A Comparison of Techniques for Classification of Multiple Neural Signals

A number ofmultiunit neural signal classification techniques are compared in their theoretical separation properties and in their empirical performance in classifying two channel recordings from the ventral nerve cord of the cockroach. The techniques include: the use of amplitude and conduction time measures, template matching, the principal components method, optimal filtering, and maximin discrimination. The noise encountered under different situations is characterized, permitting the comparisons to be made as functions of the experimental conditions. Recommendations are made as to the appropriate use of the techniques.