Jitter test

[Julf]

I don't wish to get into a discussion of this - it has already been covered by Amir here

Unfortunately that thread still doesn't address the audibility of jitter at all. Even Amir's piece that starts that thread just states "Invariably, by the time I get to this point of the argument with someone, the conversation turns into “yes but… is it audible?” As unfair as it might be, I am going to punt that question."

As to the discussion about the theory, if you disagree with the "simplistic examples", I am sure we would all welcome your views on what is wrong with them.

 

[jkeny]

Julf,
I'm not interested in your line of argument so I will desist from discussion with you. When you ever move beyond your simplistic views of jitter it may be worth engaging but otherwise its a waste of time.

Vincent, this Archimago guy is a self-confessed obsessive so why not tell him about the Jim LeSurf IQ-test which will be a better tool for investigating what he is trying to investigate - the possible effect of OS load on audio playback.

 

[Julf]

I'm not interested in your line of argument so I will desist from discussion with you. When you ever move beyond your simplistic views of jitter it may be worth engaging but otherwise its a waste of time.

Thank you - appreciate your factual and helpful approach.

 

[opus111]

This simplistic discussion of jitter & simplistic examples (started in that article with the so called good perspective) are so misleading as to be verging on disingenuous.

The disingenuity is palpably obvious when you get to the end where he claims 'due to space' for omitting very important stuff like what kind of DAC is being employed. Yet he had space enough to pontificate about total irrelevancies like:

Your ear drum is 30 to 120 um thick.

and outright falsehoods like:

If you move your head by the thickness of 2.5 atoms of gold then you are introducing 1 ps of jitter!

 

[amirm]

Unfortunately that thread still doesn't address the audibility of jitter at all. Even Amir's piece that starts that thread just states "Invariably, by the time I get to this point of the argument with someone, the conversation turns into “yes but… is it audible?” As unfair as it might be, I am going to punt that question."

In that article I wanted to introduce the concept and hence that comment. Since Jitter can have infinite scope, it is kind of hard to characterize it. We can however get our arms around a single tone jitter and its frequency in this excellent summary from Julian Dunn Paper and Prism Sound Presentation of the same:

So as we see, the threshold for low frequency jitter can indeed be very high. This is because masking is strongest around our main tone. Since jitter creates sideband distortion spikes around our music at its frequency, then the lower the jitter frequency, the lower the odds of hearing it. At the other extreme, when jitter frequency becomes high, then it separates from the music signal and is no longer masked. That makes its audibility simply subject to what kind of dynamic range we want to achieve. Assuming 120 db, we get the above chart as we put all of this together.

 

[Julf]

At the other extreme, when jitter frequency becomes high, then it separates from the music signal and is no longer masked. That makes its audibility simply subject to what kind of dynamic range we want to achieve. Assuming 120 db, we get the above chart as we put all of this together.

Thanks - that is a very interesting chart. Is it based on actual listening tests, or theoretical/mathematical calculations?

 

[amirm]

Thanks - that is a very interesting chart. Is it based on actual listening tests, or theoretical/mathematical calculations?

It is both . The principal of masking is based on listening tests. That empirical data is then used as the foundation of determining whether the jitter distortion spikes rise up above that curve. For the later part of the graph (straight line), it is simple math of attempting to keep the distortions below 120 dB. That is also experimentally determined by examining the highest dynamic range we would want to reproduce. See http://www.madronadigital.com/Library/RoomDynamicRange.html.

 

[Julf]

It is both . The principal of masking is based on listening tests. That empirical data is then used as the foundation of determining whether the jitter distortion spikes rise up above that curve.

Let me see if I understand what you are saying. So it is based on a listening-test-derived masking curve (that is not specific to jitter, but covers the audibility of any "additional" signal such as noise, distortion or intermodulation products), and a calculation of how much jitter is needed to reach that level?

For the later part of the graph (straight line), it is simple math of attempting to keep the distortions below 120 dB.

Just to be sure - are you saying that jitter products that are down at -119 dBFS are audible?

 

[amirm]

Let me see if I understand what you are saying. So it is based on a listening-test-derived masking curve (that is not specific to jitter, but covers the audibility of any "additional" signal such as noise, distortion or intermodulation products), and a calculation of how much jitter is needed to reach that level?

Correct. Here is a good visualization:

The shoulders of that graph are determined using listening tests and confirmed by physiology of our ear/hearing system. The bar in blue therefore is considered to be inaudible (relative to the average population hearing threshold).

Just to be sure - are you saying that jitter products that are down at -119 dBFS are audible?

Assuming you set your listening level to 0 dbfs, then -119 is just at the threshold of quietest rooms we have and above absolute hearing threshold.

 

[Julf]

Correct.

Thanks! Appreciated!

Assuming you set your listening level to 0 dbfs, then -119 is just at the threshold of quietest rooms we have and above absolute hearing threshold.

But wouldn't applying that model to "ordinary" IM distortion (that is non-harmonic, just like jitter) imply that IM distortion would be audible down at a level of 0.0001%?

 

[amirm]

But wouldn't applying that model to "ordinary" IM distortion (that is non-harmonic, just like jitter) imply that IM distortion would be audible down at a level of 0.0001%?

