Dithering analog summing and hardware inserts

When researching my DAC and how it operates I found this in regards to audio signal hitting DAC ...

Cs4398 DAC adds the type of dither below to the output , or that's how I read it ...

Dynamic Range (Note 1) 24-bit A-Weighted

Note 1:One-half LSB of triangular PDF dither is added to data.

Product Manual for Cs4398 24 Bit 216khz DAC
alsa-project.org/~james/datasheets/www.cirrus.com/en/pubs/proDatasheet/CS4398-2.pdf

Is this quote accurate ?
" rounding off or truncating is adding variable amounts of DC before the quantization step, and does nothing to decorrelate the error from the signal."


Paul From lynx:

"When a DAW says 32-bit float, that refers to the resolution for internal processing; fades, plug ins, reverb tails, etc.
The audio that gets delivered to the audio card is already at 24-bits. The number of bits representing the waveform is never more than 24. There is, essentially, no 32 bit hardware readily available, and it is arguable what the benefits would be to have more dynamic range than what is offered by 24-bits.

The DAC does not need to dither, because the audio is already 24 bits when its delivered to the DAC." End of 2016

I rarely work at 32 bit float so I am more interested in how this affects 24 bit fixed point data... 24 bit offers enough Headroom for me, and if Paulnfrom lynx is correct with the above quote I feel better about my workflow.
Will keep researching for myself, I posted hoping the info may help someone.

ragan call apogee and ask them about the DAC is my suggestion

I don't want to speak for Dave, or quote him without his permission, but he said to the effect the extra 8 bits in 32 bit float keep track of the decimal point enabling much more dynamic precision in the computer, but the digital audio is always 24 bit when outputed to an interface, and there's no truncation involved in that math process. 

Not to say the analog conversion of those 24 bits couldn't or wouldn't sound more pleasing with the addition of dither on the way. It well could.

I think the lynx guy is saying that the DAW doesn't pass 32 bit to the DAC. The DAW outputs 24 bit.. there for LYNX don't need to dither... if so, that doesn't negate the fact that the DAW should have dithered its 32 bit output to the DAC in the first place.

cheers

Wiz

Well, I heard back from J_J - he seems to be having some difficulty signing up for an account here (I seem to recall something similar and having to email JK to clear it up), but here is his answer to me -


I think that someone at MOTU may not quite understand what's going on, or how their hardware actually works.

How the hell is the DAW program going to not feed the interface 32 bit float? The DAW would need to be set to 24 bit fixed point operation. The dither needs to be provided between the DAW program and the interface hardware. The PR/Customer Service guy doesn't understand how the system works. Where does he think the dither is applied?

Obviously I don't mean the bits themselves are somehow physical sound waves, John.

I know you like to play the scolding contrarian but there's no need to straw man the (potential) point Don was relaying.

Perhaps my humor is a bit too dry...?

I start to get a bit "battle worn" as an engineer once we hit the "this is best practice - although you can't actually hear it doing anything" zone.

Although I agree - if it's best practice, it only takes a second to add a dither plugin (it's taken 100,000 x longer to debate it in this thread!)
I do feel a bit daft though - adding a plugin to solve an issue I can't actually detect on a standard double blind ABX. 

Q. How do people in this thread deal with DC Off Set?

Do you remove it on a track by track basis?
Do you just remove it from final mixes at Mastering?
Do you even check if it's present in a file?
Are you able to hear DC Off Set?

I'm no expert on dither but I believe a point that has been made a few times in this thread has been that, even if you can't hear it now in the "early" stages of tracking/mixing, it's possible you might hear it later after mastering when things are no longer in your control and buildup of those quant. errors over multiple rounds of truncation could become audible. Maybe you won't hear it after one round. But the buildup that occurs from not dithering at each occurrence of truncation would eventually cause it to become audible. Which, if my understanding is correct, would make the first round of truncation the WORST one to not dither even if it's the one where you may stand the least chance of hearing the difference at the time of that particular dither application.

Or something like that....

OK!

I'm not a mathemetician like J_J, but that quote strikes me as hogwash. Yes the low 8 bits are used for mathematical overruns, but where the hell does he think those overruns come from?

From computing the wave form, that's where. It IS "part of the waveform".

If this was analog there wouldn't even be an argument, you could just slap a distortion analyzer on the output, but since it's digital some people think they can ignore  details willy-nilly and it doesn't matter for some reason. This is a problem that afflicts a lot of things in digital audio - simply ignoring things that are inconvenient or might make the coding more difficult because "they don't matter" because "people won't notice."

Well, EVERYTHING matters and some people DO notice. Just because programmers (whose ears are not trained to hear fine detail) can't hear the effects doesn't mean that they don't exist, doesn't mean that people with better ears can't hear them, and doesn't mean that cumulative errors don't stack up.

The other thing is that "Mr Paul" is either intentionally  distorting the question or (hard as it may be to accept) simply doesn't understand it. Nobody said that the DAC hardware needed to dither. That's not the question, not what we're talking about, and frankly is a pretty nonsensical idea. The point is that in a 32 bit system (floating or fixed), dither MUST be applied BEFORE it hits the (24 bit) DAC. In the software. NOT in the hardware.

The idea that dither should somehow be implemented in the DAC is a red herring - it just distracts people from the real issue.

You

are

forgetting

what I've already said several times in this thread.  


The MOTU AVB products report as receiving 32 bit float words from the computer, both by the computer and by the literature on the DAC chip in question.

Anything Lynx says has to do with their product. Period. No inferral past that.

And

The received 24 bit word, if that is all that is transmitted, is either truncated or dithered.

It is dithered to 24 bits if the DAW is correctly implemented. See: Bell Labs.

Many are not. See: decades of observations made by Bob O and others.

You may not hear any of this today. You may crank levels around drastically tomorrow and find that everything suddenly seems edgy because you turned it up a bunch of low level truncation error distortion.

Or

You can just turn on dither everywhere, install it in your templates, and comfortably never think about it again.

Or

You can not turn it on, and keep wondering and debating.


I guess this is the equivalent of an office water cooler discussion. 

So can we assume that, in the unique case of the UAD Apollo, the dithering is occurring in real time within the Apollo when UAD plugins are printed in real time as something is tracked but that, when the Apollo is used simply as DSP during mixing, it should be treated as any other plugin associated truncation and thusly dither should be manually applied at any subsequent DA output?