How to measure PLL's jitter characteristics?

[Hughes]

How to measure PLL's jitter characteristics?

In Hsiang-Hui Chang's paper: "A Wide-Range Delay-Locked Loop With a Fixed Latency of One Clock Cycle", I found a jitter histogram as show in the attached figure. I don't know how to get this histogram. And what does the vertical axis stand for?

Any inputs will be appreciated.

 

[eternal_nan]

Vertical axis is voltage. You get a plot like this by setting your scope to trigger on rising edge, zooming in on the edge in the waveform that imediattely follows the trigger edge and turning on the "Infinite Persistance" (sometimes called "Analog Persistance") mode of the scope. That will make it overlay the waveforms obtained from maximum triggering. The width of the persistance trace will give you the period jitter.

 

[radiohead]

A better way would be to simply measure the phase noise and convert this frequency domain information into time domain jitter. It would be more accurate. There are papers that describe this technique.

 

[Jim cage]

Is this measurement true?
saturating the ADC of the scope (for example, 20mV/diV will give 200mV full scale which will yeild high quantization noise).

Jimmy

 

[andy2000a]

some oscilliscope have historgram . and you can "overlap" Clock signal find jitter

 

[icemc2008]

According to the period jitter measured by scope,(using the infinit persistance), if I want to compare the simulation and measurement. for example, I have the PLL schematic at ADS, how to simulate or calculate the output oscillation signal to compare with measurement?

Is there a function which could transfer simulation result in time to a period histogram? or I have to export the result, and do it manully?

Thanks

 

[dick_freebird]

I like to "clock the clock with the data" (data is the trigger, clock is the waveform)
and set the mode to envelope. With a high number of samples you arrive at a
P-P jitter result that's pretty real. You also get to see things like what of it is
deterministic (banding that develops in the shorter term). If you have per-sample
delay measurement and histogramming you can even get stats. Of course your
test equipment needs to be better than your test article and somewhere near
the bleeding edge, that kind of falls apart.