[art]

Hey gang:

I am really busy, working on getting a proto up and running, so I have not had time to follow this thread. So, let me quickly comment on this one post. Don't know when I will get time to check back, so be patient.

(I will be checking my e-mail, so if anyone really needs to get hold of me.......)

Please read this article, if you have not done so already -
http://www.stereophile.com/features/368/

What I see there makes no sense. The two graphs that show jitter - Figures 2 and 3 - show the sidebands at very high levels around -60 and -50 dB respectively. If the jitter really were that high it would be clearly audible and highly objectionable. I cannot accept that any CD player or any other digital system could have such high levels of artifacts, even very early players from the 1980s. At -50 that would be worse than the hiss from a pre-recorded cassette, and we all know that CDs are infinitely quieter than any cassette.

The data in the Stereophile article is also wrong because the stated jitter levels for those graphs are listed as 145 picoseconds and 561 picoseconds. Contrast that with Figure 4.28 in Ken Pohlmann's book Principles of Digital Audio (4th ed.) where he shows the spectrum for 2 nanoseconds of jitter as being below -120 dB. Now, Pohlmann's example is 2 ns peak to peak, where Stereophile uses RMS. But even still, Pohlmann's jitter is four to 14 times higher than Stereophile's, yet the artifacts are 60 to 70 dB lower. So ether Pohlmann doesn't know how to measure jitter properly, or Stereopile doesn't know how to measure jitter properly. Guess who my money is on.

--Ethan

Ethan,

Anyone who has worked with me knows that I have little use for Stereophile. I do not have time to research the reference that you cite. But I can tell you that from >35 years of either telecom or audio, that if think that jitter of 2 nSec is around -120 dB, then no wonder you don't seem to understand a damn word I am saying.

When I measure jitter, I measure the jitter on the 256 x Fs (or 384 x Fs) clock that feeds the DAC chip. 2 nSec of jitter is one helluva lotta jitter. Period.

I don't know (or care) how Stereophile does it. I can only assume that your reference uses some totally different scheme.

The people who design oscillators/clocks usually use my method. That is what it common in the RF world. Frequently, we use measure something called phase noise. I can assure you that 2 nSec of jitter would be tons of phase noise. The only clocks that have that much are ones derived from SPDIF sources. I do not know of how anyone could make a stand-alone oscillator with 2 nSec of jitter. Unless they really tried hard.

Typical stand-alone oscillators have jitter numbers in the single pSec range. Part of that depends on the range over which it is measured. Harder to get good numbers very close in (the single Hz range), as crystal Q is the big part of that function. Really good crystals cost $$$$$$.

All for now.........gotta run.

Pat

[kyrill]

If a default clock in a transport is replaced by a high precision clock and the pws ONLY for that crystal is replaced by a regulated more stable pws, will the effect on the signal mostly be jitter related?

I hear and heard tremendous positive differences in the musical signal even with better than mainstream transports. ( Teac VDRS, Naim)

the differences is not mostly on noise or noise artefacts.

I do think that a more precision clock mostly affects the jitter, so a better  clock primarily lowers the jitter. The difference is not subtle at all. ppl who can't hear it I dont understand

They believe weird enough their eyes ( scope) better than their ears, how can that be when listening is the ultimate end product?

Is in our paradigm an emphasis on :  "what you see is true" what you hear ( probably) not? Will those ppl not seriously hurt themselves in the audio business?

[AphileEarlyAdopter]

So, your opinion is jitter manifests solely as noise artifacts without any effect on primary signal?  Is this scientifically accepted?  I don't know the answer.

Good summarizing question.
BTW, I dont have the time to look into the graphs and numbers today. To me, the whole thing was a 'no brainer' because I heard the effects of jitter, just by changing the digital cables 10 years ago. Anybody, who uses a digital connection can check this out, unless, your DAC or receiver already has high enough jitter, that you dont notice the newly introduced interface jitter.

