ORIGINAL: no criminal intent

This is a really informative thread that has cleared up alot of things. Just one small question remains however. Should I leave dither on or not.

zumba

Mmm, you're asking for too much, methinks.

If Cakewalk were real intelligent with this they would beat everyone else to it and work out some easy sounding guidelines like this for example:

"The important thing to remember is whether to dither hither and thither (i.e., leave it on) or just hither or thither (i.e., in Sonar or at the mastering stage)."

But it appears that before we get to that level of simplicity several technical aspects need to be analyzed by those having way more knowledge than the likes of me (and don't anybody come questioning my "credentials" in regards to my ignorance).

Richard

FYI,
For you mere mortals interested in reading the "Nika Aldrich" explaination of "dither", here is a link for you. Try not to gobble up everything as absolute fact as I'm seeing from some in this thread. Take into consideration some of the steps that a DAC performs in the DtoA conversion. Then take into account that "Dither" is a form of distortion and it is accumulative everytime you apply it.

http://www.users.qwest.net/~volt42/cadenzarecording/DitherExplained.pdf

I have no idea where the intermodulative distortions are coming from in Pattors informative questioning. Even Nika refers to "harmonic distortion" and like I said, these are the things that the interpolator and anti-imaging filter section within a DAC addresses in the DtoA process, but I'm sure if you look at the audio signal thru a spectral analyzer "Pre-DAC", you would see a lot of these "harmonic distortions", like seen in Nika Aldrich's report. Maybe I'll do some studies tonight and show some bit reduced sinewaves and run them thru a Spectral Analyzer post D/A conversion for a comparitive study.

ORIGINAL: Rednroll

ORIGINAL: pattor

Uh....rednroll

The quantization distortion ain't just a pure harmonic distortion. The quantization can produce all kinds of distortion. How does the D/A handle the intermodulative ones?

Hmmm??? Stair stepping isn't just pure harmonic distortion???

Again, you are confusing the encoding and decoding mechanisms of digital audio where this "steppings" occures which is dealt with by the anti imaging filters in the output DAC and quanrization distortion due to bit depth reduction. They are two seperate issues.

Quantization distortion is not just stepping and isn't pure harmonic distortion. Because no anti aliasing filters are used during bit reduction, the harmonic distortion passes the nyquist frequency and aliases back into the audible spectrum aka the signal bandwidth. This distortion can not be removed by the filters in the DAC because they are all over the frequency range and not just out of band frequencies like the "stepping" in digital audio which is correctly removed by the DAC. (Assuming it is a good DAC of course).

That's funny my calculus classes always showed that when I added an infinite amount of harmonics to a sinewave, I ended up with a square wave (ie stair step) Ok, please explain?

Then you should also know that a true square wave needs infinite bandwidth which just isn't possible in digital audio. Hence the anti aliasing filters at the relevant places.

Intermodulative distortions? Ok, to be intermodulative doesn't it need to be "modulated" with another signal? Please do explain what that other signal is in a bit reduction process?

If I remember correctly, the sample rate and the signal that is being bit reduced.

You seem to keep wagging the dog, instead of answering with backing facts supporting your "theory" where we all could learn. I find it humorous that you come around with another question as a response. Have you ever thought about running for political office? Ok, I give up....so tell me how does dither deal with this "intermodulative" distortions you are referring too?

By linearising the quantae values. This is explained in Nika's article.

Please first explain where this "intermodulative distortion" comes from. Maybe you're referring to intermodulative distortion when the audio signal gets modulated with that noise of the dither being applied huh?

See above.

UnderTow

ORIGINAL: UnderTow

ORIGINAL: SteveJL

Great discussion everyone. I would like to offer my $.02 on this subject.

It has been discussed extensively here, and Cakewalk staff have also chimed in, and the bottom line remains the same:

Dither is NOT required for monitoring, as the output file is 24-bit and it goes through 24-bit DACS, PERIOD.

Dither is ONLY required during the FINAL Mastering stage.

