|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
212.42.175.73
In Reply to: Noise characteristics of the formats. posted by Al Sekela on February 20, 2007 at 14:24:44:
Dear Al,Pointing to noise as the main contributor to the differences between the formats may have some relevance with Redbook versus SACD, although I think the main difference lies in the inherent sonic signature within the different architecture of the A-D/D-As used in the single bit, heavily oversampled, SACD versus the multibit original Redbook A-D/D-As, as I think thorough comparisons between discs made with single and multi bit systems will confirm.
The case is different with analogue, because unlike the digital formats, the analogue formats do not cut off the information at the noise floor, or have low resolution at small signal levels, so perhaps you should revisit your theories on why the formats sound different.
In fact, I think part of the secret of why LPs sound better is buried in the fact that the sound continues well below noise level and despite what some digital theorists will have you believe, the human ear is perfectly capable of hearing information well below the noise floor (just think of sitting in a noisy restaurant, having a conversation for example, it is perfectly possible to take part), so whatever theory may be applied to the digital formats, analogue is different, and this shows in the sound as well as the reasons why and it certainly does not link noise limiting resolution.
Sincerely,
Peter Qvortrup
Follow Ups:
The case is different with analogue, because unlike the digital formats, the analogue formats do not cut off the information at the noise floor, or have low resolution at small signal levels, so perhaps you should revisit your theories on why the formats sound different.Thanks to dither, digital formats DO preserve signal information below the noise floor. When dither is properly applied, quantization noise is additive and uncorrelated, just like analog white noise.
Dear Dave,Please see my response to theaudiohobby elsewhere in this thread.
The terms have historical connotations that may not apply to the topic at hand. Is correlated 'noise' more like a form of distortion? How do correlated and uncorrelated noise sources limit our ability to resolve musical details?You are right in pointing out that any difference in the physical processing chains for the different digital formats will introduce differences in sonic character.
"In fact, I think part of the secret of why LPs sound better is buried in the fact that the sound continues well below noise level "hmmm....ticks,pops and surface noise are not a noise floor just uncorrelated noise, as is the case with uncorrelated noise, if the signal is strong enough in comparison to the noise, you will hear signal, there is cut off where this condition no longer holds i.e. the noise overwhelms the signal and that is a noise floor.
"(just think of sitting in a noisy restaurant, having a conversation for example, it is perfectly possible to take part), "
As Earl Geddess pointed out a while back, the cocktail effect does not apply in this instance, afterall in the case of LP or any other analogue audio format, there is no other stimuli apart from the signal, not so the case in the noisy restaurant, there are various other stimuli such visual that can help intelligibily.
Music making the painting, recording it the photograph
Dear TAH,What you say does not disprove the fundamental truth in my statement, fact is that analogue media have audible and recognisible signal artifacts well below their noise floor, which digital media do not.
I am not sure what kind of turntable you have but I have not had much of this kind of noise for many years and even on 78 where there can be quite a lot of clicks and pops they rarely last long enough to overwhelm the signal as you suggest, in fact this is no different from sitting at a concert with people coughing behind you in the quiet passage of a symphony, the musical line remains unbroken.
Cocktail effect?
There are no other stimuli, for example, from someone sitting with their back to you in a restaurant, but you can still hear what they say even if it is below the ambient noise in the room.
It is well established that the ear has marvellous separation of individual sounds from noise once it has had a chance to home in on this, something we have not been able to replicate with any test equipment I know of.
All this really means is that we lack the complete understanding of what causes the differences in sound between the media, which is hardly surprising given the primitive measurements we use as proof of progress.
Sincerely,
Peter Qvortrup
At this point, it behooves me to revise my original comment, i.e. a signal may be retrieved from a given noise floor provided the noise is uncorrelated with the signal and this applies to many scenarios of which include the noisy restaurant, digital audio or analogue audio (LP or tape). In other words, the ability to retrieve signal from groove noise etc is not unique to analogue audio.
Music making the painting, recording it the photograph
Dear TAH,OK, so let us look at what these differences are then, since you persist.
Firstly, analogue is quantised, but at a ridiculously low level, Planck time etc, this is true physical quantisation and is quite different to the quantisation we speak of with digital.
