Close in jitter?

[Blumlein 88]

https://www.theguardian.com/lifeandstyle/2012/may/18/experience-quietest-place-on-earth

This is the fellow who stayed for the 'record' 45 minutes. He didn't go insane or seem all that bothered by it.

He mentions when he wrinkles his forehead he could hear it sliding over the skull. I can hear that when sitting in a quiet (though not anechoic) room. Then again maybe I just have a noisy forehead/skull interface.

 

[RayDunzl]

If it were masked, a second larger signal would be required.

I'd think masked.

Example:

The first thing I hear when I insert earplugs is my breathing. I rarely hear it without, during the day.

Did I turn off my filters or suppress the masking noise?

(I get to be wrong, so, have at it)

 

[Cosmik]

I don't remember where, but I have seen an estimation of how much your pulse moves your ear drums. The pulse of blood thru the hearing mechanism does move the ear drum and membrane. It was an amount of movement that was several times more than the air movement that would match up with sub-picosecond jitter levels. And as I believe I mentioned earlier, those levels of jitter would be swamped by the random brownian motion of the air. These effects just can't physically exist in a way we could hear. So you need some carefully done blind tests of significant numbers of listeners to give this much credibility. If it is true, then you have one heck of a mystery as to how it can be so in the physical world of air and sound.

I don't understand why you would leap headlong into a blind listening test. If you established that the error due to the jitter is actually 'silence' even before any masking is applied, why would you need to carry out a blind listening test? If you are going to indulge every audiophile superstition with a listening test that is controlled for better than sub-Brownian levels of background noise, you may not have much time left for listening to music!

 

[Blumlein 88]

I don't understand why you would leap headlong into a blind listening test. If you established that the error due to the jitter is actually 'silence' even before any masking is applied, why would you need to carry out a blind listening test? If you are going to indulge every audiophile superstition with a listening test that is controlled for better than sub-Brownian levels of background noise, you may not have much time left for listening to music!

Well just because you can't conceive of how something is so, doesn't automatically prevent it from being so.

So until the idea extremely small amounts of close in jitter are audible are shown as real, I would go right along as if they are not. This does get back to the idea that those who claim it is so are those responsible for the blind tests to get us to seriously consider it. For me I can see no path it is possible. If some well done blind tests show people can in fact perceive it, then a big unexpected and unexplainable idea needs investigating. Otherwise I would not engage in testing it.

As is the case in very nearly 100% of all such circumstances, if those people with the claim bother to test, they will find they were mistaken.

So I am not leaping headlong into anything. It is the only convincing path open to those claiming this. I am not one of them.

I still think it is audiphoolery retreating into an area not easily testable similar to a God in the Gaps logical fallacy.

 

[Cosmik]

Well just because you can't conceive of how something is so, doesn't automatically prevent it from being so...

But on that basis, you have to investigate every idea that the audiophiles dream up: cryogenic treatments, green pens, magic stones, quantum phono plugs.

Progress is only possible if we accept certain things as 'axiomatic'. I would suggest that one of them is that if a noise is inaudible in isolation, it will be inaudible in the presence of a much louder signal. If you try to show otherwise using an elaborate listening test, someone can come along and suggest that a computer monitor in the next room created electromagnetic disturbances that were different depending on the WAV file name it was displaying and it was *this* that people's brains were detecting when they gave you your positive result. You might counter with "There is no mechanism for human brains to detect electromagnetic activity from monitors". And they will counter with "Just because you can't conceive of how something is so, doesn't automatically prevent it from being so". Back to the drawing board!

And the same with a negative result. Someone will always be able to find something that you didn't control for and which may have spoiled the experiment. Starting with "It is not possible to show that a person who knows they are in an experiment loses their ability to discern minute changes in sound; changes that they do notice when listening for pleasure".

