Head-Fi.org › Forums › Equipment Forums › Sound Science › Jitter Correlation to Audibility
New Posts  All Forums:Forum Nav:

Jitter Correlation to Audibility - Page 9

post #121 of 251
Quote:
Originally Posted by BlindInOneEar View Post
 

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?

Point taken. I altered the protocol for no burn-in listening, just side-by-side comparisons across longer segments. I returned to the first track stv014 made, Range_j.flac, which did not have the jitter jacked up so high, and applied this test, listening to the drum set as a musician would.

 

All the snare hits are different.

 

foo_abx 1.3.4 report
foobar2000 v1.2.8
2013/09/22 11:36:16

File A: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\Range.wav
File B: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\Range_j.flac

11:36:16 : Test started.
11:36:48 : 01/01  50.0%
11:36:59 : 02/02  25.0%
11:37:10 : 02/03  50.0%
11:37:43 : 03/04  31.3%
11:38:02 : 04/05  18.8%
11:38:19 : 05/06  10.9%
11:38:41 : 06/07  6.3%
11:40:04 : 07/08  3.5%
11:40:40 : 07/09  9.0%
11:41:18 : 08/10  5.5%
11:42:26 : 09/11  3.3%
11:45:48 : 09/12  7.3%
11:46:10 : 10/13  4.6%
11:46:19 : 11/14  2.9%
11:46:27 : Test finished.

 ----------
Total: 11/14 (2.9%)

 

Of course, this time I had the enormous benefit of knowing what to listen for (snare), that I could apply across the track. No hunting around for artifacts.

 

I hear the jittered drum set as *slightly* less precise. The jittered version thuds instead of snapping tightly. It's a *tiny* difference, but you can find it anyplace that the cymbal-crazy drummer isn't covering it up. If I had to reverse-engineer from the treatment, I would hypothesize that the applied jittering is smearing/obscuring the attacks slightly.

post #122 of 251

New test files:

 

path30n.flac

path30j.flac

range_n.flac

range_j.flac

range_jl.flac

 

This time, to avoid any sampling artifacts (pre-ringing) in the jittered version near 22.05 kHz, which UltMusicSnob may be able to detect (according to the previous upsampling ABX tests), I upsampled the input files first to 96 kHz with "resample -r 96000 -q 4 -ff 0.48 -f 3". Also, the sample format is now 24-bit PCM, to avoid the possibility of audible dither noise in a quiet track like path_30.wav. The jitter spectrum (implemented as phase modulation on a 768 kHz upsampled signal) is:

Files with an 'n' suffix have no simulated jitter. 'j' indicates jitter, and 'jl' is lower jitter, that is, like shown above, but reduced overall by a factor of 3. With path_30.wav, there is actually only a fairly low error signal, because the input file does not have much high frequency content.

 

Lower jitter versions will be created on successful ABX results. :normal_smile :


Edited by stv014 - 9/22/13 at 11:45am
post #123 of 251
Quote:
Originally Posted by stv014 View Post
 

New test files:

 

path30n.flac

path30j.flac

range_n.flac

range_j.flac

range_jl.flac

 

Lower jitter versions will be created on successful ABX results. :normal_smile :

 

Many thanks for the work you have put in, So the normal jitter version is for simplicity about  ? ns and the lower jitter version is    ?

post #124 of 251
Quote:
Originally Posted by stv014 View Post
 

 

Files with an 'n' suffix have no simulated jitter. 'j' indicates jitter, and 'jl' is lower jitter, that is, like shown above, but reduced overall by a factor of 3. With path_30.wav, there is actually only a fairly low error signal, because the input file does not have much high frequency content.

 

 
Hmmm, would it be worth uploading some Emerson , Lake and Palmer synth noodling ?
post #125 of 251
Quote:
Originally Posted by nick_charles View Post
 

Many thanks for the work you have put in, So the normal jitter version is for simplicity about  ? ns and the lower jitter version is    ?

