Pulse-number distribution for the neural spike train in the cat's auditory nerve.

Abstract

Pulse-number distributions (PNDs) were recorded from primary afferent fibers in the auditory nerve of the cat, using standard extracellular microelectrode recording techniques. Pure-tone and broadband-noise stimuli were used. The number of neural spikes (pulses) n was measured in a set of contiguous intervals, each of duration T seconds. The quantity n varies from one interval to another. These data were then used to determine the PND, which is the probability p(n,T) of occurrence of n spikes in the time T, versus the number n. The estimated mean and variance of p(n,T) were obtained. Two different values of T were used. An unexpected observation was that the count mean-to-variance ratio R is relatively constant and independent of the stimulus intensity. Use of the PND as a statistical measure of the underlying neural point process has a number of virtues. For example, the PND readily exhibits the existence of spike clusters (e.g., pairs) for some units. The PND is essentially unaffected by time jitter and time quantization and provides a statistically significant measure for units firing at low rates. A study of the scaled and unscaled pulse-interval distributions (PIDs), under conditions of spontaneous firing, demonstrates that the occurrences of neural events are generally not describable by a renewal process. Our investigation shows that none of the point processes customarily used to model the auditory neural spike train is consistent with all of the data. It appears that the encoding of acoustic information into nerve spikes in the peripheral auditory system takes the form of a cluster point process similar to the Neyman-Scott type. For pure-tone excitation, the PND will be well represented as a multinomial distribution in this case.