Jitter measurement

reference

relationship.

and

is

My favorite technique is to use a digital oscilloscope in X-Y mode. A stable reference that's locked to the signal to be measured gets split and filtered into a sine and cosine with reasonably inexpensive RF components such as those from minicircuits.com (or passives for the lower frequencies). The reference sine and cosine feed 2 channels of the scope in XY mode resulting in a circle. The signal with the jitter goes to the trigger. With one point per trigger, the scope shows a spot, a smear, or a "smile" showing the amount of jitter within one reference cycle (degrees within the 360 degree circle). The scope I used a dozen years ago had a timebase jitter of about 75 ps, limiting the measurement.

The requirement to lock on to the incoming signal means a little extra work. A VCXO with a low frequency cutoff tracks the incoming signal below the cutoff but only delivers the VCXO noise above the cutoff. With this reference, you can watch the jitter.

Or use a SERDES and an FPGA to take the incoming signal and resolve it down to less than 1ns of time interval resolution per clock.

reference

relationship.

and

is

Having a look at the noise of the VCO control voltage may be one way, but if not done carefully may introduce additional jitter. The usual method is to measure the slopes of the carrier. EG have it running on 10MHz and measure from 10Hz to 1MHz beside the carrier. Then you can integrate the slope to get ps_rms-jitter.

Rene

--
Ing.Buero R.Tschaggelar - http://www.ibrtses.com
& commercial newsgroups - http://www.talkto.net

Can you make a second system?

In theory, the 10k from two different GPSes should be the same frequency. The scope display of both outputs should have no slip action so you can blow up the timescale to see the jitter.

--
--
kensmith@rahul.net   forging knowledge

Residual phase variation relative to what? In other words, what is your reference point for the measurement?

If your reference is the 10kHz signal from the GPS receiver (your PLL input), then use the 10kHz signal as the scope trigger, and display the

10MHz output. If there's no residual phase variation, you'll see a clear edge of the 10MHz output every time the scope triggers.

-- Mike --

reference

What is unknown is how the 10kHz signal is derived. It is possible that the pulse count is 10,000 per sec but the duty cycle varies, or a number of other possibilities.. The data sheet isn't revealing, and the manufacturer doesn't want to know legacy receivers, so there is no detail on the derivation. (Having seen how the 50/60Hz was derived from colour-burst crystals in the old Fairchild

5369 devices, I am leery of making assumptions In that family, M cycles were counted in the high state, and N in the low state, with MN. IIRC they also varied M across one output cycle to achieve "proper" division - the 5369EYRN produced 50Hz from a 3.579545 crystal by this form of "fudged" non-integer division.)

For this reason, I see it as conceivable that two devices may show jitter but still deliver zero slip.

That is why I used the OCXO from the frequency counter timebase in a CRO comparison - the jitter or variation in the counter timebase would certainly be independent of the GPS-derived oscillator's variation. That disclosed no discernible perturbation to a slow but steady slip, indicating that neither was particularly bad.

What I am trying to achieve is some improvements in the area of the loop filter and the oscillator itself. With a methodology for quantifying the variation, this becomes a more scientific process.

Please see my reply to Ken Smith's post. I'm not after a relative measure, I'm after an absolute - i.e. compared to a pure jitter-free ideal source.

Certainly the case. And that would disclose any irregularities in the 10kHz stream also

But as I mentioned in the other reply, I am after a method to quantify the variation.

In article , budgie wrote: [....]

Yes, I've done stuff like this to get non-integer frequency ratios. If you have reason to believe that this is the sort of thing they have done, it may be worthwhile to try to figure out what they have really done.

Does the 10KHz hold steady if you trigger from the 1PPS pulse? If so, you will know that the pattern repeats from one second to the next. This will help to assure you that there isn't going to be much below 1HZ.

You can buy a really good OCXO for a few hundreds of US$.

or:

Assuming you can make three of your PLL circuits, you could use them to help look at the noise in the 10KHz etc. (You can use two but it takes more work and you have to assume some stuff)

If you lock multiple identical PLL circuits onto the 10KHz, you can see the noise they introduce by comparing their outputs. Exactly how to compare them, I haven't thought through yet. (See below)

If you set the loop filter in the PLLs to be very slow, you will know that they can't track high frequency noise on their inputs. If you have more than one, you can see how much high frequency noise they make. By using all combinations of the three, you can solve for the noise for each one.

If you compare two with two different bandwidths, you may be able to see the noise in the input. (If the PLLs are too noisy you won't be able to see it) If you know the noise performance of your PLLs you can solve for the noise in the input signal that comes through in the band the two PLLs don't share.

