The information capacity of nerve cells using a frequency code

Abstract

Approximate equations are derived for the amount of information a nerve cell or group of nerve cells can transmit about a stimulus of a given duration using a frequency code (i.e., assuming the mean frequency of nerve impulses measures the intensity of a maintained stimulus). The equations take into account the variability of successive interspike intervals, and any serial correlations between successive intervals, but do not require detailed assumptions about the mechanism of impulse initiation. The errors involved in using these approximations are evaluated for neurons which discharge either completely regularly, completely at random (Poisson process) or show a particular type of intermediate variability (gamma distribution model). The errors become negligibly small as the stimulus duration or the number of functionally similar nerve cells increases. The conditions for applying these equations to experimental data are discussed. The application of these equations should help considerably in eliminating the enormous discrepancies between some earlier estimates for the information processing capabilities of single nerve cells and systems of nerve cells.