I am simulating a ~25GHz LC oscillator using harmonic balance and pnoise, then calculating the phase noise at 10MHz offset (this offset is added as a specific point to pnoise). Even with the most aggressive accuracy settings (harm=40,steadyratio=1e-3,reltol=1e-6,vabstol=1e-9,iabstol=1e-12), the phase noise measurements are extremely sensitive to tstab. More specifically, the measurements are sensitive to where the oscillation begins.
Please correct me if I am wrong BUT if I want to sweep temperature or do Monte Carlo, this initial phase will vary and add +/-0.25dBc numerical noise to my results.
I attached to waveforms to show the problem. Figure #1 shows thelast cycles of initial transient simulation while Figure #2 shows the phase noise measurements. As you can see the measurements for 5nsec and 5.3nsec are close since the oscillations for these begin with a very similar phase. By the way, I am usingSpectre Version 18.104.22.1687.isr19
Here is the second waveform.
I would suggest contacting Customer Support on this - it will probably need taking a look at your setup in more detail. It might just need a longer tstab, but it's hard to tell without seeing the circuit and doing some experiments.