Thinking about comparing:

We need a phase detector that has a low noise and a high gain and that doesn't have its over-range point near the point where the two edges line up.

Assuming we have VCOs running at a multiple, we can make a phase shifted signal for use in an XOR type phase detector. I see a problem with this because the noise from the power supplies appears on the output of the XOR and this could seriously limit the noise floor.

Perhaps a system with some sort of clean tri-stating would work. You only enable the output during a short period near the edge.

--
--
kensmith@rahul.net   forging knowledge

In article , Mike wrote: [....]

Here's a silly idea. I post it only because there may be the spark of a good idea in here.

Get about 6KM of coax. Feed the 10KHz into one end and compare the two ends of the cable with a scope.

At 10KHz a modest fraction of what you put in will make it to the far end.

--
--
kensmith@rahul.net   forging knowledge

This is basically a giant delay-line discriminator. I wonder if the phase stability of a big reel of coax is better or worse than a quartz crystal's? A quartz crystal discriminator with a Q of 10**6 would do much the same thing in a smaller space for much less money, and would avoid the problem of multiples of 2pi delay in a long cable. The sensitivity can be surprisingly high even at very small deviations.

Cheers,

Phil Hobbs

I don't have any reason to believ they have done this, but I don;'t know exactly how they derive 10k from the GPS signal so I can't rule it out.

The 1pps edge is synced to the 10kHz, but that really only requires that every second there are 10,000 cycles on the 10k line - not that the individual cycles are themselves uniform. Sort of like the way OUR electricity generation frequency is +/- a certain frequency tolerance, but as midnight approacheth the frequency is driven to ensure that at midnight the total number of cycles in the

24-hours is correct.

At one stage I wondered about timing each cycle of the 10k reference pulses using a higher speed clock and period counting, but the gating would need to be synced to the 10k pulse - and at the end of the day this would only remove the uncertainty over the "regularity" of the 10k stream.

I *think* the one in the frequency counter is fairly clean and wobble/jitter-free. Certainly the side-by side (and Lissajou) comparisons on the CRO with the PLL unit suggest neither has any issues - if either one had a problem then it would be evident, even though it wouldn't be clear which.

Comparing is something I have mused over. I can't discern any wobble or jitter visually on the CRO comparison, but that only means that it is below the threshold of visual discernment (sic). Similarly, I can't see any signs on the loop filter output. Neither of these "easy" comparison approaches lets me quantify, and hence precludes their use in optimising.

This PLL is far different from those of wide(r)-band synthesisers I am more familar with. It doesn't have a wide acquisition requirement, so the loop filter can be over-damped and is ultra-slow.

Being an XOR comparator (which obviously has its own contribution to jitter), at quadrature lock its output (and the filter output to the varicap) is half-rail. The varicap control line has a switch to select SETUP or OPERATE functions. In setup, a half-rail divider voltage feed the VCO which is trimmed to 10MHz. This is just to centre the PLL operation. The 10MHz VCXO has about a 80Hz range from rail to rail (non-linear of course but we're not trying to generate linear FM/PM signals) so that's about 16Hz/volt average slope

I don't for a moment accept the premise that what I have to date is perfect, but maybe I'm trying to go too far with a simple setup. Maybe it IS *good enough* as a cal source for the workshop.

[...]

Do you mean is synced by circuits or appears to have a stable timing relationship. I'm going to assume the latter.

If they are using a non-integer divider or the like, this ensures that there isn't going to be modulation below the 1Hz mark from that. You seem to have missed this point in my posting.

[...]

No, this just means you didn't look closely enough :>. Everything has noise. In this case it may be so low that you don't care about it but you can be sure it is there somewhere.

[... phase compare ...]

Most analog scopes still have a X10 horz button. Try using it. Since you are repeating fast, the trace should still be bright. Also, it can help to turn off the lights.

The flicker in the lights beats with the sweep of the scope and causes some visual artifacts. These usually look like troubles that are not there and not hide troubles.

[...]

Over damping is good. You want a large phase margin to avoid noise peaking near the gain cross over.

You could use a tri-stating phase detector to reduce this. Another trick assuming that you have the divide by twos at the bottom of the counter is to use an XOR to construct the inverse of the ideal locked signal.

---- ---- Q(n-1) ---- ---- ----

-- -- -- -- Q(n-2) -- -- -- -- -- ---- ---- XOR -- ---- ----

Compare the XOR to the output of your phase detector and you'll get the idea.

In fact the VCO may be fairly linear. (not good enough for audio) If the varicap is connected to the turned circuit through a smallish capacitance to reduce its effect, this tends to make things more linear. If instead it is in parallel with a largish capacitance, the circuit is less linear.

--
--
kensmith@rahul.net   forging knowledge