Enhanced coding in a cochlear-implant model using additive noise: aperiodic stochastic resonance with tuning

Analog electrical stimulation of the cochlear nerve (the nerve of hearing) by a cochlear implant is an effective method of providing functional hearing to profoundly deaf people. Recent physiological and computational experiments have shown that analog cochlear implants are unlikely to convey certain speech cues by the temporal pattern of evoked nerve discharges. However, these experiments have also shown that the optimal addition of noise to cochlear implant signals can enhance the temporal representation of speech cues [R. P. Morse and E. F. Evans, Nature Medicine 2, 928 (1996)]. We present a simple model to explain this enhancement of temporal representation. Our model derives from a rate equation for the mean threshold-crossing rate of an infinite set of parallel discriminators (level-crossing detectors); a system that well describes the time coding of information by a set of nerve fibers. Our results show that the optimal transfer of information occurs when the threshold level of each discriminator is equal to the root-mean-square noise level. The optimal transfer of information by a cochlear implant is therefore expected to occur when the internal root-mean-square noise level of each stimulated fiber is approximately equal to the nerve threshold. When interpreted within the framework of aperiodic stochastic resonance, our results indicate therefore that for an infinite array of discriminators, a tuning of the noise is still necessary for optimal performance. This is in contrast to previous results [Collins, Chow, and Imhoff, Nature 376, 236 (1995); Chialvo, Longtin, and Müller-Gerking, Phys. Rev. E 55, 1798 (1997)] on arrays of FitzHugh-Nagumo neurons.