No in the general sense because that would violate masking. Only when masking is done away with, i.e. when the distortion is way away from the source signal, can we make that judgment.

All of these single figure distortion metrics are faulty in that they do not take into account psychoacoustics. Doing so means analyzing things based on frequency and relationship to the source. THD has the same problem in how it puts the same weight on the distortion products.

The other way to look at this is that we are trying to set the threshold of inaudibility. That is, what number can we pick where we can demonstrate we can't hear the distortion. To do that we need to be past provable thresholds. Inverted tests that rely on listening tests have the problem that we don't know if we excited the system with worst case signal. The mathematical data based on threshold listening tests don't have this problem. So in some sense, we may be overstating the fidelity we need. That is fine with me. As an engineer, my job is to design a system where I can verify if I achieved an objective goal.

 

[Julf]

As an engineer, my job is to design a system where I can verify if I achieved an objective goal.

Totally agree.

 

[Julf]

The other way to look at this is that we are trying to set the threshold of inaudibility. That is, what number can we pick where we can demonstrate we can't hear the distortion. To do that we need to be past provable thresholds. Inverted tests that rely on listening tests have the problem that we don't know if we excited the system with worst case signal. The mathematical data based on threshold listening tests don't have this problem. So in some sense, we may be overstating the fidelity we need. That is fine with me. As an engineer, my job is to design a system where I can verify if I achieved an objective goal.

Right. The threshold analysis helps provide a lower limit - a limit we can use to say "well, as long as we stay below these values, jitter should definitely be inaudible". They might still be much lower than the actual, practical limits of audibility (as we know, studies such as the 2005 Ashihara study puts the limit as high as 250 ns), but until we have better information, the threshold analysis-derived figures are of course a prudent target. Of course, specifying just the jitter amplitude without addressing it's spectral behaviour is also a rather simplistic approach.

 

[amirm]

Right. The threshold analysis helps provide a lower limit - a limit we can use to say "well, as long as we stay below these values, jitter should definitely be inaudible". They might still be much lower than the actual, practical limits of audibility (as we know, studies such as the 2005 Ashihara study puts the limit as high as 250 ns), but until we have better information, the threshold analysis-derived figures are of course a prudent target. Of course, specifying just the jitter amplitude without addressing it's spectral behaviour is also a rather simplistic approach.

The Ashihara study unfortunately is not instructive for the reason you mention at the end. He tested random jitter which mathematically translates into random noise. Random noise just increases our, well, noise floor . So he tested the sensitivity of different noise levels, not jitter. The bothersome jitter is one where distortion spikes are created and for that, they need to be non-random. Even the graph I gave you is not as good as we want as jitter is rarely a single tone for us to plug in there.

 

[Julf]

The Ashihara study unfortunately is not instructive for the reason you mention at the end.

Well, I guess it is a reasonable study into the audibility of the noise generated by random jitter

Even the graph I gave you is not as good as we want as jitter is rarely a single tone for us to plug in there.

True, but non-random jitter probably only has a couple of dominating frequencies, so in that sense it should behave like a few single-tone jitter patterns.

 

[jkeny]

Well, I guess it is a reasonable study into the audibility of the noise generated by random jitter

I applaud Amir for his tenacity in giving you some necessary lessons on jitter. Why, in one post quote the Ashihara study as representative of the audibility of ALL jitter & when corrected you still persist in trying to justify it? I really hoped that you might want to get to the truth & learn something from Amir's patience & concise examples.

True, but non-random jitter probably only has a couple of dominating frequencies, so in that sense it should behave like a few single-tone jitter patterns.

Your use of "probably" reveals that you are guessing. Can you give any evidence for your guess?

 

[Julf]

I applaud Amir for his tenacity in giving you some necessary lessons on jitter. Why, in one post quote the Ashihara study as representative of the audibility of ALL jitter & when corrected you still persist in trying to justify it?

I guess you missed the smiley at the end of my sentence...

I really hoped that you might want to get to the truth & learn something from Amir's patience & concise examples.

I definitely appreciate Amir's patience and willingness to discuss the issues.

Your use of "probably" reveals that you are guessing. Can you give any evidence for your guess?

I think "speculating" is a more appropriate word than "guessing". My speculation is based on my assumption of the result of a convolution of two non-random signals.

 

[jkeny]

I guess you missed the smiley at the end of my sentence...

Nope didnt miss it

I definitely appreciate Amir's patience and willingness to discuss the issues.

I would say he is doing more than discussing the issues

I think "speculating" is a more appropriate word than "guessing". My speculation is based on my assumption of the result of a convolution of two non-random signals.

And you think that "jitter probably only has a couple of dominating frequencies" covers the issue with regard to the playback of music signal?

 

[DonH50]

I would assume deterministic jitter has spurs at all products of the signal and clock, as well as all aliased (ADC) and image (DAC) frequencies...

p.s. Amir -- Do you have the loudness contours in numerical format? I have only seen plots and would like to piddle with some things but need to enter the plot data. And am lazy.

 

[amirm]

No I don't Don. I have had to recreate them before but wound up doing it with Photoshop and scaling manually!