An evaluation of autoregressive spectral estimation model order for brain-computer interface applications.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Autoregressive (AR) spectral estimation is widely used for modeling electroencephalogram (EEG) signals.
• EEG frequency domain phenomena are key for brain-computer interface (BCI) control.
• Previous studies on optimal AR model order for EEG may not apply to SMR-BCI control.

Purpose of the Study:

• To determine the optimal AR model order for EEG spectral analysis in sensorimotor rhythm (SMR)-based BCI control.
• To evaluate if existing criteria for AR model order selection are suitable for SMR-BCI applications.

Main Methods:

• EEG data from SMR-BCI control tasks were analyzed.
• Various AR model orders and evaluation criteria were applied to the EEG spectra.
• The impact of different AR model orders on BCI control performance was assessed.

Main Results:

• The optimal AR model order for SMR-BCI control generally requires a higher order than commonly used in existing studies.
• Model evaluation criteria significantly influence the determination of the optimal AR model order.
• Higher AR model orders can lead to improved SMR-BCI control performance.

Conclusions:

• The optimal AR model order for SMR-BCI control is typically higher than previously reported.
• Careful selection of AR model order and evaluation criteria is essential for maximizing SMR-BCI performance.
• Future research should consider higher AR model orders for optimizing SMR-BCI systems.