Whats The Deal With The Summing?

  • Thread starter Thread starter illacov
  • Start date Start date
regebro said:
It's MATHS.


and you then reduce it back to the standard internal bith depth.


You don't have to divide after the summing! When using floating point maths, it IS only addition. Tada! :) However, it introduces a glitch, and that is that you now may need to do rounding.

The rounding problem is actually exactly the same problem as you have with any division / bit reduction. So in practice the difference suddenly becomes nil. ;)

The solution to division/bit reduction/rounding is usually known as "dithering". This is nothing new, nothing strange, and nothing magic about this.

What error corrections? Do you think the computer suddenly will think 1+1 is 5?

Now if you want to listen or not is not up to me to decide, you can go around with your head in a bucket of jello, that's your problem. But this is the truth. The real deal.

but I guess we are dealing with dogma and religion here, as usual.
:rolleyes:

There it is again. You make a statement and declare it to be the undeniable truth and call everyone who disagrees to be dogmatic. Huh?

Everyone else has thier head in bucket of jello.

My problem is that my head in a bucket of jello.

Okie Dokie.


Classic Post Lenny

SoMm
 
There it is again. You make a statement and declare it to be the undeniable truth and call everyone who disagrees to be dogmatic.

Yup. And I also added an explanation on WHY it was the undeniable truth. You are welcome to read it.

Classic Post Lenny

Thank you.
 
Most big films are mixed on digital consoles. I think they sound great.

The Sony Oxford console sounds great. It's digital.

There are many pros out there who could mix on a Neve or an SSL, and yet choose to do otherwise. Some of them even mix on digital. In fact, most major console manufacturers are primarily making digital mixers now. Why is that?

I don't think the answer is as simple as "digital summing sounds bad." I think the answer has more to do with the professional market's overall comfort level with technological changes. Maenwhile, the new kids coming up in the business are caring less and less about whether it's analog or digital...
 
TexRoadkill said:
So far my experience has been that DAW mixes tend to sound a bit more flat and stale compared to analog mixes. Whether that is simply the mixbuss or everything else combined I don't know but I think it's the latter.

i think the latter has to do with poor conversion....
shitty conversion will SUCK the life out of anything...YMMV
 
Son of Mixerman said:
Its pretty scary to think of 24 tracks represented by a number


You know, it's also awfully scary to think that 24 tracks will be represented by a single constantly-changing voltage value too! Or that all those instruments will be represented by a single (or maybe 2, or 3) mechanical moving piece of paper! Or that the sound of all those wonderfully-recorded tracks will eventually be represented by - of all things! - a single damn electrical signal (that caries a value, a NUMBER) in a nerve in our brain! Or that everything will be subject to how our brain interprets that signal - are all the chemicals and hormones in optimum amounts to regain the full 24-track experience? Is there enough blood going through the aural processing centre? Is the listener on drugs which will fuck around with that value represented inside our mind?


Case in point: representative values are everywhere. I wouldn't get too caught up by worrying about the inherent problems in what is the MOST ACCURATE and the only essentially perfect part of the number chain.

If you want to improve the mathematics in your signal chain, start where it's worst: take care of your health and hearing.
 
charger said:
most major console manufacturers are primarily making digital mixers now. Why is that?

Are they really?

I didn't know that Neve, API, SSL, Neotek, etc "are primarily making digital mixers now"... I thought they all made predominantly analog consoles, with an occasional digital product thrown in, if at all?

I learn something new everyday...:D
 
charger said:
There are many pros out there who could mix on a Neve or an SSL, and yet choose to do otherwise. Some of them even mix on digital. In fact, most major console manufacturers are primarily making digital mixers now. Why is that?

Because they cost less to make, therefore they sell more?

I posted this on another thread a while ago:

http://mixonline.com/ar/audio_chris_lordalge/

Here's the relevant bit:


--------------------------------------------------------------------------------

What is it you love about this [SSL 4000 G Plus] console so much?

Bottom line, this console has attitude. In 10 more years, these consoles will be like old Neves. They're classic. This one was installed in about 1985; it's modified as far as you can go, and it's in great shape. And it's got light meters, which I like. I'm all about the meters.

What do you mean?

I've got to see what's going on. I don't do this by ear only! I use the meters to balance things left and right, and to see what's going on with each fader so I can optimize the console.

They're set to show input?

No, the output of the fader — fader to mix — EQ'd and everything. It's all about maximizing the signal strength. You hit the tape machine a certain way, the tape machine is hitting the console a certain way, you're hitting the mix bus a certain way. You're at the sweet spot. All consoles have a pretty small sweet spot where it really sounds good.

That's something you don't hear mentioned much lately.

