Dither, noise, self-dither

[macc]

Something's been bugging me, I hope people can make it clearer for me I'll keep it short.

Even a great signal chain has a noise floor at what, -80 to -85 or something. Actually having typed that I just checked mine, and at unity gain it's -79.1dB. Not too bad.

So, well... what's the point of adding dither?

[macc]

FWIW I am pretty sure this is a really stupid question. I just want to check with you knowledgeable types.

Ta

[Cellotron]

It's very easy to hear the reason if you compare the results of a 24bit file captured back through an analog chain that's amplitudes in around an area of -90dBFs to -80dBFs - first truncating to 16bit and then second adding dither prior to requantizing.

More often than not, regardless of the noise floor of the chain - in these very low level areas the truncated version will fitz and distort as it heads into the noise floor - where as the dithered version will sound smoother as things head into the noise floor (albeit with some additional noise added to the least significant bit).

In other words ime the idea that things can be "self dithering" tends to be a myth - especially when you realize that more often than not a fade will be created digitally in the DAW - in which case even if there was a random enough bit of noise at the least significant bit to allow for some "self dithering" - this would get broken at the one place that dithering is most often needed.

Best regards,
Steve Berson

[macc]

Originally Posted by Cellotron

In other words ime the idea that things can be "self dithering" tends to be a myth - especially when you realize that more often than not a fade will be created digitally in the DAW - in which case even if there was a random enough bit of noise at the least significant bit to allow for some "self dithering" - this would get broken at the one place that dithering is most often needed.

This made a couple of things click - thank you!

[jayfrigo]

There's self-dithering on initial A to D conversion, and there's dithering during processing or delivery when reducing wordlength. Even if there is enough noise of the right kind to self-dither at the ADC, this doesn't exempt you from needing dither for subsequent wordlength reductions to prevent truncation distortion and loss of subjective resolution below the LSB (Least Significant Bit).

However, re: AD conversion, even with the noise present during conversion, there's no way to guarantee that there will be a sufficient amount of the right kind of noise, particularly a high frequencies, to ensure complete de-correlation, and the benefits resulting from proper dithering.

[masteringhouse]

I agree with Jay, it has to be the right type of noise and well distributed. Also if you plan on using any kind of noise shaping a proper type of dithering algorithm will be needed.

[macc]

Originally Posted by masteringhouse

it has to be the right type of noise

This was exactly the stupid thing I was missing that made it a stupid question thumbsup It all fell into place after that.

Thanks again all.

[kjg]

self-dithering... is that the same as auto-ditheration?
seek help from a decent shrink macc!

Originally Posted by Cellotron

...realize that more often than not a fade will be created digitally in the DAW - in which case even if there was a random enough bit of noise at the least significant bit to allow for some "self dithering" - this would get broken at the one place that dithering is most often needed.

That's it. Any processing comes after the "dithering", and changes the noise in level, freq content, dynamics etc. Even if there was noise that was suitable in the recordings/mix at one point, it might very well not be so suitable any more after processing in mixing or mastering, or just plain faded out at the moments you need it most.

Dither should be the right type of noise, added after the last stage of processing. When going the self-dithering route you are assuming the right type of noise was there in the first place, and then assuming that it will still be the right type of noise after processing.

Not very likely, especially when doing any ITB fades.

Regards,
Klaas-Jan

[Ethan Winer]

Originally Posted by macc

FWIW I am pretty sure this is a really stupid question. I just want to check with you knowledgeable types.

It's never stupid to question conventional wisdom. I do it all the time.

Some people hate when I link this article, but I persist anyway:

Dither Report

Closely related:

Artifact Audibility Report

--Ethan

[Bob Olhsson]

"Self dithering" is BS.

Dither by definition is the least amount of random noise required to prevent truncation distortion. While dither sounds like noise, noise is not dither unless it has the correct spectral content and level. Jim Johnston of Bell Labs told me that it would require at least ten dB. more random noise to prevent distortion than the amount of properly distributed dither required.

[jayfrigo]

Originally Posted by Ethan Winer

It's never stupid to question conventional wisdom. I do it all the time.

Some people hate when I link this article, but I persist anyway:

Don't confuse "conventional wisdom" with measurable, scientific fact. What you choose to do with the facts is up to you.

Conventional wisdom is that distortion from truncation is bad (I agree; Ethan apparently does not). However, scientific fact is that distortion from truncation exists. Whether you want to use dither to prevent it is up to you, and has been argued in other threads, but is off-topic in this one.

[jayfrigo]

Originally Posted by Bob Olhsson

"Self dithering" is BS.

Dither by definition is the least amount of random noise required to prevent truncation distortion. While dither sounds like noise, noise is not dither unless it has the correct spectral content and level. Jim Johnston of Bell Labs told me that it would require at least ten dB. more random noise to prevent distortion than the amount of properly distributed dither required.

If you look at noise spectra, including TPDF dither (essentially white noise), and noise from recording (including technical and environmental), you can see the difference right away, and it gives you an idea why one will not replace the other (without being significantly increased as per JJ's comment).

[Ethan Winer]

Originally Posted by jayfrigo

Don't confuse "conventional wisdom" with measurable, scientific fact.

Jay, I know that what dither does is real and can be verified by measuring distortion. I just think it's importance is blown way out of proportion because what it does is typically 90 dB below the music. In some types of very soft music it might be just barely audible. On typical pop music it is not. At least I can't hear it. In the half-year or so my Dither Test linked above has been online, with more than 2,200 reads, not one person has ever sent me the correct answer.

--Ethan