I think dither was raised over the last few pages not as a method to increase resolution for audio signals, but as a technique for improving measuring instrument resolution (along with suitable post signal processing). I think the dither discussion has drifted off track or been misinterpreted if people are suggesting it is a valid technique for standard audio signals.
Last edited:
Sy,
Don't apologise. I do not mind personal attacks at all. Moderators might, but I ain't one. I always enjoy when people are letting their real intentions speak...
Well, let me posit what the ubiquitous and easily available resource called wikipedia has to say first:
Now allow me to again quote Sy from post #14111:
Based on the above it is clear that Sy claims that in fact that adding noise increases ENOB. There is no false attribution, nothing added.
And I object to the thesis that an increase in noisefloor improves ENOB. In fact, I would argue that it observably dis-improves ENOB.
So, if I am being called a Liar for pointing such a fundamental inconsistency in the statement, so be it. The Catholic church called Galileo a heretic liar for pointing out a similar issue of inconsistencies in public statements, not that I am comparing myself to Galileo.
Quantization Distortion is well understood. In an ideal digital system it has a fixed value, due to the limited number of values available (+/- 0.5 LSB).
The minimum dither that can be used to reduce this distortion is greater in impact on ENOB than Quantization Distortion itself.
So you clearly postulate something else, or your understanding of dither and quantization distortion is not quite what you make it out to be.
Past that, I do not care what you believe, but if you make statements of fact that are not supported by either commonly accessible facts OR evidence presented by you I fear that such an idea must be put to suitable scrutiny, must be evaluated and if it turns out to be snakeoil or bunkum, this must made clear as well.
If you have any evidence to support your beliefs you are welcome to present them and I shall be first to apologise for being wrong in my understanding of the matter. However, huffy statements and refusal to engage in dialogue and to clear up things do not support your position in the least.
Ciao T
Don't apologise. I do not mind personal attacks at all. Moderators might, but I ain't one. I always enjoy when people are letting their real intentions speak...
Well, let me posit what the ubiquitous and easily available resource called wikipedia has to say first:
Wikipedia.org said:
Wikipedia.org said:
Now allow me to again quote Sy from post #14111:
Based on the above it is clear that Sy claims that in fact that adding noise increases ENOB. There is no false attribution, nothing added.
And I object to the thesis that an increase in noisefloor improves ENOB. In fact, I would argue that it observably dis-improves ENOB.
So, if I am being called a Liar for pointing such a fundamental inconsistency in the statement, so be it. The Catholic church called Galileo a heretic liar for pointing out a similar issue of inconsistencies in public statements, not that I am comparing myself to Galileo.
Quantization Distortion is well understood. In an ideal digital system it has a fixed value, due to the limited number of values available (+/- 0.5 LSB).
The minimum dither that can be used to reduce this distortion is greater in impact on ENOB than Quantization Distortion itself.
So you clearly postulate something else, or your understanding of dither and quantization distortion is not quite what you make it out to be.
Past that, I do not care what you believe, but if you make statements of fact that are not supported by either commonly accessible facts OR evidence presented by you I fear that such an idea must be put to suitable scrutiny, must be evaluated and if it turns out to be snakeoil or bunkum, this must made clear as well.
If you have any evidence to support your beliefs you are welcome to present them and I shall be first to apologise for being wrong in my understanding of the matter. However, huffy statements and refusal to engage in dialogue and to clear up things do not support your position in the least.
Ciao T
"Listen up"
Homepage of Alexey Lukin
exagerated examples of what dither does for human perception of digital audio exsample files with truncation/various dither is used with 8 & 12 bit resolution so "anyone" can hear the effect of "below the (quantization) noise floor resolution"
Homepage of Alexey Lukin
exagerated examples of what dither does for human perception of digital audio exsample files with truncation/various dither is used with 8 & 12 bit resolution so "anyone" can hear the effect of "below the (quantization) noise floor resolution"
Last edited:
Except I didn't.
Tell you what, since it repeats with dreary regularity, every time you do the usual game of attributing things to me that I never said (the English word for that is "lying"), then demanding I defend them, I'll just recall the old joke about the Comedian's Club and just say, "31."