Well, everybody's thinking because they're digital, they don't have to worry about a sweet spot. Well, yeah, you do. Digital craps out so quick it's disgusting. In an all-digital console, the sweet spot is tiny — like a postage stamp. But there's a small window on any console where the headroom is right and where it really sounds the best. That's where I stay, in that window.

Oh yeah, and remember how valves were superceded by transistors? Valve gear was worthless for a period there, but we have now learnt our lesson. And this will probably apply to digital too. Actually it already started happening a couple of years ago, but a lot of people haven't realised it yet.
 
vox said:

Floating point arithmetic is a way to not have to have parts of the application where you have 24 bits and other parts where you have 48 or 56 bits, and doing conversions between those. Instead you run 32 bit all the way. Less processor power is used, and each plugin does not have to convert down to 24 bits at the end.

The manufacturers that decided to use floating point thought it sounded better than 24 bits,

The article above basically argues that 48 bit fixed point is better than 32 floating point. Well, nobody doubts that. Big straw man argument there. :D

As afinal argument he uses a "3-digit processor". It's a good way to get his point across. But he ends up with the numbers 21498 and 21200, to show the difference between 48 bit float and 32 bit fixed. What he forgets is that both those numbers at the end is gonna be dithered down to a 16-bit format for CD. And then those differences will simply disappear, because they are down in the details that get dropped.

Strange how he forgot to mention THAT! :D
 
I didn't know that Neve, API, SSL, Neotek, etc "are primarily making digital mixers now
SSL makes these digital consoles:
http://www.solid-state-logic.com/film/products/avantfilm/index.php
http://www.solid-state-logic.com/music/products/c200/
http://www.solid-state-logic.com/music/products/mtplus/
http://www.solid-state-logic.com/broadcast/products/c100/
http://www.solid-state-logic.com/broadcast/products/c200/
http://www.solid-state-logic.com/broadcast/products/AysisAir/
http://www.solid-state-logic.com/broadcast/products/MTproduction/
(they still make the analog consoles too)

API doesn't make consoles any more, as far as I can tell from their website. They make analog outboard though.

AMS-Neve makes these digital consoles:
http://www.ams-neve.com/prod/cap.htm
http://www.ams-neve.com/prod/lib.htm
http://www.ams-neve.com/prod/dfc.htm
http://www.ams-neve.com/prod/lmmc.htm
http://www.ams-neve.com/prod/libp.htm
http://www.ams-neve.com/prod/logics.htm
http://www.ams-neve.com/prod/libl.htm
In addition, they make 2 analog consoles:
http://www.ams-neve.com/prod/btc.htm
http://www.ams-neve.com/prod/88r.htm

Neotek makes the Elan and the Elite, and they are analog.

Euphonix:
http://www.euphonix.com/ - since the title on their website is Digital Audio Mixing systems, I'd guess that's their focus.

Sony:
http://bssc.sel.sony.com/Professional/webapp/SubCategory?m=0&p=10&sp=84&sm=0&s=&cpos=
http://bssc.sel.sony.com/Professional/webapp/ModelInfo?m=0&sm=0&p=10&sp=87&id=56147

Calrec:
http://www.calrec.com/pages/products/range.htm
They make 4 digital consoles, 4 analog consoles, and 1 small analog mixer.

Studer
http://www.studer.ch/products12.htm (digital)
http://www.studer.ch/products11.htm (analog)
They make 2 analog consoles and 8 digital.

Soundtracs http://www.soundtracs.com/soundt/docs/front/front_3.htm
Soundtracs makes only digital consoles now.

Amek:
http://www.amek.com/
Amek makes all analog consoles. Pretty ones, too, that were designed at some point in their past by Neve.
 
The article above basically argues that 48 bit fixed point is better than 32 floating point. Well, nobody doubts that. Big straw man argument there.
Actually, people doubt this all the time. Pro Tools TDM/HD (the versions that use DSP chips) use 48-bit fixed processing, and other companies (and users) whose products use 32-bit floating point processing argue its superiority all the time.
 
Well, at least when it comes to summing, and summing less than 16000 channels, 48bit will be more accuracte than 32bit. I don't know enough about filters to say.

But probably it's all just bullshit. 48bit processing *should* produce more accurate results, unless you find yourself needing to dither down all the time, and with 48 bit that shouldn't happen too often...

I'm so annoyed I can't find that article that had all the specs about protools summing. :mad: Does prottool really use 48 bit all the time? Or it it just the summing? Because if it's all the time, the otehr just aint got nothin' on it. If it's just the summing, then they have an argument. Especially if the filters are just 24 bit.
 
regebro said:
Well, at least when it comes to summing, and summing less than 16000 channels, 48bit will be more accuracte than 32bit. I don't know enough about filters to say.

But probably it's all just bullshit. 48bit processing *should* produce more accurate results, unless you find yourself needing to dither down all the time, and with 48 bit that shouldn't happen too often...

