ar-t
Audio Loudspeaker
 Offline
Posts: 3
|
 |
« Reply #68 on: January 08, 2011, 09:11:27 pm » |
|
OK.............jitter...............a real long subject.
Jitter is a tough parameter to characterize. To engineers, like myself, jitter has its roots in phase noise. Which, is nothing more than a form of modulation. But, just like any other kind of modulation, unless you know the amplitude and frequency content, it is just a number. Numbers, without context, are just numbers. They look good on a spec sheet, but that is about all they are.
"Don't you give jitter specs for the stuff your company makes?" Do you want us to believe them?"
Uh...........yeah............numbers that look good on a spec sheet. (OK, someone challenged us to a numbers fight, and they lost. But don't tell them that is the only reason it is there.)
But, let's look at how those numbers come about.
Part of the problem is how you measure it, and how it gives the results.
The over-priced gizmo that we use, has an internal 100 MHz clock, as its standard. They spec it as being better than 3 pSec.
Sounds good, right?
Uh, no!
We can calculate the phase noise of that clock (ok, we have to make some assumptions), and converting that value to 11.2896 MHz (44.1 kHz x 256), it isn't that hot. The reference clock would measure around -70 dBc, at 10 Hz offset. (This means the noise, at 10 Hz from the frequency is only down -70 dB.) For a 11.2896 MHz clock, the same jitter number would give us a value of around -90 dBc. (We have some other gizmo, that tells us our clock is in that ball park. But, that is not the point.)
So, the first thing you have to understand is that you also have to know the frequency of the signal that you are measuring the jitter of.
If I told you our clock measured 700 pSec of jitter, you would say it is horrible. But, if you convert that 11.2896 MHz, down to 44.1 kHz (IOW, word clock), that is exactly what you get!
So, when some self-appointed expert tells you that anything under 2 nSec is not audible, and he fails to mention that he is using 44.1 kHz as the frequency that is being manipulated, to determine jitter sensitivity, set him straight. 2 nSec works out to be -80 dBc, for a 11 MHz clock. Which, is a decent clock.
Sort of.
"Whaddya mean 'sort of'. Explain."
When you are measuring phase noise, or any other form of modulation, you have to know the frequency spectrum. The freeware we use, to convert phase noise to jitter, only goes down to 10 Hz. Which is the lower limit on a lot of test equipment.
Actually, some equipment doesn't even go that low. A lot of oscilloscopes, that have built-in jitter measurin' abilities, only go down to 12 kHz!!!!!!!!!
Why? Because that is what a lot of telecom stuff is spec'ed for.
If you look at the companies that make oscillators/clocks, you will see, in the fine print, if you look really close, because they print it really small, it will say that jitter frequency is > 1 kHz.
Well, jitter measured only down to 1 kHz............anything will look good.
Let's take the 100 MHz clock, in our expensive gizmo, and limit the low frequency limit to only 1 kHz.
25 fSec. Yeah, 0.025 pSec.
Ridiculous, right?
So much for specs.................
The other thing that you have to know, and it is probably the most important.............is whether the jitter is Gaussian or deterministic.
Gaussian: just white noise. Exists everywhere, and can not be avoided. Not too horrible. Except............maybe............well, some of us feel the really close in stuff, you know, below the 10 Hz limit, on most measurements, is more harmful than the higher frequency stuff. But, let's forget about that. At least for now.
Deterministic: bad stuff. Two things can make up this sort of jitter. One is any form of tone, you know........something that is not random. Like 60 Hz, getting into the signal.
OK, this may not be good, but there is one form that is definitely worse.
The other is tones that are data-correlated. IOW, its frequency content has a relationship to the signal who's jitter you are trying to measure.
Yeah, that is the bad stuff. And guess what?? That is the kind you get, when you are listening to SPDIF!
Yep, without going into a lot of theory, the process of recovering the clock from a SPDIF signal gives you lots of data-correlated jitter. Definitely not bueno.
If you have trouble believing that, here is what you can do:
(Only for crazed DIYers, who can replace any chips they may blow up, by inadvertently shorting out some pins.)
Get a battery-powered listening device, of some sort. (Yes, a headphone type would be good.) Find the place on your SPDIF RX chip, that has the PLL filter. Listen to it.
Then play some music, and listen some more. If you can. It may make you ill.
OK, for those of you who don't have those abilities, you will hear a highly-distorted version of what you are listening to.
Yep.............not bueno. But, that is what you will get.
So, is that the only way to get data-correlated jitter? No. Take for an example a line of fine equipment, made by some big company (which, I may or may not have mentioned in my previous post), their gear has a fair amount of data-correlated jitter.
Why?
Well...........let's just say sticking the clock in the same chip as the SPDIF driver may be a good way to save a fraction of a penny, but does bad stuff, for jitter.
Actually, they are not the only ones! Yeah, we have measured other gear, that has data-correlated jitter. Similar manner, just different method. (Anyone who tells you "Hey, we use this brand of SPDIF TX chip, because it has a built-in reclocker, and that negates the use of an external one"................yeah, guess what...........wrong!)
"Well, what do we do?"
If you insist on using SPDIF, know that you will always have a certain amount of deterministic jitter. Even with reclockers and ASRCs. Highly reduced, but not totally eliminated. (Some of it gets converted into other nasty stuff, but let's not go there.)
But, more importantly, don't get hung up on jitter numbers. As you can see, there are all sorts of ways to measure and characterize them. None of them are good at telling you what will come out the end of your DAC. Guys like me use them, in the design process, to help in the evaluation stage. But even we don't all agree on how to measure jitter, or which methods yield useful data. OK, maybe some of us tout our GPT is better than the next guy's GPT. Unless you have a good knowledge of how it is measured, don't get too excited about it.
Having said all of that.................
A certain audiophile mag likes to show plots, of the jitter, of various pieces of gear. I would not get too hung up on them. But............if one piece of gear you are thinking about has a really nasty looking plot, with all sorts of stuff, sticking up, all over the place..................yeah, you may want to look at buying something else. The rest of them...................yep, you may have to listen to them, to know.
OK, I'm typed out, and you are all probably as red-eyed as I am, having both written and read this. You can send hate mail, to our CSB! (She is out of the hospital, but not back to working with us. Translation: no one is there to read it.) I may or may not check back. If I don't, take it as we are busy.
Happy listening, guys................
|