 

The lower jitter (jl) version has ~35-36 ns RMS, and ~260-270 ns peak to peak, however, note that much of the peak to peak level is from low frequency noise, which is not as audible as modulation by high frequency tones. With the noise components removed, leaving only the sidebands, the level is 32 ns RMS and 126 ns peak to peak. The other (j) files use the same modulator signal, but multiplied by 3.


Edited by stv014 - 9/22/13 at 12:29pm
post #126 of 251
Quote:
Originally Posted by nick_charles View Post
 

 

Many thanks for the work you have put in, So the normal jitter version is for simplicity about  ? ns and the lower jitter version is    ?

This is two tests of the same two files, pathn versus pathj, so there's replication of the result.

 

I did previous tests listening for any kind of noise or breakup effect on the sustained chords. Those are all null, failure to positively ABX. So if there's some sort of calculable noise contribution of the jitter to the signal, it was inaudible to me.

 

Short sharp snare hits are not available in this track, so I had to re-learn how to detect a difference.

 

Since the Range sample cued successfully on the snare hits, I switched to listening to the piano note attacks, and that provided the successful cues. The non-jittered piano was just a little more delicate and focused, while the jittered version had just a little more thump and spread in it. Soft chords, listened to at soft levels, worked best to detect this difference.

 

First test with positive result:

foo_abx 1.3.4 report
foobar2000 v1.2.8
2013/09/22 23:04:33

File A: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\path30n.wav
File B: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\path30j.wav

23:04:33 : Test started.
23:05:30 : 01/01  50.0%
23:05:38 : 02/02  25.0%
23:05:58 : 03/03  12.5%
23:06:10 : 03/04  31.3%
23:06:17 : 04/05  18.8%
23:06:31 : 05/06  10.9%
23:06:50 : 06/07  6.3%
23:07:13 : 06/08  14.5%
23:07:29 : 07/09  9.0%
23:08:15 : 07/10  17.2%
23:08:23 : 08/11  11.3%
23:08:30 : 08/12  19.4%
23:08:40 : 09/13  13.3%
23:08:52 : 10/14  9.0%
23:09:07 : 11/15  5.9%
23:09:21 : 12/16  3.8%
23:09:42 : 13/17  2.5%
23:11:51 : 13/18  4.8%
23:12:04 : 14/19  3.2%
23:12:59 : 15/20  2.1%
23:13:13 : Test finished.

 ----------
Total: 15/20 (2.1%)

 

 

Second test to replicate:

foo_abx 1.3.4 report
foobar2000 v1.2.8
2013/09/23 00:01:19

File A: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\path30n.wav
File B: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\path30j.wav

00:01:19 : Test started.
00:01:44 : 01/01  50.0%
00:01:50 : 02/02  25.0%
00:02:18 : 02/03  50.0%
00:02:26 : 03/04  31.3%
00:02:48 : 04/05  18.8%
00:03:04 : 05/06  10.9%
00:03:24 : 06/07  6.3%
00:03:48 : 06/08  14.5%
00:04:21 : 06/09  25.4%
00:04:42 : 07/10  17.2%
00:04:55 : 08/11  11.3%
00:05:54 : 09/12  7.3%
00:06:47 : 10/13  4.6%
00:07:59 : 11/14  2.9%
00:08:32 : Test finished.

 ----------
Total: 11/14 (2.9%)

post #127 of 251

Well, it seems a "lower jitter" version will indeed be needed. Did you try the updated "range" samples ?

post #128 of 251

Reduced jitter level, similarly to range_jl.flac (9.5 dB lower than shown on the graph above):

 

path30jl.flac

post #129 of 251
Quote:
Originally Posted by stv014 View Post
 

Well, it seems a "lower jitter" version will indeed be needed. Did you try the updated "range" samples ?

I haven't done updated range yet, thanks for path30jl.

