Speech and Natural Language Dept., Bolt Beranek and Newman, 70 Fawcett St., Cambridge, MA 02138
A bank of temporal bandpass filters was simulated using model cochlear nucleus neurons. Average firing rate equations were used to model chopper neurons as slow average rate integrators (lowpass filters with low cut-off frequencies) and primarylike neurons as fast average rate integrators (lowpass filters with high cut-off frequencies). Using the principle that the difference of two lowpass filters is a bandpass filter, the filter bank was constructed so that the output at a particular point was the difference in output rates of a primarylike and a chopper neuron at that same point. Different bandpass shapes were generated by varying model neuron parameters. In the linear case, only three parameters for each chopper/primarylike pair are needed to completely specify a simple closed-form solution for the bandwidth and center frequency of the temporal filter. [Frisina et al., Hear. Res. 44, 99--122 (1990)] described bandpass filterlike responses from a spectral analysis of single unit PSTH recordings in cochlear nucleus. An explanation of this phenomena in terms of saturation neuron firing rates was given by [Hewitt et al., J. Acoust. Soc. Am. 91, 2096--2109 (1991)]. The model presented here relies on a fundamentally different scheme for generating bandpass filter.