Signal processing technique to extract neuronal activity from noise

A method of extracting extracellularly recorded action potentials from background electronic noise is described. Segments of traces containing stimulus-induced activity are Fourier transformed and the increase in the total power density over that of control noise segments is used as a measure of stimulus-induced neuronal activity. We show first, with observations from the amphibian visual system and mammalian auditory system, that our technique yields similar quantitative information to that obtained from the conventional spike counting method when the recording arrangement is optimal. Moreover, the size and centre of a visual receptive field can be determined even when the evoked action potentials are buried in the background noise. To investigate the potential of this technique further, we have used it to study the auditory responses in the amphibian midbrain. The power spectral density, we demonstrate here, is proportional to the stimulus intensity over a wide range, and varies systematically with stimulus frequency and the direction of sound source. Other possible applications of this technique, together with the theoretical basis for it, are discussed.