Subject: Re: [linux-audio-dev] gain curves, interpolation, etc.
From: Bill Schottstaedt (bil_AT_ccrma.Stanford.EDU)
Date: Thu Nov 30 2000 - 16:48:01 EET
> I'm still not sure the roundoff error is really much of a problem in
> practice. But, it's fun to think about stuff like that, and important
> to keep in mind, particularly when dealing with floating point
> calculations.
Back in the 70's, the Samson box (a synthesizer used a lot here at
ccrma) had 20 bit fixed-point amplitude envelopes and 28 bit fixed
point frequency envelopes; our software would check envelope
segments for cases that fell "between the cracks" (i.e. a slope
of 1.4) and split them into two pieces, each being at an exact
(nominal) integer slope. In the first version of CLM, I used
floats (C 32-bit), but quickly ran into trouble; I changed it
to be an offset from the start of a segment + an accumulating
delta; this also failed in some cases (primarily where the
envelope was describing a frequency trajectory in Hz -- the
amplitude envelopes normally were between 0 and 1, and were
less problematic); in the current CLM, I use doubles throughout,
and haven't yet encountered a case that these can't handle.
> Um, wow, isn't there prior art? I could swear I used gain curve software
> on an amiga in the late 80's.
And I remember it from the late-70's; I wonder what they're actually
trying to patent. I heard through the grapevine that the Lucent
patent on the entire notion of an amplitude envelope did not
survive a court challenge.
This archive was generated by hypermail 2b28 : Thu Nov 30 2000 - 17:34:48 EET