A Reworked SOBI Algorithm Based on SCHUR Decomposition for EEG Data Processing

In brain machine interfaces (BMI) that are used to control motor rehabilitation devices there is the need to process the monitored brain signals with the purpose of recognizing patients intentions to move his hands or limbs and reject artifact and noise superimposed on these signals. This kind of processing has to take place within time limits imposed by the on-line control requirements of such devices. A widely-used algorithm is the Second Order Blind Identification (SOBI) independent component analysis (ICA) algorithm. This algorithm, however, presents long processing time and therefor it not suitable for use in the brain-based control of rehabilitation devices. A rework of this algorithm that is presented in this paper and based on SCHUR decomposition results to significantly reduced processing time. This new algorithm is quite appropriate for use in brain-based control of rehabilitation devices.