Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Nika Aldrich]

Lee,

Yes, sampling is dependant upon two coordinates - the amplitude (y) and the positioning (x).  Amplitude error clearly manifests itself as amplitude error (duh).  Timing error, however, is more complicated.

If the samples are taken at varying intervals of time - say there is error in the sampling mechanism - then the resultant samples will be interpreted as having happened at the correct times, and the data will travel through the rest of the system like this.  The result is that bad amplitude data will be captured, and this data will be like any other amplitude data error.  In other words, bad timing data during the capture process manifests itself as bad amplitude data.  This is because bad timing data causes bad amplitude to be captured.

OK, now on the output side - this doesn't cause bad amplitude data, per se, but it causes the same results.  Jitter on the output means that the samples are turned into analog voltages at the wrong times, thus recreating an erroneous version of the waveform.  This erroneous version, however, is of the same variety that happens when bad timing is used on the input - the waveform that is represented by the samples is erroneous in the same way.  The timing error on the input causes the waveform to be misshapen - stretched or compressed - representing bogus frequencies (that's the noise part) and the more error there is the more amplitude those other frequencies have.  The same occurrs on the output - the waveform - by means of being converted at the wrong times - ends up compressed or stretched in the exact same way.  In other words, the manifestation of jitter on the input or the output both produces a result that is akin to simple amplitude error of a certain variety.

Am I making sense?  Ask if you have more specific questions.

Nika

[Lee Flier]

Thanks Paul, I think you're getting to the crux of my questioning.

Paul Frindle wrote on Mon, 21 February 2005 16:52

You are very correct - we are in fact significantly more sensitive to this (what is essentially high speed wow and freq modulation) than people seem to think. The loss of detail and timbre in the extreme top end and loss of imaging depth is evident when this happens.

Yes, I agree, as I am very sensitive to this myself.

Quote:

But there is a misunderstanding that this is somehow caused by the quantisation of timing samples alone - the effect would not improve significantly even with infinite sampling rate

Hmmm... like I said to Nika, I can see where this would be the case if the clock has systemic error but not in the case of random error.

And I don't necessarily think it's timing *alone* that causes these problems... all I'm saying is that ANY error that occurs in sampling would seem to be highly significant if you've only got two samples from which to reconstruct your waveform.  You seem to agree with me on that point... but yes, you're correct that where I must be misunderstanding is where you say increasing the sample rate, even infinitely, wouldn't improve things.

Quote:

- if you could modulate the timing of analogue signals in the same way the effect would be audibly identical.

Yes, but of course with analog signals it isn't necessary to do that.

Quote:

What is of course needed is due care taken with timing signal handling in the first place - with due regard to the much ignored fact that the clock signal IS ANALOGUE in every respect  - not only just electically, but also in its function

Well yes... I agree that a lot of the "problems" with digitizing audio occur on the analog side... but that seems more than a little irrelevant to me considering you can't separate the two from each other!  Sometimes digital theorists seem to be saying "Digital audio would be perfect if only we could eliminate that pesky analog stage - that's the problem!" but you can't, and why would you really want to?  And if you have to twist the analog signal into a horrible pretzel in order to digitize it, then it's little comfort that the digital domain "faithfully" captures and reproduces all of these horrors.  I'm exaggerating of course, but there's a point in there somewhere.

[Lee Flier]

Nika Aldrich wrote on Mon, 21 February 2005 17:11

If the samples are taken at varying intervals of time - say there is error in the sampling mechanism - then the resultant samples will be interpreted as having happened at the correct times, and the data will travel through the rest of the system like this.

Yes, that's exactly what I'm saying.  And then once it is output by the DAC it would still be wrong (and possibly audibly wrong), but it would have no way of knowing that once the samples have been taken wrong initially.

Quote:

The result is that bad amplitude data will be captured, and this data will be like any other amplitude data error.

But all I'm saying (and now Paul has agreed) is that it isn't ONLY bad amplitude, it will be the wrong frequency as well.  In any case Paul seems to have understood me and identified what I think was the relevant point in my question, so maybe his post and my reply will make that a little clearer.

