Head-Fi.org › Forums › Equipment Forums › Dedicated Source Components › An interesting paper on jitter audibility
New Posts  All Forums:Forum Nav:

An interesting paper on jitter audibility - Page 3

post #31 of 112
Quote:
Originally Posted by hciman77 View Post
But in this case it was a listening test i.e the subjects (including musicians) listened to two signals one with and one without jitter and were unable to discern differences until the jitter was at very very very very high levels.

It almost seems as if some folks want jitter to be a problem, I just cannot fathom it, surely we should rejoice if jitter was a non-issue.
I wasnt referring to this test. I mean in general.
post #32 of 112
Quote:
Originally Posted by drarthurwells View Post
Chaos theory at work. An extremely small (or minor) event can have catastrophic effects down the chain-of-causality road - a very small amount of jitter can go a long way in later effects on the sound.
Don't try to Jeff Goldblum me, buddy! I know what chaos theory is, and it doesn't say anything about jitter. This is a fancy marketing technique that is used all the time in audio. Tell me, how exactly does chaos theory govern jitter? I'd really like to know.

Quote:
Repeated trials over many days of listening to more-jittered and less-jittered music will eventually bring this initial vague peception into full focus as we later realize the jittered music is really bad.
Yep, folks, this is another classic argument that shows up again and again. It smacks of truthiness. It seems like it could true, yet somehow, there is absolutely no evidence of this effect.

Quote:
Human perception can be very unreliable - any listening tests that don't allow much listening experience to both A (very slightly more jitter) and B (very slightly less jitter) may not show discrimination abilities between the two.
More of the same.
post #33 of 112
Quote:
Originally Posted by hciman77 View Post
But in this case it was a listening test i.e the subjects (including musicians) listened to two signals one with and one without jitter and were unable to discern differences until the jitter was at very very very very high levels.

It almost seems as if some folks want jitter to be a problem, I just cannot fathom it, surely we should rejoice if jitter was a non-issue.
Jitter is no different than other aspects of high-end audio. If your system is not set-up well acoustically with superb resolving components and cables, then you probably cannot hear half of what we are talking about, whether its detail, imaging or jitter..... This is why drawing any conclusions from this study is tenouous at best. The systems used have questionable quality. Every system is at some level of perfection, there is just no way of telling where these were, at the bottom or the top. More likely in the middle IMO.

I suppose if you said you had a Bose system in your whole house and you could not tell the difference in jitter, then that would be a data point. For those that have Bose whole-house systems, you will probably not hear even high levels of jitter.

Steve N.
post #34 of 112
Originally Posted by drarthurwells:
Chaos theory at work. An extremely small (or minor) event can have catastrophic effects down the chain-of-causality road - a very small amount of jitter can go a long way in later effects on the sound.


ezkcdude: Don't try to Jeff Goldblum me, buddy! I know what chaos theory is, and it doesn't say anything about jitter. This is a fancy marketing technique that is used all the time in audio. Tell me, how exactly does chaos theory govern jitter? I'd really like to know.

Art: If you knew anything at all about Chaos theory you would know it is about events in general and is not restricted to a particular event. So it can apply to the case where any slight change can eventuate in large scale changes. Jitter is slight change that can lead to problems in sound quality at a large level of change.

Art: Repeated trials over many days of listening to more-jittered and less-jittered music will eventually bring this initial vague peception into full focus as we later realize the jittered music is really bad. Human perception can be very unreliable - any listening tests that don't allow much listening experience to both A (very slightly more jitter) and B (very slightly less jitter) may not show discrimination abilities between the two.

ezkcdude: Yep, folks, this is another classic argument that shows up again and again. It smacks of truthiness. It seems like it could true, yet somehow, there is absolutely no evidence of this effect.

Art: Twenty five years ago I bought a top Sony ES model CD for my system. A friend, who also had high end equipment, participated in some tests. First I cued up a CD on my Sony CDP with the exact CD being played on a nearby FM classical station, which I had tuned to with my high end FM tuner. I blindfolded my friend who tried to pick the better CDP sound from the same CD on FM, as I switched back and forth at the same volume.

