Unsupervised frequency-recognition method of SSVEPs using a filter bank implementation of binary subband CCA

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Steady-state visual evoked potential (SSVEP)-based brain-computer interfaces (BCI) offer effective command detection.
• Current methods require calibration, increasing user time and fatigue, especially with more commands.

Purpose of the Study:

• To develop a novel unsupervised method for accurate SSVEP stimulus frequency detection.
• To improve the performance of SSVEP-based BCIs by eliminating the need for calibration.

Main Methods:

• Implemented a novel unsupervised technique, binary subband canonical correlation analysis (BsCCA), in a multiband approach.
• BsCCA uses two subbands, computing correlation coefficients and employing artificial reference signals for enhanced frequency recognition.
• The SSVEP signal is decomposed into subbands, with BsCCA applied to each, and results combined via a weighted sum.

Main Results:

• Evaluated on a 12-class SSVEP dataset from ten subjects, BsCCA significantly outperformed state-of-the-art methods.
• Achieved an average information transfer rate (ITR) of 77.04 bits/min, with a maximum individual ITR of 107.55 bits/min.
• Outperformed standard CCA (69.29 bits/min) and NCCA (69.44 bits/min).

Conclusions:

• The proposed unsupervised BsCCA method significantly enhances SSVEP-based BCI performance.
• BsCCA offers a practical and efficient solution for real-world BCI applications, reducing calibration time and user fatigue.