Close in jitter?

[Blumlein 88]

I was going to ask why some consider close in jitter to be a big problem before another thread was closed. If Thomas thinks this is inappropriate then just remove this one.

So why the idea close in jitter is a big problem? As close in jitter manifests like other jitter, and due to masking the results of it would seem likely to be a lesser problem. Yet some are claiming it is what disturbs stereo imaging. What is the thinking behind this? Is there some rationale behind it or just some conjecture born of sighted listening?

Here are a couple plots of a quarter sample rate tone of one of my DACs. 1.1 hz from the central tone this signal is already down about 80 db.

Here in a wider 200 hz wide plot of the same signal we see by a 20 hz offset it is below -100 db and below -120 db at a 40 hz offset.

 

[amirm]

Audibility of jitter has two parts:
1) the amount of it
2) Its spectrum

Vast majority of close-in jitter is random and hence manifests itself as simple increase of noise floor. It is indistinguishable from noise (of similar spectrum) in that regard. That's important because the rise of noise does not at all yield to the common misconception of timing error, stereo imaging problems, etc.

It is also massively masked as I post in the other thread.

 

[Mivera]

Something you might want to read:

https://www.by-rutgers.nl/PDFiles/clock_jitter_spec-1.pdf

 

[Mivera]

The J-test was developed for "interface jitter" when using AES/EBU or SPDIF interfaces, not clock jitter. The testing isn't even accurate if you're using a USB interface, or other type of built in transport.

Refer to the AES whitepaper I shared to learn about the difference between clock jitter and interface jitter.

 

[Mivera]

Audibility of jitter has two parts:
1) the amount of it
2) Its spectrum

Vast majority of close-in jitter is random and hence manifests itself as simple increase of noise floor. It is indistinguishable from noise (of similar spectrum) in that regard. That's important because the rise of noise does not at all yield to the common misconception of timing error, stereo imaging problems, etc.

It is also massively masked as I post in the other thread.

I guess you know more about jitter than Bruno Putzeys. Clearly he sounds clueless based on reading that AES published whitepaper.

"Only signal-related components are present: the audio signal and its HF aliases. There is no other noise present, so SNR is not affected by jitter"

 

[Jakob1863]

Julian Dunn, in his 1991 paper, that Amirm mentioned in the other thread, based his threshold curve indeed on the perception of sidebands introduced if jitter effects modulated a fundamental. Therefore he concluded that due to masking jitter of low frequeny would be inaudible unless very high jitter amplitudes were present.
But it is based on single tones (signal) and jitter of a single frequency, which is very different from reality.

The interesting part was, that the people at grimm, afair noticed nevertheless an audible difference if using their CC1 masterclock together with different ADCs or DACs, although the measured jitter related difference was below 200 Hz. Katz even conducted a controlled listening test for the Tape Op magazine and the result was a reliable but hard to detect audible difference, when used together with some configurations.

I don´t know if they put such an effort in it to figure out, if the reason was indeed due to low frequency jitter (even in the wander region) or if other possible effects could play a role (exchange of external clock units might introduce EMC related differences)

 

[Mivera]

Julian Dunn, in his 1991 paper, that Amirm mentioned in the other thread, based his threshold curve indeed on the perception of sidebands introduced if jitter effects modulated a fundamental. Therefore he concluded that due to masking jitter of low frequeny would be inaudible unless very high jitter amplitudes were present.
But it is based on single tones (signal) and jitter of a single frequency, which is very different from reality.

The interesting part was, that the people at grimm, afair noticed nevertheless an audible difference if using their CC1 masterclock together with different ADCs or DACs, although the measured jitter related difference was below 200 Hz. Katz even conducted a controlled listening test for the Tape Op magazine and the result was a reliable but hard to detect audible difference, when used together with some configurations.

