[SB]

http://www.jstage.jst.go.jp/article/ast/26/1/50/_pdf

Thought I would post this up, jitter is not audible until 30,000ps in a sighted test with direct control and 250,000ps in a blind test.

 

[b0dhi]

Incorrect. The correct conclusion is that it isn't detectable by conscious comparison by untrained listeners. "Audibility" is a seperate issue.

Also, they had a mouse interface. They don't mention any keyboard control, which is strange considering how important it is to be able to perform the test with eyes closed. They also used a 100ms crossfade when the user wanted to switch samples. Also, the test only addresses frequency or amplitude domain effects of jitter. Time domain effects are not explored with their digital simulation method. The utility of this study is limited.

 

[leeperry]

...and injected jitter into digital files is different from genuine clock jitter IMO.

you'd have to use the same exact device, one w/ a HIGH jitter clock and one w/ a uber-low, and then A/B...and using high quality headphones, you'd find out that the lower the jitter the wider the stereo stage/coherence.

 

[Speederlander]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif ...and injected jitter into digital files is different from genuine clock jitter IMO.

Physically, what is the difference in the final signal?

 

[ert]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif Incorrect. The correct conclusion is that it isn't detectable by conscious comparison by untrained listeners. "Audibility" is a seperate issue.

FTA "A total of 23 audio professionals or semi-professionals participated as the listeners. They were audio engineers, audio critics, sound engineers, and musicians." What is your definition of "untrained listener"? Also, I'm not sure that your distinction between "audibility" and "detectable" has any practical implication.

An interesting study. They seem to discuss some more details of the simulation wrt actual observed hardware jitter in other papers.

 

[AdamWysokinski]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif Incorrect. The correct conclusion is that it isn't detectable by conscious comparison by untrained listeners.

I wouldn't call these "23 audio professionals or semi-professionals [..] audio engineers, audio critics, sound engineers, and musicians" (p. 51) untrained listeners. Besides, comparisons in ABX tests are as conscious as in other blind tests.

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif "Audibility" is a seperate issue.

Oh, is it? How can you determine sonic qualities of something that is not detectable by the hearing system?

 

[nick_charles]

Quote:

Originally Posted by SB /img/forum/go_quote.gif http://www.jstage.jst.go.jp/article/ast/26/1/50/_pdf Thought I would post this up, jitter is not audible until 30,000ps in a sighted test with direct control and 250,000ps in a blind test.

There are several different types of jitter, random jitter as in the paper you cite is indeed far less audible, Benjamin and Gannon used correlated jitter in their 1998 study and found the thresholds to be from about 20ns in music but getting as high as 300ns in some cases. Correlated jitter creates distortion sidebands while random jitter raises the noise floor.


Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif Incorrect. The correct conclusion is that it isn't detectable by conscious comparison by untrained listeners. "Audibility" is a seperate issue. Also, they had a mouse interface. They don't mention any keyboard control, which is strange considering how important it is to be able to perform the test with eyes closed. They also used a 100ms crossfade when the user wanted to switch samples. Also, the test only addresses frequency or amplitude domain effects of jitter. Time domain effects are not explored with their digital simulation method. The utility of this study is limited.

Ashihara et al state that they used audio pros and semi-pros.

I use a mouse driven DBT in FooBar to test CD players and filters and I have no problem getting 20/20 between my CD players with my eyes open and detecting the difference between filters in the same mode, if a difference is big enough you will detect it.


Quote:

Originally Posted by leeperry /img/forum/go_quote.gif ...and injected jitter into digital files is different from genuine clock jitter IMO. you'd have to use the same exact device, one w/ a HIGH jitter clock and one w/ a uber-low, and then A/B...and using high quality headphones, you'd find out that the lower the jitter the wider the stereo stage/coherence.

Can you cite evidence to back up your assertion, remember where we are

What sensible reseach we have on jitter fails to give a consensus on what jitter sounds like anyway.

Of course if jitter is normally inaudible that might explain it, but that is speculation on my part.

But here is a point for general consideration. Why has no high end digital device manufacturer ever done any controlled blind listening tests varying jitter. Many talk about how low the jtter is on their kit, and why it matters, but not one to date has backed up the assertion that this matters with good empirical data.

In a different thread I also posted the Ashihara paper link as Dan Lavry did not believe the 250ns figure I mentioned. I expected Mr Lavry to come back with a critique of the paper, its methods or conclusions.

Mr Lavry had earlier stated that jitter of 177ns would basically mean lost or repeated samples an effect which you would think would be highly audible.

Perhaps human hearing is just not that good ?


EDIT: I am running some tests to see what is the minimum amount of total signal loss I can detect, 1 sec was easy so was 0.1sec I am now down to 0.006 seconds and it is getting harder, when I get to the point where it is really difficult I will run some blind tests. 0.001 secs is audible , down we go.

So far I can DBT 1/10,000 of a second drop ( a bit under 4 samples) out but it was tough (9/10) and a lot of concentration. At 0.00009 I was at pure guessing so for me the threshold for a single dropout is about 3.8 samples. Of course several dropouts is different. i will try dropouts interespered with good samples now.

 

[punkaroo]

Maybe my ears need a cleaning, but I've done a listening test using an iBasso D10, and hooked it up via USB and optical. I could not tell a difference in SQ using either or. I was shocked to say the least. I was really expecting optical to be on top, but nope!

