A computational model for rate-level functions from cat auditory-nerve fibers

A computationally tractable form of the rate-level model proposed by Sachs and Abbas (1974) is presented. The first stage of the model is a compressive nonlinearity whose input-output function is chosen to reflect current data on basilar-membrane displacement. The output of this nonlinearity is converted to driven discharge rate by the saturating nonlinearity originally used by Sachs and Abbas (1974). In fitting the model to data four model parameters are chosen to minimize the mean squared error between rate functions generated by the model and the data. With parameters chosen in this way, the model provides good fits to the range of rate-level shapes from flat saturations to sloping saturations. One important parameter in the model is the 'threshold for compression'. For low- and medium-spontaneous rate fibers with similar best frequencies (BFs), the minimum mean squared error compression threshold is roughly constant at about 30 dB above the thresholds of the most sensitive (high-spontaneous rate) fibers at that BF.