Separation of multi-unit nerve impulse trains by a multi-channel linear filter algorithm.

We have applied a technique involving multi-channel linear filtering to the problem of sorting and identifying singly unit neuronal activity in records containing several simultaneously active units. The technique is effective even when the waveforms of two or more nerve impulses are superimposed. We have applied it successfully to data involving up to 6 units, and in theory it is applicable to substantially more. Current limitations are that the waveform of the impulse from each unit must first be determined by identifying and averaging several isolated examples, and that the number of recording electrodes must approximate the number of units present.