Careful it is very easy to pose an example that violates the sampling theorem i. e. the amplitude of a transient containing information at 20K can not change fast enough to create >22.05k sideband information. Using pulses as stand-ins for transients does not cut it.
This problem is very easy to simulate mathematically. Remember I also would like to see this issue illustrated. The arguements so far still rely on "I hear what I hear."
In a conversation with Bob Adams I asked about creating a psuedo-random dither sequence of some FFT length that has both a flat spectrum and is still TPD at one LSB or what ever the perfect linearizing one is. Unfortunately the problem is mathematically tedious or actually does not yield to iterative techniques. This would eliminate the averaging and be an interesting experiment.
Last edited:
Hmm, that's 1981 I just posted a CD plot from Sony/CBS (on a technical CD that was only meant for audio professionals) from 1983 - no dither. I remember the AES papers from the cheese-head twins as being less that 30yr. ago. You did say over 30yr. ago?
Last edited:
Mr. Lipshitz says some people knew about dither before they published their paper. Obviously Mr. Curl is one of them.
Hi,
I did not take any. This was many years ago.
Sadly the classic Bitstream DAC's that do not include on board averaging in the form mostly of a switched "dirty sand" capacitor filter are history.
A 'scope on the output of an SM5872 (Marantz CD63/67/6000) while playing a square wave from the CBS Test CD is instructive. Edit, even more instructive are sawtooth waves, but thos are not on the CBS test CD...
The 'scope I used was analog, yes. Philips, tube based (it had solid state rectifiers, but PSU Regulation was tube & Gas Reference, 50MHz. I hated it when it broke beyond reasonable repair. It can also be done I think with a decent digital 'scope in single shot mode.
Ciao T
I did not take any. This was many years ago.
Sadly the classic Bitstream DAC's that do not include on board averaging in the form mostly of a switched "dirty sand" capacitor filter are history.
A 'scope on the output of an SM5872 (Marantz CD63/67/6000) while playing a square wave from the CBS Test CD is instructive. Edit, even more instructive are sawtooth waves, but thos are not on the CBS test CD...
The 'scope I used was analog, yes. Philips, tube based (it had solid state rectifiers, but PSU Regulation was tube & Gas Reference, 50MHz. I hated it when it broke beyond reasonable repair. It can also be done I think with a decent digital 'scope in single shot mode.
Ciao T
Last edited:
Mr. Lipshitz says some people knew about dither before they published their paper. Obviously Mr. Curl is one of them.
No idea what you are talking about, must be mainly the goal of discrediting JC.
Here the reference that dither in digital audio was known then:
Both quoted from Mr. Lipshitz and Mr. Vanderkooy Paper from 1984.
Mr. Lipshitz says some people knew about dither before they published their paper. Obviously Mr. Curl is one of them.
Since JC claims to know Tom Stockham I suppose the truth will never be known.
But on another point, I hope folks made the observation that making the plot I posted was very difficult for an average researcher in 1983. First of all there were no CD rippers at least easily available and essentially only mainframe computers could do 64K double precision FFT's. I can see what Harry Pearson said about the malignant absense of noise. The actual noise floor is -156db or so but there are plenty of spurs.
Using which definition of resolution? What kind of signal? What kind of dither?
Numerology.
That's it. You can model it and see the same. From nowhere/infinity to horrible spurs
Again, please be correct in technical terms. In theory, it is 98.08 dB (6.02N + 1.76). When you start adding FFT narrow band analysis with narrow binwidth, long memory and averaging, you can get to fantastic 130 - 140dB and confuse the issue
Pavel,
What happens if I decide to use that format to only record up to 2,000 hertz?
I have already mentioned that the ear has critical bands similar to a DFT system and can recover signals below the wideband noise and have suggested between that and the handling of crest factor you need to allow 9 bits to be safe on those issues, so we pretty much agree about the 130 to 140 issue.
(Crest factor is why signals such as applause are so difficult to get right!)
What I am trying to oversimplify is that even with PCM you can still apply PWM (Which I prefer to call ratiometric the original and more easily understood name) to gain additional apparent resolution. That as bandwidth goes down apparent resolution can go up.
Then if you are clever you can generate a signal to modulate the LSB in a frequency dependent manner to increase the apparent resolution, with more at lower frequencies and none at the high end. Of course it is ideal to use a modulating signal that our ears pretty much ignore. (Like noise!)
This then leads to why you want a 16 bit A/D that actually can be 22 bits, so you can do more linear modulation, then we go on to oversampling and finally digital filtering. Maybe even why some CD's really do sound great at "16" bits while the original where actually 9 bits and not quite as good as a high end cassette!
As to PWM not being equal to PCM !!!
