gda@creare.uucp (gda) (04/02/85)
>This opens the path to more ambitious home-audio projects. What I >have in mind is a programmable audio-processor that can do things >like equilization (including phase), 'holography', etc. The preliminary >design below is subject to discussions (and change without notice). A large part of the time I spent at AR was spent working on digital equalizers, and all the time we were wishing Sony and Phillips would "open" their systems and make our job easier. Now it's happened, and none of us are working on the stuff any more... >Based on AMD's 29510 16bit multiply accumulator, each channel should >be able to perform 256 *+ operations on each sample (about 85 ns cycle). Some helpful hints: Use a TMS32010 (or more than one!) rather than buiding up multiplier/accumulators. The small loss of processing power is more than made up by the flexibility of a programmable system. You can even get a development system for an IBM PC - call Sky Computer about their Sky320 board. It has the TMS32010 (and soon the faster TMS32020), parallel input and output ports, and a cross-assembler and "C" cross-compiler. We built an equalizer prototype with a multiplier/accumulator (a TRW bipolar model) and it took months (and it wasn't a "home project"); also, the fan required to cool it masked any effects of the equalizer... >Some kind of PC will be required to download the tables and there >is a need for signal-processing software to compute the tables for >a desired frequency response. Designing the filters can also be tricky. Most beginners try to invert the FFT of the impulse response; this usually fails because there are zeros outside the unit circle (in the z-transform). You can't multiply zero by infinity to get one. We mostly used the LMS algorithm to design our filters. Another consideration about equalization is the resolution. You'd like to have fairly detailed correction at low frequencies, but not such detail at higher frequency. Most of the corrections will be related to room acoustics and are highly position dependent. At frequencies above 1kHz, the wavelengths become smaller than your head. Our early experiments created a "magic spot", a basketball-sized sphere, floating in mid-air, in which the sound was dramatically different from everywhere else. You might like to look up our paper from the Rye Conference on Digital Audio (or something like that). It was given by Robert Berkovitz, I think in 1982 (?). The conference was connected with the Audio Engineering Society. Good luck ! Gray Abbott Creare Inc. {...dartvax!creare!gda}