I have a netlist where, by setting initial conditions, the first few ns of transient simulation are perfectly periodic with period t, before numerical noise leads to a mode where things are periodic with period 2t. I'm confident numerical noise is the issue because tightening reltol and maxstep increase the length of the initial periodic state to 10ns or beyond (50+ 5GHz cycles), but with looser settings the transient sim runs only one or two ns before flipping to the longer mode. The basic issue is that there is a loop gain which amplifies any difference between two nodes (set to zero in the initial condition) and if this happens, the periodicity of the system will exactly double. So in some sense the system is unstable, but transient simulation can easily run for many 10s of cycles in the desired mode before leaving.

The problem I have is: I want to do pss (and pxf/pnoise) analysis on the first periodic state, but for some reason no parameters I give to pss make it converge instantly, even though the first few ns are clearly periodic as shown by careful examination of the transient simulation, and even though the appropriate period is set. Sometimes pss will converge quickly, sometimes slowly, usually it fails. I've tried a couple of different simulators, both give the same result.

Is there any way I can tell pss not to iterate at all? I'd like to be able to just run transient, point to some part of it, say "here, use this" and forcibly do pxf/pnoise analysis on that data. Is this possible?

Or are there any tricks that might help pss converge on the desired mode?

Fair enough.

Actually, I think the circuit is inherently unstable, but this is because of the testbench, not because of the "electronic" (devices+interconnect) part of the circuit.

Anyway if the issue persists I will try and box it up and see what support might have to say about it.