Yes, probably. The dither improves SFDR (spurious free dynamic range), and this is IMO THE reason why it audibly helps.
Hi,
This is interesting, but there are several unsupported claims on this page that attempt to spread FUD. Why not post 20 Bit to 16 Bit examples if "When reducing word length from 20+ bits to 16 bits, similar artifacts arise.", surely then it would be equally audible to everyone?
Further, Extrabit includes a limiter, which non of the other processors do and it is NOT stated anywhere that this limiter was bypassed.
I would suggest instead that anyone who wants to hear what truncation sounds like and what dither sounds like simply downloads a 24 or 20 Bit Musictrade release (FLAC) of Doc & Lena (this has a lot of good, acoustic piano) and then applies both straight truncation and dithering to 16 bit without sample rate conversion.
Further, it still leaves the: "dither degrades resolution quantified by standard established methods observation" unanswered.
Ciao T
This is interesting, but there are several unsupported claims on this page that attempt to spread FUD. Why not post 20 Bit to 16 Bit examples if "When reducing word length from 20+ bits to 16 bits, similar artifacts arise.", surely then it would be equally audible to everyone?
Further, Extrabit includes a limiter, which non of the other processors do and it is NOT stated anywhere that this limiter was bypassed.
I would suggest instead that anyone who wants to hear what truncation sounds like and what dither sounds like simply downloads a 24 or 20 Bit Musictrade release (FLAC) of Doc & Lena (this has a lot of good, acoustic piano) and then applies both straight truncation and dithering to 16 bit without sample rate conversion.
Further, it still leaves the: "dither degrades resolution quantified by standard established methods observation" unanswered.
Ciao T
The main problem of undithered signal is a correlation between the quantization noise and the signal. This depends upon the ratio of the sampling frequency to the input frequency. If this ratio is integer, like 32, we get a lot of spurious spectral lines. This is improved by dither. BUT, I am not saying (and never did) that the dither improves resolution. No! It does not. It only improves SFDR at expense of added noise.
Sy,
We are going around this again.
You said:
"Yes, dither adds to the noise (it IS noise!) BUT it increases resolution."
As resolution is stated commonly in ENOB and dither degrades ENOB you statement is false.
There is no lie involved or attribution.
If you desired to express something else that what you typed, you could have clarified.
As it stands to claim that increasing noise increases resolution can only be regarded as a most extraordinary claim.
If what I attributed to you was not what you typed, then who did type?
Otherwise, in English and USA politics we find often that politicians make claims of not having said something that they in fact did say, of not having done what they in fact did do and of claiming in fact things for themselves that they never actually have done...
Do remind me what we do call that?
My English vocabulary seems to leave me a little short...
Ciao T
We are going around this again.
You said:
"Yes, dither adds to the noise (it IS noise!) BUT it increases resolution."
As resolution is stated commonly in ENOB and dither degrades ENOB you statement is false.
There is no lie involved or attribution.
If you desired to express something else that what you typed, you could have clarified.
As it stands to claim that increasing noise increases resolution can only be regarded as a most extraordinary claim.
If what I attributed to you was not what you typed, then who did type?
Otherwise, in English and USA politics we find often that politicians make claims of not having said something that they in fact did say, of not having done what they in fact did do and of claiming in fact things for themselves that they never actually have done...
Do remind me what we do call that?
My English vocabulary seems to leave me a little short...
Ciao T
Hi,
True. However we need to consider if use a steady state signal, or music. Music has a constantly changing amplitude/frequency content.
To test how much of a problem this is and how much dither improves the situation I would suggest that noiseloading may be a much better test.
This we completely agree upon.
So we trade an increase in noise across the board (lest we apply noise shaping, which has other potential issues and challenges) for a reduction of spuriae mainly at a few spot frequencies, which have clearly high measured visibility, but who'se actual excitement with the intended signal (music/speech) and the resultant audibility is highly debatable.
In terms of classic definitions the improved SFDR is one measure that must be compared to the reduction in SNR/ENOB. Listening tests would be needed to decide.