[Nika Aldrich]

Lee,

It is important to understand how jitter manifests itself on a signal.  I would advise reading the article on dither at Dan Lavry's site:

http://www.lavryengineering.com/white_papers/jitter.pdf

Through page 6 should be enough for the sake of this conversation.  You'll see that random errors in timing manifest themselves as merely added frequency content.  It does not CHANGE the frequency of the material but rather ADDS frequency components to it.  If the jitter is random then the added frequency content will have random components to it - the amplitude of which is based on the amplitude of the variations (+/- how many ps or ns?)

Indeed our ears are very sensitive to this type of variation on a signal, but it is not a CHANGE in frequency content - what we hear is the ADDED frequency content that is based on the frequency of the jitter.  A change in sample frequency won't change the way in which jitter manifests on in-band material.

Let me know if this makes any sense?  I'm a bit rushed right now - kid's in the bath.

Nika

[Lee Flier]

Nika, yes I understand what jitter does and what dither does... that doesn't answer my question but that's OK, go take care of your kid.

[Paul Frindle]

Lee Flier wrote on Mon, 21 February 2005 22:29

Thanks Paul, I think you're getting to the crux of my questioning. Paul Frindle wrote on Mon, 21 February 2005 16:52 You are very correct - we are in fact significantly more sensitive to this (what is essentially high speed wow and freq modulation) than people seem to think. The loss of detail and timbre in the extreme top end and loss of imaging depth is evident when this happens. Yes, I agree, as I am very sensitive to this myself. Quote:   But there is a misunderstanding that this is somehow caused by the quantisation of timing samples alone - the effect would not improve significantly even with infinite sampling rate Hmmm... like I said to Nika, I can see where this would be the case if the clock has systemic error but not in the case of random error.

Random error will produce timing inaccuracy and freq modulation just like a systemic or cyclic error. However a cyclic timing error will be more readily identifiable than a random one - but having said that a random error is still a bloody nuisance as it is just as grave but non-specific and therefore goes unnoticed more easily.

The further error caused by the quantisation of the samples is not what you will hear first. There is no real quantisation of time or phase in properly implemented sampled system - despite the way it looks when you view the samples undecoded. What you will hear is the equivalent of freq modulation - if the modulation itself is a high freq we are particularly susceptible to it (IMO) because there is no analogue of this situation in the real world - it's totally unnatural.

Quote:

And I don't necessarily think it's timing *alone* that causes these problems... all I'm saying is that ANY error that occurs in sampling would seem to be highly significant if you've only got two samples from which to reconstruct your waveform.

Again it doesn't really matter how many samples you seem to have in any given wave form - an error is an error. For instance if the data is completely the wrong magnitude value for a given time it does not matter how many times it is repeated. You should hear what happens to a DSD signal if data errors occur - 64FS doesn't save you believe me.

A timing inaccuracy will cause the same (freq modulation) result as far as you will hear - regardless of sampling rate - or even if it were analogue and not sampled at all.

Quote:

- if you could modulate the timing of analogue signals in the same way the effect would be audibly identical.

Quote:

Yes, but of course with analog signals it isn't necessary to do that. Neither is it 'necessary' to do this with digital systems!

Quote:

What is of course needed is due care taken with timing signal handling in the first place - with due regard to the much ignored fact that the clock signal IS ANALOGUE in every respect  - not only just electically, but also in its function

Quote:

Well yes... I agree that a lot of the "problems" with digitizing audio occur on the analog side... but that seems more than a little irrelevant to me considering you can't separate the two from each other!  Sometimes digital theorists seem to be saying "Digital audio would be perfect if only we could eliminate that pesky analog stage - that's the problem!" but you can't, and why would you really want to?  And if you have to twist the analog signal into a horrible pretzel in order to digitize it, then it's little comfort that the digital domain "faithfully" captures and reproduces all of these horrors.  I'm exaggerating of course, but there's a point in there somewhere.

Yes the point would be that digital presentations and storage can give you the promise of total security and repeatability but the interface with the real world is necessarily and forcibly analogue (both practically and philosophically). One such often overlooked signal is the sync which, although looks like squarewaves, is actually an analogue timing signal - which is terribly important to the whole process. This is a physical issue that would beset any system capable of changing speed this quickly - and the solution to it has a physical analogue in the real world. Imagine having a remote varispeed connected to a hypothetical analogue tape machine that could vary it's speed at a rate faster than microseconds! Then imagine what it would sound like if the wire connecting the remote to the machine started picking up noise from all around! Thats pretty much the equivalent of many digital set ups.
The answer of course would be to slow down the response of the machine's ability to change speed - a fly wheel if you will. That system in a digital set-up is called a phase locked loop  - that should be installed WITHIN your converters, but often isn't or is present but lacking in the required accuracy itself.

