I. Introduction
Brain–computer interfaces (BCIs) provide the possibility of a direct, nonmuscular communication, and control channel by recognizing patterns of brain activity [1]. The most common recording technique used for these devices is the electroencephalography (EEG). This is due to its high temporal resolution, portability, and relative low cost [2]. However, this technique is characterized by low spatial resolution, since the neural activity is propagated through the brain tissue and scalp which acts as a low-pass filter and smears the activity [3]. Moreover, it is prone to contamination due to muscular artifacts and electromagnetic noise, resulting in a low signal-to-noise ratio (SNR). Furthermore, modifications in the recording settings, e.g., changes in conductivity [4], result in signal variations that may also affect the decoding performance.