When I search the AES library for "Dither" I get:
Authors: Blessener, Barry; Lee, Francis F.
Affiliation: Massachusetts Institute of Technology, Cambridge, MA
AES Convention:40 (April 1971)
Author: Blesser, Barry A.
Affiliation: Blesser Associates, Raymond, NH
JAES Volume 26 Issue 10 pp. 739-771; October 1978
and!
A dynamic range of 118 dB is determined necessary for subjective noise-free reproduction of music in a dithered digital audio recorder. Maximum peak sound levels in music are compared to the minimum discernible level of white noise in a quiet listening situation. Microphone noise limitations, monitoring loudspeaker capabilities, and performance environment noise levels are also considered.
Author: Fielder, Louis D.
Affiliation: Ampex Corporation, Redwood City, CA
AES Convention:69 (May 1981)
Authors: Blessener, Barry; Lee, Francis F.
Affiliation: Massachusetts Institute of Technology, Cambridge, MA
AES Convention:40 (April 1971)
Author: Blesser, Barry A.
Affiliation: Blesser Associates, Raymond, NH
JAES Volume 26 Issue 10 pp. 739-771; October 1978
and!
A dynamic range of 118 dB is determined necessary for subjective noise-free reproduction of music in a dithered digital audio recorder. Maximum peak sound levels in music are compared to the minimum discernible level of white noise in a quiet listening situation. Microphone noise limitations, monitoring loudspeaker capabilities, and performance environment noise levels are also considered.
Author: Fielder, Louis D.
Affiliation: Ampex Corporation, Redwood City, CA
AES Convention:69 (May 1981)
Pavel,
A transient is in theory of infinite bandwidth, therefore after the LPF filter should show up as the maximum frequency component passed and as mentioned jitter (or ratiometric system process gain) decreases as frequency increases. So we agree at maximum frequency dither does not provide enhancement.
One other issue mentioned was a Vanderkooy paper which if I recall it correctly did a long math treatment but ignored the physical implementation. If you have a sample and hold grab your input signal, how can you tell what the A/D conversion process is?
Finally I started by using a 16 bit Discrete Level and Discrete time sampling system as being too limited for my idea of audio use. Virtually everyone assume I was talking about CD use. Since most of my work is on live use, that is not what the image in my mind was.
So there are still quite a few translation issues to be worked out.
SY he is just plain wrong!
Show me how it works in A/D, to get below 1/2 LSB resolution. How you get the right digital representation of input signal of right amplitude at right time, with resolution higher than 1/2 LSB.
You need exact and precise A/D conversion to get the signal for CD playback.
I am not interested in something periodic, delayed, averaged and placed at wrong time at D/A output. No empty words, but facts please (as you like it).
Last edited:
Simon, sure, I would correct it as a non-periodic signal limited in bandwidth. I do not want to speak about infinity etc. I am sure you know what I mean.
Music is a flow of transients, not steady state periodic sinuses.
Pavel,
I think we are having trouble communicating. I can design a waveform that sounds like noise but will allow low frequency components of the signal to bounce up and down enough from their nominal value that they will get multiple encodings over their period. When reproduced (as signal and out of critical band noise) this is almost the same as doing more step sizes to increase the resolution. It does not require the input sample to be periodic, only to be long enough to get multiple samples over the base period.
ES
Ate, drank, fooled around, and when exhausted and broke designed equipment! Why the rhetorical questions?
Not digital, Maybe Barry was not in the inner circle and losing SNR below the 16 bit level was politically incorrect to the "perfect sound forever" crew. When the raw resolution of A/D's got better the issue is less important.
I have the AES digest from 1982, dither was a very small footnote. Serious evaluation of different methods had not been done. Saying digital has problems and dither won't fix it is not a particularly useful input without even any data.
Last edited:
??? I am completely puzzled. I quoted this:
And this, which is referenced above:
It was a JAES Paper from 1978! Not 1982.
Why are you insisting that he did not mention dither??
Hello Thorsten
>>Having tried it I prefer to do the EQ in analog hardware and to convert to Digital afterwards, strictly for subjective sonic reasons. But I also like multibit DA's without dither and followed by tubes that produce a completely scandalous around 0.2% 2nd HD at digital full scale, so WTFDIK?
When you refer to digital full scale what is the RMS level of the analog output signal.
Regards
Arthur
>>Having tried it I prefer to do the EQ in analog hardware and to convert to Digital afterwards, strictly for subjective sonic reasons. But I also like multibit DA's without dither and followed by tubes that produce a completely scandalous around 0.2% 2nd HD at digital full scale, so WTFDIK?
When you refer to digital full scale what is the RMS level of the analog output signal.
Regards
Arthur
Did you read what I said? He had a small mention of dither in a later paper of the same conference. His keynote paper did not mention it as essential (or at all). Even in 1982 it seems the industry (AES) as a whole did not consider it very important. In fact Barry discovered it at MIT and didn't bother publishing it, his technique as Lipshitz mentions was apparently never developed further. Apparently his 1978 paper did not light any fires. Ed found only 3 papers pre 1982, but I don't know if he was exhaustive. Besides dither is so close to things like stocastic resonance that I don't believe there is not a lot more relevant papers in other disciplines. So forget it, I have no idea how the AES decides what's important
EDIT - Ed found a mention in 1971, I find it strange it wasn't studied fully well before 1981.
Last edited:
What *is* this fuzzy distortion, later quoted to be 50% or 1000%, that isn't measurable, and must be at least pretty hard to hear (since most people can't hear *any* difference in many serial passes through an ordinary stock 44.1/16 bit A/D/A)? Is it some kind of modulation of signal, or are we talking about simple noise? Or what?
The word "transient" if fraught with peril - A/D/A conversions are necessarily bandlimited. Could you suggest a setup where I could see this fuzzy distortion for myself? (Got lotsa scopes.)
Thanks,
Chris
In my simulations it looks like noise, I suppose added noise is a fuzzy distortion. I don't see a controversy here. A simple truncation experiment works. Take a floating point 1k sine wave at -80dB make two copies and add -80dB rms gaussian noise to one (not ideal but proves a point) and tuncate them to 16bits with no dither. The one with added noise does not have the THD and the other looks exactly like the input fed through an ideal simulated A/D. So the official CBS/SONY CD test disk (circa 1983) was incapable of testing the true low level performance of a DAC since the test waveforms are packed with in band tones/distortion that could have been eliminated with dither.
Last edited:
- Status
- Not open for further replies.