steadyratio impact on pss

Generally speaking I would leave steadyratio alone. I rarely use it to aid convergence problems. In general the convergence criteria (which is covered in that manual, but will be covered in more detail in the Spectre RF documentation from the SPECTRE release) is the same as the normal convergence criteria the local truncation error at each tilmestep, but multiplied by the steadyratio. It's tighter because you're trying to ensure that the first and last time point are essentially the same (with a small error). With it set at 1 then the criteria is the same as at each timestep - determining how close the solved value is to the predicted value. In this case we know that the first and last values should be the same for it to be periodic (and we already simulated the first time point, so it's not a prediction), so it's reasonable that it should be tighter.

Increasing the value could lead to it incorrectly identifying the circuit as settled when it hasn't actually settled - remember that the periodic steady state is trying to find the settled periodic behaviour of the circuit (for both shooting and harmonic balance).

I suggest you take a look at <SPECTREinstDir>/doc/spectreRFTheory/spectreRFTheory.pdf for more detail (searching for streadyratio in this document gives the equations used). That's got more detail than the document you're looking at.

Andrew

If you decide to increase steadyratio, I suggest that you only increase it as much as needed. Find the lowest value of the "Conv norm" in the spectre.out log file and multiply steadyratio by just a little more than this value. Do not use this method if the resulting steadyratio value would be larger than 1.

Hi Andrew,

I just copy the description from spectreRFTheory.pdf here: 

The steadyratio parameter specifies the maximum allowed mismatch between node voltages or current branches from the beginning of the steady-state period to its end. The steadyratio value is multiplied by the lteratio and reltol parameter values to determine the convergence criterion. 

So can i say, larger steadyratio relax (but not tight) convergence request for PSS?

And, i also find that for different errpreset, steadyratio is different also. 

E.g., for shooting PSS

Steadyratio = 0.001 @ liberal

Steadyratio = 0.001 @ moderate

Steadyratio = 0.01 @ conservative

From such a relationship, it seems the larger steadyratio, means better accuracy and tighter the mismatch?

Is there any contradiction in the definition?

BR

There's no contradiction, because reltol is tighter for conservative. The tighter reltol cancels out the looser steadyratio. However, the tighter reltol will also impact the accuracy of the iterative solver as well as the local truncation error, and so overall it will end up more accurate (also every point in the period ends up being more accurate, not just the end points).

In the early days of Spectre RF, steadyratio used to be might tighter for conservative which resulted in it being extremely difficult to converge; the choices have been stable for many years now and were the basis of many tests on lots of different circuits.

Andrew

Thanks Andrew, so i can say larger Steadyratio means relaxing convergence of PSS, right (assume other parameters are fixed) ?

sjwprcker said:

so i can say larger Steadyratio means relaxing convergence of PSS, right (assume other parameters are fixed) ?

Yes.