To Dither, or not to Dither..?

[Mick C]

The Waves L Series limiters include dither options. The DAW I use is Studio One and it automatically applies dither when mixing down to stereo at CD 44.1 / 16 bit. So should I switch off the dither in the Waves L3 limiter I’m using before I do a mix down, and let Studio One take care of it?

Currently the L3 is set to 24bit, as is the bit rate of the Studio One session, so it feels like letting the L3 dither at 24 bit within a 24 bit session would be ok, and then Studio One can do its own dithering down to 16 bit when mixing down to CD?

[cyrano.mac]

You may call me simple, but why would a limiter need dithering options?

I'm really surprised. Hope one of the smarter folk can explain...

[MOF]

I think the dither setting is for when the L3 is the last part of the recording process and it will only apply when changing bit settings for export.
I’ll be surprised if it has any effect on 24 bit files staying as 24 bit.
Logic has a dialogue box for bounce downs, exports etc which allow you to independently set dither, MP3 rates etc for simultaneous processing of the master file.

[CS70]

The Waves L Series limiters include dither options. The DAW I use is Studio One and it automatically applies dither when mixing down to stereo at CD 44.1 / 16 bit. So should I switch off the dither in the Waves L3 limiter I’m using before I do a mix down, and let Studio One take care of it?

Yes. Even if not tragic, dither replaces the least significant bit in every sample with random noise, where the randomness is given by a certain distribution, so no need to do it twice.

Currently the L3 is set to 24bit, as is the bit rate of the Studio One session, so it feels like letting the L3 dither at 24 bit within a 24 bit session would be ok, and then Studio One can do its own dithering down to 16 bit when mixing down to CD?

You dither only when you cut the word length. If your data is 24 bit and you bounce at 24 bits, dithering is not needed and introduces unnecessary noise (replacing good information with noise).

When u go down in sample word length (say to 16 bit), if you just truncate, you need noticeable errors. If you add a bit of noise, the errors smooth out and it sounds better.

[CS70]

You may call me simple, but why would a limiter need dithering options?

I'm really surprised. Hope one of the smarter folk can explain...

It's quite common for digital limiters, as it's the last effect and most often in these situation you want to dither.

[Hugh Robjohns]

The Waves L Series limiters include dither options. The DAW I use is Studio One and it automatically applies dither when mixing down to stereo at CD 44.1 / 16 bit. So should I switch off the dither in the Waves L3 limiter I’m using before I do a mix down, and let Studio One take care of it?

Yes. Dithering should only be applied at the point where word-length reduction takes place.

...it feels like letting the L3 dither at 24 bit within a 24 bit session would be ok, and then Studio One can do its own dithering down to 16 bit when mixing down to CD?

Yes, this is an appropriate solution in your situation.

H

[Hugh Robjohns]

You may call me simple, but why would a limiter need dithering options?

It's because the limiter is traditionally the final process in the signal path, so many (developed from the era of CD production) also included the facility to reduce the word-length to 16-bits for the CD master.

And where you reduce the word-length you also need to introduce dither, so this became a popular added feature (often with different options for noise-shaped dithers, too).

[Hugh Robjohns]

...dither replaces the least significant bit in every sample with random noise...

Small point of pedantry... Although you can perform dithering with random noise, most systems actually use the audio data in the truncated part of the sample to modulate, in a noise-like way, the new least-significant bit in the shortened sample word. That's what maintains the 'signal resolution' and allows very low level audio content from the 24-bit data to still be heard under the noise floor of the 16-bit version.

If you add a bit of noise, the errors smooth out and it sounds better.

Technically, the quantising distortions that result from truncation of the sample word-length are fully removed. The result is entirely analogue-like performance where the signal can be faded smoothly into and below the noise floor, just like on an analogue mixer, with perfect 'resolution' and no distortion or breakup (which is what happens without dither)

[Mick C]

The Waves L Series limiters include dither options. The DAW I use is Studio One and it automatically applies dither when mixing down to stereo at CD 44.1 / 16 bit. So should I switch off the dither in the Waves L3 limiter I’m using before I do a mix down, and let Studio One take care of it?

Yes. Dithering should only be applied at the point where word-length reduction takes place.

...it feels like letting the L3 dither at 24 bit within a 24 bit session would be ok, and then Studio One can do its own dithering down to 16 bit when mixing down to CD?

Yes, this is an appropriate solution in your situation.

H

Thanks Hugh,
So basically I should turn off dither in the L3 because it’s unnecessary as Studio One adds it’s own dither at mixdown
But actually having the L3 dither turned on at 24 bit won’t create any issues either, because it’s effectively doing nothing in this context, as Studio One is doing the dithering work (ie the L3 dither isn’t doing any harm either).
Have I got it right ?

[CS70]

Small point of pedantry...

Good points Hugh - thanks, difficult to balance the compelling need of detailed precision with the basic nature of the question

[Hugh Robjohns]

So basically I should turn off dither in the L3 because it’s unnecessary as Studio One adds it’s own dither at mixdown...

I've not explored Studio One, but assuming you are correct about it applying dither at the mixdown output, then there's obviously no need for it in the L3 as well.

But actually having the L3 dither turned on at 24 bit won’t create any issues either, because it’s effectively doing nothing in this context

Correct.

H

[Hugh Robjohns]

...difficult to balance the compelling need of detailed precision with the basic nature of the question

Yes, but the devil is always in the details, and often small 'simplifications' often set the foundations for big misunderstandings later. So in my view it's much better to correct them asap.

The suggestion that the LSB is replaced with noise implies the removal and loss of low-level audio data... whereas the true detail that the LSB is modulated in a noise-like way using the data from the truncated bits represents the true situation: low level audio data is not discarded, it's just buried into the noise floor of the shorter word-length samples, so works just like an analogue system. This is a critical difference, I'm sure you'll agree.

