Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Questions regarding Dithers & Noise Shaping (Read 3741 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Questions regarding Dithers & Noise Shaping

To keep it relatively short;

If you have a "master" recording at, say, 96khz/24bit which has already been edited as desired and is "ready" for reproduction on CD-----

Since the target rate is 44.1khz/16bit, if you were to first reduce word length from 24-bit to 16-bit with very strong noise shaping,  could this not, in theory, push nearly 100% of the dither noise into the 40-48khz domain, which would then be 100% removed upon a later downsample to 44.1khz? (from 96khz)

Would the dither, then, essentially be "free", adding to the perceived dynamic range and increasing SNR, while adding "no noise" since it was removed in the downsample? (due to being well above nyquist of the target downsample rate?)

If this is the case, why do people ever complain about dither noise in the product(however insignificant it may have been), if it has been nearly 100% removed in the first place?

Questions regarding Dithers & Noise Shaping

Reply #1 –
Quote from: SirChristof on
Since the target rate is 44.1khz/16bit, if you were to first reduce word length from 24-bit to 16-bit with very strong noise shaping,  could this not, in theory, push nearly 100% of the dither noise into the 40-48khz domain, which would then be 100% removed upon a later downsample to 44.1khz? (from 96khz)

Would the dither, then, essentially be "free", adding to the perceived dynamic range and increasing SNR, while adding "no noise" since it was removed in the downsample? (due to being well above nyquist of the target downsample rate?)

If this is the case, why do people ever complain about dither noise in the product(however insignificant it may have been), if it has been nearly 100% removed in the first place?


You can try what you suggest with 8 bit output and see yourself that it is just not possible to remove dithering noise.
However I don't know why it is not possible to do as you suggested. Probably because 96 kHz sampling rate allows you to variate dithering noise across 2 samples and thus move it into higher region. When you downsample you have to put the same dithering noise on 1 sample but then you just don't have the possibility to put it in high frequency region.
 

Questions regarding Dithers & Noise Shaping

Reply #2 –
Quote from: SirChristof on
To keep it relatively short;

If you have a "master" recording at, say, 96khz/24bit which has already been edited as desired and is "ready" for reproduction on CD-----

Since the target rate is 44.1khz/16bit, if you were to first reduce word length from 24-bit to 16-bit with very strong noise shaping,  could this not, in theory, push nearly 100% of the dither noise into the 40-48khz domain, which would then be 100% removed upon a later downsample to 44.1khz? (from 96khz)

Would the dither, then, essentially be "free", adding to the perceived dynamic range and increasing SNR, while adding "no noise" since it was removed in the downsample? (due to being well above nyquist of the target downsample rate?)


No, because downsampling from 96kHz to 44.1kHz (like most processing) produces data beyond the 16th bit which requires dithering down to 16bits!

Tha can't get owt for nowt.

(You can't get something for nothing).

Cheers,
David.

Questions regarding Dithers & Noise Shaping

Reply #3 –
There is such a thing as 'subtractive dither' where the dither added before the operation, to randomize quantization error, is removed after the operation. I don't believe it gets much use. If one is not forced to work in 16 bit, the single application of noise shapped dither when converting to 16 bit is so unlikely to be audible that the extra effort  for subtractive dither is without real benefit.

Questions regarding Dithers & Noise Shaping

Reply #4 –
Quote from: 2Bdecided on
No, because downsampling from 96kHz to 44.1kHz (like most processing) produces data beyond the 16th bit which requires dithering down to 16bits!
Tha can't get owt for nowt.
Spot on!
AndyH-ha already mentioned subtractive dither. Sony had its own low-noise solution, explained in this AES preprint from 1992:
Quote
Super Bit Mapping: Psychoacoustically Optimized Digital Recording
Super bit mapping uses psychoacoustic principles and critical band analysis to maximize the sound quality of 16-bit digital audio. 20-bit recordings are quantized to 16 bits using an adaptive error-feedback filter. This filter shapes the quantization error into an optimal spectrum as determined by the short-term masking and equi-loudness characteristics of the input signal. Through this technique, the perceptual quality of 20-bit sound is available on a standard compact disc.
Preprint Number:  3371    Convention:  93 (September 1992)
Authors:  Akune, Makoto; Heddle, Robert; Akagiri, Kenzo
The Sonic Solutions workstation (which was and afaik still is being used by SONY music for editing and mastering) had several dither options, including SBM1 and SBM2
Quote
SBM2 is an adaptive version which, in addition to equi-loudness characteristics, adapts to the instantaneous masking characteristics of the input signal.
It doesn't sound like dither noise, it is much quieter, but apparently had a few shortcomings compared to "correct" dither versions.
If someone is interested I could probably find some time to make a short audio sample with different dither versions including SBM.

Questions regarding Dithers & Noise Shaping

Reply #5 –
Quote from: 2Bdecided on
Tha can't get owt for nowt.

(You can't get something for nothing).
The basic principle of engineering

SirChristof: Another way of stating this is that you are stuck with a certain amount of quantization noise - you can't "trick" the theory into getting rid of it. Dither is merely a way of ensuring that the noise is uncorrelated with the signal, and noise shaping is a way of pusing it out of the band of interest.

Questions regarding Dithers & Noise Shaping

Reply #6 –
Quote from: AndyH-ha on
There is such a thing as 'subtractive dither' where the dither added before the operation, to randomize quantization error, is removed after the operation. I don't believe it gets much use. If one is not forced to work in 16 bit, the single application of noise shapped dither when converting to 16 bit is so unlikely to be audible that the extra effort  for subtractive dither is without real benefit.


It's been a _long_ time, but IIRC subtractive dither only buys you 3 or 6dB. That's all. It's also a real pain to implement: any error or loss of sync and you're effectively adding noise, rather than subtracting it.

Noise shaping can gain you much more than 6dB advantage, at least perceptually.

Then there's Apogee UV22.

Cheers,
David.