[art]

Here are some jitter measurements, that some might find interesting. I am not going to try to explain every nuance, as that is not the point. The point is to show there are many ways to characterize jitter, and no single one tells the story. As I have said, many times, over many years, on many sites, jitter numbers without both amplitude and spectral content, are just numbers.

The folks who design gear can use any of the many methods, to help to refine the design process. From my standpoint, once the design process is complete, the actual results are of little use. You make a series of refinements, measure their effect, and from that, make a decision, on how to progress. Other than that, I don't get hung up, on the specific numbers.

Having said that, there are subtle points that such measurements can reveal. And that is what I want to bring to your attention. So, here is a set of histograms, that will display these points.

First, it is important to understand one key point: the type of jitter. This is really more important, than the amount. This breaks down into two different types.

Gaussian jitter is just what it sounds like: random noise. It really isn't all that harmful, within reason. Some of us believe the real low frequency noise is worse than the stuff above 1 kHz, but let's not go into that. That is not the point I am trying to make. Besides, it is harder to measure. (Won't go into that, either. Just trust me.)

Deterministic jitter is the one we don't want. Basically, something that is not random. Repeatable................predictable.. .......something that has a period to it.

This type can be something innocuous, like mains frequency components, getting into things. Or, worse, it can be data-correlated. This is the stuff that mucks up SPDIF, as it present, in abundance, in the recovered clock. As a result, folks came up with reclockers, and then ASRCs, to get rid of it. (Whether they are effective or not: not the point.)

So, on to the funny looking pictures.

First, we have a design, that we are evaluating. Not going into what design, who it is for, whether it will ever go to market, etc. Again, not important.





The first picture shows the basic design, in the idle state. We can see the 1-sigma jitter is only around 13 pSec. In the context of what is being measured, this isn't too bad. But, you will notice there are 2 peaks to the jitter. Hmm.............you wouldn't know that, if you only had some jitter machine, that spit out a number. Or, that is has 43 pSec of deterministic jitter. And how the total jitter has gone way up. Now, if this was a recovered clock, this would be good. But, since this is a source, we might want to look deeper. The random jitter is really low, so there is a good chance it has a decent clock. Just some other circuit problems.

The same unit, but with music playing:



YUK!

The random, deterministic, and total jitter, stay about the same. But, there are 2 distinct peaks, and the 1-sigma goes up. If I were a betting man, I would bet this unit could be helped by reclocking the output.

OK, not fair..............I know what the answer would be. So, for everyone else, here it is:



Well, isn't that interesting? The random and 1-sigma are down to under 5 pSec. That is good. And the deterministic is gone!

So, let's play some music, and see what happens:



Those last two numbers go up some, and the total jitter is about the same as the stock version. But, there is no deterministic jitter. So, I would conclude that the reclocker did its job.

Ok, the punch line, if there is one.............

The SPDIF TX chip, in this design, has a built-in reclocker. Some folks, and I won't say who heads up this faction, will go "Ha, ha! Told ya so! Those guys don't know any more about SPDIF TX, than they do RX. And are clueless on how to prevent jitter." The other camp will go "Well, that is just your opinion. I would take theirs, over yours. They say we don't need one, and we are going with them."

(Yeah, but you still pay us to make measurements, don't you!)

I'll let you guys decide whether or not the internal reclocker does its job, or not. I think you know my answer.

OK, putting all that aside................one unit, two iterations, some jitter numbers the same, others wildly different. So, which one do you accept? Which one, if any, goes on the data sheet? And, what, if anything, does it mean?

Well, unless you know all of the parameters, and how they came about, it doesn't really tell you anything. Except some look better in print, than others.

And none, repeat, none, will tell you what the output of your DAC will look like. But, someone else can hook up some machine, get some numbers, put them in a magazine, and you know what...............

You still won't have the answer.

What you measure, how you measure it, the format the answers are in, what is in the answers you get..............can't reduce that to a number in a brochure. The bottom line is getting hung up on jitter numbers, numbers than none of us can even agree on how to measure and quantify, are just numbers.

Stop worrying about the numbers. Do any of you buy an amp, only if it has less than 0.00392% THD? No, I don't think so. Ditto for jitter. Look at the pretty pictures in the magazines, and other than going "Man, that one really looks bad", don't lose sleep over it. Jitter, like distortion, will always exist. Some kinds are worse than others. And no single number can give the answer.

Oh, before I forget...............here is a pic, from some unit that..............well, we'll just call it a reference. You decide how good of a reference it is.