Here is why,
The case with digital audio is that there is quantisation in two dimensions, time and amplitude. Dither uses the stochastic resonance effect on the amplitude dimension, but leaves the time dimension rigidly quantised.
Anyway, the digital noise floor is not the same as the analogue floor. If you conceptually look at it, at any moment in time, and considering signal voltage for convenience. The analogue floor, again at any given time, could according exist at any voltage level, there are some levels which are highly improbable, and some which are physically limited however within reason the voltage level could be anything at all.
Adding hither and dither.
By adding dither at the recording end the previously undetectable signal is encoded into the digital data by way of a kind of PWM (pulse width modulation), the higher level the undetectable signal, the more often will it toggle the LSB when ordinarily it wouldn’t do so.
If you apply a low level signal to an ADC, one which never reaches an amplitude which would toggle the LSB (lowest significant bit), you won’t see it. If you add some noise such that there is some proportionality between the signal level and how often the LSB is toggled per unit time, and then you integrate the output, you can detect some kind of representation of it. The smaller the signal is, the longer and longer the time which is required to build up a picture of it.
It isn’t magic though, the dither signal pushing the LSB to toggle means that occasionally the ADC input will clip when it wouldn’t have done with no dither. With CD you have a 16 bit window, but the same applies to any bit window, at one end is no recorded information, at the other end is clip.
By adding dither at the playback end you can lessen the unpleasant effects of noise which is correlated with the signal, and make something which measures like true analogue noise, whether it sounds the same as analogue noise is another matter.
Sample by sample, there can be no information which occupies any level smaller than the LSB of the data, but if the output is integrated, and the “previously undetectable signal” was periodic (or DC), which then had a PWM effect on the ADC data, and you do an FFT of the signal taken over many, many samples you’ll see some kind of representation of it. Once again, by adding dither to the data going into the DAC occasionally it will clip when with no dither it wouldn’t have. So dither increases detectability, but reduces dynamic range, and a compromise is required.
The increase in detectability relies on the signal being periodic or DC, so that you can build up a picture of that signal over more samples than would be required to record a signal which is within the normal 16 bit window.
With music, as far as we are concerned, you need to look at the situation without any kind of averaging or integrating, because it must be assumed there is little or no correspondence between one sample and the next. If this assumption is accepted then dither makes no difference, all that can be recorded must be done so within the 16 bit amplitude window.
Analogue has no such constraints, effectively it has no quantisation in either the frequency or time dimensions and that in our view explains why a good analogue recording will never be bettered by any digital medium regardless of theoretical "resolution", noise or bandwidth.
Sincerely,
Peter Qvortrup
Andy Grove
Peter,The core issue in extracting signals from noise is the same regardless of whether we're talking analog or digital. In both cases, the instantaneous amplitude consists of both a signal component and a noise component, whether the source of the noise is tape hiss or dither. And at any instant in time, it is impossible to separate them. In both analog and digital cases, a short duration transient signal at or below the noise floor will be indistinguishable from the noise. And in both cases, extracting a signal from the noise requires averaging, and the longer you average over, the greater the chance of detecting the signal. In fact, fundamental signal detection theory makes no distinction between analog and digital signals. On audio playback, your ears and/or brain do the averaging required to extract the signal from either analog or digital noise. In other applications, circuits or software do it. But the principle is the same regardless.
Quantization does not prevent signals smaller than the LSB from being encoded. Quantization simply adds correlated noise to the signal, which can mask the signal. The purpose of dither is to add uncorrelated noise prior to quantization, so the resulting quantization noise will be correlated to the dither instead of the original signal. Once that is done, averaging will raise the signal to noise ratio and allow the small signal to be detected. Without the dither, the signal and quantization noise remain correlated and thus averaging cannot separate them. Similarly, correlated noise in the analog domain will also mask small signals, for the same reason.
But there is nothing special about dither, any uncorrelated noise source will do. If the analog signal to be converted already has uncorrelated noise above the quantization noise floor, than no dither is required. And if that analog signal has components below the quantization noise floor, they will still be encoded, even without the dither.
In short, regarding the detectability of signals below the noise floor, what matters is whether the noise is correlated or uncorrelated, not whether it is analog or digital.
