I'm designing an IF filter and would like to find out the IP3 performance of my circuit. Firstly, I used the PSS analysis, but couldn't find +3dB/dB slop region at low power area.

After I read the article "Accurate and rapid measurement of IP2 and IP3" from the Analysis section, I started to use PSS+PAC and got a nice result. However, I was still curious why these 2 results are quite different, so I simulated and compared these 2 methods again in the same waveform viewer. I noticed that in the moderate power area, these 2 graphs were somewhat similar, while in the low and high power areas 2 graphs were different.

Reading from the manual I understand that at the low input power area, they are some error mechanisms in PSS analysis that make the graph incorrect, while in the high input power the small signal assumption of PAC is violated. (I'm using the same sweep input power parameter in both PSS and PAC)

Furthermore, I tried increasing "Number of harmonics" in PSS setup, surprisingly, the PSS graph trend to match the graph of PAC in the low power area. For a simple testing circuit like openloop opamp, 500 Number of harmonics make the PSS results almost match to PAC.

More complex circuits seem to need more number of harmonics to make PSS match to PAC, unfortunately, Spectre has a limit at 2,000 number of harmonics. After that it can't finish the simulation at all. (IC446)

My question is "what's the mechanism of this relation?" Since in SpectreRF's theory said, there is no relation between number of harmonics and the accuracy of PSS analysis. Is this an indirect effect or I just did something wrong?  :-/

Thank you in advance for your help.

Amorn

Thank you very much, Ken, for pointing out this relationship. Then I read spectre documents again and found a sentence said the same like yours that maxstep is defined by number of harmonics unless there is another accuracy parameter specified.

I tried changing maxstep in PSS option and noticed the accuracy improvement by deceasing maxstep value. However, I couldn't see any maxstep change in spectre's logfile when I altered the number of harmonics. It is alway fixed at T/50 for moderate errpreset as mentioned in user guide. Does it mean that there is a change inside simulator but doesn't report in the logfile? I'm just wondering.

-Amorn