I had hoped to be able to post some nice shots, of what reflections look like, and how important they are, but that blankety-blank autofocus did not work, and the manual focus isn't much better. So, that will have to wait.

Until then...............happy listening!

One last thing............just happened to think it might be important to mention what exactly what is being measured. Obvious to me, but I do this, all day long.

"Yeah, that would help!"

OK, this is the rise time, of some SPDIF gizmo. Rise time, under 1 nSec, safe to say it is some kind of LCV logic chip. Keep this in mind, for the next set of pictures. (Not sure when I will get time to make and post them, but rise time is important, when you start to worry about SPDIF cable length.)

Pat

[jtwrace]

Great stuff! 

Will read it later tonight.

[TRADERXFAN]

Thanks. I learned quite a bit.

[Chris Adams]

Pat, Thanks for that info. Don't know if I understand all of it, but intriguing and I'm curious to read more.

[art]

Don't worry. Most of the folks I do this kind of work for don't understand all of it, either!

But, as long as you understand this is a complex subject, that can not be reduced to one number, you are ahead of the game. That is what I am trying to convey.

Pat

[Quiet Earth]

I'm happy to see that the entire digital audio community is not obsessed with the word "jitter" and some number assigned to it which qualifies good sound or bad. I was really starting to wonder there for a while . . . .

Thank you for standing up for common sense Pat.   

[art]

Oh, dear................this will probably add to my enemies list.

OK, here is a histogram of something that I have a lot of respect for. Not saying what it is, but for what it does, at the price it does it, and what it accomplishes, it has earned that respect.

Just one small problem........

When you make millions of something, sometimes you have to worry about that extra penny it would cost, to put the clock and the SPDIF chip, in one chip.

Yeah, can you say "deterministic jitter"? (Done with it playing some music.)



Luckily, for all of you DIYers, you know how to fix this.

(And get rid of that ferrite bead, on the output, when you are in there!)

OK, I know someone is going to freak out when they see that >500 pSec jitter number. So what? Forget about it.

"WHY?" It looks awful!"

Yeah, well..........before you get carried away..........

Let's go back to how this is derived.

The uber-expensive gizmo we use has a s/w package, that calculates the amount of Gaussian jitter................how much is deterministic, etc., blah, blah. And guess what?

It isn't honest.

Well, let me clarify that.......

it doesn't lie, it just gives you answer, based on what you put in.

So, if you run one of these tests, for too short a period of time, it can give some awful numbers. Or, put another way, you have to run it a long time, to even get something believable.

Like this:



Same piece of gear, but idle. Does anyone notice the deterministic jitter went away?

Well, not always.

Yep, depending on how long I run the test, I get a number anywhere from ZERO, to >25 pSec.

So, which one do you believe?

As a designer, I would believe the fact it shows deterministic jitter, when it is playing music, as something would need to be addressed, in the design process. The fact it does exist is more important than how much it is. Because, not only is the number meaningless, without context, but it can way off, depending on how you handle things.

I suppose a sleazy vendor, who could afford one of these, could massage the data, to give the answer that makes the competition look bad, and he wouldn't be lying.

Just not 100% honest.

So, numbers................yeah, look good in print. Best to minimize the value, but as parameter to decide what to buy.............yeah, not all that helpful.

Anyway..........

Also, you will notice the rise time around 4 pSec...............yeah, HC logic.

Pat

[art]

I'm happy to see that the entire digital audio community is not obsessed with the word "jitter" and some number assigned to it which qualifies good sound or bad. I was really starting to wonder there for a while . . . .

Thank you for standing up for common sense Pat.   

Well, actually, I am obsessed about jitter. Which is why we managed to acquire one of these uber-expensive gizmos.

I just know what to believe, what not to believe, the why or why not, and more importantly..........

How to get rid of it. And when to ignore it.

Oh, one last thing...............

I wrote about this, a few days back, on some other forum............

That number, that raw number...............that magic 2 nSec number, you hear about that "You can't hear jitter below that level...............". You know, that number.

Well, that number is valid at 44.1 kHz. IOW, word clock. If you translate that up to bit clock (which is 11.2896 MHz, in a lot of stuff I work on), the same amount of jitter is less than 8 pSec, of jitter.

Yep! Unless you also know the frequency of the signal, that you are measuring.....................guess what...............jitter number means even less.

"You mean that as the frequency of the signal that we are trying to measure the jitter of goes up, the jitter number goes down?

Yep.

And, then....................the icing on top.................

