Dither

Music Manic · Post by **Music Manic** » Mon Aug 02, 2004 12:08 pm

What type of dither does Scope use?
Thanks

ScofieldKid · Post by **ScofieldKid** » Thu Aug 19, 2004 12:51 am

I'm curious about that as well.

The thread about VDAT has been good for explaining that SFP is 32-bit INT internal.

So the obvious questions are if you -> 24-bit
or if you -> 16-bit, what is actually happening there.

So for example, if I play from Scope into ASIO in Samplitude at 24-bit/44.1, whom is doing what do get that to there...????

Post by **garyb** » Thu Aug 19, 2004 1:40 am

dither is a matter of choice. check the dither module.

blazesboylan · Post by **blazesboylan** » Thu Aug 19, 2004 1:57 am

It's not when you send a standard 32 bit (FLOAT, not int) signal into ASIO though. I think the question was: what's the default dither algo?

firubbi · Post by **firubbi** » Thu Aug 19, 2004 9:13 am

http://www.cadenzarecording.com/dither.html

marcuspocus · Post by **marcuspocus** » Thu Aug 19, 2004 9:50 am

excellent link

BingoTheClowno · Post by **BingoTheClowno** » Thu Aug 19, 2004 9:55 am

I guess we have to find out what chips on the board are responsible for AD conversion and look up their specs. Does anyone have a big image of any CW card, where the chips part numbers are legible?

<font size=-1>[ This Message was edited by: BingoTheClowno on 2004-08-19 10:56 ]</font>

blazesboylan · Post by **blazesboylan** » Thu Aug 19, 2004 1:19 pm

Dithering is also done by the SHARCs when changing bitrate to send from 32 bit float to 16 / 24 / 32 bit int ASIO, 16 / 24 bit ADAT / SMUX, and so on.

(For that matter it's also done in image processing, when scaling images.)

ScofieldKid · Post by **ScofieldKid** » Thu Aug 19, 2004 3:42 pm

Exactly. We are trying to divine if when you go from 32-bit INTEGER to 24-bit INTEGER, if that conversion will be done by truncation, or if it will be done by dithering. If it is being done by truncation, then we want to watch out for that big time.

astroman · Post by **astroman** » Thu Aug 19, 2004 4:29 pm

unless you have a high quality - highly dynamic signal (say solo violin in a classical orchestra recording) I bet you don't hear any (quality) difference at all, even from truncation.
You possibly may notice a slight difference due to the converter's playback (a 24 bit converter fed with a 16 bit signal), but not at 'regular' mixing levels.
We're not talking SACD mastering, do we ?

cheers, Tom

blazesboylan · Post by **blazesboylan** » Thu Aug 19, 2004 4:31 pm

SCOPE uses 32 bit floating point internally, NOT 32 bit integer.

Regardless of the results of this interesting investigation, I have no desire to "avoid" the SCOPE dithering algos. They sound good to me. I don't really care whether or not they're sub-optimal. If the dithering turns out to be some kind of shortcut, who's going to go throw their SCOPE cards away? Not me.

Stige · Post by **Stige** » Thu Aug 19, 2004 4:40 pm

On 2004-08-19 17:31, blazesboylan wrote:
SCOPE uses 32 bit floating point internally, NOT 32 bit integer.

I'm afraid I have to disagree.
But I'm too tired at moment to type, so please consider this as an opinion among others.

blazesboylan · Post by **blazesboylan** » Thu Aug 19, 2004 5:03 pm

Hmmm, maybe you're right. (Maybe we're both right...?!?)

SHARC ADSP-21065L

Sorry for spreading mis-information, I was sure that the SHARCs were floating point only...

SFP could be done entirely using 32 bit ints after all. I'd like to see the reference, but regardless you all have my apologies!

Johann

alfonso · Post by **alfonso** » Thu Aug 19, 2004 5:15 pm

Things are a bit confusing.
I've always thought that SFP was 32 integer, and what I know for sure is that VDAT records in that format.

Confusion is brought from what is written on two different pages of CWA site.

I quote what's said in the Introduction page for Scope platform:
"The employed SHARC DSP was specifically designed to perform high-resolution audio processing. It always processes the audio with a 32-bit resolution, and algorithms and parameters are computed with a floating-point resolution of even 40 bits. "

This seems a confirmation of the 32 integer resolution...

But in the Scope Project page (the link 3 places lower after Introduction), in the Highlights paragraph, processing is reported to happen at 32bits floating point.

This creates some confusion....

blazesboylan · Post by **blazesboylan** » Thu Aug 19, 2004 5:49 pm

It probably is both then.

The older SHARC can handle 32 bit int and 32 bit float.

The newer SHARCs also support 40 bit float, so maybe the old ones do too... Dunno.

Anyway it makes sense that the data flies around as integers and the math processing is at least sometimes done as floating point. There are many algorithms that nobody in their right mind would optimize for integer...

<font size=-1>[ This Message was edited by: blazesboylan on 2004-08-19 18:51 ]</font>

j9k · Post by **j9k** » Fri Aug 20, 2004 1:45 am

whether certain calculations are done with floats is impossible to tell but the sync and async signals are all done with 32 bit integer.

if you want to find out how the signals are being transfered between your recording software and the scope card record a signal into a 2track editor and look at it. also try attenuating the signal on one side of the exchange and boosting it on the other.

what would your audio sound like it was dithered over and over as you build your mix or do multiple passes for processing.

j9k

ScofieldKid · Post by **ScofieldKid** » Fri Aug 20, 2004 2:48 am

Yeah, I understand what Blaze is saying about not really worrying to much about this... but since there are a lot of audio geeks here...

I was thinking on the way home today that truncation in PCM could blast you in both FREQ and in AMPLITUDE. So a simple test might be to generate a 440Hz signal, and then push it to a near-peak level. Then check out the values of the samples in 32-bit INT. Then the fun part, take that, and just blast it to 24-bit INT, and see what you end up with. If it's truncating, you could end up with next-to-nothing on output.

Someone correct me if this I'm wrong. I'm just thinking about amplitude for this experiment...