Toward a model-based predictive controller design in brain-computer interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Control Systems

Background:

• Current brain-computer interface (BCI) applications often rely on ad hoc, non-model-based filters.
• Model-based predictive control (MPC) offers a potential pathway to enhance BCI performance.
• Designing an optimal MPC is a crucial first step for advanced BCI systems.

Purpose of the Study:

• To present a foundational step in designing a robust and optimal model-based predictive controller (MPC) for BCI applications.
• To explore the utility of model-based features extracted from electroencephalography (EEG) for controller design and task discrimination.

Main Methods:

• Extracted model-based features from human scalp electroencephalography (EEG) data during motor imagery tasks.
• Utilized a simple autoregressive (AR) model to generate parameters for controller design.
• Evaluated the performance of these parameters for both controller design and motor imagery task discrimination.

Main Results:

• The parameters generated for MPC design demonstrated effectiveness in motor imagery task discrimination.
• Task discrimination performance using these parameters was comparable to commonly used features like band powers and direct AR parameters.
• Achieved task discrimination errors ranging from 8-23%.

Conclusions:

• Model-based features derived from EEG can be effectively used for both BCI controller design and task discrimination.
• The proposed approach using an AR model for MPC parameter extraction shows promise for improving BCI performance.
• Optimal MPC has significant implications for developing high-performance BCI applications.