An input-output model for the gerbil cochlea

Signal processing in individual channels of the auditory periphery are modelled based on I/O data. Each channel comprises the outer and middle ear apparatus in cascade with a single cochlear filter served by several cochlear afferent nerve fibers. The input is an auditory stimulus and the output is a sequence of neural spike events recorded from cochlear afferent nerve fibers. Tuning properties of the cochlear filter are identified. The channel is modelled as a linear dynamical system in cascade with a memoryless nonlinearity (a Wiener system), followed by an additional element which generates Poisson distributed spike events. We show that the Poisson block can be replaced by an additive noise term. Thus we can consider the more general problem of Wiener identification with output noise, and simplify the analysis of the identification. The impulse response of the linear system shows increased damping and broadened frequency spectrum as stimulus amplitude is increased. By modulating the rms input amplitude and strobing the I/O data during estimation, we study dynamical changes in the linear system's tuning properties. We thus show that changes in the cochlear channel's linear response occur rapidly and reversibly as a function of the rms input amplitude. Our results suggest that a good I/O model for the cochlear channel might consist of a family of Wiener-Poisson models, parameterized by the rms input amplitude.