[*Scotty*]

The there is little the audio consumer can do to directly reduce the level of jitter in a transport-DAC system using an SPDIF interface. As Art stated a few pages ago,SPDIF is inherently flawed. I mentioned earlier in this thread a procedure involving replacing the crystal in stand alone DACs,CD and
DVD players with a VCXO which is immune to the destabilizing effects of the DAC which it provides the time base for, this mod will directly impact the amount of jitter the in the DAC.  Most affordable products we can buy are missing this basic design feature. Of course realizing any sonic benefit from this improvement depends on the output stage not negating your efforts. Obviously the rest of your system should also be a cut above mass-fi or you have wasted your time or money.
Scotty

[Geardaddy]

The there is little the audio consumer can do to directly reduce the level of jitter in a transport-DAC system using an SPDIF interface. As Art stated a few pages ago,SPDIF is inherently flawed.

What about eliminating S/PDIF jitter issues via an I2S converter as offered by companies such as Empirical Audio?  People's thoughts on that?

[woodsyi]

That works too but it's system dependent whereas Pace Car adds a really good reference clock that will slave the transport and the I2S DAC.  To me I heard more details and precision with Pace car then Off ramp. 

[Geardaddy]

That works too but it's system dependent whereas Pace Car adds a really good reference clock that will slave the transport and the I2S DAC.  To me I heard more details and precision with Pace car then Off ramp. 

Understood....I noticed in your system that you listed a multplicity of digital sources.  What has been your experience (particuarly in light of this thread....)?

[opaqueice]

So, your opinion is jitter manifests solely as noise artifacts without any effect on primary signal?  Is this scientifically accepted?  I don't know the answer.

Well, one way to define noise and distortion is to take the signal that's actually produced and subtract the signal that should have been produced from it.  Whatever's left is noise or distortion (the difference being basically that noise is uncorrelated with the signal, while distortion is not).  So with that definition, yes - but it's kind of an empty statement.  Not sure if that answered your question.

One thing I can say is this - if I start with a sine wave and add some jitter with a specified spectrum, I can just sit down and calculate precisely what that will do to the output signal.  I gave a simple example above (in the interests of full disclosure - I didn't check rigorously that this is right, but I'm pretty sure it is):  add some jitter peaked at 100Hz to a digitized pure 1000Hz tone, and when you DAC it out you'll get a 1000Hz analogue sine wave plus two sidebands at 1100 Hz and 900 Hz, and that's all.

But of course in a real system the spectrum of jitter isn't known and is probably correlated with the signal.  Still, the test I proposed would be useful for at least getting an idea of what the thresholds are.

[darrenyeats]

One thing I can say is this - if I start with a sine wave and add some jitter with a specified spectrum, I can just sit down and calculate precisely what that will do to the output signal.

Opaqueice, what you said in your post is based on the fact that any signal (including the example of a pure sine wave distorted by jitter) can be expressed as a sum of pure sine components. Have I got that right?
Darren

[opaqueice]

One thing I can say is this - if I start with a sine wave and add some jitter with a specified spectrum, I can just sit down and calculate precisely what that will do to the output signal.

Opaqueice, what you said in your post is based on the fact that any signal (including the example of a pure sine wave distorted by jitter) can be expressed as a sum of pure sine components. Have I got that right?
Darren

Yes, that's correct.  However I realized there are two subtleties I hadn't appreciated when I posted earlier:

1) If the signal is a pure tone with frequency f and the jitter has peak-peak amplitude A and has a pure tone spectrum with frequency fj, it is true that there will be sidebands frequencies f-fj and f+fj and with amplitude fA, as I said earlier.  However there will also be sidebands at f-2fj, f-3fj, etc., which I hadn't noticed.  Their amplitudes will be down by powers of fA, however (e.g. (fA)^2 etc.) and that's very small for audio frequencies and reasonable jitter.  So that's immaterial.

For a more complex signal (not a pure tone), just use take the Fourier transform to decompose into a sum of pure tones, and then apply the analysis above.  You can do the same for a more complex jitter spectrum (assuming you know what it is).