I'm so annoyed I can't find that article that had all the specs about protools summing. :mad: Does prottool really use 48 bit all the time? Or it it just the summing? Because if it's all the time, the otehr just aint got nothin' on it. If it's just the summing, then they have an argument. Especially if the filters are just 24 bit.

PT does its internal calculation 48bit fixed but dither back down to 24 EACH time
 
OK, finally, here is the deal on ProTools summing:

Yes, the summing is done with 48 bits internally. It's done by multiplying the value with the fader setting (including panning). This is apparently done with integer multiplication, which mean sthat you get no rounding errors. However, the result is larger than the original. But since the original is 24 bits, and you have 48 bits to play with, that is not a problem.

ProTools apparently has a unity gain of 2**14 for the faders in the summing. That means that the minimum value you can set your faders to is around -80 db, WITHOUT LOSS OF SOUND QUALITY. So you have to lower the volume so much that you can't here it before the sound quality suffers. It also means that when mixing with unity gain, your signal will use up 14+24 = 38 bits. That leaves 10 bits headroom. The ten bits of headroom means you can mix 1024 channels at unity gain before hitting the headroom ceiling.

So, when mixing you can, with protools, mix up to 1024 channels, and have faders set anywhere between -80db and 0db without the resulting 48 bit sound having any loss of soundquality from the summing.

After that, you need to get those 48 bits down to 24 again. This is done with dithering. Dithering, as wel all know, adds one bit of noise. This means that you will get one bit of noise added at around -120db. Note that this dithering is only done ONCE.

It seems to me that protools are doing all this completely correctly. There simply can be no practical loss of sound quality when doing it like this.

Once again we are forced to the same conclusion: If mixing through an analog console sound better than mixing with the internal ProTools mixer, it si NOT because there is something flawed in the digital summing process.

Now, will you give this myth a rest? There is nothing flawed with digital summing? OK?
 
Here is a link from digi supporting what Lenny is saying.

http://www.digidesign.com/digizine/archive/digizine_january04/contents.cfm?body=techtalk/

Here is the link to where 3D audio has gone around with looking into the problem. Many of the people in that forum are experts with more than just a calculator. While I will concede that the basic math behind summing is simple addition, the results of the real world data seems to suggest something else is going on. Look specifically at the FFT plots from Spectrafoo of the differences in summing. Nika was showing how simple summing is, (Lenny seems to be right) because were dealing with ones and zeros. But the summing issue seems to lie in other areas that cascade into the mix bus. Lynn noted that a simple twist of a pan pot or fader move on a 16 bit fader instantly kicked the output of that channel into a 24 bit number, add a plug-in and depending on which one it gets kicked into either a 32 bit float or 48 bit fixed number. Where I think PT and maybe other companies are struggling is not the summing but rather the method in which they reduce it back down to the 24 bit mix bus after the bit extension for DSP.
Nika confirmed that the math IS bad in alot of the digital mixers. I knew this but had forgotten. There was a time and to some degree even today that some programs and mixers rounded down during bit reduction while others were truncating. The problem as Lenny said Isn't the summing.... well not completely, its the subtraction thats killing the summing bus. So the summing math isn't simple addition in the end is it. The summing bus is using addition, subtraction and/or whatever dither algorithm the program used to get the bit reduction done. Nika pretty much said the math is wrong, and that it comes down to when and where you dither. Space and Processor power puts limitations on how and when dithering occurs and it all costs money to implement correctly. He noted that the reason the Sony Oxford is so pricy and sounds so good is that it has 440 processors on board. One third of those are dedicated to dither. Basically dithering has to be done everytime there is a bit reduction irregardless to where i.e. Plug-ins summing etc
and that plug-ins, eq's or any DSP that does not dither introduces distortion and errors that do mirgrate into the mix bus. Digital mixers make things worse. And... that trying to dither a 32 bit floating point if so hard most comapnies don't bother adding that to the math. 48 bit fixed is more accurate and dither everything that is undergoing bit reduction.


Also recommended reading:

http://www.cadenzarecording.com/papers

http://www.3daudioinc.com/cgi-bin/ultimatebb.cgi?ubb=forum&f=19

Well I learned a whole bunch of stuff and loosened up some cobwebs. Digital summing only flawed on certain systems because of poor implementation due to various constraints like money. Digi knows all of this and thats why they have made changes with the HD. Another problem is the Sigma Delta ADC and DAC choices for each system. Go read the DAW SUM thread at 3D Audio.

Buy a Sony Oxford..set it and forget it.

SoMm
 
I suspect that as we get more and more processing power, the DAW manufacturers will stop fiddling about, and basically only convert "upwards" in the whole chain, and then end up with some ridicolous 256 bit summing or something, to make sure that it's never a problem, ever. :)
 
Back
Top