 

[Jakob1863]

@Cosmik ,

main difficulty in our field is the missing axiomatic, because it is strictly based on observation. (Mathematics is the only exception therefore able to really proof things. But we know that even within those axiomatic systems theorems exist that can´t be proofed) We have some knowledge about the physiology of our hearing mechanism and neuroscience provides some more insight about transfer of signals to and processing in the brain. But the combination of hearing mechanism and processing brain is a nonlinear system with some feedback loops build in.
Up to now there exists no model of the system that fits to reality sufficiently, therefore we still have to rely on perceptual evaluation.

Most things we know about our hearing sense were found out with controlled listening experiments, but usually done with just small samples. Statistics tells us to be very careful to analyze the results and to draw further conclusions about the underlying population distributions. And quite often a newer experiment finds considerably different results compared to earlier studies.

From a scientific viewpoint it is absolutely valid to say a novel hypothesis about the audibility of effect "ZSX" does not fit to the known model and to request real evidence before beginning to further investigate or even accept it.
But to dismiss it right from the beginning because it does not fit the current state of knowledge isn´t a valid approach at all.

 

[Jakob1863]

I am not sure if you are saying that makes the test bad or good. <snip>

Neither nor...

The "monophonic" part of their experiments was the contradictionary element compared to grimm´s assertion, while in the case of Benjamin/Gannon it was the "tone like" element.

Wrt the question about if it matters if the DAC already masks it; well it depends on the hypothesis/question to investigate. If i´m trying to find out if there is something to grimm´s assertion i´ve to ensure that my equipment is on par for the task.

 

[Cosmik]

From a scientific viewpoint it is absolutely valid to say a novel hypothesis about the audibility of effect "ZSX" does not fit to the known model and to request real evidence before beginning to further investigate or even accept it.
But to dismiss it right from the beginning because it does not fit the current state of knowledge isn´t a valid approach at all.

I wish you the best of luck with that. I'll start you off with "Deep Cryogenic Treatment of the cable increases the depth, clarity and timing of the music." When you're finished with that, I have a whole bunch more

 

[Jakob1863]

Let me clarify that when I said "modicum" of effort I meant for people in the industry like Bruno. For everyday person unless someone has already created the files for you, then it is impossible to do. But the claims we are examining are that of the experts put forward. They seemingly have time to go and build such circuits but can't or won't spend a day or to in Matlab to create different profile jitters to test? Or performed controlled listening tests that are documented? Both tell me that all they have done is sighted testing and with the strong bias of what they think the effects should be, they "heard" the same thing. They should know better -- certainly a lot better than putting such things in AES papers.

As said before, i really appreciate all your posts with informations about the papers/test instruments/theory etc. and it surely helps a lot of readers to better understand the matter, but i know it and i´m quite confident that the people at Grimm audio, at least Putzeys/Grimm knew it too.

And, as said before, controlled listening tests should have been done to gain some evidence on the matter. I simply don´t know if the grimm people did some controlled tests or just sighted listening and afaik they did not put anything in the AES paper. The pdf i´ve quoted from was just a techpaper from grimm.
I´ll try to find out .....

 

[Jakob1863]

I wish you the best of luck with that. I'll start you off with "Deep Cryogenic Treatment of the cable increases the depth, clarity and timing of the music." When you're finished with that, I have a whole bunch more

But, why should i? I´d simply request further evidence before beginning to ......

On a personal level things are a bit easier (or not?) . Axiomatic or not axiomatic, should i do everything (means copy/buy/do) because somebody somewhere and somewhow could detect something? Or should i still better try to figure out if it matters to me. Isn´t is the same with nearly every somewhat luxury thing people are able to be fanatic over it? Everyday experience told us (usually) that you can´t trust in everything people are telling you, but otoh that you can´t be an expert in everything (and even experts learn something new) and therefore has to figure out whos advice is more trustworthy (humans behave erroneous but not always) and meets your personal preferences.

Overall the "grand debate" is running for roughly 40-50 years (or even longer, as the term "hifi" was coined ~1927 and the term "audiophile" a couple of years later) and engineers quite often said at their time that humans can´t hear "this" ; 10 - 20 years later a newer generation of engineers quite often says "of course the gear from the past wasn´t that good" but today you can´t hear "this" . Roughly 10 - 20 years later engineers quite often say......

