Spectral mixing in nervous systems: Experimental evidence and biologically plausible circuits

Abstract

The ability to compute the difference frequency for two periodic signals depends on a nonlinear operation that mixes those signals. Behavioral and psychophysical evidence suggest that such mixing is likely to occur in the vertebrate nervous system as a means to compare rhythmic sensory signals, such as occurs in human audition, and as a means to lock an intrinsic rhythm to a sensory input. Electrophysiological data from electroreceptors in the immobilized electric fish and somatosensory cortex in the anesthetized rat yield direct evidence for such mixing, providing a neurological substrate for the modulation and demodulation of rhythmic neuronal signals. We consider an analytical model of spectral mixing that makes use of the threshold characteristics of neuronal firing and which has features consistent with the experimental observations. This model serves as a guide for constructing circuits that isolate given mixture components. In particular, such circuits can generate nearly pure difference tones from sinusoidal inputs without the use of band-pass filters, in analogy to an image-reject mixer in communications engineering. We speculate that such computations may play a role in coding of sensory input and feedback stabilization of motor output in nervous systems.