Dear Dave,I did think that someone might come back with an answer like that, even though I hoped we had built in sufficient information to illustrate that there is a distinction between the digital and analogue cases when MUSIC or another “random” pattern is the signal, because signal processing theories are generally blind to the application, as they were most likely derived by university lecturers and telephone engineers, without music as a specific application.
Let me quote from Terry Pratchett's Discworld, Moving Pictures,
"Reality is not digital, an on and off state, but analogue, something gradual. In other words, reality is a quality that things possess in the same way that they possess say, weight."
Ignoring the amplitude domain/dimension, because for this argument, time and frequency are sufficient.
Digital: The time domain is quantised, and therefore the set of samples, from time zero to time infinity can be counted according to integers, 1,2,3, no matter what the sampling rate. The set of samples has a one to one correspondence to Aleph 0, the set of countable numbers.
Analogue: Analogue is not a discrete process, therefore cannot be counted by integers, and would correspond to the set of real numbers Aleph 1 (or c) depending on the acceptance of the continuum hypothesis, but that makes no difference to this argument.
Cantor proved that Aleph 1 (or c) is greater than Aleph 0.
Therefore digital can never be analogue, no matter what the sampling rate, EVEN IF IT WERE INFINITE.
Cantor also showed that any subdivision of the real number line also has a correspondence to the entire line. That may have some interesting implications, and might conflict with certain sampling theorems.
Maybe one could enter an argument that at some level the analogue time domain would be subdivided by Planck time, but one would have to examine the derivation of Planck time, and how it relates to this set theory argument.
Looking at it this way points to the analogue noise floor having “no grain”, it is continuous, the digital noise floor would be composed of discrete frequency bands. Dither may/would modulate noise out of bands, but again I’m not sure whether it’s possible to translate from a correspondence with Aleph 0 to c, I don’t think it is at first thought.
This is essentially the issue, so whilst much of what you say is correct according to conventional theory it does not actually cover the full subject, but to comprehensive prove this one would have to spend a lifetime in research and there is unfortunately no money in that and commercial pressures against it are comparable to what was wielded against hemp in the 1930s, so even if one came up with an answer it is likely to be suppressed by powerful commercial interests.
My argument is essentially based on: “How do you average music to detect it?” Maybe there are periodic elements in there which can be averaged. With digital you get one sample, every 22 odd microseconds, with analogue the time domain is continuous, so the question is, what period do you have to average over for it to make sense?
There is much further theory that can be used, in number theory for example, but we would have to study it further, and to transfer, for example, transfinite numbers into information theory or signal theory or whatever is the kind of thing which takes 20 years and you win a Nobel prize for, provided you live long enough!
Going back to the textbooks and digging out the equations and then quoting them assumes that whoever did the original assumptions and subsequent calculations actually did them both including the maths with a view to achieve the best possible end result i.e. sound, and even if they did, then by the time products were designed using these theories and their calculations, I think that commercial expedience (meaning achieving low cost primarily) would have been higher on the agenda, as is mostly the case with this type of technology (question to solve is/was, "How do we get round these problems quickly and cheaply?" not, "How do we do this properly?"), so whilst the text books provide a "solution" and arguments for why, they do not necessarily hold anything close to the final answer to the real requirements of highly variable and complex waveforms contained in music, because if they did then what we have from digital reproduction media would be closer to or even better than the best analogue reproduction solutions they were intended to replace and live up to the original slogan "Perfect Sound Forever".
So take you pick, either the theory with its associated assumptions and calculations are at worst wrong or at best incomplete, or the products that are designed based around them have be designed cutting corners.
Just in passing allow me to give you a simple question that current theory has problems answering,
"Did you ever try a silver cable?”
Standard answer to that is,
“I don’t need to it will make no difference.”
However when you try it between transport and DAC it makes a significant difference, in ways which cannot be explained with any current available "tools"
This discussion has got us thinking though, about number/group/set theory and Georg Cantor’s work, so we shall dig deeper when time allows.
Peter/Andy,First, an answer to your simple question: I use silver ICs. I believe I hear the difference and have a preference, but don't know why (from an engineering standpoint). I'll admit the differences I hear between ICs seems smaller than other cable differences, which are also small but not insignificant.
Second, I don't believe that digital offers audibly perfect reproduction. I think there is a different set of issues and tradeoffs in digital reproduction vs. analog. But as I mentioned in my prior post, I don't think the difference lies in averaging.