Sometimes they were right, sometimes not. Exaggeration does not help; audiophile generalization does not help but nonaudiophile dismissal doesn´t too.

 

[Cosmik]

But, why should i? I´d simply request further evidence before beginning to ......

On a personal level things are a bit easier (or not?) . Axiomatic or not axiomatic, should i do everything (means copy/buy/do) because somebody somewhere and somewhow could detect something? Or should i still better try to figure out if it matters to me. Isn´t is the same with nearly every somewhat luxury thing people are able to be fanatic over it? Everyday experience told us (usually) that you can´t trust in everything people are telling you, but otoh that you can´t be an expert in everything (and even experts learn something new) and therefore has to figure out whos advice is more trustworthy (humans behave erroneous but not always) and meets your personal preferences.

Overall the "grand debate" is running for roughly 40-50 years (or even longer, as the term "hifi" was coined ~1927 and the term "audiophile" a couple of years later) and engineers quite often said at their time that humans can´t hear "this" ; 10 - 20 years later a newer generation of engineers quite often says "of course the gear from the past wasn´t that good" but today you can´t hear "this" . Roughly 10 - 20 years later engineers quite often say......

Sometimes they were right, sometimes not. Exaggeration does not help; audiophile generalization does not help but nonaudiophile dismissal doesn´t too.

What more evidence do you need? Audiophiles spend £621 on a cable because cryogenic freezing makes it sound better! I could easily spin some yarn to do with crystal stresses and alignments, don't you worry...

Going back to jitter: I am saying that if a known error due to jitter was calculated to be well below the standard threshold of hearing (established in the past using listening tests), then there would be no point in doing a listening test with that error played alongside some music. If we do think it is worth doing an empirical experiment, it would appear that we are casting doubt on the listening tests that established the threshold of hearing..? If we cannot build upon earlier science, why should anyone rely on our proposed new listening test? What use can it possibly be?

 

[amirm]

Here is a great example of what I talked about earlier. That is, high-speed engineering tools are inappropriate for audio measurements due to horrific dynamic range. Jkeny on WBF just put this tektronix scope forward as a badge of honor with respect to measurements. Here are its specifications for different models:

Notice the last line: the effective dynamic range is represented by just 6.8 bits at 500 Mhz sampling! And keeps going down to just 5.6 bits. When you are whaling at 0.5 Ghz to 3.5 Ghz, you lose accuracy in the analog to digital converter. Of course you get tons of bandwidth. But for audio, we don't need that bandwidth. We just need to know what happens in audio bands that is just 0.00002 Gigahertz. That "slow" speed enables us to use 24-bit converters which achieve about 20 bits in real life.

We use that incredible dynamic range in frequency domain which in turn allows us to go way, way deep into the noise floor. Something that is impossible to do with the above scope.

 

[Jakob1863]

What more evidence do you need? Audiophiles spend £621 on a cable because cryogenic freezing makes it sound better! I could easily spin some yarn to do with crystal stresses and alignments, don't you worry...

Just evidence due to controlled listening tests or opinion of a listener whose listening abilities i know from the past.....

Going back to jitter: I am saying that if a known error due to jitter was calculated to be well below the standard threshold of hearing (established in the past using listening tests), then there would be no point in doing a listening test with that error played alongside some music. If we do think it is worth doing an empirical experiment, it would appear that we are casting doubt on the listening tests that established the threshold of hearing..? If we cannot build upon earlier science, why should anyone rely on our proposed new listening test? What use can it possibly be?

In general basis of scientific knowledge should rely on replication/repetition of experiments, but is unfortunately seldom done in the audio field. So, casting doubt on earlier results isn´t per se an evil thing.
Furthermore, the assertion written in the pdf from 2006) was, that extraordinary low frequeny jitter (wander) of low level could provoke detectable differences in stereophonic reproduction of music (instead of test tones).

