Now down to the measurements.
Pardon me, here I would like to quote what I had already written elsewhere:
(It's just pure laziness!)
"The pivotal point of these jitter tests is the measurement of the Time Interval Error (TIE) value.
A good short description could be found, for example, here:Jitter & Wander Tutorial
A short definition could be this -
"The time difference between a real clock and an ideal uniform time scale, after a time interval following perfect synchronization between the clock and the scale. "
That is, the Lecroy scope is set to collect a huge amount of data points. (10 -200 million points). This memory should be also very fast - we are putting in data at the maximum sample rate. In this case, at 100psec intervals.
Then the run is stopped, and the data edges are searched for. An algorithm looks at the data structure, defines the UI, and from there recovers the frequency of the fundamental clock embedded in the data.
When this ideal fundamental clock is reconstructed then each individual edge is examined again, and from it's actual crossing time value the corresponding ideal value is subtracted. The resulting difference is what is called TIE.
All this is nicely illustrated in the link above.
-As it is shown there, one can depict a time trend graph, showing how these error values are evolving for each consecutive data transition edge.
This is what the TIE time trend plots are showing.
-Also an FFT can be executed on these values, to show their behavior in the frequency domain. These are the phase noise plots.
-Then histograms can be collected, for all transitions, to see their statistical distribution.
From this then ISI, BER, Rj Pj DDj etc values can be extracted. Here I've calculated only the rms jitter value, like if all jitter were random, Rj."
Obviously, it's not the case generally. But. With respect to a general "industrial" situation, here I can do something:
By stopping the stream, like "stop play" or "pause", then restarting "play" again, I can take histograms with/ without data.
This way the random / deterministic data elements can be, "kind of", separated.