PSS warning

When doing a PSS analyses I'm getting a following warning:

"Warning from spectre during periodic steady state analyses 'pss'.
The high-order finite difference method PSS simualtion failed to
converge. You can turn off the method by setting 'highorder' to no in
the PSS options. Outputing results without high-order finite difference
refine"

How does this warning message affect the accuracy of the PSS results
and a following PNoise analyses?

 

Starting from IC50, a new algorithm (the "Multi Interval Chebyshev method")
was introduced to refine the PSS results at the end to improve the accuracy of
linear and weakly non-linear circuits. Essentially it can reduce the numerical
noise floor significantly in these cases.

It is controlled by a parameter "highorder" on the PSS analysis. In IC50
versions, it was on by default whenever errpreset was set to moderate or
conservative. From IC5032 onwards it is off by default - and you need to turn
it on if you need this improved accuracy.

Essentially it's only really needed if you need very high dynamic range -
which is typically the case if the circuit is responding in a weakly
non-linear way. If the circuit is compressing (distorting) significantly, the
concern about low numerical noise floor is not so much of an issue, since
there are real harmonics in the output which dominate.

The algorithm tends not to converge so well when the circuit is distorting a
lot, and what happens when it fails to converge is that it reverts to the PSS
solution it found earlier - so you get the original PSS solution - which would
generally be sufficient in these cases - because usually it fails to converge
when the numerical noise floor wouldn't be an issue.

Note that in IC50XX releases, the MIC method was only available for driven
circuits; from IC5141 onwards, it's available for autonomous circuits (e.g.
oscillators) too.

In general this improvement in numerical noise floor during the PSS then
affects the accuracy of the following small signal analyses - but the same
rules apply - if it fails to converge then generally the numerical noise floor
won't be an issue.

There are parameters which can be used to help the MIC algorithm converge
(psaratio), but mostly there's not that much benefit to be had in setting
these.

Regards,

Andrew.