ASA 130th Meeting - St. Louis, MO - 1995 Nov 27 .. Dec 01

2pPP2. Noise discriminability. II. Leaky integrator models.

Donald E. Robinson

Martin E. Rickert

Dept. of Psych., Indiana Univ., Bloomington, IN 47405

Several two-interval, same--different experiments involving the discriminability of samples of noise have been reported previously. A striking feature of these data is the effect of temporal position: samples that are altered near the end are more discriminable than those in which the change occurs earlier. This effect occurs over a wide range of durations and bandwidths. Here two models are reported that describe these results. In each model the waveforms from the two temporal intervals are jittered in amplitude, filtered, and squared or rectified. In one model, the resulting waveforms are passed through a leaky integrator and subtracted from one another. This difference waveform is squared and passed through a second leaky integrator. A sample of the output of the second-stage leaky integrator taken at the end of the integration period is used as a decision variable from which hit and false alarm rates are obtained and d[sup '] is computed. In the other model, the squared waveforms are multiplied by an exponential weighting function before subtraction. A signal-to-noise statistic is then used to obtain an estimate of d[sup ']. The fitting parameters for each model are the variance of the internal noise process and the time constant of either the second-stage integrator or of the exponential weighting function. The results of simulations will be compared with data from two experiments, and the relation of these models to others involving leaky integrators will be discussed. [Work supported by AFOSR.]