In some very sensitive (badly designed) systems just picking up the sync cable in your hand is enough to change the sound of your programme - and it won't matter how much you may have coughed up for that expensive sync generator sitting only a metre away

[Johnny B]

Paul mentions a good point above, you would think that some even multiple of 8 would have been used early on such as the 64 he mentions. 44.1 seems like an invitation to more math problems, but I'm sure there was a very good argument advanced for it at the time. Hindsight is always 20-20!

What will it take to get the timing right?

I'd like to hear a few frank opinions, so how about providing a few examples of some well-designed systems and some not so well-designed systems where things like just grasping the cable can induce noise?

And how about some examples of the best chips?

In fact, who is making the best chips this week?

[Paul Frindle]

Johnny B wrote on Tue, 22 February 2005 07:49

Paul mentions a good point above, you would think that some even multiple of 8 would have been used early on such as the 64 he mentions. 44.1 seems like an invitation to more math problems, but I'm sure there was a very good argument advanced for it at the time. Hindsight is always 20-20! What will it take to get the timing right? I'd like to hear a few frank opinions, so how about providing a few examples of some well-designed systems and some not so well-designed systems where things like just grasping the cable can induce noise? And how about some examples of the best chips? In fact, who is making the best chips this week? Grasping the cable isn't even necessarily inducing noise - it's changing what's already there. At the point where messing with cables changes anything the system is already compromised. There are two absolutely equal and fundamental factors in sampling: a) What is the value of the signal? b) At what relative time was this value present? Both are equally important to obtaining and reproducing the programme. Now we talk endlessly about 'resolution', noise and distortion which are value related issues, but the timing is just as significant and you cannot find a single spec on your converters that refers to this. If your converters lack proper clock recovery/regeneration that apparently diminuitive bit of cable is as sensitive to disturbance as your whole audio connection regime and has just as much bearing on the quality of your overall system. You can begin to see how popular perceptions are 'coloured' by hype? Why doesn't our precious AES come up with some guideline specs and standards relating to timing integrity - so that you could make informed choices in the first place? Rather than bogging ourselves down with the latest 'buzz-tech' whilst real problem remain right under our noses?

[Johnny B]

Paul,

I do like to try and keep my eye on the "buzz tech" because you never know if some kind of breakthrough is about to emerge.

And yes, the AES should have some published standards, so should the EBU and the IEEE.



 

[rsdio]

I understand the conceptual viewpoint of treating DSD as a 1-bit PCM stream with dithering.
I have not read this entire thread, so perhaps the following has already been mentioned: There is a paper which basically proves that dithering is not fully realizable in a 1-bit system because there are only two codes. Sure, you can compute a process that is similar to dither, but you need at least a 2-bit code system before dither becomes anywhere near as effective as it should be.
Sorry I don't have a link, but my laptop got stolen, along with my bookmarks and PDF downloads. Since there seem to be a number of well-versed experts contributing to this thread, I would assume that at least one of you has read the mathematical proof. Hey, maybe one of you wrote it!

[lambda]

rsdio wrote on Sat, 16 April 2005 14:05

There is a paper which basically proves that dithering is not fully realizable in a 1-bit system because there are only two codes. Sure, you can compute a process that is similar to dither, but you need at least a 2-bit code system before dither becomes anywhere near as effective as it should be. Sorry I don't have a link, but my laptop got stolen, along with my bookmarks and PDF downloads. Since there seem to be a number of well-versed experts contributing to this thread, I would assume that at least one of you has read the mathematical proof. Hey, maybe one of you wrote it!

http://sjeng.org/ftp/SACD.pdf

That is the link. The two people in the link are actually from the university I attend, but in the wrong department (I am in Pure Mathematics).

[Bob Olhsson]

Paul Frindle wrote on Tue, 22 February 2005 03:41

... Why doesn't our precious AES come up with some guideline specs and standards relating to timing integrity - so that you could make informed choices in the first place? ...

Because it's funded by audio manufacturers. If everybody could easily buy something really good, they would start worrying about more important issues than buying trendy audio gear.