance of only using training data to estimate the spatial filter is better than the spatial filter estimated by a single-trial signal,and the performance of the spatial filter estimated by a single template is better than that of the universal sine-cosine signals template.In addition,we explored another spatial filtering algorithm,maximum signal fraction analysis(MSFA),which also has equivalent performance with TRCA.On the basis,we further proved that CCA,MSFA and TRCA are also equivalent under the condition that the data lengths,the number of channels and the numbers of training trials are sufficient.On the basis,we further proved that CCA,MSFA and TRCA are also equivalent under the conditions that the data length,the number of channels and the number of training experiments are sufficient.A benchmark SSVEP dataset was used to compare the performance of these algorithms and their combinations with ensemble spatial filter and/or filter bank analysis according to different lengths of data,numbers of channels and numbers of training trials.The experimental results suggested that in terms of classification accuracy,the equivalent relation holds between CCA and MSFA regardless of these parameters and among the three algorithms only if the signal-to-noise ratio of training data is high enough.If the signal noise ratio(SNR)of the training signal is low,CCA and MSFA performed better than TRCA.Both filter bank analysis and ensemble spatial filter are very effective methods,which can further improve performance.Further research has been done to suggest that the differences between the three spatial filtering algorithms,CCA and MSFA have the same performance in terms of classification accuracy,better robustness and stronger anti-noise capability than TRCA.When it comes to operating speed,MSFA is the best of them.