Dither should ONLY be applied ONCE through the whole process, as it is distortion, and is cumulative.

Final PERIOD <g>.

Wrong!

Let me quote Ron Kuper for you:

Before 5.2, SONAR had a bug where dither would *not* be applied if you were using 24-bit hardware. In 5.2 we fixed this bug, but it now means that if you are playing through 24-bit hardware, your CPU load will increase due to the overhead of adding dither.

Also new to 5.2 is that if you are recording from 24-bit hardware, but your file type is 16-bit PCM, that SONAR will dither in that stage as well, also adding CPU load.

...

Dither is applied live during playback, when converting the internal 32-bit or 64-bit float values to whatever format is used by the audio h/w. Before 5.2 this would only occur if the audio h/w was using 16-bit PCM. In 5.2 it now also does it for 24-bit PCM.

...

As I mentioned earlier in this thread, in 5.2 we fixed a bug where dither was incorrectly not being applied when streaming out of 24-bit hardware. In 5.2 we now apply dither

...

You technically want to dither any time you are reducing the word length (bits per sample). So since SONAR internally works in 32-bit or 64-bit float, you would dither when converting this audio to 16-bit or 24-bit PCM.

Does that settle it?

Please guys, if you do not fully understand the subject, refrain from posting your misguided opinions as facts. Using bold font won't make you any less wrong either ...

UnderTow

In the same thread, Ron goes on to say: http://forum.cakewalk.com/tm.asp?m=747870&mpage=5&key=

ORIGINAL: Ron Kuper [Cakewalk]

Dither needs to be turned off and on depending on what we're doing?

You technically want to dither any time you are reducing the word length (bits per sample). So since SONAR internally works in 32-bit or 64-bit float, you would dither when converting this audio to 16-bit or 24-bit PCM.

However -- while you are tracking and mixing there is no absolute need to dither, since you are "merely" dithering for your own ears as you track and mix. You could consider dithering only when you produce your final mix or stems, since at that point you are comitting the sound to a new file.

I will personally continue to leave the option off, as it makes no sense to me to leave it on all the time. But I will continue to observe the discussion, and would like to see Cakewalk get more involved also.

As was said there is no ABSOLUTE need to dither. This goes right back to the original point of this thread:

As patton suggested, if you do the entire mix without dither, your ears have (to some extent) be exposed to the harmonic distortion caused by truncating the Sonar 32/64-bit floats to the 16/24-bit finite words that you soundcard can handle. This we can all agree on I hope.

Had you dithered the whole time while mixing, your ear will (again to some extent) been exposed to the dither noise.

You get one or the other!!!

But pattons point was if you do it without dither and then dither at the end your final mix MAY sound odd cuz all the harmonic distortion may have affected some of you mixing dicisions and when its gone the overal tonal quality may seem different after all that hard work.

So I say if you aren't going to dither during the mix, don't do it ever cuz you mixed your song as if the harmonic distortion was going to be there at the end!!!

If you dither the whole time then make sure you dither at the end cuz thats what your ears were hearing the whole time.

If you are going to dither anyway why not be consistent and mix that way.

ORIGINAL: Nika

My ears are burning. I haven't had the time to go back to the beginning and figure out what the ongoing discussions are, but if someone could catch me up on remaining issues/questions I would be happy to throw in my two cents.

Nika

Hi Nika,

Welcome to this forum.

I don't know if there are any remaining issues. I think most of the stuff has been handled. The only thing I personaly am not entirely sure of is how the quantization distortion is related to the signal frequency. From my understanding the distortion is more correlated to the signal at lower frequencies. Is that correct and how exactly does that work? Or is this purely related to the fact that distortion of higher frequencies are aliased back into the audible band and thus less correlated with the actual higher frequency signal?

