[music-dsp] fixed point fft?

Joshua Scholar joshscholar at yahoo.com
Sun Apr 20 23:39:01 EDT 2003


----- Original Message -----
From: "Vesa Norilo" <warder at dlc.fi>
To: <music-dsp at aulos.calarts.edu>
Sent: Sunday, April 20, 2003 2:31 PM
Subject: Re: [music-dsp] fixed point fft?


> Maybe it's the ability to claim bit-perfect CRC-passing compression.
> Although I would think that the 50th bit in the PPC FPU would still
> remain below at least 16-bit precision in the final result.
> And anyway, isn't IEEE floating point math pretty well defined?
>
> Vesa
>

Yeah I wanted to be able to say "lossless" and mean lossless.

It would also be convinient to be able to replicate a model in the
decompressor and be sure it never went insane and made a different threshold
decision than the compressor.

I don't mind writing in assembly language to get that result.

But really I just want compatibility on consumer platforms, intel, powerpc
and whatever goes in those PDAs (ARM?).

If IEEE processors have some minimum compatibily mode that's actually
perfectly reproducable on all machines then that would be THE BEST SOLUTION.

I'd rather not waste time looking for fixed point or integer algorithms if I
don't have too.

I know Intel has options for turning off the guard bits and setting rounding
modes.

Who would I ask?

Joshua Scholar





More information about the music-dsp mailing list