Extension of common spatial pattern (CSP) algorithm to multi-task case by Jacobi Rotations for single-trial EEG classification

Low information transfer rate (ITR) is one of main problems that a brain-computer interface (BCI) faces. One method to increase ITR is to extend two-class mental tasks to multiple tasks. Accordingly an efficient method for feature extraction is needed to ensure good classification performance. This paper generalizes well-known common spatial pattern (CSP) algorithm from two task conditions to multi-task case by Jacobi Rotations. The detailed mathematical derivation of the algorithm is given, followed by a computer simulation. The algorithm is then applied to four data sets recorded during motor imagery of three mental tasks. The simulation shows that the algorithm can correctly extract signal components specific to each task, while the classification experiments verify the validity and effectiveness of the method.