Adaptive Algorithm for Estimation of Model of Neural Activity in Brain-Computer Interfaces

In this paper, a method for forming the neuro-feedback (NFB) signal is suggested. The implementation of the method is based on the design of an adaptive model of subjects’ neural activity in the form of a vector autoregression (VAR) model for capturing the spatio-temporal dynamics of an electroencephalogram (EEG). The adaptive model parameter adjustment algorithm is iterative and is designed using gradient methods for solving goal inequalities. At each stage of training, the control signal is first calculated. Then, based on the control signal, the model coefficients are modified. With the improvement of the EEG recording in the sense of its proximity to the desired one, the coefficients of the EEG model are recalculated. The proposed scheme has physiological analogues and is not directly related to the EEG model used in calculating the control effect. The proposed algorithm has advantages over other known algorithms because of its adaptability to each specific subject.