 

How close are we to realistic jitter yet, as a function of actual clocks in hardware? I've assumed (because of gigantic clock rates all over normal PC motherboards) that jitter was never an issue at all, but I don't know what sort of jitter levels are actually seen in listening devices. I'd be inclined to only look at DAC jitter, because the recording jitter I can't do anything about anyway, and data cable jitter (if I understand it correctly) is an utter non-issue altogether.

post #130 of 251

It would be very poor for hardware jitter (a fairly cheap sound card or DAC could have more than 100 times lower jitter of this type). However, there are different kinds of jitter (both in terms of spectral characteristics and exactly how they affect the output signal, and the "modulator" may even include the audio signal in digital or analog form). That is one of the reasons why I created "pessimistic" samples at first.

post #131 of 251
Quote:
Originally Posted by UltMusicSnob View Post
 

I haven't done updated range yet, thanks for path30jl.

 

How close are we to realistic jitter yet,

 

The jitter model STV014 is using was based on the very worst measured jitter of any commercial user-end digital device to have passed through the lab at Stereophile and was boosted to rather more drastic levels - see posts 51 & 55 for reference. It is also worth reading the Benjamin and Gannon and Ashihara et al papers

 

as a function of actual clocks in hardware? I've assumed (because of gigantic clock rates all over normal PC motherboards) that jitter was never an issue at all, but I don't know what sort of jitter levels are actually seen in listening devices.

 

There is an open debate about how best to measure/characterize jitter but Stereophile measures jitter on all digital devices it reviews. In short under their measurement system a typical device would demonstrate well under 1ns of jitter (normally 100 - 300 ps) some poorer devices might show 1 to 4 ns and the poster child for bad jitter is the McIntosh MS750 Music server which for simplicity clocks in at about 14ns.

 

 

So what you have ABX'ed so far is a bit worse than the worst of the worst. For currently established thresholds of audibility 

 

 

 

http://downloads.bbc.co.uk/rd/pubs/reports/1974-11.pdf - A BBC paper from 1974

 

 

E. Benjamin and B. Gannon, ‘‘Theoretical and audible effects of jitter on digital audio quality,’’ Preprint of the 105th AES
Convention, #4826 (1998) - sorry you need to buy it from the AES but the contents have been summarized here and elsewhere

 

Detection threshold for distortions due to jitter on digital audio; Kaoru Ashihara, Shogo Kiryu, Nobuo Koizumi, Akira Nishimura, Juro Ohga, Masaki Sawaguchi, and Shokichiro Yoshikawa, Acoust. Sci. & Tech. 26, 1 (2005) https://www.jstage.jst.go.jp/article/ast/26/1/26_1_50/_pdf

 

 

I'd be inclined to only look at DAC jitter

 

 

Yep, stop there !


Edited by nick_charles - 9/23/13 at 9:01am
post #132 of 251
Quote:
Originally Posted by UltMusicSnob View Post

I haven't done updated range yet, thanks for path30jl.

How close are we to realistic jitter yet, as a function of actual clocks in hardware? I've assumed (because of gigantic clock rates all over normal PC motherboards) that jitter was never an issue at all, but I don't know what sort of jitter levels are actually seen in listening devices.

The cheap quartz crystal resonators used in PC motherboards while are fairly stable, depending on the quality of the driving circuit, and can easily get into the 10s of nanoseconds jitter. On the flipside, vast majority of dedicated readily available oscillators are easily sub 1ns and in fact are now getting into sub 1ps jitter numbers. Cost is a predominant factor in PC board design, so I normally don't expect as much focus on this during in the design phase (or let's just say there are other overruling concerns above audio quality). I would only hope any sane engineer designing a stand-alone audio unit, however, would avoid using a resonator as a clock source.
post #133 of 251
Quote:
Originally Posted by Digitalchkn View Post

The cheap quartz crystal resonators used in PC motherboards while are fairly stable, depending on the quality of the driving circuit, and can easily get into the 10s of nanoseconds jitter.

