Abstract

As the field of Brain-Computer Interfaces (BCI) is rapidly evolving within both academia and industry, the necessity of improving the signal processing module of such systems becomes of significant practical and theoretical importance. Additionally, the employment of Electroencephalography (EEG) headset, which is considered as the best non-invasive modality for collecting brain signals, offers a relatively more user-friendly experience, affordability, and flexibility of design to the developers of a BCI system. Motivated by the aforementioned facts, the thesis investigates several venues through which an EEG-based BCI can more accurately interpret the users' intention. The first part of the thesis is devoted to development of theoretical approaches by which the dimensionality of the collected EEG signals can be reduced with minimum information loss. In this part, two novel frameworks are proposed based on graph signal processing theory, referred to as the GD-BCI and the GDR-BCI, where the geometrical structure of the EEG electrodes are employed to define and configure the underlying graphs. The second part of the thesis is devoted to seeking practical, yet facile-to-implement, solutions to improve the classification accuracy of BCI systems. Finally, in the last part of the thesis, inspired by the recent surge of interest in hybrid BCIs, a novel framework is proposed for cuff-less blood pressure estimation to be further coupled with an EEG-based BCI. Referred to as the WAKE-BPAT, the proposed framework simultaneously processes Electrocardiography (ECG) and Photoplethysmogram (PPG) signals via an adaptive Kalman filtering approach.