 

[leeperry]

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif Can you cite evidence to back up your assertion, remember where we are

well, I said "IMO"...I don't quite have the same kind of equipment the XXhighend coder has to measure jitter...I merely use my two ears to judge on that

all I can say is that there was some files floating around a few months ago w/ different types of jitter injected to them..this was supposed to be a "live" simulation of DAC jitter, they all sounded identical tbh on my STX.

then I've ditched my Asus STX for the Asus ST(the only difference being the clock conditioner that supposedly gives very low jitter), and the difference was clearly night and day

I A/B'ed them several times(and so did several friends of mine)...and the stereo coherence was a big mess/mushy as hell on the STX and much wider/cleaner/clearer on the ST.

several ppl also modded their STX/ST w/ a much tighter clock, and said the exact same thing...but well, these are empirical/subjective tests indeed...fully agreed! and the CMI8788 DSP is spec'ed in its datasheet for a 750ps max jitter, so jitter adds up

and a lower jitter is not necessarily better, the ST was very fatiguing to my brain...too precise, too accurate..this card clearly makes you realize that uber low jitter + 124dB(actual measured) SNR is not that nice to listen to, after all.

Also, the CMI8788 DSP requires only one clock(24Mhz) so getting 44.1Khz out of it *WILL* create jitter(hence the need for a conditioner on the ST, that was engineered after the STX).

the Envy24 DSP needs two discrete clocks to operate:

SQ is not messy, not fatiguing..it's just great, with the burson discrete op-amp it just flows

still, injecting jitter into digital files needs to be run on an uber-low system..otherwise jitter adds up, you can -more than likely- hear a difference between 20ps jitter and 250 on headphones, but between 200 and 350? God only knows.
Quote:

Originally Posted by punkaroo /img/forum/go_quote.gif I was really expecting optical to be on top, but nope!

the italian guys who make that uber-low jitter USB interface said that optical is hopeless jitter-wise, it's on their site..check it out

besides USB is terrible jitter-wise w/ some low quality chips, but you can definitely improve it a lot.

Onkyo boasts about exactly that anyway: SE-U33GXV (B) USB

PS: ah, there used to be more infos about the jitter correction of this card, looks like they took it off..

 

[jcx]

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif ...In a different thread I also posted the Ashihara paper link as Dan Lavry did not believe the 250ns figure I mentioned. I expected Mr Lavry to come back with a critique of the paper, its methods or conclusions. Mr Lavry had earlier stated that jitter of 177ns would basically mean lost or repeated samples an effect which you would think would be highly audible. Perhaps human hearing is just not that good ? ...

I believe he lost track of the context - 250 nS jitter is only ~1% of the 44.1 sample time - which is where this jitter # applies - not at SPDIF bit edges

the only limit on digital domain jitter is that all the bits arrive in sequence and can be correctly assembled into the original value for the DAC every ~22uS

further, jitter spectrum is important to understanding effects - 250 nS rms jitter with a 20KHz bandwidth limit will not corrupt > 1 MHz self clocked serial data

 

[nick_charles]

Quote:

Originally Posted by jcx /img/forum/go_quote.gif I believe he lost track of the context - 250 nS jitter is only ~10% of the 44.1 sample time - which is where this jitter # applies - not at SPDIF bit edges the only limit on digital domain jitter is that all the bits arrive in sequence and can be correctly assembled into the original value for the ADC every ~22uS further, jitter spectrum is important to understanding effects - 250 nS rms jitter with a 20KHz bandwidth limit will not corrupt > 1 MHz self clocked serial data

Thanks for the explanation !

 

[leeperry]

I forgot this link: Burson Clock
Quote:

By reducing the jitter error, you will hear clearer positioning, also details are further refined vocally and instrumentally. Sound stage and positioning will improve

it's exactly what you experience when going STX>ST or by putting a killer clock on the STX..wider SS, and clearer separation. Jitter *is* audible.

and here the XXHighend player coder did some jitter measurements: Measuring XXHighEnd ...

he claims that he has found ways to lower jitter within the windows kernel, and many ppl on his forum and computeraudiophile.com assure that they can ABX his different "algorithms"

 

[b0dhi]

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif Ashihara et al state that they used audio pros and semi-pros.

If the test depends on conscious comparison and detection, it is a skill. It is therefore required to train the listeners if the result is to be extrapolated as "audibility" and not just conscious comparative detection skill.

Audio-pros, musicians and audiophiles are not people trained in detecting jitter-caused distortion. Being slightly more likely to be able to do so than the general population due to one's profession/hobby is hardly "trained".


Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif I use a mouse driven DBT in FooBar to test CD players and filters and I have no problem getting 20/20 between my CD players with my eyes open and detecting the difference between filters in the same mode, if a difference is big enough you will detect it.

Ofcourse. The point was that requiring the eyes to be open and having to navigate the interface with a mouse significantly reduces the ability of the listeners to concentrate. I have done ABXs and failed due to the conditions of the ABX requiring me to use a mouse (p>0.4). After fixing that condition, I was able to pass (p<<0.01). It makes all the difference when one is dealing with extremely small differences at the limit of one's perception.

 

[nick_charles]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif IfThe point was that requiring the eyes to be open and having to navigate the interface with a mouse significantly reduces the ability of the listeners to concentrate. I have done ABXs and failed due to the conditions of the ABX requiring me to use a mouse (p>0.4). After fixing that condition, I was able to pass (p<<0.01). It makes all the difference when one is dealing with extremely small differences at the limit of one's perception.

Hmm, I think this may fall into the what works best for you category, at my age I close my eyes and I start falling asleep

I think I concentrate better eyes open - shrug, of course having reflected on it it will now be impossible for me to test this fairly

 

[b0dhi]

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif Hmm, I think this may fall into the what works best for you category, at my age I close my eyes and I start falling asleep I think I concentrate better eyes open - shrug, of course having reflected on it it will now be impossible for me to test this fairly

Lol

You have a fair point. I'd simply assumed others used the same method as myself, but ofcourse that's not true. As a sidenote, I read some time ago that the visual system also participates in audio processing in some way, so it would seem that the question of whether to close ones eyes or not is more complex.