Dither (was: Re: Burroughs)


Subject: Dither (was: Re: Burroughs)
From: Richard Wentk (richard@skydancer.com)
Date: Wed Jan 19 2005 - 12:40:36 EST


At 05:57 18/01/2005, you wrote:

>If you like UV22 then try the POW-r noise shaper many systems provide, it
>sounds even better,
>
>It sounds like you are talking about the ProTools TDM mixer-plugins. If
>you mix more than 68 channels then more than one DSP-chip is required and
>they used to truncate the communications between the multiple DSP-chips
>(each rnning an instance of the mixer-plugin) to 24-bits.
>3-4 years back they fixed this not by dithering but keeping the full
>48/56-bit mixer resolution.

That's a good thing. ;)

>ProTools TDM has a 24-bit path and a 48/56-bit mixer. If you use their
>dithering mixer and use plugins that work at 48-bit and dithers to 24-bit
>correctly (like the Sony Oxford series) then it is one of the cleanest
>setups around today. I also like the fact that they openly declare how
>their systems are designed, most manufacturers will not tell you anything.

I've seen arguments about whether 24-bit (48/56) fixed point is better than
32-bit floating. I'd personally edge towards 32-bit float, but it's an open
question.

I expect at some point in the future we'll have 64-bit float, and that will
settle the issue definitively. With that kind of resolution you'll be able
to hear the quarks in the baryons in the copper cable rubbing up against
each other. There's not really anywhere else to go after that. ;)

>Note that ProTools LE (non-TDM-hardware) use 32-bit floating point like
>everyone else (it seems) that rely on the native CPU in the computer to
>crunch numbers. 32-bit floating point has a 24-bit path. Due to the
>limitations in CPU-power many "native" digital workstations dont dither at
>all places required. Instead they rely on the rounding errors of floating
>point math to mask the truncation.
>
>This is all a real mess.

But getting less messy. Wouldn't FP rounding errors be less objectionable
than outright fixed-length truncation?

Richard



This archive was generated by hypermail 2b27 : Sat Dec 22 2007 - 01:46:05 EST