I don´t know if they put such an effort in it to figure out, if the reason was indeed due to low frequency jitter (even in the wander region) or if other possible effects could play a role (exchange of external clock units might introduce EMC related differences)

Yes and all this stuff happened years ago now. Today it's well established in the industry how important ultra low phase noise is for DAC's. This is why all of the best DAC's use ultra low phase noise clocks. Even what's considered run of the mill today, the Crystek 575, still has 82 femtoseconds of jitter. However as Bruno's whitepaper says, we need to pay attention to "close in phase" not full spectrum jitter, or high frequency phase noise. . I think Herbert Rutgers (Who builds clocks for Grimm and Mola Mola) Explains what matters best in this whitepaper:

https://www.by-rutgers.nl/PDFiles/Audio Jitter.pdf

and here's the clock the Grimm CC1, and LS1 active speaker clocks are based on:


https://www.by-rutgers.nl/rutgerS-Clock.html

 

[Blumlein 88]

Henk ten Pierick developped a measuring method with which he could rank tubes, tube-amps, solid state amps, op-amps, name it, on sound quality. He is very mysterious about it but with a number of experienced listeners we agreed with his ranking, every time!

So the above is referenced several times in articles related to your links. Another mysterious special measurement method that was kept from the public. It works though just trust us.

Let me see I remember someone elsewhere stringing people along about a special group of test signals that correlated exactly with the listening evaluation of audiophiles. You just couldn't get straight answers on it. Hints they were prime frequencies and spaced in a special way etc. etc. etc. Any attempt to get what the super special test signal was always ended in no answer just that whatever you were proposing didn't get at the heart of the matter. Which is hard to argue with since you couldn't find out what it was.

So what is special about the 100 hz and lower level 5 khz signal that is so illuminating about this close in jitter?

 

[Mivera]

Like Bruno said in his AES published white paper:

"Where does the carrier end and where does jitter begin? The question is nearly a philosophical one that only listening tests can settle. For now, we should content ourselves with the measurement capability of the ana- lyser, ie. the width of the notch placed at fmax."

Sometimes you just need a good set of ears. If you don't have a good set of ears, don't concern yourself with it. I can assure you Stevie Wonder doesn't lose any sleep over the quality of his sunglasses tint.

 

[amirm]

Sometimes you just need a good set of ears.

Do you have such ears?

 

[oivavoi]

Is there anybody who has been able to ABX differences between low jitter levels and super-low jitter levels - i.e. differences that can be detectable in a blind test?

EDIT: I realize that "low" and "super-low" are not very clear terms. I guess what I'm wondering basically is what jitter levels one should aim for when buying gear.

 

[amirm]

Something you might want to read:

https://www.by-rutgers.nl/PDFiles/clock_jitter_spec-1.pdf

I read it. It is a pretty simple paper that for its introduction actually follows the same mathematical simplification of sinusoidal jitter as in Julian Dunn paper. Indeed Julian's paper is the only reference:

I have also post his conclusions. As you see there, he is railing against industry one-number phase specs which is the same thing I explained at the outset of this thread. That is, jitter specs are meaningless with respect to audibility. You must have their spectrum analyzed so that we can then use psychoacoustics to determine audibility. His solution of using a different one-number score still misses that completely. You must show the spectrum as I and John Atkinson show in DAC reviews.

Using the above method is of course a shortcut instead of conducting formal listening tests. Such tests are important but not ever done. So we are stuck with interpreting measurements.

He also makes a gratuitous comment about audibility:

First of all, we don't perform audio jitter measurements as he is assuming, i.e. getting a number. Instead, we perform a full spectrum analysis so there is no one "50 ps" jitter number in our work. Here is an example of what I have posted on budget DACs:

Looking at Signstek measurements, we see all of those distortion products. We can analyze them per above. I don't rely or ever spec a single number jitter.

Single number jitter specs are used in IC (component) marketing literature which is what his beef is about. Not this type of spectrum analysis.

Back to his quote, there is no data presented regarding audibility of "50 ps" anything. Yes, we need to do a spectrum analysis to say anything about audibility. But this road goes both ways. You can't say it is audible just as well as you can't say it is not. You do the spectrum analysis and then you know.