He could not tell which was which in AB testing. Then I did the same with a tape made of a CD with the same CD being played in synch on the CDP I used to make the tape with. Again he could not reliably tell which was which in AB switching.

Then he listended repeatedly to a short passage from the CDP. He knew it was the CDP. Then we did the same with the tape. Then we repeated this AB where both A and B involved repeated listening to one before switching to the other. This allowed him to learn the sound. He then was able to state that the CDP had the better sound, but more importantly, was now able to reliably tell which of the two was playing while listening to the AB switching blindfolded.

AB only works when both A and B are for extended periods to allow adequate perceptual learning. Eyewitnesses to crime often make the wrong ID of suspects but are 100% accurate if they know the criminal before the crime. A long period of perceptual experience is best for perceptual learning to yield valid perceptual discriminations (visual or auditory).

And BTW, what does it mean when you say I "Jeff Goldblum you"?

Is that something you can tell me in public?
post #35 of 112
Thread Starter 
Quote:
Originally Posted by audioengr View Post
Jitter is no different than other aspects of high-end audio. If your system is not set-up well acoustically with superb resolving components and cables, then you probably cannot hear half of what we are talking about, whether its detail, imaging or jitter.....
Lets keep this to jitter, let us not get diverted.

So to hear jitter at say 250ns which is close to 500 times worse than even a pretty modest CD player one needs a really high end system ?

Quote:
This is why drawing any conclusions from this study is tenouous at best. The systems used have questionable quality. Every system is at some level of perfection, there is just no way of telling where these were, at the bottom or the top. More likely in the middle IMO.
Call it a data point by all means but it has a reasonably high level of control for each subject i.e one independent variable and blind testing. As for the quality of their systems this is (as you say) unknown but speculation on them is just that speculation. You cannot prove that there were no high end systems in the sample and I cannot prove that there were any high end systems in the sample. I have emailed the lead author to ask if they can give us some insights into their subjects and their systems, I await their answers.
post #36 of 112
an excerpt from the Audio Precision Application Note 5: Measurement Techniques for Digital Audio


[Jitter] Audibility considerations

It is one thing to be able to identify and measure sampling jitter. But how can we tell if there is too much?
A recent paper by Eric Benjamin and Benjamin Gannon describes practical research that found the lowest jitter level at which the jitter made a noticeable difference was about 10 ns rms. This was with a high level test sine tone at 17 kHz. With music, none of the subjects found jitter below 20 ns rms to be audible.

This author has developed a model for jitter audibility based on worst case audio single tone signals including the effects of masking.
This concluded:
“Masking theory suggests that the maximum amount of jitter that will not produce an audible effect is dependent on the jitter spectrum. At low frequencies this level is greater than 100 ns, with a sharp cut-off above 100 Hz to a lower limit of approximately 1 ns (peak) at 500 Hz, falling above this frequency at 6 dB per octave to approximately 10 ps (peak) at 24 kHz, for systems where the audio signal is 120 dB above the threshold of hearing.”

In the view of the more recent research, this may be considered to be overcautious. However, the consideration that sampling jitter below 100 Hz will probably be less audible by a factor of more than 40 dB when compared with jitter above 500 Hz is useful when determining the likely relative significance of low- and high-frequency sampling jitter.
post #37 of 112
Thread Starter 
Quote:
Originally Posted by jcx View Post
an excerpt form the AudioPrecision Application Note 5: Measurement Techniques for Digital Audio


[Jitter] Audibility considerations

It is one thing to be able to identify and measure sampling jitter. But how can we tell if there is too much?
A recent paper by Eric Benjamin and Benjamin Gannon describes practical research that found the lowest jitter level at which the jitter made a noticeable difference was about 10 ns rms. This was with a high level test sine tone at 17 kHz. With music, none of the subjects found jitter below 20 ns rms to be audible.

This author has developed a model for jitter audibility based on worst case audio single tone signals including the effects of masking.
This concluded:
“Masking theory suggests that the maximum amount of jitter that will not produce an audible effect is dependent on the jitter spectrum. At low frequencies this level is greater than 100 ns, with a sharp cut-off above 100 Hz to a lower limit of approximately 1 ns (peak) at 500 Hz, falling above this frequency at 6 dB per octave to approximately 10 ps (peak) at 24 kHz, for systems where the audio signal is 120 dB above the threshold of hearing.”