So which standard should be taken into consideration? Is there one? As said before, the earliest experiment i know about was done by the BBC in 1974; the guys used music but in _monophonic_ reproduction. They found that below ~500 Hz of jitter frequency the level got enormous and referred - as amirm already pointed out - to a study about the effect of wow/flutter dated from 1955 (i have to reread that as i can´t remember if they were using a stereophonic system). The other was the Benjamin/Gannon study from 1998, but they used test-tone like music and jitter with quite higher frequencies (~1750 Hz).

It´s not surprising as they used roughly the same model that Julian Dunn´s estimation was based on before.
In addition only a few participants were involved and if you want to establish thresholds you have to draw conclusions from the small sample to the huge population existing. You´ll be warned not to do it in statistics, but if you do it despite the warning you have to estimate a distribution (in most cases due to the variance in the sample) that is shaped like a broad normal one. The one percent tails reach far out on each side of the mean.

Threshold experiments are about probabilities not about switching from inaudible ot audible like a staircase.......

PS. really large sample sizes were used by Licklider in an experiment in (?1952?) about hearing thresholds of adults, they examined afair about 22000 humans; codec listening tests are another example of larger samples used.

 

[DonH56]

We have the bigger brothers of those scopes. How ENOB translates to real-world performance is complex and there is a lot of processing behind the ADC as well as a good analog front end (AFE) before it. That said what you see on screen in terms of dynamic range and resolution is ultimately limited by the ADC, though of course you can oversample and process if you don't need the full bandwidth.

For time jitter, what matters is timebase accuracy and stability, aperture error, and all that jazz, and it is very good in a good wideband 'scope. That is why we can measure jitter well below 1 ps on a multi-GHz carrier with our 33 GHz/100 GS/s DSO. There are also all sorts of averaging and processing going on over the 50 M or so samples we capture. (Sweep, wait, wait, wait, results appear, do it again...) So you can probably use one of those to measure jitter.

So you can use one of those DSOs to measure jitter, though typically to get precision jitter you pay dearly for an optional package, and can gain decent resolution through processing, but it ain't going to show you 100 dB dynamic range without help. At least I don't think so; I have never actually used one of our DSOs for audio measurements. Something about hauling a piece of test equipment home to play with, when it costs more than my home, gives me pause when I have to sign it out... For high dynamic range measurements we use different instruments, and those are generally more narrowband frequency-domain types of test equipment.

 

[DonH56]

It may be worth the reminder that, for random jitter (RJ), the degradation in SNR related to RJ is a function of the signal frequency, not the clock. The sampling rate (clock frequency) drops out of the equation in the analysis so SNR depends upon signal frequency, signal amplitude, converter (ADC or DAC) resolution, and jitter amplitude. Clock frequency is irrelevant (theoretically). The derivation is straight-forward and can be found many places. It has a nice closed-form solution that leads to the famous SNR = 6.021*N + 1.8 dB for an N-bit converter. The SFDR, alas, leads into Bessel functions that make my brain hurt (being a simple hairy-knuckled engineer) and has no closed-form solution.

FWIWFM - Don

 

[amirm]

Posting this graph as another example of very high jitter in analog turntables. This is from Hi Fi News:

We see random jitter around 3.15 Khz that rises almost to the peak of of our center frequency for digital.

Here is another example of one of the worst DAC measurements I have (Yamaha AVR using its HDMI input):

This is at 12 Khz. Scaling this down to what would be at 3.15 Khz which was used for the turntable would further shrink the random jitter area as to make it needle sharp compared to turntable.

Every way we measure, random jitter is far better for digital versus analog. So if analog folks are not complaining about this type of jitter, then it is not an issue for digital.

 

[The Smokester]

...Every way we measure, random jitter is far better for digital versus analog. So if analog folks are not complaining about this type of jitter, then it is not an issue for digital.

I appreciate that (most) DACs' performance deviations are below measured human hearing limits. However, I wonder if DAC and turntable frequency distributions are really of the same character.

Given my ignorance (lack of ability to measure), let's hypothesize that the turntable deviations are due to variations in the rotational velocity of the platter. Then the deviations will be highly correlated because the inertia of the platter won't allow sudden (and random) changes from one end of the distribution to the other. Two sequential measurements (done at or above the Nyquist frequency) can't be very far apart. So they are correlated. Also, let's hypothesize that the two channels will be exactly correlated in frequency so both see the same shift in frequency at any given time.

