Abstract:

Determination of the distance to targets from echo delay is an important aspect of echolocation in bats. Neurophysiological studies suggest that the tonotopically organized delay-sensitive neurons (DSN) found in the cortex of bat encode target distance. These studies also indicate that in the tonotopic area, the DSNs in each active ensemble share the same range acuity and the acuity systematically improves over a period of time. A biologically plausible model is suggested here to understand how the bat assembles responses to echoes from multiple targets. This model consists of three stages in cascade: (1) an auditory periphery system (APS); (2) a self-organization neural network (SONN) and; (3) a cortical representation neural network (CRNN). The inputs of APS are the pulse-echo (PE) signals and the outputs are the time-frequency spike trains of PE. SONN self-organizes to the spike trains of PE in frequency index. CRNN has the same architecture of SONN but with a specific neuron response function, y(T,(delta),f)=I((delta),f)(Phi)[a[inf 1]T((delta)-a[inf 2][sup n](beta)[inf 1])/a[inf 2][sup n]](Phi)(f-(beta)[inf 2]), to achieve the multi-resolution. In this function, (Phi)(.) is the negative second derivative of a Gaussian function, I((delta),f) is the spike amplitude of frequency component f arriving at time (delta), T is the time onset of echoes after vocalization, a[inf 1] and a[inf 2]>1 are constants, n is a negative integer, (beta)[inf 1] is the minimum best delay, and (beta)[inf 2] is the best frequency. The response maps of CRNN at different T form the multi-resolution representation of the acoustic scene. Simulations are carried out for the proposed model using phantom targets and the results are compared to neurophysiological findings. [Work supported by NSF.]