dBu to 0 dBFS

[JohnRoberts]

This is why I tried to avoid this veer, and am surely repeating myself (I'm starting to get used to that around here.).

For low bit/word length that classic number of quantization levels is adequate to characterize a path performance..... But modern high bit/word length systems do not follow the classic behavior all the way down to the LSB (actually a good thing).

There have been numerous tutorials written to explain this better than I will but one important observation, the analog (output) side of D/A conversions is not discrete steps, but a continuous (smoothed) analog voltage, so don't expect to count stair steps.

Modern convertors typically perform an over-sampling conversion at their input. The analog stream is converted at a very high sample rate, but shorter digital word length, then that digital stream is mathematically crunched (decimated) down to a lower sample rate with longer word length. The noise floor is dominated by this oversampling stage and not only analog sounding (hiss instead of grainy quantization) but manipulated to shift noise above the audible band.

When characterizing the behavior of modern high performance convertors try to resist the attraction of digital math, and look to old school analog metrics, like S/N, THD, etc, measured the old fashion way. 

Yes listen to the digital noise floor, you will probably hear other gear in your chain, or an analog sounding hiss.   

I am old enough to remember when digital noise floors were something you could identify and had to worry about. Lots of time and effort wasted to manage gain structures to keep signal above the dreaded quantization floor.

JR

 

[mattiasNYC]

It is, because it just contradicts the mathematical definition of resolution.

No, it isn't. I wrote "the way we commonly think about it. " for a reason.

 

[abbey road d enfer]

No, it isn't. I wrote "the way we commonly think about it. " for a reason.

Well, you wrote: "... the point was merely to illustrate that we increase dynamic range, not "increase resolution" ...
That's what I disagree with. And that's not the way I (and a large number of knowledgeable people) commonly think about it.

 

[mattiasNYC]

But the signal that you wish to capture won't be captured with any more detail in a 24 bit system compared to a 16 bit one. The difference in both cases is the LSB. But whenever you get an error you end up with quantization distortion. That's the error. With more bits you push that down (from full scale), but you're not otherwise gaining any information or any greater detail or resolution of the signal you actually care about and want to capture cleanly.

In other words a 1k sine wave at -20dBFS will be equally well described in a 16 bit system as it is in a 24 bit system. The latter doesn't show the 1k sine wave with any greater resolution.

The way people commonly think of the term resolution it refers more to "grainyness" rather than smoothness, to "steps" rather than continuity.... etc... of the source we want to capture/recreate.

 

[benb]

I'm hesitating to say anything about recent posts (which have veered off from the OP's question), but instead I'll just post this link:
http://www.ti.com.cn/cn/lit/an/snoa232/snoa232.pdf

 

[abbey road d enfer]

But the signal that you wish to capture won't be captured with any more detail in a 24 bit system compared to a 16 bit one. The difference in both cases is the LSB. But whenever you get an error you end up with quantization distortion. That's the error. With more bits you push that down (from full scale), but you're not otherwise gaining any information or any greater detail or resolution of the signal you actually care about and want to capture cleanly.

  I just disagree; "detail" is not a mathematical concept, resolution is. You cannot freely exchange one for the other.

In other words a 1k sine wave at -20dBFS will be equally well described in a 16 bit system as it is in a 24 bit system.

Not true; the difference in distortion justifies the difference in "describing as well". Look at the ethymology of the word distortion.

The latter doesn't show the 1k sine wave with any greater resolution.

Yes it does. Just read the definition of resolution. It may not make much audible difference  (-120dB THD is as much inaudible as -90), but there is a measurable difference. In video/imaging, the difference in resolution clearly translates into a difference of "detail". That's because video systems, contrary to Hi-Rez audio systems, operate with a restricted bit depth compred to the discrimination capability of the visual system.

The way people commonly think of the term resolution it refers more to "grainyness" rather than smoothness,

Both these terms are audiophool language; they have nothing to do in a scientific discussion.

to "steps" rather than continuity....

steps are a graphic concept associated with visualization of digital conversion; they have their use in explaining how increased resolution results in better precision, but they tend to imply things that don't exist in reality.

 

[JohnRoberts]

Can we at least keep this to facts, not what "people" think the facts are?

JR

PS: I agree with abbey....  i couldn't  open benb's link (may be a personal problem) but there are many good tutorials about this stuff around the WWW.

 

[kambo]

i uploaded benb's pdf doc.

