Sampled pxf analysis is the method of choice for this. You get the jitter by dividing the result by the slope (derivative) of the signal at the crossing point. I have used this method extensively in the past (using tdnoise and special noise sources at a time when sampled pxf did not yet exist) and the results have always been very close to those of transient simulations. Sampled pac analysis is still rather buggy at the moment but sampled pxf seems to work ok, probably because it is in some way part of the tdnoise analysis.

Frank Wiedmann wrote on Aug 23^rd, 2007, 12:05am:

can we extract DJ and RJ separately from the total jitter?

Hi Frank,

   Thank you for your explanation of DJ simulation. How about an oscillator with a modified power supply, how to simulate the jitter?

Hi Frank,

   Thank you for your reply. Maybe I don't describe the problem clearly. For example, I have a ring oscillator, but the power supply is not clean (i.e. switched-capacitor DC converter or the ripple due to large current drop). I read some documents, your threads, and Ken's paper about the bang-bang noise. I think the jitter due to power supply should belong to deterministic jitter. I am not sure if we can simulate the DJ of an oscillator due to power supply ripple directly by PSS and PNoise because I am afraid PSS will not be converged since it is like a two-tone system. We can turn off the device noise and just look at the DJ due to the power supply ripple.
Regarding the PXF simulation you mentioned in this thread, do you know if there is any application notes or more detailed descriptions. It looks oscillator is an autonomous system, but when the jitter is generated by power supply, it becomes a kind of driven circuit?
Thank you.
Jacki

Hi Frank,

   I see you talk about the sampled PXF, what is the unit of the output from the sampled PXF? Does the transfer function has the output unit as sec/V? Because you say we can get the jitter by dividing the transfer function results by the slope of the output signal (I guess PSS can do it, do we need to define the time event as in PNoise simulation?). I think the slope is dv/dt, if so, the output should have the unit square(sec)/V? Please correct me, I am new to use the PXF transfer function. Thank you.

Hello Frank,

   Thank you. It seems the sampled PXF gives the sensitivity of the output jitter to the supply ripple, and it is a small-signal condition. Since the result from PXF divided by the slope dv/dt has the unit sec/V, how to calculate the jitter at the output due to the power supply ripple? Also is the PXF giving the transfer function in Frequency domain? If so, it should be a frequency dependent transfer function (at least plot from 0 Hz to f0, f0 is the beating frequency)? how to choose the value from the PXF result?
Assume we already know the sensitivity for the output jitter to supply ripple, but the ripple on the power supply is an unknown value, how to calculate the jitter? I am guessing if I need to average the ripple value, and use the averaged ripple voltage multiply the sensitivity to get the "RMS jitter".
   Please give me a hint, and I hope I will not bother you on this problem again. Thank you. Have a nice weekend.

Thank you very much Frank. It seems PXF can be used to estimate a known power supply ripple with absolute frequency. I think the transfer function divided by the slope at the transition time gives the transfer function from voltage to jitter, and if we want to know the jitter, we need to know the amplitude of the ripple first. As you mentioned, for an unknown ripple, it is better to run a transient simulation. Have a good day.
Regards, Jacki