If these are true, won't the stereo image remain fixed? Only an (probably) inperceptable pitch change will occur.

Perhaps the turntable measurements have been averaged so correlations, if they exist, aren't seen.

Anyway, I can imagine that the correlations in DACs are quite different (particularly in dual-DAC systems) and that could, in theory, effect stereo balance differently.

Still, please reread first sentence.

 

[amirm]

Given my ignorance (lack of ability to measure), let's hypothesize that the turntable deviations are due to variations in the rotational velocity of the platter. Then the deviations will be highly correlated because the inertia of the platter won't allow sudden (and random) changes from one end of the distribution to the other. Two sequential measurements (done at or above the Nyquist frequency) can't be very far apart. So they are correlated. Also, let's hypothesize that the two channels will be exactly correlated in frequency so both see the same shift in frequency at any given time.

At macro level things seem that way. At micro level though, the mechanical system is highly random and unpredictable. Let's think of variations in the ball bearing with lubricant not being distributed evenly. As the platter spins, there are now tiny variations due to friction of the bearing for the spindle being modulated that way. Let's say you have a belt drive. That belt at moment to moment will stretch and relax. Some of these will get filtered out due to inertia of the platter.

Regardless of the reason, these show up in measurements as I have posted. And the skirts around the main tone show them to be random due to highly unpredictable nature of them at micro level. At macro level, yes, you are right that we don't perceive any sudden changes. Our instrument is just very sensitive relative to our ears in discovering the effects of tiny and random variations of speed.

If these are true, won't the stereo image remain fixed? Only an (probably) inperceptable pitch change will occur.

Our perception of stereo imaging doesn't have this kind of resolution. We don't need it in nature/need to survive. We don't triangulate location of a prey for example down to 0.00001 degrees. We just know it is at one o'clock for example.

Perhaps the turntable measurements have been averaged so correlations, if they exist, aren't seen.

They are not averaged per-se but measured over a duration. Same is true of digital clock measurements. We can't tell what that clock is doing in one sample. We have to let it run for hundreds and thousands of samples to measure its distributions. What we see there is occuring at atomic level just like the micro perturbations I explained for analog above.

Anyway, I can imagine that the correlations in DACs are quite different (particularly in dual-DAC systems) and that could, in theory, effect stereo balance differently.

Still, please reread first sentence.

I have . Please see my next post.

 

[amirm]

A few years ago I did a test of audibility of clock jitter using an Audiphilleo USB to S/PDIF converter. This unit came with a convenient switch that changed the clock jitter/noise profile as such:

Blue is its normal, high-precision clock. The one in pink is the degraded "jitter simulator." We see that above 1 Khz or so, the two clocks already have phase noise/jitter that is below -100 db so what happens above that is inconsequential from audibility point of view. As we get lower though, we heap on some 40-60 db of noise.

I did a listening test first sighted. The high-precision clock sounded smoother, warmer, more analog like.

I then closed my eyes, randomly switched back and forth and then tried to identify the better clock. No luck. All the difference I thought I heard, were no longer there.

I then took the test to work and tested two other people. They had no perception of difference between the two and gave me a blank look when I asked if they heard an improvement.

This of course agrees with psychoacoustics (and perils of using sighted tests). Close-in jitter is subject to very strong masking and hence highly inaudible.

Mind you, there may be an audible difference there but relative to the effort that I put in there, it shows no such problem.

Furthermore, the "good clock" in audiophilleo is still much worse than the clocks advocated that are subject of this thread. I use that unit all the time in my music and it doesn't display any of the audible ills that people say such close-in jitter has. Great music sounds superb and as good as any DAC I have heard anywhere at any room, any location.

I have a better unit (Berkeley) but can't use it due to incompatibility with Dirac Live EQ.

 

[amirm]

And this graph that shows contributions to phase noise in clock oscillators.