A Stimulus-Free Brain-Computer Interface using Mental Tasks and Echo State Networks

We propose an EEG classification algorithm for the mental task BCI paradigm that uses Echo State Networks (ESN). In this approach, ESN are used to model the dynamics of EEG during each of several mental tasks. Classification is performed by applying several of these models and assigning the class label associated with the ESN that produces the lowest forecasting error. Experiments performed on 14 subjects using a portable EEG system achieve information transfer rates as high as 15 bits-per-minute with four tasks and 21 bits-per-minute for two tasks.