Dithering Revisited

No, it’s not the name of an English PBS drama . . .
 
Recent discussions on dithering in the forum led me to believe that I should set my Sonar system dithering to Triangular and forget about it. However, I want to sketch out my situation, and check to see if this really is the best choice for me.
 
I mix down in Sonar using the Voxengo Elephant limiter on the master bus. I use Elephant’s dither settings when exporting/mixing down, and so I set dither to "none" during this function. However, my Sonar system options are set (under Audio Options | Advanced) to triangular dither.
 
I use a few mic’ed/DI recorded tracks and some soft synths in each piece; not all audio tracks are rendered out before mix down. I record at 44.1 kHz/24 bit.
 
I tried export/mixing down both with the system dither set to None and Triangular. When I compare these two exported WAV files using the phase invert test, I am hearing an audible difference. To me, this suggests that one choice is better than the other for sonic quality. The question is: which one??
 
How/when is Sonar applying triangular dither to a mixdown of 44.1/24 bit when the export action specifies no dither and the dithering is handled by Elephant?

If that's true, then I should leave the system setting to "none" and only use the dithering of the Elephant limiter.
 
But someone else had claimed that setting the system dither to triangular was beneficial for certain functions . . . can't remember now why . . .
 
Anyone know?

Okay . . . some of this is starting to sound familiar.
 
So, in a final mix export, am I using any internal rendering? If my sample/bit rate is 44.1kHz/24 bit, but am exporting to 44.1/16 bit, is ALL the dithering applied by the Elephant limiter? My audio driver bit depth is 24 bit, with the 64 bit double precision engine box checked, and I am using WinXP 32 bit version of Sonar.
 
My inverted phase test indicates that at least SOME internal triangle dithering is occurring . . . the difference is slight, but audible. But where? How? Am I dithering unnecessarily if I set the system dither setting to triangle?
 
Does someone from Calkwalk have an answer here?

You are correct in asserting that the invert phase null test will not work on two files exported from the same mix using a random quantize dither limiter - I did the test too.
 
So, that means that the only thing I have to go on is the actual sound of the two exported files: one with system dither settings on triangular, and the other with system settings on "none."
 
The difference is slight; but the one with triangular system dither sounds a bit more brittle. Of course - as mentioned earlier - this is only the system dither settings; the dithering is always set to "none" in the export dialogue.
 
We still have a debate however, since we are seeing different opinions from learned forum members. Anyone else have a reply?

pdarg

...
We still have a debate however, since we are seeing different opinions from learned forum members. Anyone else have a reply?

What differing opinions?  As far as I can tell, all the conclusions in this thread have been the same.

Naw, there's no such thing as too many options (unless it's something illogical).

Having options is what Sonar's about.

BitFlipper has stated that the system dither option should be set to none.

Most other opinions have stated that the triangular dither option should be selected for the system dither, but turned off for export.

This is the heart of the debate.

Or someone from Cakewalk who can answer this . . .

pdarg


Or someone from Cakewalk who can answer this . . .

Somewhere in a thread from many moons ago former CTO Ron Kuper posted on this when I asked the same question.  I tried to search in this new forum ... but I just can't naviagate around to find it.  Sorry!

-D

I just did another test - removing the Elephant limiter and just testing both a Sonar dithered export and a non-dithered export with the system dither turned off and also set to triangle.

Even in cases where no export dither was applied, the inverted phase null test indicates a difference between a file exported with system dither at triangle, and one exported without.

This means that choosing a system dither setting of triangle or none makes a difference; so, the question is: which setting is correct for this export/mixdown process?

pdarg


I just did another test - removing the Elephant limiter and just testing both a Sonar dithered export and a non-dithered export with the system dither turned off and also set to triangle.

Even in cases where no export dither was applied, the inverted phase null test indicates a difference between a file exported with system dither at triangle, and one exported without.

This means that choosing a system dither setting of triangle or none makes a difference; so, the question is: which setting is correct for this export/mixdown process?

 
You have to be careful how you go about performing a null test.  All variables (except for the ones being tested) should be eliminated.
 
That includes modulating effects such as Chorus, Flangers, Phasers as well as softsynths that use these effects and/or ones that trigger different samples at the same velocities.

So . . . is it correct to set system dither to triangular AND use export/final plug-in dither etc. together on export?? It seems to me that this is using TWO instances of dither in one process - however, if there is some internal processing with plug-ins, softsynths etc., perhaps it is needed.

Ah . . . now we're getting somewhere.

So, if I turn off Sonar's dither in the export audio dialogue, and use the dither on the Elephant plug-in, this would be the same as setting my system dither settings to "none" - at least for the purposes of exporting audio - correct?

However, if that is the case, why am I hearing a difference between system dither set to triangle and none - when turning off Sonar's dither in the export dialogue and using Elephant's dither?

This suggests to me that the system dither settings are doing something apart from the export dither . . . this is why I am still confused.

Jose7822


Pdarg,

I think you're drowning yourself in a glass of water.

The definition of Dithering according to Wikipedia is "an intentionally applied form of noise used to randomize quantization error".  This randomization is the difference you hear between the files.  So two files will never cancel out as long as dithering is being applied. 

The other thing is that, system dither is being applied to files going from a higher bit depth to a lower one, as is the case when bouncing or feezing tracks in Sonar (assuming the Render Bit Depth setting is set to something lower than 64 bit and that the 64 bit engine is engaged).  This means that, the more destructive processing you do in Sonar, the more dithering layers you apply to the audio file.  However, this all becomes moot when processing at 64 and 32 bits.  At 24 bits the noise is still low, but it is possible for it to reach audible levels.

So again, leave the system dithering ON and change the one in the export window as needed.  Sonar takes care of the rest.

Another thing to point out is the system dither (at 24 bits) is ~48dB lower in level than the export dither (at 16 bits). Therefore I would worry more about the export dither setting, and don't sweat the system dither setting.

drewfx

Okay, so let me ask you this: If I record in 44.1/24, use some mic'ed tracks in conjunction with softsynths, etc. - but do not use any frezzing or bouncing in my Sonar mix, is it necessary to have the system dither set to triangular?