Design, fabrication & evaluation of a biomimetic filter-bank architecture for low-power noise-robust cochlear implant processors

Abstract

This thesis presents a new bio-inspired filterbank-architecture termed “OZGF-with-LI” for the signal processing in cochlear implants (CIs), and its ultra-low-power analog very-large-scale-integration (VLSI) implementation. The OZGF-with-LI system provides a potential solution to the noise susceptibility of current CI users by simulating the lateral inhibition (LI)—a biological spectral-enhancement mechanism that may partly account for the high noise robustness of the human auditory system but is missing in current CIs—as a soft local “winner-take-all” link between different frequency regions in the input spectrum. It performs multi-channel syllabic compression via automatic gain control (AGC) while preserving well the spectral contrast and hence the original spectral features of the system input—an advantage over the compression scheme used in current CIs, which degrades spectral contrast and thus impairs the ability of CI users to pick out the spectral peaks (features) as the identity of various speech sounds embedded in noise. Two perceptual tests via acoustic simulation of CIs verified the benefits of the proposed system: the simulated LI provided a substantial benefit for listening to speech presented in noise. The proposed architecture was explicitly designed to be amenable to low-power analog VLSI implementations. This thesis reports such a silicon integrated-circuit (IC) prototype of the OZGF-with-LI system fabricated in the commercially available 0.35μm AMS CMOS process, with a power consumption of 28μW and an input dynamic range of 92dB. This system has potential for use in fully implantable CIs of the future, which have very stringent requirements on the power consumption for signal processing.