Cheers,

UnderTow

ORIGINAL: Rednroll

Look, we can bicker back and forth and you can continue to insult me. I'm bigger than that, I will just run my own tests and see what the data shows me. If I'm wrong, I will gladly admit it and take it that I've learned something new. I have a bench setup with a spectral analyzer, I'll see what it shows me. I'll compare data running a SA within the app (ie Pre D/A) , and then I'll compare that with the audio post D/A being fed directly into the SA and post the results.

We are not trying to insult you!

Please...

Are you really going to do a comparison between DATA being streamed to an DA and data streamed from an AD in order to prove things I wish you the best of luck, because not only are you going to do two conversions instead of one, you'll also be very dependant to distortion factors that might not have to do with the very quantization distorsion at all.

The main reasons to go from 24 bit to 32 bit float are the increased headroom flexibility

Ohhhhh...you get increased headroom? Hmmmm??? I did not know that, and here silly me always thought that 0dB was always 0dB no matter if it was 8bit, 16bit, 24 bit, 32 bit or 64 bit. I just thought you gain extra values inbetween -inf to 0dB to reduce distortions due to quantization error, by having more values along the Y-axis to get a closer approximation of the amplitude of the audio when sampled? But hey, if you say I get extra headroom that's nice to know. Wait until I let the rest of the guys know, that 0dB is really not the limit in digital audio. Thanks for all that correct information.

ORIGINAL: Rednroll

The main reasons to go from 24 bit to 32 bit float are the increased headroom flexibility

Ohhhhh...you get increased headroom? Hmmmm??? I did not know that, and here silly me always thought that 0dB was always 0dB no matter if it was 8bit, 16bit, 24 bit, 32 bit or 64 bit. I just thought you gain extra values inbetween -inf to 0dB to reduce distortions due to quantization error, by having more values along the Y-axis to get a closer approximation of the amplitude of the audio when sampled? But hey, if you say I get extra headroom that's nice to know. Wait until I let the rest of the guys know, that 0dB is really not the limit in digital audio. Thanks for all that correct information.

You don't know how floating point audio works? Are you serious? And you think you can post your opinions as fact in this thread? You are joking, right?


UnderTow

Oh, and here is the 16 bit to 8 bit dither comparison you asked for to make it more obvious.
http://www.stashbox.org/uploads/1145924529/16-bit-to-8-bit-comparison.jpg

Notice I also threw one in there where I applied the "dither" multiple times. Kind of like you've been recommending to do, instead of just once like I've been saying.

You don't know how floating point audio works? Are you serious? And you think you can post your opinions as fact in this thread? You are joking, right?

The only thing to really joke about is I said the differences between working at 24bit vs. 32bit, and I mentioned nothing about "floating point", but somehow you decided to throw that in there in your explaination for extra headroom.

Here's what I said, try reading it again because I'm starting to see a trend where you're not fully paying attention to things you read:

You know I need to ask the obvious question with all this. If the bit reduction adds noise and distortions, and the dither adds noise and distortions, then why work at higher bit debts at all? If we're getting all these distortions then what is the benefit of working at a higher bit debt? Why not just work at 24 bit and avoid it all together?

Yep, I know how "floating point" works, but that is not what I said.

ORIGINAL: Rednroll

You don't know how floating point audio works? Are you serious? And you think you can post your opinions as fact in this thread? You are joking, right?

The only thing to really joke about is I said the differences between working at 24bit vs. 32bit, and I mentioned nothing about "floating point", but somehow you decided to throw that in there in your explaination for extra headroom.

I mentioned 32 bit floating point and I mentioned advantages to which you respond with something that seems to show you don't understand floating point. Hint: You can go over 0 dB FS with 32 bit float. And this is absolutely correct information.

Now are you honestly going to say that when you mentioned working at higher bit depth you didn't mean 32 and 64 bit float? I'm sure you are aware that Sonar doesn't support anything above 24 bit int. The higher bit depths are in float format. Then again ... maybe you don't know this. Lack of knowledge doesn't seem to stop you posting your ideas as fact ...

You are back pedaling.

UnderTow

You are back pedaling.

