A real-time BCI with a small number of channels based on CSP

For implementations of electroencephalography (EEG)-based brain computer interfaces (BCIs), common spatial pattern (CSP) is one of the most popular methods. Although many variations of CSP have been introduced, the original form is still quite effective and efficient. However, its application is usually based on multi-channels, which requires a relatively long time for electrode installation. The aim of this study is to build a BCI with CSP in a simple and practical way, with a few electrodes and Linear Discriminant Analysis (LDA) as the classifier. Two experiments were carried out to compare effects of different numbers of channels and overlap of sliding windows. By optimizing these basic parameters, it is possible to build an easy-to-setup and effective BCI. The results showed that for four subjects with either 15/10 channels, the online trial accuracy could reach 83.9% on average. Comparable results could be obtained with only 5 channels.