You always have to do a pss analysis first for the periodic small-signal analyses pac, pxf, and pnoise because the pss analysis determines the periodic operating point for these analyses.

If you want to simulate jitter, use sampled pxf with an absolute frequency sweep. As mentioned in http://www.designers-guide.org/Forum/YaBB.pl?num=1187679312/1#1, you convert the result to jitter by dividing it by the slope of the signal, in the same way as for tdnoise. As you correctly noted, the result is a transfer function, so there is no need (and no possibility) to specify an input spectrum.

Yes, this is right. By the way, for the division by the slope, you can take the last part of the calculator expression in reply #9 as a reference.

In most cases, you will not know excatly in advance what the disturbances on the power supply will look like. The transfer functions should help you to identify and fix possible weaknesses in your design which might only exist in certain frequency ranges.

If you know the disturbances on the power supply exactly, the easiest way to simulate the resulting jitter is probably to model them with time-domain sources, do a transient analysis, and evaluate the result with the help of the eyeDiagram calculator function.

Hi Frank. I have one more question. Is there a way to simulate jitter caused by both device noise and power supply noise in 'pnoise' analysis? thanks.

Hi Frank, I guess I have to bother you with one more question. I get the transfer function plot from 'sampled pxf' analysis, but there seems to be a transfer function for each harmonic (from harmonic=-10 to harmonic=+10)? I guess I only need to look at the one for harmonic=0, right?

I wonder the 21 transfer functions (including harmonic=0) are obtained because of 'maximum sideband' set to 10? In 'pnoise', the 'maximum sideband' should be reasonably large in order to account for enough noise folding. What about the 'maximum sideband' in 'pxf' analysis, also be set very large?

In 'pss' analysis, a parameter is 'number of harmonics' (default set to 9 for shooting and 10 for harmonic balance). What does this parameter do in 'pss'? I feel that it is not important for a driven circuit, but useful for an autonomous circuit.

Another point that I should probably add is that the random jitter due to device noise has a Gaussian distribution and the simulated jitter value is the standard deviation or rms (root mean square) value of this distribution.

The "maximum random jitter" depends on the percentage of events in which this value may be exceeded (often defined as bit error rate) and is usually some multiple of the standard deviation. For example, if you look at the figure in http://en.wikipedia.org/wiki/Jitter, you can see that in 0.3% of the cases, the jitter will exceed ±3 times the standard deviation. For lower values of the bit error rate (BER), see for example the table in http://www.sigcon.com/Pubs/edn/RandomJitter.htm.

I have also seen pretty good matching with transient noise (if all the settings are correct there). The advantage of pnoise over transient noise is not only that it is often faster but also that you can see where the noise comes from in the Noise Summary.

In Virtuoso Analog Design Environment, select Results->Print->Noise Summary.

I don't know of a general way to set the noise of a part of the circuit to zero. However, some components may have parameters that permit setting their noise to zero.