Complex-valued spatial filters for task discrimination

The method of common spatial patterns (CSP) has been widely adopted for the discrimination of mental tasks using EEG data. In this paper, some limitations of the standard CSP implementation when considering data where phase relationships play a significant role are highlighted. Furthermore, a variant of the CSP method based on the analytic representation of signals is proposed to make up for these drawbacks. The advantages of the proposed method over the standard CSP implementation are demonstrated using simulated data and tests with real EEG data. Specifically, it is shown that the complex-valued spatial filters and the derived spatial patterns can improve the discrimination process and give a more adequate representation of the tasks being considered, respectively.