analog
RESEARCH

Low jitter clocks, for digital audio.



The dirty truth about jitter measurements.
This could also be called "What is all this femto clock stuff, anyway?"

Jitter numbers, without context, are really useless.

"Huh? What does that mean? I see jitter numbers as 'x pSec' all the time. And you are trying to tell me they are useless?"

Yes. Because they are.

First, you have to know the frequency of the signal we are saying has so much jitter. 1 nSec @ 44 kHz may not sound like much, but if that signal is 11.2896 MHz, then it is a whole lot.

Then you have to know the frequency range over which the jitter is measured. It is easy to say "<1 pSec of jitter", if the range is 12 kHz-100 kHz. Which it is, in the telecom world. For digital audio, if measured in that range, it is most likely a bad measurement.

You also have to know whether it is data-correlated, or if it is random. But, let's skip that part, for right now. Let's concentrate on the first two.

Refer to the phase noise plot, shown above. We state that it has a jitter number of 0.4 pSec. But, what is important is how we specify it.

We always spec jitter from 1 Hz to 1 kHz.

Why is that? Because we have found that range to be the best indicator of how it will sound.

Why not go higher? Well, it really doesn't tell you anything. There is (or shouldn't be) anything above the noise floor, from around 1 kHz up. So, why give a spec that includes the noise floor? The noise floor is not all that important, for digital audio.

Why not go lower? Well, lots of reasons:

1.) It is harder to measure. The lower that you go, the longer it takes to get a good measurement. The above plot took 2 hours to complete. So, it isn't practical, especially in a production environment.
2.) It makes the numbers look worse! And we all know (or should know) that in a numbers game, this is a case where lower is better. Why would you want to spec something that makes it look worse? Well, depends who you are, and what you are trying to do.

We sell low jitter clocks. We spec them so that you know what you are getting. We spec them in the most meaningful manner, not the manner that will make them look better than they are.

Case in point.............let's consider the same clock, but we will move the lower frequency limit from 1 Hz up to 10 Hz:

How 'bout that. The jitter drops from .39 pSec to 0.063 pSec. Or, in the latest parlance, 63 femtoseconds. Huh. How did that happen?

Easy. When you derive the jitter, from a phase noise plot (which is really the way it should be done, for a clock, jitter is a function of area under to curve. So, make the curve smaller, and the jitter number drops.

But, why stop there? Let's raise the lower limit up to 100 Hz. Just like the guys who are selling "Femto Crocks". I mean clocks, not crocks. Clocks. Anyway:


Gee, now it drops to 28 fSec! Same clock, just a different jitter frequency range.

But, just like the area under the curve is important, so it the frequency of the signal. (This next part will require that you pay close attention, so you have been warned.)

Let's take a signal, any signal. And measure its phase noise and jitter. We get x pSec of jitter and -y dBc pf phase noise. Let's run that signal through a flip-flop, to divide it by 2. When we do that, the phase noise drops by 6 dB. By inverse, if we run it through a 2x multiplier, it raises the phase noise by 6 db.
But the jitter stays the same.

You can not lower the jitter, by dividing by 2, anymore than you can increase it by multiplying by 2. The phase noise changes, but the jitter does not. That is because the frequency of the signal is in the denominator, of that "area under the curve" equation.

So, unless you know the signal frequency, you can not really tell how good the jitter number is.

Let's use some practical examples:
If you have 2 clocks, where one is twice the frequency as the other, which is the better one? It depends. You have to correct for frequency.

If you are comparing phase noise (which is what guys who design this stuff do), and they have the same phase noise number, then the higher frequency one is the better one. (Since it should be 6 dB higher.) (This is also why a certain expensive clock, made by some of our buddies in a certain peninsular state, somewhere in the SE United States is not such a good deal, if you buy the 22 or 24 MHz ones.)

Or, if you want to compare jitter, and the numbers are the same, then once again, the higher frequency one is better.

But, what if they are different? Well, you have to convert them to a common frequency. Twice the frequency, divide the jitter number in half. Or, subtract 6 dB (20 * log2.) Or whatever the number may be to get them to the same frequency.

So, what is the point of this? Well, the laws of physics. As you go higher in frequency, it becomes harder to make a really low jitter clock. Because the Q of the crystal also is a function of the frequency. Same type of crystal, and as you go higher, there is a inverse relation to Q and frequency. Frequency goes up, Q goes down. But, since F is in the bottom of that equation, for jitter, the jitter goes down by the same amount that it should go up as the Q drops.

I know, confusing, if you do not do this all day long. Anyway, the point is: There are lots of ways to spec jitter and stuff, and if you don't know all the tricks, then you don't know all the ways they can make them sound that they are better than they really are.
And why certain 45 and 49 MHz ones should be a better deal, than their 22 and 24 MHz counterparts. At least on paper. In real life, well, we charge people to find that out, so no more hints!)

All this assumes you have the means to actually measure this stuff. And guess what? Most companies do not! (Which is why they pay us. And you should pay us, to buy screened clock parts. You really do not want to take your chances with the other guy. We know how many should go in the trash.)

"So, how do you measure this?"

Let's eliminate how NOT to measure them.

The worst way, and the way a lot of "experts" measure it is with a really fancy 'scope. OK, that fancy 'scope can measure jitter, but not to the degree we need for digital audio. And it measures jitter to the telecom world spec. Which is from 12 kHz to 100 kHz. In other words, it is not sensitive enough, and it measures a range that means nothing. (And the funny part is that expensive 'scope costs as much as a fancy phase noise rig. Or, you can buy the cheap version that we use, for the same price as the cheap version of that 'scope.)

No, you need an instrument designed for measuring phase noise. For starters. Unless you have lots of experience, there are some very easy mistakes, that can cause the phase noise numbers, down at 0.01 Hz, to be off as much as 30 db (too high.)

Of course, the Masters of the Universe, over on the DIY forums, will tell you this is nonsense. They will also tell you that jitter isn't important, and if it was, there is no way anything down that low in frequency can matter. (In fact, if pressed hard enough, they will tell you pretty much everything sounds the same anyway. So, none of it matters. If it doesn't matter, then why do they waste all their spare time on DIY forums? Still waiting on an answer for that one.)

So, you can chose to believe the Masters of the Universe, or you can believe the folks who do this for a living. (Do any of those guys have jobs? They sure spend a lot of time of DIY forums. Telling us how smart they are, which they are. And also showing us how really stupid they can be, at the same time. Some feat, eh!)

All of this leads to the question all of you want to know: "What does jitter sound like?"

Well, you do not hear the actual jitter. Since almost all of the jitter is Gaussian noise, and the jitter is a modulation of the clock frequency, you really don't hear it. (I suppose if you had a really bad tone, like a ton of line spurs, that might be directly audible. But, let's leave that condition aside.)

It turns out what you really end up hearing is the lack of modulation (jitter). When you eliminate the jitter, you end up eliminating all the stuff that makes "digital" sound, well, digital! Bass is tighter, highs are smoother, and the soundstage is more apparent. (No, don't ask us to explain. You either believe it is possible, or you don't. If you don't believe, why should we waste our time? If you do believe, then you have to try a better clock.)

In either case, we sell clocks that are affordable, and high performance. All with guaranteed specs. So, what is there to lose? While you have lots to gain.



Contact us at:

stuff
at
analogresearch
dot
co




©2014 Analog Research