2) More seriously, the analysis above assumes the DAC does nothing to attenuate jitter:  that is, if the jitter j(t) is defined to be the difference between the arrival times of the leading edges of the words and their average arrival times, I'm taking the output of the DAC to be Sj(t) = S(t+j(t)), where S(t) is the unjittered signal.  So here the DAC is simply using the arrival times as its clock.  Of course all modern DACs use some form of averaging to deal with this - a PLL for example.  That will vastly improve the situation, and really cannot be ignored.   So in that case we should think of j(t) as the jitter in the DAC clock AFTER all the filtering has taken place.

EDIT - oh yeah, one more thing - this is assuming all the frequencies in question are well below Nyquist.  Otherwise the continuous treatment I used there needs to be modified, and there will be aliasing effects.

[Ethan Winer]

Rather, we should focus on jittered tracks themselves with actual jitter levels.

I agree, but I don't know if actual jitter can be added to music in an editor program precisely enough to say with certainty that what was added really is the same as jitter that occurs naturally in A/D/A converters.

So, your opinion is jitter manifests solely as noise artifacts without any effect on primary signal?  Is this scientifically accepted?

Yes, and Yes. See the figure below. Again, the test I propose is to add artifacts that are much nastier than jitter could ever be. I can't see how that is not at least as good as generating actual jitter, and if anything it biases the test toward making the artifacts seem even more audible. So I don't understand the objections to this test. Again, jitter is just a subset of the larger debate of what is audible and what is not. To my thinking, that's the more important issue to settle.

Well, I don't think most hard core "audiophiles" are going to accept this, no matter what.  But for the rest of us it will be interesting.

Right, people with strongly held beliefs will not change their opinion no matter how compelling the arguments. You are also correct that I'm not trying to influence those people because I know it's futile. Some people are convinced they know what they heard, even when it makes no sense logically. Sort of like people who swear they were abducted by aliens. I'm sure they really believe they were abducted! But that doesn't make it so.

My goal is to present what I hope are logical and compelling arguments for the benefit of folks who have not yet made up their mind. By the way, I consider myself a hard core "audiophile" too.

if think that jitter of 2 nSec is around -120 dB, then no wonder you don't seem to understand a damn word I am saying.

Are you saying the figure below from Pohlmann's book is in error?

I don't know (or care) how Stereophile does it.

Me either, which is why I'm hesitant to use tracks from a Stereophile test CD unless they explain in detail exactly how they generated the dither.

Typical stand-alone oscillators have jitter numbers in the single pSec range.

Pat, do you have a formula that equates ps or ns of jitter with a dB level of the sidebands created? Not frequency weighting or correlation which might affect the jitter's audibility, rather just the relationship between time shift and level.

--Ethan

[opaqueice]

Pat, do you have a formula that equates ps or ns of jitter with a dB level of the sidebands created? Not frequency weighting or correlation which might affect the jitter's audibility, rather just the relationship between time shift and level.

I'm not Pat, but in my previous post, the formula I get for the amplitude of (signal-jittered signal)/signal is simply fA, where f is the signal frequency (around 10kHz in this case) and A is the amplitude of the jitter in units of time (so 1 ns here).  That's for level, so (fA)^2 for power.  However that gives 10^{-5}, or 10^{-10} for power.  Your graph shows -80dB - and I assume that's dB of power?  If so my formula is off by 20dB, which is not so good. 

fA is valid when the jitter has a pure tone spectrum - when it has a white noise spectrum that needs to be adjusted.  Perhaps there's a factor of the square root of the bandwidth?  I don't have time to check now.

They report the jitter amplitude using a peak-peak figure.  That's a terrible way to do it, because it's not an average.  The longer the sample is you measure, the higher the peak-peak value will be.  It's like estimating home prices in an area by picking the most expensive one.  If the jitter is gaussian it might not matter much, but for white noise it might really matter.