 

That depends on the motherboard. This one is not bad, for example, although others could of course be much worse. But motherboard audio can have other problems bad enough to be audible, such as noise (particularly from interference).

post #134 of 251
Quote:
Originally Posted by stv014 View Post
 

 

That depends on the motherboard. This one is not bad, for example, although others could of course be much worse. But motherboard audio can have other problems bad enough to be audible, such as noise (particularly from interference).

 

Vast majority of the time the audio section is sourced by say a 25MHz at cut crystal running through some sort of a multi-purpose clock generator/buffer that spits out all sorts of clocks for CPU, memory, peripheral busses, I/O bridges, etc.  It is reasonably clean, except it's PLLs bandwidths are often above the audio frequencies. The higher speed digital standards don't really care about what happens to clocks at audio frequencies.  I have personally measured plenty of motherboard clocks that wander around like crazy at low frequencies -- but they pass all the jitter requirements set out by that fancy 5Gbps high speed standard.

 

I am not even getting into crosstalk/EMI interference.

post #135 of 251
Quote:
Originally Posted by stv014 View Post
 

Reduced jitter level, similarly to range_jl.flac (9.5 dB lower than shown on the graph above):

 

path30jl.flac

Are we highly confident that there is no other possible source of differences in these tracks?

 

The tests below are for path30n versus path30jl.

I have learned what these clips sound like, and I am learning (maybe) what jitter sounds like.

But even discounting for some sort of learning factor, path30jl was ABXed with precisely the same characteristics as path30j (soft piano attack), and with less difficulty.

Cue again was focus and delicacy (better focus in n).

As you can see, both tests took close to three minutes, and were conducted minutes apart.

 

First test, through Beyerdynamic 770 Pros ("full, rich, dark")

 

foo_abx 1.3.4 report
foobar2000 v1.2.8
2013/09/23 19:15:24

File A: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\path30n.flac
File B: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\path30jl.flac

19:15:24 : Test started.
19:16:19 : 01/01  50.0%
19:16:25 : 02/02  25.0%
19:16:31 : 03/03  12.5%
19:16:56 : 04/04  6.3%
19:17:10 : 05/05  3.1%
19:17:31 : 05/06  10.9%
19:17:40 : 06/07  6.3%
19:17:46 : 07/08  3.5%
19:17:57 : 08/09  2.0%
19:18:14 : 09/10  1.1%
19:18:25 : Test finished.

 ----------
Total: 9/10 (1.1%)

 

 

Second test, switched to my OLD Beyerdynamic DT 48s (highly detailed, missing bass extension badly for techno)

foo_abx 1.3.4 report
foobar2000 v1.2.8
2013/09/23 19:22:44

File A: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\path30n.flac
File B: C:\Users\KiarkAudio\Documents\Ravel Listening Tests\path30jl.flac

19:22:44 : Test started.
19:23:02 : 01/01  50.0%
19:23:07 : 02/02  25.0%
19:23:13 : 03/03  12.5%
19:23:26 : 04/04  6.3%
19:23:38 : 05/05  3.1%
19:24:06 : 06/06  1.6%
19:24:24 : 07/07  0.8%
19:25:05 : 08/08  0.4%
19:25:18 : 08/09  2.0%
19:25:36 : 09/10  1.1%
19:25:38 : Test finished.

 ----------
Total: 9/10 (1.1%)

 

If the level of distortion is scaling linearly to the severity of jitter applied, it is not apparent. I expected less difference, but that was not my experience.

 

Replications, anyone, please? Pick any soft chord attack, pick the more delicate, more focused one. I might also describe the 'n' as more 'realistic'; the jittered version seems slightly off-balance EQ wise, like a virtual piano instead of a real one.


Edited by UltMusicSnob - 9/23/13 at 5:45pm
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › Jitter Correlation to Audibility