Also, something to not forget, analogue sources (microphones, tape machines etc.) have considerable noise of their own, interestingly in the case of condensor microphones and tape machines we find this noise is actually usually approximating "red noise" and for a long time it has been argued that such noise counteracts the problems of quantisation noise especially well, note all this is anecdotal and there is little documented evidence.
However, this is, as they say, a whole other kettle of fish...
Ciao T
True. However we need to consider if use a steady state signal, or music. Music has a constantly changing amplitude/frequency content.
To test how much of a problem this is and how much dither improves the situation I would suggest that noiseloading may be a much better test.
This we completely agree upon.
So we trade an increase in noise across the board (lest we apply noise shaping, which has other potential issues and challenges) for a reduction of spuriae mainly at a few spot frequencies, which have clearly high measured visibility, but who'se actual excitement with the intended signal (music/speech) and the resultant audibility is highly debatable.
In terms of classic definitions the improved SFDR is one measure that must be compared to the reduction in SNR/ENOB. Listening tests would be needed to decide.
Also, something to not forget, analogue sources (microphones, tape machines etc.) have considerable noise of their own, interestingly in the case of condensor microphones and tape machines we find this noise is actually usually approximating "red noise" and for a long time it has been argued that such noise counteracts the problems of quantisation noise especially well, note all this is anecdotal and there is little documented evidence.
However, this is, as they say, a whole other kettle of fish...
Ciao T
Hi,
Remember, noise-shaping redistributes the noise-density towards areas of the audio spectrum where it is thought to be less audible (remember your own note on audibility earlier), the net amount must be equal.
So you normally end up with a lot of noise at very top end of the Audio Spectrum, where having a lot of extra noisefloor has implications, especially for those who insist on using NE5532/34 and TL072 et al in DAC analogue stages, but much less for those who prefer to use AD811 or LM6171/6181 et al or even tubes...
DSD/SACD is notable for what is probably the most aggressive use of noiseshaping, the Lipshitz/Vanderkoy paper on DSD is quite amusing.
Ciao T
Remember, noise-shaping redistributes the noise-density towards areas of the audio spectrum where it is thought to be less audible (remember your own note on audibility earlier), the net amount must be equal.
So you normally end up with a lot of noise at very top end of the Audio Spectrum, where having a lot of extra noisefloor has implications, especially for those who insist on using NE5532/34 and TL072 et al in DAC analogue stages, but much less for those who prefer to use AD811 or LM6171/6181 et al or even tubes...
DSD/SACD is notable for what is probably the most aggressive use of noiseshaping, the Lipshitz/Vanderkoy paper on DSD is quite amusing.
Ciao T
Please check the reference I gave and tell me the same thing about Bob Katz. This is old and well-known stuff, and two minutes of Googling will turn up thousands of references.
If you conflate and redefine terms and measures, come to what appears to be an incorrect conclusion, and say, "That's my view of it," that's at least honest. I may disagree with it and point out the flaws, but it doesn't reflect on your integrity and honesty, it just means one of us is not understanding the other, and we can proceed to have a reasonable discussion. When you instead attribute the results of an erroneous argument to me, then I have no other word for it. That mode of argument is disgusting and, IMO, immoral. I will not engage you on that basis other than to say 31.
If you conflate and redefine terms and measures, come to what appears to be an incorrect conclusion, and say, "That's my view of it," that's at least honest. I may disagree with it and point out the flaws, but it doesn't reflect on your integrity and honesty, it just means one of us is not understanding the other, and we can proceed to have a reasonable discussion. When you instead attribute the results of an erroneous argument to me, then I have no other word for it. That mode of argument is disgusting and, IMO, immoral. I will not engage you on that basis other than to say 31.
yes that is what SY said. and that is ALL he said.
You extrapolated what SY said and then attributed to him something that he had not actually said, ie
I'm going to ask that this pointless argument stop here. edit: feel free to discus the phenomenon and what dither does/does not do, just drop the "he said, so therefore meant" crap
Tony.
