Spectre: is it possible to define a deterministic start time for the PSS beat period?

Dear dontpanic,

dontpanic said:

Some context: we would like to have a deterministic start time for the beat period to (1) aid convergence and (2) perform measurements on the PSS timedomain results

I do not believe there is a technique to "force" the pss algorithm to start at a specific time with precision. One can override the value of tstab for the pss analysis, but choosing the very first time point that will provide pss convergence is not possible. I think this is expected as the entire purpose of a pss analysis (using either a shooting or harmonic balance algorithm) is to find a periodic solution.

You can view the periodic solution in the frequency and time domain following the analysis and it will show a period of the solution it finds. I'm not sure how this poses an issue with your use of the Calculator. 

Others who monitor this forum may have more information than I, but I've used these analyses for years and have never seen that as an option (and based on my understanding of how the algorithms work).

Shawn

For a driven circuit, it should be deterministic:

• tstart is the start time you specify in the analysis (usually 0)
• tonset is the onset of periodicity. This will be when all sources start being periodic, and so is 0 unless you have specified delays on sources, or have a pulse or pwl source without a period (in which case it's whatever time they stop changing). It's easy enough to figure out if you look at the sources
• tinit is normally a single period of the PSS fundamental.
• tstab adds to the other times - so this is your control. Note that if you've got the option set to automatically detect steady state during the transient assist, it may stop before that and start the periodic analysis.

Oscillators are more complex because it is also trying to solve the frequency and it's much harder to pick a precise start time in that case.

Note that you don't have to ensure the circuit is fully settled in the initial transient; PSS is trying to solve that efficiently in the shooting (or hb) analysis part, However, being along the way to being settled can certainly help with convergence, particularly when you have nonlinear startup behaviour. Most of the time I don't need to fine tune the edge position because in practical circuits you'll have a variety of phases and so there's no single "good" place. In trivially simple circuits it can make a difference where the start is, but most of the time you don't need to worry too much about this.

Regards,

Andrew

For a driven circuit, it should be deterministic:

• tstart is the start time you specify in the analysis (usually 0)
• tonset is the onset of periodicity. This will be when all sources start being periodic, and so is 0 unless you have specified delays on sources, or have a pulse or pwl source without a period (in which case it's whatever time they stop changing). It's easy enough to figure out if you look at the sources
• tinit is normally a single period of the PSS fundamental.
• tstab adds to the other times - so this is your control. Note that if you've got the option set to automatically detect steady state during the transient assist, it may stop before that and start the periodic analysis.

Oscillators are more complex because it is also trying to solve the frequency and it's much harder to pick a precise start time in that case.

Note that you don't have to ensure the circuit is fully settled in the initial transient; PSS is trying to solve that efficiently in the shooting (or hb) analysis part, However, being along the way to being settled can certainly help with convergence, particularly when you have nonlinear startup behaviour. Most of the time I don't need to fine tune the edge position because in practical circuits you'll have a variety of phases and so there's no single "good" place. In trivially simple circuits it can make a difference where the start is, but most of the time you don't need to worry too much about this.

Regards,

Andrew