I don't have specific experience with simulating PLLs with PSS, but I do know that PSS convergence is sensitive to the value of "tstab" when simulating switched circuits such as dividers, PFD and inverters.

The PSS start and finish points for the fundamental need to be on the rising or falling edge of a signal that is periodic with fundamental. PSS iterates to make the start and finish point the same. If the beginning and endpoints are at the max or min value of a clock then PSS cannot determine a unique point where the signal is periodic (any change yields the same answer).

I generally try to pick a tstab that starts the period on a rising or falling edge of the highest frequency clocked/limited signal.

A completely different suggestion:
You could try using "transient noise" analysis with APS to simulate noise. The down side is that transient noise cannot tell you who the individual contributors to the noise total, but (on the positive side) you can probe noise anywhere in the circuit.

Good luck

RFICDUDE wrote on Oct 16^th, 2012, 4:37am:

Thanks RFICDUDE,

Does that means that if I want to simulate PSS for a PLL, I should set "tstab" much larger than the PLL lock time? It really costs lots of time.

The other question is, if I simulate PSS for a divider, which takes a VerilogA frequency source as its freq input, theorically, the output freq is stable, so can I set "tstab" to a small value or just leave it blank?

Yes, trans noise is convenient to get, but I think it has little relationship with Phase Noise, right?

Thank you~

No. If the phase detector has a dead zone it will not exhibit periodic behavior. If it is a fractional-N synthesizer it will not exhibit periodic behavior. If the PLL is unstable or is subject to a non-synchronous interferrer it will not exhibit periodic behavior.

The easiest way to test this is to run a long transient analysis with strobing turned on. The strobe period should be the expected PSS period. All signals when strobed at the fundamental frequency should eventually become constant valued. If any signal is not asymptotically approaching a constant value then the PSS analysis will not converge. At the minimum you should examine the VCO control signal, the VCO output frequency, and the supply currents.

-Ken

Do you know what a strobe light is and what it is used for? Strobing is the Spectre equivalent. Just set strobeperiod=T. This will cause Spectre to only output time points that are at times that are exact multiples of T. In other words, the computed waveforms are sampled every T and the sampled waveform is output rather than the original waveform. If your signals are T-periodic, then they should appear constant valued when sampled every T seconds. If they are not asymptotically approaching constant valued then your PLL is not exhibiting a T solution.

-Ken

By their very nature bang-bang CDRs exhibit chaotic behavior and so will never converge in a PSS analysis. What I know about simulating bang-bang CDRs I have written up and published in http://designers-guide.org/Analysis/bang-bang.pdf. If you need anything more that that, you need to be more specific.

-Ken