Vinyl De-noisers
I understand that the main thrust of this discussion is not vinyl noise reduction, however; there are a few analog-only vinyl NR units which are, in my experience, far superior to their DAW plug-in offspring. Also, it is important to remember in this discussion that we are talking about recordings which have OBJECTIONABLE noise, so they are already less than optimum, we are just trying to make a bad situation better!
The best Analog NR I have used is the Packburn 323, which analyzes supersonic (>20kHz) information to decide what signals in the audible passband are correlated with surface noise and what is recording. When correctly adjusted it is capable of amazing amounts of noise reduction (especially transient noise) with precious little degradation to the HF dynamics of the recording itself. The Burwen TNR is a similar though less capable unit.
The pre-requsite (obviously) of using DAW NR is that the recording already be band-limited pre-A>D. Because of this one fact, an important clue as to the nature of the in-band noise is missing, and the NR plug-in has to guess based on other characteristics of in-band information. I have worked in a replication plant since 1986, and have purchased Sonic Solution, CEDAR, and other NR systems claiming to be side-effect-free in their NR. On critical listening, the Packburn causes less incidental damage to the recording. It is this fact that informs me that the "software models" of the DAW NR plug-ins units are far less capable. Indeed, since the supersonic clues are not present, the software models are based on entirely different working principles.
I have a modified Burwen unit at home which I switch into the phono path when needed with very satisfying results as well, once again remembering that the goal is to replace a problem with something less objectionable.
The techniques used in these units is not magic, it is just smart engineering!
Howard Hoyt
CE - WXYC-FM
UNC Chapel Hill, NC
www.wxyc.org
1st on the internet
I understand that the main thrust of this discussion is not vinyl noise reduction, however; there are a few analog-only vinyl NR units which are, in my experience, far superior to their DAW plug-in offspring. Also, it is important to remember in this discussion that we are talking about recordings which have OBJECTIONABLE noise, so they are already less than optimum, we are just trying to make a bad situation better!
The best Analog NR I have used is the Packburn 323, which analyzes supersonic (>20kHz) information to decide what signals in the audible passband are correlated with surface noise and what is recording. When correctly adjusted it is capable of amazing amounts of noise reduction (especially transient noise) with precious little degradation to the HF dynamics of the recording itself. The Burwen TNR is a similar though less capable unit.
The pre-requsite (obviously) of using DAW NR is that the recording already be band-limited pre-A>D. Because of this one fact, an important clue as to the nature of the in-band noise is missing, and the NR plug-in has to guess based on other characteristics of in-band information. I have worked in a replication plant since 1986, and have purchased Sonic Solution, CEDAR, and other NR systems claiming to be side-effect-free in their NR. On critical listening, the Packburn causes less incidental damage to the recording. It is this fact that informs me that the "software models" of the DAW NR plug-ins units are far less capable. Indeed, since the supersonic clues are not present, the software models are based on entirely different working principles.
I have a modified Burwen unit at home which I switch into the phono path when needed with very satisfying results as well, once again remembering that the goal is to replace a problem with something less objectionable.
The techniques used in these units is not magic, it is just smart engineering!
Howard Hoyt
CE - WXYC-FM
UNC Chapel Hill, NC
www.wxyc.org
1st on the internet
CBS/Sony/Philips rolled out CD without dither in many cases except or course the tape noise was sufficient in most cases to obviate the dither somewhat, in fact IIRC Lipshitz gave a paper with actual samples from released CD's. There was no comprehesive theory and audio industry wide involvement with dither in 1980, in fact Soundstream considered it a trade secret. Did Tom Stockham argue over it with JC in 1980 thus giving away something of great value to him? See Lipshitz interviewed, he tells the whole story about seeing Tom's notebooks after his 1987 paper.
So please provide evidence of an industry wide discussion concerning dither in CD's (or digital recording in general) among audio engineers in 1980-81. Also please consider how relatively primitive digital audio was in 1980. I would hardly think it was at the state where critical listening and analysis of dither was the top priority.
Last edited:
Very interesting comments re: CEDAR are you talking about their full DSP workstation? I know top restoration engineers swear by it.
If you want to talk about methods to increase it beyond the actual number of bits then you are misusing the common use of the word resolution.
That is why I refer to it as enhanced resolution.
When you don't have linearity past the given LSB you can generate additional data some of which is perceived as an improvement, but it is not as good as actually having more LS bits and doing a fancy truncation algorithm. So Pavel and I pretty much use the terms the same way as do most of those I work with.
So lets not grind things to where acetylsalicylic acid is required.
Scott,
The hand built Audio A/Ds I saw in the 70's had built in dither! It was from the Allen Bradley hot molded carbon composition resistors! They even used parallel resistors to trim in the LSB's. Until the analog noise dropped the need for dither was not apparent!
ES
- Status
- Not open for further replies.