Last edited:
Hither be Dither
Well said, Pavel. It is obvious that the addition of dither raises the noise floor relative to all bits off. Dither is merely a means to subjectively improve the sound quality when the level is low enough that the distortion correlated with the audio signal due to the lowest bit(s) turning on and off is masked by a more stochastic distribution of energy. There are as many different noise shaping schemes for the added dither as there are manufacturers, for this reason; the preference for the spectral shaping of the added dither is subjective.
What is less obvious by thought, but not by ear, is that our sensory apparatus and hear significantly below the single number published S/N specification. Tests I ran years ago with another non-linear media (cassettes) showed that when an A-weighted meter showed 70dB S/N, a trained listener could pick out single tones as much as 30dB or more below this. The reason is obvious: noise consists of a random sequence of impulses at different frequencies, but not all at the same time. Therefor in a narrow-band analysis centered around the tone, the actual energy level relative to a steady-state signal is much lower than a single number integrating wide-band energy.
What this has to do with dither: The dynamic range specification of a digital system is expressed by a single number, as is the added dither. Neither specification describes the actual sound of low-level audio residing in the lowest bit(s). When one choses a dithering scheme, you don't do it by a number, you choose it for the affect on the sound of the audio in the lowest bits. And as Pavel pointed out, the sound of proper dithering contains less spurious components correlated to the audio than does the same audio without the dither. Does it contain more noise? Sure. Will measurement systems show more noise? Yes. Can you perceive low level music more clearly with it? Yes. A spectral analysis of the music with added dither can show the noise obscuring the low-level audio, however human hearing can often pick it out of the noise.
I apologize for wordiness as usual, YMMV
Howie
Howard Hoyt
CE - WXYC-FM 89.3
UNC Chapel Hill, NC
www.wxyc.org
1st on the internet
Well said, Pavel. It is obvious that the addition of dither raises the noise floor relative to all bits off. Dither is merely a means to subjectively improve the sound quality when the level is low enough that the distortion correlated with the audio signal due to the lowest bit(s) turning on and off is masked by a more stochastic distribution of energy. There are as many different noise shaping schemes for the added dither as there are manufacturers, for this reason; the preference for the spectral shaping of the added dither is subjective.
What is less obvious by thought, but not by ear, is that our sensory apparatus and hear significantly below the single number published S/N specification. Tests I ran years ago with another non-linear media (cassettes) showed that when an A-weighted meter showed 70dB S/N, a trained listener could pick out single tones as much as 30dB or more below this. The reason is obvious: noise consists of a random sequence of impulses at different frequencies, but not all at the same time. Therefor in a narrow-band analysis centered around the tone, the actual energy level relative to a steady-state signal is much lower than a single number integrating wide-band energy.
What this has to do with dither: The dynamic range specification of a digital system is expressed by a single number, as is the added dither. Neither specification describes the actual sound of low-level audio residing in the lowest bit(s). When one choses a dithering scheme, you don't do it by a number, you choose it for the affect on the sound of the audio in the lowest bits. And as Pavel pointed out, the sound of proper dithering contains less spurious components correlated to the audio than does the same audio without the dither. Does it contain more noise? Sure. Will measurement systems show more noise? Yes. Can you perceive low level music more clearly with it? Yes. A spectral analysis of the music with added dither can show the noise obscuring the low-level audio, however human hearing can often pick it out of the noise.
I apologize for wordiness as usual, YMMV
Howie
Howard Hoyt
CE - WXYC-FM 89.3
UNC Chapel Hill, NC
www.wxyc.org
1st on the internet
The experiment was a 100 ohm or so source into a soldered in copper band loaded by about 100K. The distortion was measured, the copper band was heated to orange hot with a torch and allowed to cool with the surface now coated by nice oxides. While heated the distortion rose when cooled the distortion returned to the base level. Since there was absolutely no change under extreme "micro diode" generation, it seems unlikely that even a lower load impedance would have much effect. Also keep in mind we are looking at a level within a few db of the noise floor of 100 ohms at 1/2 of an octave centered around 2.3khz.
For those unfamiliar with the color heat scale it goes from red to orange to yellow to white. Copper melts around yellow.
P.S. Even if it hasn't made it to red... don't touch it!
- Status
- Not open for further replies.