Digi White Paper for HD 48 bit mixer

[Beck Henderer-Pena]

I just came across this...for anybody interested in the ongoing internal/external summing debate, it's a pretty interesting read.

I did a search and didn't see much reference to this up on this board...I appologize if it's something that has already been discussed at length.

Here is the link:
http://akmedia.digidesign.com/support/d ... _26688.pdf

I wish there was one for LE, but I can't find it...

Enjoy.

[JGriffin]

Yeah, i read that awhile ago, it's good. Explains a lot of this stuff about 'using all the bits' and things like that.

[joel hamilton]

I just came across this...for anybody interested in the ongoing internal/external summing debate, it's a pretty interesting read.

I did a search and didn't see much reference to this up on this board...I appologize if it's something that has already been discussed at length.

Here is the link:
http://akmedia.digidesign.com/support/d ... _26688.pdf

I wish there was one for LE, but I can't find it...

Enjoy.

Beck!
Hey man.
Yeah, we raged on that a while back. I called it a "republican spin on math."

Like , subtracting a little isnt actually subtracting at ALL if you ask them....

amazing.

Like we shouldnt be concerned at all with the "fractional loss" when "fractional gains" are all we do, every single day. It is all tiny little things that add up to the whole in this process, so OF COURSE we would be concerned with this "tiny loss" that digi talks about being "subjective" like "subjective" is a bad word.

What we fucking DO is subjective!!??!??!?

It is an interesting read though nonetheless.

[cvanwinkle]

I wish there was one for LE, but I can't find it...

Enjoy.

PT LE (& M-Powered) uses a 32-bit Floating point engine as opposed to a 48-bit fixed point. Thus, any information you find out there about 32-bit float processing (i.e. just about every other DAW on the market) should apply to PT LE. You won't find much info on this from Digi since it isn't their flagship product.

-C

[Beck Henderer-Pena]

For what it's worth...I just got an email from the guy who wrote the doc and said he is working on one for LE that will probably be done mid-next year...we'll see.

Question though: Is it really ture that any info about 32 Bit floating point would apply across the board, or do different companies implement the 32 Bit floating point thing differently? Isn't there some difference in the way the software is written that makes Logic different from LE different from LIVE?

[joel hamilton]

For what it's worth...I just got an email from the guy who wrote the doc and said he is working on one for LE that will probably be done mid-next year...we'll see.

Question though: Is it really ture that any info about 32 Bit floating point would apply across the board, or do different companies implement the 32 Bit floating point thing differently? Isn't there some difference in the way the software is written that makes Logic different from LE different from LIVE?

I hope someone can answer this.
I also hope someone can answer how "subtracting a little" is actually "not losing anything."

[Beck Henderer-Pena]

I also hope someone can answer how "subtracting a little" is actually "not losing anything."

me too!

[cvanwinkle]

Is it really ture that any info about 32 Bit floating point would apply across the board, or do different companies implement the 32 Bit floating point thing differently? Isn't there some difference in the way the software is written that makes Logic different from LE different from LIVE?

I don't think I'm necessarily the right person (or even qualified) to give a definite yes or no to the question, but here is my best stab:

Most (if not all) modern audio apps that use a 32-bit float engine would be using the IEEE Standard for Binary Floating-Point Arithmetic (IEEE 754). This is a widely adopted standard for arithmatic in computers. It is supported by the OSes, the Chips, the source code languages, and their compilers. As for differences in software from different companies, there shouldn't be any difference if all we're talking about is mixing (i.e. the summation of signals) which equates to the adding of signals (or depending on how you look at it, multiplying). In this case, math is math and should be straight across the board. A change in +6.22 dB in Charlie's DAW should be the quantifyably the same as a change in +6.22 dB in Jack's DAW, otherwise one or both of their DAW's aren't doing what they claim and thus shouldn't be used.

If you want to get into arguments about rounding error in floating point computations, someone else will have to post cause I don't want to get into that.

Since it sounds like I'm not the only person who reads white papers on the forum, I'll direct the daring to this article: http://docs.sun.com/source/806-3568/ncg_goldberg.html

With all that being said, there will be and ARE differences in implementation of various things as simple as panning law. There are multiple popular methods on how to do this which will yield different results. This is the same as if you were to put a pan pot on an analog console, put it at 2 o'clock and did the same on a different console the voltages would be measurably different.

Differences will also occur in just about any other DSP that includes filtering in the frequency domain or convolving in the time domain. Things that are conceptually simple like EQ will even fall into this category. The accuracy in which this is computed will depend on how much processing overhead the DAW manufacturer will want to allot for various functions.

So, if all we're talking about is just the static mixing of signals (which is what the Digi paper is mostly about) then there shouldn't be any differences. Dynamic mixing like applying automation to the signals will have a granularity equal to a certain buffer size. This will vary from manufacturer to manufacturer (and in most ASIO implementations will vary on your sound card's buffer size). If I'm not mistaken, this buffer size is also different in PT depending on whether you're playing back in real time or doing a bounce to disk.

Also of interest is that the current VST specification dictates that all VST interfaces (the in from the DAW and the out back to the DAW) are 32-bit float. I think this is being updated to allow for 64-bit interfacing.

---------------------------------------
If I'm wrong on any of this, someone please correct me. I hate to be spreading more of those internet lies.

-C

[Beck Henderer-Pena]

Since it sounds like I'm not the only person who reads white papers on the forum, I'll direct the daring to this article: http://docs.sun.com/source/806-3568/ncg_goldberg.html

Thanks for the reply! I'm not sure I'm smart enough for this article, but I'll give it a try...