Auditory dynamic range derived from the mean rate-intensity function in the cat.

Loudness change is perceived over a far greater range than the discrimination range of single afferents-the "dynamic range problem" [1]. However, earlier neural count models have not fully considered how variability in the dynamic ranges, thresholds, spontaneous rates, and saturation rates of these afferents may affect overall dynamic range. In the present model, all four of these characteristics appear in a logistic rate-intensity function that fits well with sigmoidally firing cat neurons (a double logistic fits well with sloping-saturating units). These equations were averaged statistically over parameter distributions for each of three spontaneous rate groups. An average over these groups (weighted by relative group size) was used to compute discriminability for a patch of basilar membrane. The upper limit on dynamic range for a patch one critical band wide at 8 kHz was 89 dB SPL, less than implied by psychophysics [2]. A two-channel model extended the upper limit by only 3 dB, and splitting the population into four channels did not improve this limit. Seven channels, optimally combined, provided a dynamic range nearly equal to that of an ideal observer that treats each fiber as a separate channel. Moreover, the contribution of sloping-saturating units to discriminability was not as great as popularly expected. Changing the variances of fiber dynamic ranges did not strongly affect overall dynamic range in any of these computations. In sum, when the distributions of neural parameters are taken into account, local pooling can greatly mitigate the "dynamic range problem."