[nick_charles]

   
In the case of the D/A conversion process, I can see how a shift in individual sample timing due to DAC clock jitter will slightly alter the analog waveform. The amount of deviation highly depends on the amount of jitter and the sampling technology used. This is the point I am trying to convey - namely that there are multiple ways to obtain jitter numbers. The A-B listening tests need to explicit in ways that jitter is being measured.

 
All three of the most important papers in this area in fact did this describing the method of creating, measuring and characterizing the jitter types, frequencies and amplitudes, most include graphs showing the distortion created by jitter of different types etc, even the Stereophile measurements show the levels of distortion and/or broadband noise. You are not limited to looking at the numbers you can see the effect graphically, while a ns or even ps number is somewhat abstract the jitter sidebands or broadband noise effect can be easily plotted and you can see how far below the fundamentals the distortion is. 

 

[stv014]

Still no interest in trying some actual experiments (remember, you can post your own samples, and suggest jitter parameters) ?

 

[Digitalchkn]

   
All three of the most important papers in this area in fact did this describing the method of creating, measuring and characterizing the jitter types, frequencies and amplitudes, most include graphs showing the distortion created by jitter of different types etc, even the Stereophile measurements show the levels of distortion and/or broadband noise. You are not limited to looking at the numbers you can see the effect graphically, while a ns or even ps number is somewhat abstract the jitter sidebands or broadband noise effect can be easily plotted and you can see how far below the fundamentals the distortion is.

 
Looking at the Stereophile article the immediate issue that I see is the measurement jitter as an RMS value. This heavily implies jitter content is purely random. Further hints at the terminology "white jitter" implies a gaussian distribution. I didn't catch any mention of deterministic elements such as those that might be cause by supply noise, etc. Needless to say, I have observed that many clock sources in practice do not exhibit white jitter spectrum at very low (audio) frequencies. This is readily seen looking at phase noise plots of various off-the-shelf oscillators.
 
I might have missed this other point, but there doesn't seem to be any discussion about the impact of jitter in a recovered clock running though a clock recover unit (e.g. a PLL). The dynamics of those devices enforce an attenuating/amplifying jitter transfer function. It seems this article essentially assumed that the sampling clock is essentially that encountered on the interface. Of course, I assume with today's technology this may not be relevant since it is fairly trivial to recover the jittered clock/data and place it onto a low jitter clock domain.

 

[nick_charles]

   
Looking at the Stereophile article the immediate issue that I see is the measurement jitter as an RMS value. 

 
Some peak-to-peak graphs of the effect of sine and random jitter from an AES conference paper from Keith Howard (2006)
 
 
 

 

 
 
How was this achieved......to gauge the effect of jitter on flute music he applied both sine and random jitter to a music segment that had a fundamental at 1500hz he used a 2K jitter signal , for the sine wave he created sidebands at -44db on signal which seems pretty bad , but to get this the jitter had to be at 1000ns peak to peak. 
 
From the same paper
 

 

[Digitalchkn]

   
Some peak-to-peak graphs of the effect of sine and random jitter from an AES conference paper from Keith Howard (2006)
 
 
How was this achieved......to gauge the effect of jitter on flute music he applied both sine and random jitter to a music segment that had a fundamental at 1500hz he used a 2K jitter signal , for the sine wave he created sidebands at -44db on signal which seems pretty bad , but to get this the jitter had to be at 1000ns peak to peak.
 
 
 

These look to be a rather extreme example. A relatively low-cost modern oscillator can easily achieve far better than Dj=1000ns pp. I should point out again that p-p jitter must be specified over a window time(# of clock cycles) as it invariably includes a random component.
 
The metric that is frequently used by oscillator manufacturers is phase noise, such as a curve shown below. This plot does describe cycle-to-cycle or periodic jitter directly, but is closer related to the clock phase error that we seem to be of most interest. There is a straightforward relationship that relates phase noise to RMS jitter within a frequency band (assuming gaussian distribution and no deterministic jitter component).
 
Note that in this example, the jitter is:
1. Not entirely "white". Low frequency jitter dominates the jitter numbers
2. There are no spurs, indicating lack of deterministic jitter
 

 

[nick_charles]

Bump so that UltMusicSnob can get in on the act and try the taste test

 

[UltMusicSnob]

I went through all 8 pages and found Nick's referent and stv014's treatments, thanks to both for providing the materials.
Jitter is something I've thought very little about, having assumed that since we can build digital clocks in the GHz range and successfully make motherboards and CPU's work, clock period variances shouldn't be something we have to worry about.
I'll plug these files (the 'Range' sample) into foobar--we're talking about uncorrelated jitter that should manifest as noise, yes? Not sure where to go with that since I listen for standard musician stuff like timbre, reverb, etc., but I'll give it a listen. Might have to ask for a classical sample later, but no need to go there yet.
 
Thanks for the bump, Nick.

 

[UltMusicSnob]

   
I was not a great track to choose (The Damned) perhaps, single pure tones are always better, but nobody else was volunteering. I have about 800 classical CDs so if somebody has some better suggestions I will see if I have a copy to upload a segment from, Bach organ ? Sustained Piano ? Brass ? Since my high freq hearing is pretty knackered I cannot from ear come up with a good candidate. But I'll try the jitter test for a laugh !
 
