Real-time multichannel neural spike recognition with DSPs

Data acquisition technology that uses relatively inexpensive digital signal processors (DSPs) for the rapid acquisition of neuronal action potential (spike) data was investigated. Three spike sorting techniques were considered. The simplest classification routine tested was peak windowing. Spikes were considered to belong to the same class when their peak amplitude fell within a user-specified interval. Another approach required computing the RMS error between stored templates of each unit´s spike waveform and the spike most recently detected. The new spike was classified according to the minimum of the RMS errors computed. In the third algorithm, two optimal coefficient vectors, called principal components, were computed from previously acquired template waveforms and stored. These principal components were used to compute two vector products for each newly detected spike waveform vector. The resulting two values were features that were used to classify the waveform. The test system that was constructed is described, and results obtained with each of the algorithms are compared and evaluated.<>