Subject: Re: Ensoniq AudioPCI 97
To: Lennart Augustsson <lennart@augustsson.net>
From: Andrew Reilly <reilly@zeta.org.au>
List: current-users
Date: 09/06/1999 10:46:16
> Well, for CD quality sound it seems to be about 25 multiply-adds per output
> sample.

That depends rather a lot on the ratio of the frequencies that
you're trying to convert between.  The unfortunate thing is that
I imagine that quite a lot of the time you'll be wanting to do
44100Hz to 48000Hz (assuming that AC97 is 48kHz?) which is about
as nasty as it gets.  To do that "properly" I think you need a bank
of 160 sync-coefficient variations (time shifts) of length 147.
That is, 147 multiply/adds per (output) sample, with a different
set of 147 coefficients every sample, in a loop of 160 samples.
That's 23520 coefficients, which is likely to do nasty things to
your caches.

I'm guessing that most software sample rate converters cheat by
using shorter filters, and in fact I think you'd get a reasonable
frequency response with a bank of 32-tap filters.  It's still nasty
work for a CPU to be doing.

Does anyone know whether the fixed-rate codecs run at 48 or
44.1kHz?  Obviously, the ability to do both would be _so_ much
nicer.

-- 
Andrew