Jitter test

[amirm]

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.

When people ask me what the profile of jitter is, my answer is "anything it wants to be" . I don't want to steal the thunder from my upcoming WSR magazine article on Jitter in AV processors but here is a sneak peak of an AVR measurement:

In an ideal case, there would be a single spike in the middle representing the test signal and nothing else. Instead, you see two graphs each with different jitter spectrums. The one in yellow has correlated jitter at specific frequencies but also has a ton of random low jitter that have served to broaden the bottom of the spike. The one in pink doesn't have the random low frequency jitter, and has different correlated jitter frequencies at the cursor position (+- 250 Hz).

Now here is the interesting part: both of these measurements are from the identical AVR! I simply repeated the measurement and would get one or the other graph!!! Clearly the activities of the internal processor in the AVR were bleeding into the DAC clock. And depending on what that processor/processing was doing, we would get different distortions introduced into the DAC and eventually out of its analog output which this is measuring.

How do we go about evaluating the audibility of this beast? We could say a lot of this is low frequency and hence inaudible but who is to say the thing doesn't do something else a few moments from now? Or if I changed some other processing in it? This is from a $1,000 premium AVR by the way.

These are over HDMI by the way where I found far, far more problems than S/PDIF.

 

[amirm]

BTW, the above is one of the reasons one can't 100% trust the article that started this thread. Plug that device into a different PC and the results could change. Indeed in my tests of processors, I found characteristics of my source PC which was generating the signal in a number of different units under test. Surely if I changed to a different source then, the signature of distortion/noise would change.

 

[DonH50]

Thanks for that, Amir, looking forward to your article!

Coupla' questions:

1. Is the jitter in either test case audible? Or can you hear it one case and not the other?
2. Are the +/-250 Hz spurs within our masking range?

I don't have an answer for either, am curious about your thoughts. - Don

 

[amirm]

The article and my testing was purely based on measurements. I didn't do any listening tests. I find that listening tests in the case of HDMI is impossible to do. Switching sources causes multiple seconds of muting, locking and unlocking of sound which totally disrupts the ability to hear small differences. I like to see switching times in milliseconds to hear such differences.

 

[amirm]

And yes, it is within masking. My eyeballing of the Dunn chart shows our threshold for 250 Hz to be 10 nanonseconds which translates into -73 dbFS. I am pretty far short of that in the pink graph at -105 dbFS.

 

[NorthStar]

The article and my testing was purely based on measurements. I didn't do any listening tests. I find that listening tests in the case of HDMI is impossible to do. Switching sources causes multiple seconds of muting, locking and unlocking of sound which totally disrupts the ability to hear small differences. I like to see switching times in milliseconds to hear such differences.

---- ...A very important note there Amir.

 

[opus111]

In an ideal case, there would be a single spike in the middle representing the test signal and nothing else. Instead, you see two graphs each with different jitter spectrums. The one in yellow has correlated jitter at specific frequencies but also has a ton of random low jitter that have served to broaden the bottom of the spike. The one in pink doesn't have the random low frequency jitter, and has different correlated jitter frequencies at the cursor position (+- 250 Hz).

What's the analyser's effective bandwidth here? I'd like to make sense of that noise 'floor' - does look to be changing (yellow eyeballing slightly higher than magenta) and if that's an FFT the processing gain could be 30-40dB meaning the broadband noise certainly won't be masked.

 

[Julf]

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

With a real music signal you will definitely get a complex spectrum, but that is why we also measure normal harmonic distortion and IM with test tomes, not real music.

As stated before, the Dunn charts are based on a mathematical model. As with any other theory, the proof is in the pudding - that model needs to be verified by controlled and repeatable observations (listening tests). It is unfortunate that there doesn't seem to be that much actual data out there - so studies like the Ashihara one, while having serious flaws, is still a step in the right direction.

Out of curiosity, what criteria do you yourself use for audibility of jitter?

 

[Julf]

The article and my testing was purely based on measurements. I didn't do any listening tests. I find that listening tests in the case of HDMI is impossible to do. Switching sources causes multiple seconds of muting, locking and unlocking of sound which totally disrupts the ability to hear small differences. I like to see switching times in milliseconds to hear such differences.

There seems to be "smart" HDMI switches that keep up the handshake with the source to avoid the switching delay, but I haven't had a chance to test one yet.

 

[jkeny]

As stated before, the Dunn charts are based on a mathematical model.

Grounded in listening tests on masking effect.

As with any other theory, the proof is in the pudding - that model needs to be verified by controlled and repeatable observations (listening tests).

Did you not understand it the first time it was explained to you - it is the the outer limit of jitter audibility at the various frequencies for random jitter.

It is unfortunate that there doesn't seem to be that much actual data out there - so studies like the Ashihara one, while having serious flaws, is still a step in the right direction.

Not if used like you have done to put a single figure of 250nS on jitter audibility without qualifying this. In fact it is so meaningless that I would not consider it a step in the right direction, whatsoever.

Out of curiosity, what criteria do you yourself use for audibility of jitter?

I usually use my ears for testing audibility

 

[Julf]

Grounded in listening tests on masking effect.

Yes. Masking curves based on listening tests applied to a mathematical model of jitter. So still theoretical, and still needs to be verified with actual real-life experiments/tests.

Did you not understand it the first time it was explained to you - it is the the outer limit of jitter audibility at the various frequencies for random jitter.

Did you not notice that I actually wrote pretty much exactly that:

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.

This might be the point where I have to quote your earlier posting:

I'm not interested in your line of argument so I will desist from discussion with you.

 

[jkeny]

This might be the point where I have to quote your earlier posting:

Yes, I thought your lessons in jitter may have taught you something of value but I see you are pretty much blinded by your world view so your reminder is timely, thanks - I will revert to not discussing with you!

 

[amirm]

There seems to be "smart" HDMI switches that keep up the handshake with the source to avoid the switching delay, but I haven't had a chance to test one yet.

There are but then I won't be measuring the HDMI on the unit I want to test . The comparison I want to do is comparing S/PDIF against HDMI in the same unit. Since S/PDIF has lower jitter then it makes a good reference here. That would necessitate having instant switching in the same processor/AVR which I have yet to find.

 

[amirm]

Yes, I thought your lessons in jitter may have taught you something of value but I see you are pretty much blinded by your world view so your reminder is timely, thanks - I will revert to not discussing with you!

Let's be cordial to each other please . More comes out of polite disagreement than ignoring/insulting each other.

 

[amirm]

What's the analyser's effective bandwidth here? I'd like to make sense of that noise 'floor' - does look to be changing (yellow eyeballing slightly higher than magenta) and if that's an FFT the processing gain could be 30-40dB meaning the broadband noise certainly won't be masked.

I plan to do that noise compensation at some point but since you are volunteering , the sampling rate was 192 Khz and 65K FFT for a delta(f) of 2.9 Hz.

 

[Julf]

There are but then I won't be measuring the HDMI on the unit I want to test . The comparison I want to do is comparing S/PDIF against HDMI in the same unit. Since S/PDIF has lower jitter then it makes a good reference here. That would necessitate having instant switching in the same processor/AVR which I have yet to find.

Hmm. Yes, that is a tricky issue. I am glad all my applications have been pure audio, so I have managed to avoid HDMI so far, but it is worrying how many of the latest media processors seem to be HDMI-only.

 

[opus111]

I plan to do that noise compensation at some point but since you are volunteering , the sampling rate was 192 Khz and 65K FFT for a delta(f) of 2.9 Hz.

I make that a processing gain of 45dB - a -120dB line across the band translates to -75dB in the full bandwidth assuming the windowing function hasn't contributed significantly.