Convention papers are not peer reviewed so even though they are usually pretty good, improper statements like this are not filtered. Had this been a Journal paper, I am pretty sure they would have asked him to back this with a listening test or take it out.

Anyway, it is a good paper talking about poor industry practices with regards to spec'ing DAC clock oscillator ratings. But has no reading on our audio measurements with respect to jitter being wrong. Or providing any new insight into what is or is not audible.

 

[amirm]

Is there anybody who has been able to ABX differences between low jitter levels and super-low jitter levels - i.e. differences that can be detectable in a blind test?

Yes. I have run such blind test with low frequency jitter -- exactly the case that Mike makes. I have a set of files that have jitter embedded in them at different levels: something like -20, -40 and -60. The last one, despite having jitter at far, far higher levels than any DAC out there, was incredibly difficult to distinguish. Low frequency jitter gets masked easily as I mentioned.

I have to run but will post more on this later.

 

[oivavoi]

Yes. I have run such blind test with low frequency jitter -- exactly the case that Mike makes. I have a set of files that have jitter embedded in them at different levels: something like -20, -40 and -60. The last one, despite having jitter at far, far higher levels than any DAC out there, was incredibly difficult to distinguish. Low frequency jitter gets masked easily as I mentioned.

I have to run but will post more on this later.

Thanks Amir!

 

[Mivera]

2 things are required for proper testing:

1: A great set of ears

2: Proper equipment.

Without those 2 items, you are just spinning your wheels and fooling yourself.

 

[oivavoi]

2 things are required for proper testing:

1: A great set of ears

2: Proper equipment.

Without those 2 items, you are just spinning your wheels and fooling yourself.

I agree. And I don't think blind testing is the answer to everything in hifi, as I've written in other threads. But still, there has to be a limit for when measurable improvements in electronics stop to make any audible impact. Whether an amplifier has a THD number of 0,002 or 0,003... I don't think that matters so much. So what I'm wondering is what audible SOTA is when it comes to jitter. If not a single person ever has been able to ABX jitter levels under a certain level for example, then there's not so much point in me pursuing jitter levels that are even lower.

 

[DonH56]

Good grief. Again.

Excess random jitter degrades SNR. The impact is readily calculated and measured. Whether any halfway decent design these days has enough RJ to be audible is another question. The other thread used 1 ps, which would turn your ideal 24-bit DAC into a lousy 23-bit DAC. I'd bet the SNR is nowhere near that of an ideal 24-bit converter, let alone the ideal SFDR.

Correlated jitter, which can take many forms, and other correlated spurs can cause distortion more readily heard than random jitter.

Close-in phase noise on a clock results in lower-frequency spurs when a signal is demodulated. In this case, those close-in components are what end up in the audio band. Clock sources typically flat-line well outside the audio band.

An additional consideration is that many high-frequencies devices have poor noise performance at audio frequencies. I have used microwave transistors that are great at 10 GHz but have a 1/f corner of 100 MHz. Not good for audio... MOSFETs have significant LF noise and so either CDS techniques (correlated double samplers) or JFETs (or BJTs) are used for audio circuits. BJTs have higher current input noise. There are always trades.

 

[Cosmik]

Excess random jitter degrades SNR.

But I appreciate what Mivera was saying earlier on. It isn't SNR as conventionally known where a 40dB SNR would be clearly audible 'in the quiet bits'. With jitter, the SNR is constant, no matter how quiet your signal. The noise is constantly masked by a much larger signal and disappears to zero in 'the quiet bits'.

 

[oivavoi]

I just realized that there was a whole long thread prior to this. Missed that. Will read up on that before I ask any further questions here.

 

[Mivera]

Jitter in digital audio is a complex issue. If you truly want to understand it, seek the advice of folks who build the very best DAC's on the market. Not self proclaimed experts on forums who have no whitepapers published, or even built any audio gear on their own.

Personally if I wanted to know more about how a BMW ECU functions, I would contact the BMW engineers who specialize in that department. I wouldn't contact some guy on the internet who reads Car and Driver on a regular basis, and attended a few marketing seminars.