Hi,

I want to find IIP3 of a Gm-C bandpass filter (3 MHz center freq.)  by giving two closely-spaced 1mV tones at f1, f2 and observe the output at 2*f1-f2 and 2*f2-f1.  

 I tried PSS with spectre RF,  but the magnitude of the output signal at 2*f1-f2 and 2*f2-f1 is changing each time  I vary tstab (the additional time for stabilization) and rerun the simulation.

 Should I rather use tran analysis+fft  or should I try QPSS?

Also, I don't understand the difference between PSS and QPSS analysis.
Please give show me a good reference to QPSS Vs PSS analysis and when to use what.

Thanks a lot.

You should start by reading the paper on measuring IP₃: http://www.designers-guide.org/Analysis/intercept-point.pdf.

You should also check the date on you copy of SpectreRF. Several years ago Spectre used to occasionally produce results that varied with tstab, but I believe that has been resolved for quite a while now.

Finally, use PSS when there is one large periodic tone, use QPSS if there are more than one and they are not harmonically related.

On amplifiers you can get away with using PSS combined with PAC to measure IP₃.

-Ken

No, the results should not vary with tstab. The PSS analysis should compute the steady-state solution. If there is only one, then tstab should not affect it.

-Ken

That is pretty old. I would not be surprised if that version of Spectre had the problem. You only need to upgrade Spectre.

-Ken

It is not necessary for the tstab interval to take you all the way to steady-state. It was initially envisioned as a way of allowing the designer specify that the circuit needed a certain amount of time to initialize itself. For example, a shift register might need a certain number of clock cycles after the reset was lifted before it was in the right state to support steady-state behavior. It was quickly found to also be a nifty way to help the convergence of the PSS analysis. It allows you to get the circuit to close to its steady-state behavior before starting the PSS. In that sense, it is kind of like nodesets and the DC analysis. If you give the DC analysis a complete set of nodesets that are close in value to the DC solution, then the DC analysis converges quickly and easily. The tstab interval is kind of like a way of automatically generating nodesets for the PSS analysis. And like nodesets, they do not have to be actual solution. Just like the DC analysis takes nodesets and uses it as a starting point in the process of calculating the true DC solution, so does the PSS analysis take the values at the end of the tstab interval as the starting point for computing the steady-state solution. Once PSS converges and produces a result, it is the steady-state solution regardless of how good or how poor the starting point was. The starting point (or tstab) should only affect whether PSS converges, not how accurate the result is.

Older versions of Spectre have a bug that makes the final solution overly sensitive to tstab in some cases. In these cases the results computed by Spectre will be inaccurate. Therefore you should upgrade to the latest version.
-Ken