[CS70]

This is a critical difference, I'm sure you'll agree.

Indeed - hence the thanks

[Hugh Robjohns]

[Mick C]

So basically I should turn off dither in the L3 because it’s unnecessary as Studio One adds it’s own dither at mixdown...

I've not explored Studio One, but assuming you are correct about it applying dither at the mixdown output, then there's obviously no need for it in the L3 as well.

But actually having the L3 dither turned on at 24 bit won’t create any issues either, because it’s effectively doing nothing in this context

Excellent Hugh. Thanks to you and all the others that contributed to answer my question.

[MOF]

Although you can perform dithering with random noise, most systems actually use the audio data in the truncated part of the sample to modulate, in a noise-like way, the new least-significant bit in the shortened sample word. That's what maintains the 'signal resolution' and allows very low level audio content from the 24-bit data to still be heard under the noise floor of the 16-bit version.

Interesting, I didn’t know this.
I’ve just looked up Logic ‘s dithering spec’s:
“Logic Pro includes the professional POWr (Psychoacoustically Optimized Wordlength Reduction) and UV22HR dither algorithms, designed to convert 24-bit recordings to 16-bit files (as required for CD burning, for example).

The following dithering options are available in the Bounce dialog:

None: No dithering is applied.

POWr #1 (Dithering): A special dithering curve is used to minimize quantization noise.

POWr #2 (Noise Shaping): Additional noise shaping is used over a wide frequency range, which can extend the dynamic range of the bounce file by 5–10 dB.

POWr #3 (Noise Shaping): Additional, optimized noise shaping is used, which can extend the dynamic range by 20 dB within the 2–4 kHz range (the range in which human hearing is most sensitive).

Note: Noise Shaping minimizes the side effects caused by bit reduction (downsampling), by moving the quantization noise spectrum to the frequency range above 10 kHz (the least sensitive part of the range of human hearing). This process is technically known as spectral displacement.

UV22HR: Allows for the best possible sound resolution when bouncing 24-bit recordings into 16-bit files.

The dithering algorithm that sounds best depends on the audio material, and also your personal taste. Audition the audio material to be dithered with each of the algorithms to determine the best setting. In some cases, you may find that the best results are achieved by using no dithering at all.

Important: Avoid dithering the same audio signal multiple times.

Published Date: Aug 9, 2019”

I presume POWr refers to the shape of the truncated curve?
Interesting that they think versions of POWr with added noise might improve the 16bit versions.
I’m hoping all this becomes a curiosity of a bygone age soon. Storage is no longer an issue, companies such as Apple are now requesting 24 bit, 96k masters for delivery to the public.

[miN2]

The DAW I use is Studio One and it automatically applies dither when mixing down to stereo at CD 44.1 / 16 bit.

So basically I should turn off dither in the L3 because it’s unnecessary as Studio One adds it’s own dither at mixdown

S1 dither can be turned off in the options, so it's only automatically applied if you have the option turned on (S1 also dithers on playback if it's on). Personally i turn S1 dither off and dither through Ozone.

[Hugh Robjohns]

There is a good argument to stick with standard TPDF (triangular probability distribution function) dither which gives a spectrally-flat noise floor which is very analogue-like and benign.

However, noise-shaped dithers can appear to give subjectively lower noise-floors -- so a 16-bit system can sound as quiet as a 20-bit system, for example -- by shifting a lot of the dither signal from the parts of the spectrum where we are most sensitive (in the 2-4kHz region) to areas where we are least sensitive (18-20kHz).

Apogee's UV22HR system piles a lot of dither energy up at 18kHz in a couple of obvious spikes:



While the Pow-R system tries to follow the human ear's sensitivity more closely:

Here's an image showing the spectral distribution of the four Prism Sound SNS noise-shaped dither options:

Note the linear frequency scale which emphasises the top-end of the spectrum, but you can see that the SNS1 and SNS2 options basically move a little of the dither energy from across most of the audio band and pile it up above 18kHz or so, this effecting a small but worthwhile reduction in the perceived noise floor level.

SNS3 and SNS4 both use much heavier 'sculpting' which removes a lot more energy from the 2-4kHz region, mimicking the natural sensitivity response of the human ear, and piles it up at the top end again, but extending down to around 14kHz (where it tends to be more audible).

Of course, this relocated dither signal energy obviously adds to whatever audio is in those parts of the spectrum, potentially causing problems with overloads, or interacting with the wanted audio and causing audible artefacts or even becoming audible in its own right. Hence the recommendation to audition different noise-shaping functions to pick the one that works best with the specific audio content.

It's definitely not a case of one-type-fits-all-situations!

I included some very informative audio examples of the effects of crude truncation, basic TPDF dither, and a very simplistic noise-shaped dither at 3, 8 and 16-bit wordlengths in an article I wrote a while back -- see the section on Quantising and dither:

https://www.soundonsound.com/techniques/digital-problems-practical-solutions

Audio files are here: https://www.soundonsound.com/techniques/digital-problems-practical-solutions-media

The noise-shaped dither on the 3-bit piano is particularly revealing and instructive as it proves beyond any doubt that properly dithered digital audio -- at whatever word-length you choose -- has exactly the same 'infinite resolution' as an analogue system. You can hear perfect, undistorted audio even at 3-bits and well below the noise floor.

Increasing the word-length does not increase or improve the resolution, it just lowers the noise-floor. Nothing else. The resolution is infinite at whatever word-length.

And in the same way, increasing the sample rate doesn't increase or improve the resolution either. At double sample rates it just increases the bandwidth, so you can record all those whistling line-lump power supplies and LCD monitor backlights more clearly ...while at quad and higher rates the inherently increasing clock jitter actively degrades the quantising accuracy...