Estimation of BCI Model Based on Non-linear Oscillator Using UKF-Based Training Method

Abstract

This research focuses on developing a mathematical model to estimate neurological signal parameters for investigating brain dynamics. The proposed model leverages the nonlinear dynamics of oscillator models, assuming that an individual’s brain state can be represented by an unknown parameter vector denoted as θ. To address the challenge of handling noisy training data, an Unscented Kalman Filter (UKF) is employed in a joint training process involving EEG and speech signals. This method allows for precise estimation of the parameters θ and ϕ, which are associated with EEG and speech signals, respectively. The validation procedure includes comparing the estimated EEG with the actual EEG and accounting for noise in the speech signal. The Unscented Kalman filter helps enhance the accuracy of synthesizing EEG signals and contributes to the modeling efficiency and reliability. Overall, the incorporation of the UKF in the modeling process improves the ability to capture the intricate dynamics of EEG signal generation based on speech data.

Keywords

Article information