Well with my laptop and Senn HD535 I could not tel j2 apart from the original but my hearing is not great !

I woulda *sworn* I hit 'Submit' for this, but maybe not, apologies if re-post.
 
The following is a foobar2000 ABX of Nick Charles' Range.wav file, with jitter added by stv014, thank you to both for the test files. Planned set of 15, but you'll see I chickened out at 14.

 
foo_abx 1.3.4 report
foobar2000 v1.2.8
2013/09/21 22:31:46
File A: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\Range.wav
File B: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\Range_j2.flac
22:31:46 : Test started.
22:32:45 : 01/01  50.0%
22:33:02 : 01/02  75.0%
22:33:26 : 02/03  50.0%
22:34:04 : 03/04  31.3%
22:34:35 : 04/05  18.8%
22:35:10 : 04/06  34.4%
22:35:37 : 05/07  22.7%
22:36:04 : 06/08  14.5%
22:36:38 : 06/09  25.4%
22:37:38 : 07/10  17.2%
22:38:08 : 08/11  11.3%
22:38:38 : 09/12  7.3%
22:39:22 : 10/13  4.6%
22:41:23 : 11/14  2.9%
22:41:40 : Test finished.
 ----------
Total: 11/14 (2.9%)
 
Not much to listen for here, especially in such a busy and trebly mix. Too sizzling for my taste, I'd want a do-over if a recording studio handed this to me as a client.
If you want to give it a shot, here's how I got to it: find the shortest, sharpest, loudest transient you can. No cymbals, the 'short' is important. I used the second-to-last snare hit in the clip. Use the 'A' version to burn that sound into your mental ear, then compare directly to X and to Y, either one first. To my ears, the jittered version was just slightly blurred.
 
Nick, if y'all want to pursue this, I'd be thinking about something that might expose graininess and breakup due to jitter on sustained notes, instead of hunting down transients. Classical piano, slow, with a held chord somewhere? I don't have any software to deliberately jitter with, so I'd need to pick that up somewhere.

 

[BlindInOneEar]

Congrats, UltMusicSnob, for doing so well on that test.  I can't help but notice that you had to "game" the test to achieve your results.  That is, you had to "listen" in an unusual way pretty much wholly removed from how one would normally listen to music.  With all respect, is this a tacit admission that the "added jitter" track would have been indistinguishable absent the gaming?

 

[stv014]

It was the second version of the sample with the exaggerated amount of jitter (the noise component raised by 20 dB relative to the graph, and the sidebands by 40 dB). Although I did not think it is a track that would reveal jitter very well, so that is why I made the simulated jitter much worse.

 

[UltMusicSnob]

  Congrats, UltMusicSnob, for doing so well on that test.  I can't help but notice that you had to "game" the test to achieve your results.  That is, you had to "listen" in an unusual way pretty much wholly removed from how one would normally listen to music.  With all respect, is this a tacit admission that the "added jitter" track would have been indistinguishable absent the gaming?

Well, I will insist on the caveat that *all* ABX testing is of a sort pretty much wholly removed from how one would normally listen to music. The protocol can't be completed otherwise. The *only* time I ever listened like that in real life was when I was trying to hear John Lennon say "I bury Paul" at the end of "Strawberry Fields".

   That said,
 
Yes, my first research question is usually "Is differentiation possible at all???", and so I use the tools available to hunt for the differences.
It was particularly difficult in this case, as I don't have a good sense of what problematic jitter *ought* to sound like, and it matters what testers are listening for.
 
Since I can pick out a difference on one snare hit, a further refinement would be to listen more 'casually', and see if the drum set sounds different throughout.
 
I'm guessing that the added jitter track would have been indistinguishable for this particular music, but it's faintly conceivable that interested listeners could learn to hear the difference without the procedures I described.

 

[nick_charles]

  Nick, if y'all want to pursue this, I'd be thinking about something that might expose graininess and breakup due to jitter on sustained notes, instead of hunting down transients. Classical piano, slow, with a held chord somewhere? I don't have any software to deliberately jitter with, so I'd need to pick that up somewhere.

 
How about a segment of Beethoven Piano Sonata No.8 in C Minor, Op.13 'Pathetique' near the beginning ?
 
Pathetique

 

[nick_charles]

  It was the second version of the sample with the exaggerated amount of jitter (the noise component raised by 20 dB relative to the graph, and the sidebands by 40 dB). Although I did not think it is a track that would reveal jitter very well, so that is why I made the simulated jitter much worse.

 
UMS is clearly not a mere mortal though !

 

[UltMusicSnob]

   
How about a segment of Beethoven Piano Sonata No.8 in C Minor, Op.13 'Pathetique' near the beginning ?

Exactly, that's the kind of thing I mean.

 

[nick_charles]

  Exactly, that's the kind of thing I mean.

 
Posted in post 117 as a 30 second segment (flac)  fyi Jeno Jando on the Naxos label. If STV014 would be kind enough to do the jitterbug thingy ?
 
Can probably dial it down a notch