In the view of the more recent research, this may be considered to be overcautious. However, the consideration that sampling jitter below 100 Hz will probably be less audible by a factor of more than 40 dB when compared with jitter above 500 Hz is useful when determining the likely relative significance of low- and high-frequency sampling jitter.
I have quoted this segment in other threads. Sadly Julian Dunn (the author) is no longer with us. The point being is that his model was just that a model and not empirically validated via controlled listening tests. The discrepancy between models and listening tests being very interesting. Perhaps masking has more effect than predicted by the model.
post #38 of 112
Quote:
Originally Posted by drarthurwells View Post
And BTW, what does it mean when you say I "Jeff Goldblum you"?

Is that something you can tell me in public?
Have you ever seen Jurassic Park? If not, rent it.
post #39 of 112
Thread Starter 
Quote:
Originally Posted by hciman77 View Post
Call it a data point by all means but it has a reasonably high level of control for each subject i.e one independent variable and blind testing. As for the quality of their systems this is (as you say) unknown but speculation on them is just that speculation. You cannot prove that there were no high end systems in the sample and I cannot prove that there were any high end systems in the sample. I have emailed the lead author to ask if they can give us some insights into their subjects and their systems, I await their answers.
I have received an answer from the lead author. I have asked for permission to reproduce his email, if not I will summarise the key points later.
post #40 of 112
drarthurwells: Just a heads up, in that long reply to ezkcdude you seem to have completely forgotten to provide any actual evidence, or even a coherent hypothesis, of how chaos theory applies to jitter being something that's audible, or to cite a source that does.
post #41 of 112
Thread Starter 

From the lead author of the jitter paper

This is the reply I received - in its entirety.
================================
Quote:
Dear Jim,

Thank you for the e-mail. I suppose that you read our paper titled
'Detection threshold for distortions due to jitter on digital
audio(http://www.jstage.jst.go.jp/article/ast/26/1/50/_pdf).' Before this
paper was published in the Acoustical Science and Technology, we had
published another paper 'The maximum permissible size and detection
threshold of time jitter on digital audio.' Unfortunately, it was written in
Japanese.

In our first experiment, which was reported in the Japanese paper, we used a
fixed listening condition and fixed materials. All of 14 participants were
university students without any special training. The audio system that we
used consisted of the following equipment.

D/A converter --- SEK'D ADDA2496S
preAmp. --- Luxman C-7
main Amp. --- Luxman M-7
loudspeakers --- DIATONE DS-205

They costed about $10,000. I don't know if they belong to high-end or not.

All participants could distinguish between sounds with and without time
jitter when the jitter size was 9216 ns. A few could when it was 1152 ns. No
one could when it was as small as 576 ns.

There was a question, however, if the result would depend on the listening
environments and the skill of the listeners. That is why we carried on the
second experiment. This second experiment is reported in the paper, the one
that you probably read.

Listeners in the second experiment were all professionals, audio engineers,
recording/mixing engineers, musicians, etc... Sound materials were selected
by the listeners so that each listener could use his (her) familiar
materials. The experimenter (we) visited the listeners’ studios or listening
rooms so that we could use listeners’ own DAC, amplifiers, loudspeakers and
headphones. The system configurations, therefore, varied among listeners.
They were mostly mid-end or above, I suppose.

As you can find in the paper, some listeners could distinguish the sounds
when time jitter was 500 ns. It could not be detected, however, when the
jitter was as small as 250 ns.

In both experiments, there was considerable difference in listeners’
performance. I don’t know, however, if it was because of their audio
experience. We had expected much better performance in the second experiment
because the listeners were professionals and they could use their favorite
environments and materials.

Our conclusion up to now is that the normal hearing listeners' detection
threshold for time jitter in program materials is several hundred ns.

I appreciate that you are interested in our paper. Thank you for asking
questions.

