Information-theoretic evaluation of brain-inspired sonar signal representation

Abstract

Many bat species, such as horseshoe bats, move their external auditory periphery (noseleaf and pinnae) during emission/reception of their biosonar signals. This peripheral dynamics has already been shown to enhance sensory information encoding at the level of the analog echo waveforms. However, the bats' brains do not process any analog waveforms directly but rely on spike codes generated in the auditory nerve instead. Hence, it is desirable to evaluate the effect of the peripheral dynamics at the level of these spike codes. As input for this analysis, natural foliage echoes akin to the forest environments experienced by bats were recorded using a biomimetic robot with a dynamic periphery similar to that of horseshoe bats. The echoes were converted into neural spike trains through a signal-processing model of the bats' cochlea/auditory nerve. The effect of the dynamic periphery was then investigated using information-theoretic techniques. As a first step, entropy was estimated to quantify coding capacity. The results showed an increased entropy within signals corresponding to a dynamic periphery when compared to a static periphery. For the next steps, the variability in characteristic spike train features such as spike timing, spike rates, and spike intervals will be characterized with additional information-theoretic methods.