I. Introduction
A BRAIN–COMPUTER interface (BCI) is a data communication channel that transmits information from a human brain to a computer. Instead of relying on brain activity transmitted through the pathways of peripheral nerves and muscles, it is observed from measurements taken on the scalp (electroencephalogram) or directly from the cortex (electrocorticogram). The unique features of rapid response and relatively inexpensive implementation have made the electroencephalogram (EEG) a practical tool for establishing nonmuscular communication and control channels [1]. The electrocorticogram (ECoG) produces a higher signal-to-noise ratio (SNR) than the EEG, and can provide higher spatial resolution. However, the invasive nature of the ECoG limits its practice. Current invasive BCI systems fall into two groups: those based on event related potentials [2] and those based on neuronal activity [3]– [5]. Noninvasive BCI systems fall into four groups based on: slow cortical potentials [6], steady-state visual-evoked response [7], P300 evoked potentials [8], or movement-related potentials [9]– [13].