Best wish

--
ASHIHARA Kaoru
================================

So in expt 1 they had a range of 9000 -> 1152 for detection i.e a range of 1 - ~9 in terms of detection ability, none detected it at 576ns. This was on a loudspeaker system. Worst case say it was theoretically detectable at 577ns say they just missed it by 1ns. That gives us a range of ability of 1 to ~16.

In expt 2 a range of 2 microseconds to 500ns i.e a range of 1 to 4 in terms of detection ability, none detected it at 250ns. Worst case say it was actually detectable at 251ns and they just missed this by 1 ns. This is a range of about 1 to ~8.

Let us say that a really bad CD player has jitter of 1ns. So for this to be a problem, detection ability on this variable has to be over 250 times better than the best result found in the paper.

Even if audiophiles are capable of far far better jitter detection (and I do not concede this point at all until it is empirically shown in blind listening tests) who will, with a straight face suggest that they are a factor of over 250 times better than the best AE, Musician or Studio professional using their own listening kit (including studios) and music they are familiar with.
post #42 of 112
Quote:
Originally Posted by Toe View Post
drarthurwells: Just a heads up, in that long reply to ezkcdude you seem to have completely forgotten to provide any actual evidence, or even a coherent hypothesis, of how chaos theory applies to jitter being something that's audible, or to cite a source that does.

Well he doesn't really have to. One of the tenets of chaos theory is that in chains of correlated events a small change early on in the chain can have much larger consequences farther down the chain. This is related to the problem of statistical error propagation. The question then becomes, is a chain of audio components long enough for significant error propagation effects to be detected? Discuss ...
post #43 of 112
hciman,

Thanks, for posting this. It is very enlightening. It shows two very important things: 1) Jitter is detectable, but only at much higher levels than you would guess from these forums and 2) The evidence from the second paper is even more striking because it does show that trained professionals, whose living is based on their ears to a large extent, do appear to be better at detecting jitter. This second point is interesting, because it is often claimed by jitter-busters, and now we can at least have some evidence for that. However, if one wants to cite this aspect of these Japanese researcher's work, then one must also acknowledge that, in absolute terms, jitter should not be audible in the amounts that exist in our stereo systems (unless it is artificially generated). I would suggest that whatever changes people hear when they measurably improve jitter in their systems are not due to the jitter itself, but rather to other changes in the signal chain. Chaos? I don't think so.
post #44 of 112
Thread Starter 
Quote:
Originally Posted by ezkcdude View Post
hciman,

2) The evidence from the second paper is even more striking because it does show that trained professionals, whose living is based on their ears to a large extent, do appear to be better at detecting jitter. This second point is interesting, because it is often claimed by jitter-busters, and now we can at least have some evidence for that.
I think suggests is a better word than shows, remember that in the 2nd expt subjects had their own systems and choice of material. Thus there are three fundamental ways that the 2nd expt differs from the first

1) Listener type (students vs pros)
2) System familiarity (standard system vs own system)
3) Material familiarity (forced material vs chosen material)

Any 1 variable or combination of variables could contribute towards the better performance, of course one could easily isolate these variables and test them separately. Are pros better ?, maybe, it would be nice to think so in the sense that I would prefer CDs mastered by folks who can hear the difference between good and mediocre.
post #45 of 112
Quote:
Originally Posted by audioengr View Post
I have posted some things on prosound bulletin boards. This is where I got to be so jaded about them. I found that most of them only care about measurements, not even what is audible...

Steve N.
Don't mean to be rude but I find the above statement very odd coming from someone who is (or claim to be) an audio engineer.

An audiophile can self-proclaim to have golden ears and defy all logic (I can hear the difference because I believe there is a difference, despite all measurements indicate otherwise.) But an engineer, being a man of science, should correlate his/her observations with the measurements before coming to a conclusion.

To say that measurements do not reflect what's audible is saying audio engineering is black magic; and black magic is not engineering because it's not based in science. Now it's possible the measurements (good or bad) do not correspond to what you hear, simply because the instrument is not made to measure the particular aspect of the audio spectrum of your interest, but that is not an excuse to ignore measurements in its entirety and its implications.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Dedicated Source Components
Head-Fi.org › Forums › Equipment Forums › Dedicated Source Components › An interesting paper on jitter audibility