Most equipment only measures this stuff, down to a 10 Hz level. Below that, if you let the LF stuff in, the jitter number goes through the roof.

And guess what...............all you DIYers, spending $$$$$ on fancy clocks.............look at the fine print:

Jitter <5 pSec, for jitter > 1 kHz.

Translation: with a 1 kHz cut-off, anything looks good.

Take the same 44 kHz, with 2 nSec..............kick it up to 11.2896 MHz (now 8 pSec), and limit the low end to 1 kHz...............

Only around 0.2 pSec.

Yeah, jitter numbers. Good on paper. Worth about as much as that other kind of paper!

I think Bob Pease is the guy who gets credit for this:

"What is in the data sheet is not as important as what is not in the data sheet."

Translation: They ain't gonna tell anything bad about their stuff! Only the good stuff. The rest, yeah, they don't want you to know.

Goes double, if not triple, for jitter numbers.

Pat

[art]

I'm sure it will be a matter of time, before someone accuses me of some self-serving garbage.

So........

Ok, yeah!

Hardly a day goes by, without having to answer e-mail that starts out with "I WANT THE LOWEST POSSIBLE JITTER, AND...................."

And, you don't know anything about what you are asking about.

"I don't want to have to use a BNC adapter. Why can't you make a cable with RCA?"

Well, because if you are really that worried about jitter, caused (or made worse) by reflections, then you wouldn't use a RCA, anywhere. Look, it doesn't matter: a RCA or a RCA adapter, they do the same thing! The only way around it is to not use one. Period.

"I want to put a BNC into my DAC. What one should I buy? I don't care how much it would cost."

Well, that's nice. Too bad you are probably going to hook it up to the twisted pair, that is presently in your DAC. No mention of "Do you know of a small 75R coax, that I can use?" No, just the usual audiophile neurosis about buying the most expensive BNC, and expecting it to be a silver (well, actually, golden) bullet.

Then there is that lousy pulse transformer, at the end of that twisted pair. And it goes right into the RX chip, without any sort of buffer stage.

And you still want me to believe you really care about the lowest possible jitter?

Don't make me laugh: I am good at cracking myself up. So, don't need the help.

Look, I am not going to lie to you: all of this stuff does matter. If it didn't, we could not sell a darn thing. Paying attention to small details is what takes "blah" to "Hey, this sounds good!" But getting hung up on some concept, that, frankly, the layman has no knowledge of, is self-defeating.

You are never going to get rid of jitter. It will always exist, in the Gaussian form, in every thing, in the chain. It will start at the clock, and build up, along the line. You can not get rid of it. Even reclockers and ASRCs will have their own inherent jitter. OK, some (or most) may get rid of what goes into them, but some will always come out.

As I previously stated, are you going to buy anything, just by looking at the distortion numbers? No, of course not. (Ok, the folks who buy Bose, or read Consumer's Report might, but who cares about them?)

That distortion number does not tell you anything. What is its spectral content? How does it effect a musical note, when does it effect it, and how will you ear respond to it? You have no idea. Nor does anyone else.

Jitter is the same thing. Low numbers good; high numbers bad.

But, depending on how you measure, and characterize it, I can give you a number anywhere from 2 nSec, to 0.2 pSec, for the same piece of equipment! It just depends how I measure it, and massage the data, to give me the best number.

So, having said all of that, here is the bottom line:

If you are really worried that much about jitter, then never use SPDIF!

If you want to use SPDIF (and we want you to, or else we don't have much to sell), then you are going to have to learn it will always be there, and quite possibly more than you want. No matter how badly you want it to go away.

How much will there be? How much will it muck things up? Not something that can be answered, in 5 words or less. Might as well ask me what speakers will work in your room, based on whether you like thin or thick crust pizza.

Now, there are ways to hopefully mitigate the effects of jitter. Speaking in broad generalities:

RCAs are bad. The spec calls for RCAs, so expecting the industry to change is a futile effort. You are going to have to look at small, niche companies, that might not be in business, a year down the line. Pick your poison.

As bad as SPDIF is, the AES/EBU version is worse. Just because they use it in the Pro Audio world does not make it better. The folks who thought up that variation know as little about RF as the guys who thought up the SPDIF, in the first place.

(Actually, it was originally intended to only be a test port. Then, the marketing boys found out about it.)

Optical is worse! TOSLINK, for obvious reasons. Other than hooking up your HT receiver, so you can get a signal into it, and you aren't worried about how the kid's DVDs will sound, it is useless. ST optical has the potential to be good, but the way it is implemented, it isn't.

OK, a bit o/t..................