I don't really think the issue is continuity either. I wish I had the time and resources to really study it, but I suspect the difference between analog and digital reproduction mostly boils down to a few things: mastering, the effect of component selection on frequency response, differences in the nature of noise and distortion, and the audibility of steep reconstruction filters.
For example, variations in frequency response are much greater in analog reproduction, allowing listeners more latitude to tailor their playback equipment to match their systems and preferences. Mastering is also different. Although I haven't participated in one, I've heard of many demos where LP playback was digitally recorded and then replayed on the same system. The usual outcome is that the digital reproduction comes close to bridging the gap to analog playback but falls a little short, especially on less tangible qualities of reproduction. I suspect the remaining audible difference reflects our ability to detect differences in the noise "signature" among formats and/or sensitivity to a sharp cutoff in frequency response in or near the audible range.
Anyway, that's my guess. Thanks for sharing yours.
Hi Dave,Very interesting thread with intelligent, informed arguments put forward in a mature manner - excellent reading even if the bulk of it required me to do a couple of searches in order to at least partially understand the main disagreement between you and Peter/Andy.
All I wanted to add was that like yourself, I have a preference for silver interconnects and can understand why they should sound 'better' than OFC copper when carrying an analogue signal, but exactly why a silver digital connection between transport and DAC should be preferable to OFC copper is something that has me scratching my head, especially when the 'bits is bits' brigade would have us believe there should be virtually no difference between ANY digital connection.
Best Regards,
Chris Redmond.
"What you say does not disprove the fundamental truth in my statement, fact is that analogue media have audible and recognisible signal artifacts well below their noise floor"No, they do not.
"There are no other stimuli, for example, from someone sitting with their back to you in a restaurant, but you can still hear what they say even if it is below the ambient noise in the room."
ambient noise is not a noise floor, much in the same way that groove noise is not a noise floor, just uncorrelated noise.
Music making the painting, recording it the photograph
"No, they do not"Yes they do.
"ambient noise is not a noise floor, much in the same way that groove noise is not a noise floor, just uncorrelated noise."
Background noise in a room is most certainly a noise floor whether correlated or not. HOwever; information that is correlated can be heard below a normal noise floor. This is more difficult if the noise floor is correlated with the music signal.
Ever listened to analog tape? There is a constant hiss from the tape mechanism. This is the tape noise floor. Now you can easily hear sounds that are softer than the tape hiss, can you not?
I think Peter's point is that there is information below the nominal noise floor of vinyl and that as this is musical information and therefore correlated it can be aubible even through the noise (tape hiss is another example). Obviously, the lower this noise is the better because it has less chance for masking.
An example of correlated noise floor is the example given by Crowhurst regarding negative feedback in amplifiers. He demonstrates that feedback generates a forest of peaks that look like noise; however, these peaks are coming from the signal itself and as such are frequency and level dependent. So the "noise floor" shifts in frequency and level with the signal. With this kind of noise floor you will surely run into a situation where infromation below this floor is masked.
thanks for the correction vis--vis noise floors, that said the principle here is noise decorrelation, if the noise is decorrelated then a signal may be retrieved, the principle applies to both analog and digital alike, or is there any reason why analogue is unique?"Ever listened to analog tape? There is a constant hiss from the tape mechanism. This is the tape noise floor. Now you can easily hear sounds that are softer than the tape hiss, can you not?"
tape noise, groove noise, dithered noise, birds of the same feather but different applications, you simply repeated what I said in the post you responded to :o, aren't both groove noise or tape hiss simply different types of uncorrelated noise.
I will ignore the off-topic point on negative feedback.
Music making the painting, recording it the photograph
I guess undithered digital is somewhat different but I don't have sufficient knowledge there to really comment.The amp comment is not off-topic if you think about it. The argument is that negative feedback CAUSES a signal correlated noise floor below which it would be difficult to extract signal because of masking.
A no feedback amp would not have this issue and is perhaps why users of SET type amps (or even Class A PP amps with no feedback...I have one of those as well) feel that the low level resolution is superior because in terms of the correlated vs. uncorrelated noise it IS superior. This would never show up on a static distortion measurement.
This post is made possible by the generous support of people like you and our sponsors: