An Active Hopf-based Digital Hardware Cochlea Implementation

The mammalian auditory system is capable of extracting auditory features from very small vibrational changes remarkably by adapting to noisy environments over an extraordinary large dynamic range to facilitate intelligent processing. Theoretical studies and biophysical measurements suggest that this outstanding performance of acute sensitivity in sound frequency and sound amplitude is strongly believed to be produced in the cochlea by adapting to the electromechanical feedback signals received by the Outer Hair Cells (OHCs) in the Organ of Corti. A growing consensus speculates that the operation of the mechanisms observed in biology that extract and compress the audio input features with high accuracy can be modelled using Hopf bifurcations. We propose a simple Hopf-based active cochlear filterbank that can approximately reproduce auditory processing observed in a mammalian cochlea and the nonlinear frequency-dependent hearing threshold observed in the auditory nerve fibers. We experimentally demonstrate how the crucial role of the feedback can benefit the background noise cancelling task, by tuning the hearing thresholds based on the activities of higher feedback projections. To prove the concept in hardware, we have implemented the proposed cochlear filter in an FPGA. The tuning curve of the filter can be configured to reproduce various biological data presented in the literature. The resource utilisation and the power consumption of the filter are provided. The proposed active cochlear filterbank can easily be extended to integrate different types of feedback projections such as the thalamus feedback observed in biology. The proposed filter could be a good candidate to use as an artificial hearing sensor that can provide auditory spike trains for the existing and future spike-based intelligent computing machines.