Approximation and Optimization of an Auditory Model for Realization in Visi Hardware

The Auditory Image Model (AIM) is a software tool set developed to functionally model the role of the ear in the human hearing process. AIM includes detailed filter equations for the major functional portions of the ear. Currently, AIM is run on a workstation and requires 10 to 100 times real-time to process audio information and produce an auditory image. An all-digital approximation of the AIM which is suitable for implementation in very large scale integrated circuits is presented. This document details the mathematical models of AIM and the approximations and optimizations used to simplify the filtering and signal processing accomplished by AIM. Included are the details of an efficient multi-rate architecture designed for sub-micron VLSI technology to carry out the approximated equations. Finally, simulation results which indicate that the architecture, when implemented in 0:8 m CMOS VLSI, will sustain real-time operation on a 32 channel system are included. The same tests also indicate that the chip will be approximately 3.3 mm 2 , and consume approximately 18 mW. The details of a new and efficient method for computing an approximate logarithm (basetwo) on binary integers is also presented. The approximate logarithm algorithm is used to convert sound energy into millibels quickly and with low power. Additionally, the algorithm, is easily extended to compute an approximate logarithm in base ten which broadens the class of problems to which it may be applied.