Advanced brain computer interface

A brain-computer interface (BCI) provides a new pathway for communication and control through decoding information directly extracted from the neurophysiological signals. In the majority of current BCI systems, brain activity is measured through the electroencephalogram (EEG), due to its low cost and high temporal resolution. However, EEG signals have a poor spatial resolution. A signal of interest is mixed with a set of irrelevant signals from concurrent brain activities while they may have similar activation frequencies and amplitude. Moreover, EEG signals are non-stationary, and they may be distorted by artifacts such as electrooculography (EOG) or electromyography (EMG). These drawbacks can result in inaccurate and deteriorated BCI performances. To overcome these issues and consequently to improve the accuracy and robustness of BCI, we propose four novel and complementary algorithms, namely sparse common spatial patterns (SCSP), optimized sparse spatio-spectral filters (OSSSF), kullback-leibler-based common spatial patterns (KLCSP), and EEG data space adaptation (EEG-DSA). The data acquired from healthy and stroke subjects are used to exemplify the power of the proposed algorithms. The experimental results show that our proposed algorithms improve the accuracy and the robustness of the EEG-based BCI by generating more robust and invariant features, and by adapting the BCI model to non-stationarities. The experimental results also show that our proposed algorithms can potentially improve the usability and practicability of BCI by reducing the set up and calibration time. These improvements will consequently lead to a more intuitive and pleasing interface for daily use.Doctor of Philosophy (SCE