Towards a Compressive Sensing-Based SSVEP-BCI
Richard Junior Manuel Godinez Tello1; Jeevan K. Pant2; Sridhar Krishnan2; Teodiano Freire Bastos3
1 Federal Institute of Espirito Santo (IFES); 2 Ryerson University; 3 Federal University of Espirito Santo (UFES)
Resumo
In this research, a study of SSVEP frequency detection is carried out for its application in compressive sensing (CS)-based brain-computer interface (BCI). The CS-BCI system is based on compressing electroencephalogram (EEG) signals using random projection, transmitting the compressed data using wireless method, and reconstructing EEG signals from the received data using a CS-reconstruction algorithm. Minimum Energy Combination (MEC), Canonical Correlation Analysis (CCA) and Multivariate Synchronization Index (MSI) algorithms are applied for the detection of visual stimulus frequency from the reconstructed EEG signals. For CS-reconstruction, the l-Regularized least-squares (l-RLS), l2-regularized least-squares (l2-RLS), and the block-sparse Bayesian learning bound-optimization (BSBL-BO) algorithms are applied. Results of simulation conducted using EEG signals acquired by using Emotiv headset indicate that the MSI algorithm offers the highest mean classification accuracy (CA), whereas the CCA algorithm offers the most stable CA. Also, the l-RLS algorithm is found to balance the trade-off between the high mean value of CA offered by the BSBL-BO algorithm and the most stable value of CA offered by the l2-RLS algorithm.
Palavras-chave: Compressive Sensing; SSVEP-BCI; CCA; MEC; MSI