In this chapter, we review the signal-processing methods applied to recorded brain signals in BCIs for tasks ranging from extracting spikes from the raw signals recorded from invasive electrodes to extracting features for classification. For many of the techniques, we use EEG as the noninvasive recording modality to illustrate the concepts involved, although the techniques could be applied to signals from other sources as well such as ECoG and MEG.

Spike Sorting

Invasive approaches to brain-computer interfacing typically rely on recording spikes from an array of microelectrodes. The goal of signal-processing methods for such an input signal is to reliably isolate and extract the spikes being emitted by a single neuron per recording electrode. This procedure is usually called spike sorting.

The signal recorded by an extracellular electrode implanted in the brain is typically a mixture of signals from several neighboring neurons, with spikes from closer neurons producing larger amplitude del ections in the recorded signal. h is signal is ot en referred to as multiunit hash or neural hash (Figure 4.1A). Although hash could also potentially be used as input to brain- computer interfaces, the more traditional form of input is spikes from individual neurons. Spike sorting methods allow spikes from a single neuron to be extracted from hash.