FPGA silicon cochlea modelling

Abstract

The inner ear structure or cochlea performs important spectral analysis of incoming audio signals to the ear. Its output is then further processed by the auditory pathway and the brain to perform tasks such as speech recognition and sound localisation. Functionality of the biological cochlea is manifested as an effect of its nonlinear travelling wave mechanism and active processes. This dissertation presents an FPGA-based design and implementation of an architecture which models qualitative behaviour of the biological cochlea. Furthermore, an ADC interface is also presented. This is required for audio signals to be fed into the FPGA evaluation board. The design adopted consists of a cascade of second order IIR low-pass filter stages taned at various frequencies following an exponential distribution. The filter chain covers the auditory human range from 20 Hz to 20 KHz. Active processes in the biological cochlea are modelled by a controllable Q-parameter in the filter's transfer function. The implementation makes use of a dual fixed-point arithmetic representation, which is characterised with an improved dynamic range when compared to the conventional fixed point representation used in FPGA-based projects. As regards to the filter chain, for implementation purposes, a time division multiplexed scheme is adopted due to limited hardware resources.