No, I'm sticking with what I originally said and that is to apply dither once and only once as your last step, and I've just proven to you why, but you choose to ignore the data. Call what I'm saying as not being facts......but I can not lie about the data, it's right there before your eyes. I've shown you where multiple dither processes can and does increased the noise floor even above the distortions of non dithered audio, then I've further shown you that the dither can be added later with the same benefits and this is EXACTLY what I originally said.

ORIGINAL: Rednroll

You don't know how floating point audio works? Are you serious? And you think you can post your opinions as fact in this thread? You are joking, right?

No could you explain it to me. Wait a minute??????......... I thought we where talking about the differences in bit debts? Isn't that what I was talking about originally? I think I was trying to point out to you that you don't get any increased headroom if you go from 16 bit, to 24 bit, or 32 bit.

Liar.

Just like you don't get any additional head room when going from 8 bit to 16 bit. Now going to floating point, yes I totally understand you can work outside the limits, but eventually you have to come back to the fixed point math in playback, and then your headroom is gone. Try to read what I said if you're going to answer it.

You asked a simple question: "why work at higher bit debts at all? If we're getting all these distortions then what is the benefit of working at a higher bit debt? Why not just work at 24 bit and avoid it all together?"

I responded correctly to that question. You can not work at 32 bit int in Sonar so don't pretend that was what you were refering to.

Then you made silly comments trying to mock me but actually, yet again, demonstrating your ignorance on the subject.

No one ever mentioned "floating Point" in my question, yet you just assumed that I did.

Of course. This is the Sonar forum in case you hadn't noticed. We are talking about working practises in Sonar.

I think I'm starting to see a trend here on how you interpret what you read.

The only trend I see here is someone that is ignorant, too weak to admit their mistakes and tries to twist things arround with dishonesty about what they really meant.

Hopefully you're learning something here, and that's dither ONCE to avoid noise build up from multiple dithers and wait to dither as the last process.

The only thing you can teach me is how to be a dishonest ignorant fool. Not a lesson I am interested in.

Over and out.

UnderTow

Hey Rednroll,

Thanks for actually doing the experiment. I was wondering if you could clarify a few things?

The first graph makes sense. You see harmonic distortion of truncation compared to the dither. It would be nice had you zoomed in on the higher frequecy regions so we could get a closer look but I think we get the idea. And to me it looks as if the dither is at least 3dB below the peak distortion (still both are quite low). Also the FFT resolution was 2048 right? Was this the highest that app you are using goes.

I did the same thing in Sound Forge last night with a FFT resolution of 65k point and was able to see a 10-15dB difference between the peak harmonic distortion and the triangular dither. Again triangular dither was the lower of the two and both were at a very low dB (as you've said we probably wouldn't hear this. I would be happy to post these graphs if someone will lend me the web space or point me to some free spot.

Anyway my real question has to do with the second set of plot in the anolog domain. What are these supposed to represent? You have an analog spectrum analyser? What FFT resolution did you use? What are the two signals? One is 24-bit (no dither or truncation)? And the 16-bit file does it have dither or no dither? It would be nice to see zoomed in comparison of the 16-bit with dither and without.

Now what I don't get is how in your digital graphs the noise floor of the 24-bit signal was WAY below the two 16-bit signals and now they are right on top of each other in the analog world. Today's DACs should get you from 112-120dB right as we never get the full 24 bits. But shouldn't the noise floor of the 16-bit signal be averaging around 96dB. I am confused. Shouldn't there be a 16-24dB (-96dB-112db=16dB) difference in your two plotted graphs? Please explain as this graph obviously re-emphorses your original premise and if you could help me understand it that would be great! Shouldn't reducing a 24-bit signal to 16-bit (putting aside this whole war on dither) yield two signals with significantly different noise floors. This it what you showed in your digital plots but a very differnt thing appread in the analog plots.

I am not mocking or insulting or wanting to get into some sort of internet gunfight. I am genuinely interested in your response!

Chris.