LAUSR

The P300 wave has been successfully employed to develop brain-computer interfaces (BCI) for speller applications. However, methods to analyze the P300 require computers with high processing capability because they are computationally complex and require many electrodes. Therefore, this paper proposes a novel BCI speller system based on the P300 wave that employs a few electrodes and a processing method aimed to design ubiquitous and embedded applications. The experiments were developed with a dataset generated by our BCI data acquisition system. The BCI speller developed requires five electrodes for data acquisition, and the visual interface is an improved Donchin speller. Our BCI includes a novel processing method composed of the following modules: preprocessing, signal averaging, low computational cost convolutional neural network, and character prediction. The network has two feature extraction sections, a fully connected layer and a SoftMax layer. According to the results, the proposed BCI speller has an accuracy of 96% using just five electrodes, and it is similar to the best BCI for P300 analysis described in the literature. The processing time makes the system practical for online applications since the processing method has a low computational burden and the acquisition system has the lowest number of electrodes for P300 analysis reported in the literature. Considering the low computational burden, the low number of electrodes required, and the accuracy achieved, we conclude that our proposed BCI speller may be considered as one of the best spellers based on P300.