Early 90s, I remarked to my bidnis partner, that it would only be a matter of time, before the magazines discover jitter, and make a big stink of it. At that point, DAC sales will plummet. A few years after that, the final nail in the coffin will be when they figure out ST glass is worse that coax.

Yep, I turned out to be right, on both counts. Fast forward >15 years, and everyone forgets about that, and the DAC market heats up again. Folks start dusting off their old ones, and want to breath new life into them. Greedy capitalists (like us!) seize the chance, to make a $ or two.

And then everyone remembers that stuff called jitter, and the game starts over. Only this time, we have outlets, such as this, to help and/or muddy things more.

You decide.

Pat

[mfsoa]

As I previously stated, are you going to buy anything, just by looking at the distortion numbers? No, of course not. (Ok, the folks who buy Bose, or read Consumer's Report might, but who cares about them?)

You would be suprised - I have ask the question that if you listened to amp A and amp B, thought amp A sounded much better but had orders of magnitue more distortion, which amp would you buy?  B was the answer. 

[werd]

I think if you measure anything in the pico timelength, doing  so in any facility other than in this caliber lab http://www.lightsource.ca/ would harbor nothing more then skewd results. I see a lot of  digital measurement talk in the pico range. This happens to be the same range of time that are being measured by these folks in an atom collider experiment. 

After taking the tour of that building (live hop skip and jump away from it  )and learning about the extreme efforts invested in that building to preserve the outcome of the results makes me suspect about a lot of the testing i see being done. I think it becomes guess work after a certain point and probably not to far off into micro area of time measurement.  I am talking about the equipment itself and not the people conducting it.

Could be wrong but i am very skeptical about the results of jitter experiments but i do believe jitter is important and audible.

[satfrat]

Not something that can be answered, in 5 words or less.
 
You decide.

Pat

That's 6 words Pat,,,, 

[werd]

You would be suprised - I have ask the question that if you listened to amp A and amp B, thought amp A sounded much better but had orders of magnitue more distortion, which amp would you buy?  B was the answer. 

I would probably buy B and configure my gear for a sound stage that accompanies lower distortion benefits, like quicker transients and better leading and trailing edges. But i would keep amp A around if it all went to shit and it didn't pan out ..... 

[art]

I think if you measure anything in the pico timelength, doing  so in any facility other than in this caliber lab http://www.lightsource.ca/ would harbor nothing more then skewd results. I see a lot of  digital measurement talk in the pico range. This happens to be the same range of time that are being measured by these folks in an atom collider experiment. 

I don't know how you come to that conclusion. Our equipment is spec'ed to do better than 3 pSec, and was calibrated, traceable to NBS.

Any skewing, in our results, is far more likely to be due to operator head space.

If you think 3 pSec is that hard, let me set you straight. Our gear uses an internal 100 MHz reference. With that level of uncertainty, we can roughly deduce the phase noise of that reference, is in the area of -75 dBc, or so, at 10 Hz offset. Which is good, but not great. We can produce a clock, using cheap parts from Digi-key, that does around -90 dBc, at 10 Hz.

Indeed, if we try to use the phase noise conversion, of the s/w, it gives an answer in the -70s. Too bad it also has a linear horizontal scale. So, useless, on 2 layers.

But, it confirms our guess, as to the quality of the internal reference.

You may want to rethink your stance.

Pat

[art]

I don't know about where you live, but arguing with the teacher is a good way to get excused from the class.

This is why I hate posting stuff here. Always someone who wants to stir the pot.

Class dismissed. Over and out.

Pat

[werd]

Carry on ........ Sorry

[stormsonic]

Respect, General. Thank you.

LCV or LVC? 

[wushuliu]

Bumping up an oldie but goodie.

[art]

Oh.............good...............grief .........................someone dredged up a 7-year old topic.

Guess I shouldn't complain, but my inbox will explode as a result. Maybe I will assume a low profile for a few days.

"You will complain, Oscar, you will. It is what you always do."

True.

Anyway, a lot has changed in those 7 years. We still have that piece of equipment, but rarely use it. Great for making histograms (like above), but really poor for phase noise. Which is another way to describe jitter, in a manner that is preferred by RF engineers.

Here is how we do it today........................



That is the phase noise of an SC-cut crystal. Which is about 40 dB better than the average clock in your garden variety piece of hi-end audio gear.

BTW..................it is operating without an oven. For those who say it can't be done. (Also at the wrong temperature, as well!)

OK, gang...............you know where to find me. I guess.

Enjoy!