AN-804 Improving A/D Converter Performance Using Dither

https://spaces.hightail.com/receive/MtygJ

 

[JohnRoberts]

i uploaded benb's pdf doc.

AN-804 Improving A/D Converter Performance Using Dither

https://spaces.hightail.com/receive/MtygJ

That time I could only see the title page, but that is enough.  "Dither" is another relatively early digital conversion enhancement technique.

Dither can toggle the LSB resembling a pulse width modulation to encode less than LSB amplitude information into the data stream. Obviously this mainly works for lower frequency content due to sampling criteria.  I will ASSume the link I haven't read yet properly explains dither (there was more than one approach back when).

It is mostly academic (?) since the modern oversampling convertors are self-dithered by the noise of their front end oversampling stage (wide band noise is an effective dither stimulus). 

Again I repeat my advice to look at and trust the analog bench measurements for convertor performance. Overthinking this can kill brain cells for little reason, while more knowledge is never a bad thing, these are relatively old and settled issues.

Modern codecs are pretty damn good.

JR

 

[Andy Peters]

But the signal that you wish to capture won't be captured with any more detail in a 24 bit system compared to a 16 bit one. The difference in both cases is the LSB. But whenever you get an error you end up with quantization distortion. That's the error. With more bits you push that down (from full scale), but you're not otherwise gaining any information or any greater detail or resolution of the signal you actually care about and want to capture cleanly.

In other words a 1k sine wave at -20dBFS will be equally well described in a 16 bit system as it is in a 24 bit system. The latter doesn't show the 1k sine wave with any greater resolution.

It is important to remember that when you "capture" a signal with a 24-bit word, you actually use all 24 bits. In other words, the 24-bit value 0x3F0000 is not the same as the 16-bit value 0x3F00.

Why?

Because the next sample might be 0x3F0006, and you need all 24 bits to describe that value. Capturing that same value in 16 bits gives you 0x3F00.

So, yes, capturing a 1 kHz sine wave in 24 bits "gives you more" than capturing same in 16 bits.

 

[kambo]

i just recorded 1khz sine wave  with microphone and mic preamp,
one at 16bit/48khz, other at 24bit/48khz...

exported both to 32bit float...
they NULL

edit: converted 16 bit to 24 bit,  NULL again

 

[mattiasNYC]

  I just disagree; "detail" is not a mathematical concept, resolution is. You cannot freely exchange one for the other.

  Not true; the difference in distortion justifies the difference in "describing as well". Look at the ethymology of the word distortion.

So your argument against what I'm saying is disputing something I wasn't talking about essentially.

Yes it does. Just read the definition of resolution. It may not make much audible difference  (-120dB THD is as much inaudible as -90), but there is a measurable difference. In video/imaging, the difference in resolution clearly translates into a difference of "detail". That's because video systems, contrary to Hi-Rez audio systems, operate with a restricted bit depth compred to the discrimination capability of the visual system.

So if you create a 1kHz sine wave at -10dBFS at 24bits, export that to 16bits, bring it back into your DAW, line it up next to the original, invert phase, what do you get?

Both these terms are audiophool language; they have nothing to do in a scientific discussion.

Ok, well I appreciate your moderation of my use of language, but I was under the impression that one was allowed and justified in using language which in context made a point despite being not 100% scientific.

Why don't we just go read the formulas on digital sampling as applied instead? That way we don't have to worry about non-scientific language, no?

steps are a graphic concept associated with visualization of digital conversion; they have their use in explaining how increased resolution results in better precision, but they tend to imply things that don't exist in reality.

I know that. Why are you telling me this? Do you really not see the point I'm making?

 

[mattiasNYC]

It is important to remember that when you "capture" a signal with a 24-bit word, you actually use all 24 bits. In other words, the 24-bit value 0x3F0000 is not the same as the 16-bit value 0x3F00.

Why?

Because the next sample might be 0x3F0006, and you need all 24 bits to describe that value. Capturing that same value in 16 bits gives you 0x3F00.

So, yes, capturing a 1 kHz sine wave in 24 bits "gives you more" than capturing same in 16 bits.

Yes, but "more" of what, specifically? "More sine wave"?

 

[abbey road d enfer]

So your argument against what I'm saying is disputing something I wasn't talking about essentially. 

If it wasn't something you wanted to mention, why did you write it, then?

So if you create a 1kHz sine wave at -10dBFS at 24bits, export that to 16bits, bring it back into your DAW, line it up next to the original, invert phase, what do you get?