Another possibility is that the discrepancy comes from the jitter-rejection capabilities of the DAC they're simulating (although that seems to go the wrong way).

[Daygloworange]

Holy crap!, this jitter stuff sure is hard to make...

Cheers 

[BrianM]

I have a really stupid question.

People who download files from Stereophile to hear jitter would need to know that their own system didn't produce any jitter, wouldn't they?  Otherwise they'd be listening to digital files containing jitter, on equipment that was adding jitter of its own while they listened.  Has this been brought up and I missed it?

EDIT: I guess I don't mean download, or, I don't know what I mean.  Play the test CD, whatever.

[bpape]

Holy crap!, this jitter stuff sure is hard to make...

Cheers 

Zactly.  There is no agreement on what it is, what it impacts, or how to replicate it - but there is defninte set in stone opinion on both sides as to whether it's audible or not.

Can't wait for intermission - I need some more popcorn...

Bryan

[Bob in St. Louis]

 

[art]

Pat, do you have a formula that equates ps or ns of jitter with a dB level of the sidebands created? Not frequency weighting or correlation which might affect the jitter's audibility, rather just the relationship between time shift and level.

--Ethan

Here is something that converts phase noise to jitter:

http://www.ibrtses.com/products/PhaseNoise102.exe

(You have to enter the carrier frequency before you hit the "calculate jitter" button. It is part of the equation. Stick 11 MHz in.)

Here is some app notes on phase noise and jitter

http://www.vectron.com/products/appnotes/phase.htm

(That one has a visual display that shows the components that make up phase noise in a stand-alone oscillator. A phase noise plot of an SPDIF clock looks much worse. I may have some photos of the 2 that I can post for comparison, when I get time.)

http://www.vectron.com/products/appnotes/jitter.htm

(Maybe not as useful. Some may find it worthwhile, so I threw it in. BTW, I don't buy from them, so no intention to promote them. Just good app notes to read.)

All for now......

Pat

[BrianM]

Zactly.  There is no agreement on what it is, what it impacts, or how to replicate it - but there is defninte set in stone opinion on both sides as to whether it's audible or not.

Whether jitter (as it normally manifests) can be heard is a matter of unprovable fact.  If someone thinks they hear it when it's present, it's pretty difficult to prove they don't.  If someone is sure they can't hear it when it's present, it's impossible to prove they in fact can.  But since when does there have to be universal agreement about the precise nature of something before one forms a strong opinion about it?  The ones insisting jitter can be heard are speaking from their own experience, and the ones insisting it can't be heard are defining it according to its apparent dB level and what they know the outer limits to be.  Neither side is being unreasonable.  What seems clear is that this falls under the category of tweaks, i.e. something the impact of which you'll never get everyone to agree about.  Debating jitter is like debating interconnects - some change IS unarguably being made to your system when you switch ICs or correct for jitter, and there will always be conflicting scientific and empirical claims about the change.  I agree it's usually not a good idea to set one's opinions in stone.  I'm playing with a couple different ICs right now, and keep going back and forth in my preferences. This stuff is subtle, subjective, and depends on way too many variables to ever arrive at a settlement.

[bpape]

It was a joke Brian.  I'll agree that we don't necessarily have to know WHY we do or don't hear a difference - but that is kind of relevant to try to reproduce identical tracks with and without it for a group of people to listen to - don't ya think?

Bryan

[BrianM]

It was a joke Brian.  I'll agree that we don't necessarily have to know WHY we do or don't hear a difference - but that is kind of relevant to try to reproduce identical tracks with and without it for a group of people to listen to - don't ya think?

Bryan

As long as the group of people are together in one room listening to the same system, sure.  One problem seems to be, unlike garden variety noise, the effects of jitter are epiphenomenal, kinda like harmonic distortion (you're not really hearing anything called jitter, jitter doesn't sound like anything, it is supposed to affect all frequencies all the time, you're hearing its effect, and most likely hearing it more by its absence, etc).  But then again at these low levels, every change becomes epiphenomenal I guess.