You get the distortion residual, at about 70-80 dB below FS. Although a null test will not tell you where this difference comes from; I know where it comes from, because a converted sinewave has more distortion at 16 bits than at 24 bits.

Ok, well I appreciate your moderation of my use of language, but I was under the impression that one was allowed and justified in using language which in context made a point despite being not 100% scientific.

It doesn't work like this; you're attacking the fundamental basics of digital conversion with improper language. It's like saying that you're allowed to say that 2+2 may equal 3 or 5 on the basis that "equal" means "similar" to you.

Why don't we just go read the formulas on digital sampling as applied instead? That way we don't have to worry about non-scientific language, no?

Huh? May I remind you that resolution is a concept that exists independantly of any application. Why would it change in the context of a specific application? If you decide to give a different meaning to resolution, it's your choice, but I don't have to adhere to it.

I know that.

But you keep on using improper terminology. Scientific communication relies on using the proper words.

Why are you telling me this?

Because you have mentioned it, in a way that convinces me you don't fully grab its significance. As a moderator, I have to make sure that such ambiguities are addressed, in order to avoid proliferation of erroneous concepts.

Do you really not see the point I'm making?

No. I may be thick, though.

 

[kambo]

isnt the higher sample rate  = more resolution
that what my DAW says to me...

 

[mattiasNYC]

If it wasn't something you wanted to mention, why did you write it, then?

Ok, hold on for a second:

If you want to get very nit-picky about the language we use and how we speak about this, then we should start with the above. What  was said was:

ME:    In other words a 1k sine wave at -20dBFS will be equally well described in a 16 bit system as it is in a 24 bit system.

YOU: Not true; the difference in distortion justifies the difference in "describing as well". Look at the ethymology of the word distortion.

I then asked you what happened if you created two sine waves, one derivative of the other but at a lower bit depth, and then inverted one and summed them. Your reply is:

" You get the distortion residual, at about 70-80 dB below FS. Although a null test will not tell you where this difference comes from; I know where it comes from, because a converted sinewave has more distortion at 16 bits than at 24 bits."

And that is exactly my point. The sine wave itself is what we care about, and it is defined well. Everything else, stuff that we care about only because it's stuff we don't want, is distortion residual. We both know it comes from.

Now, if you go back to my statement above, you should be able to see two things, and the more important one is that I'm NOT saying that the "difference in distortion" DOESN'T 'justify'  "the difference in "describing as well"". I agree with that.

The other thing that is clear is that increasing our bit depth increases our dynamic range, which in turn allows this residual distortion to exist further away from our equally well described sine wave.

Someone posted a video in this section where a guy goes through digital audio theory, and in it he essentially says the same thing, and for the same reason.

It doesn't work like this; you're attacking the fundamental basics of digital conversion with improper language. It's like saying that you're allowed to say that 2+2 may equal 3 or 5 on the basis that "equal" means "similar" to you.
Huh? May I remind you that resolution is a concept that exists independantly of any application. Why would it change in the context of a specific application? If you decide to give a different meaning to resolution, it's your choice, but I don't have to adhere to it.
But you keep on using improper terminology. Scientific communication relies on using the proper words.
Because you have mentioned it, in a way that convinces me you don't fully grab its significance. As a moderator, I have to make sure that such ambiguities are addressed, in order to avoid proliferation of erroneous concepts.
No. I may be thick, though.

Whatever.

 

[mattiasNYC]

i just recorded 1khz sine wave  with microphone and mic preamp,
one at 16bit/48khz, other at 24bit/48khz...

exported both to 32bit float...
they NULL

edit: converted 16 bit to 24 bit,  NULL again

This gets to what I'm pointing out.

 

[kambo]

here is the distortion figures from REW
16bit


edit:
some reason i wasnt able to measure noise/THD with REW at 16/24 bit correctly ... 
check for correct meaurements with RMA at:
https://groupdiy.com/index.php?topic=64948.msg823210#msg823210

edit 2: actually this is correct. without added dither!

 

[john12ax7]

Kambo,  I'm pretty sure something went wrong with your experiment.  Having the same noise and distortion in a 16 and 24 bit system doesn't make sense.

 

[kambo]

Kambo,  I'm pretty sure something went wrong with your experiment.  Having the same noise and distortion in a 16 and 24 bit system doesn't make sense.

i am kind a surprised too!
i triple checked it.... even i closed/reOpended REW with the new setting...
why dont u give it a go too, i am sure u have a copy of REW...
i was testing with ASIO at 48Khz ... lynx_aurora