I am simulating an upconversion mixer, which requires fairly high linearity. When I simulate with 3-tone QPSS using flexible-balance algorithm in SpectreRF, I get lower IIP3 number compared to 2-tone QPSS followed by QPAC analysis. I am quite far from IP1dB point (at least 10dB) so I don't think the circuit compresses. Any ideas about why this discrepancy could be happening?

Well, I haven't really swept the input power as we are really interested in S/IM3 at one particular input power level. S/IM3 is around 70dB and thus any numerical errors in IM3 will be important in the final result. I don't fiddle with harmonic settings and just use default setting. In HB-QPSS, I use 5 large signal harmonics with 3x oversampling ratio.

Murat

I am now trying with 7 harmonics.  I looked at your old posting and I think you believe that QPSS/QPAC is a bit more reliable compared to 3-tone QPSS. Am I right?

My company does not have much experience with QPSS/QPAC combination, so any results from that type of simulation would not be acceptable anyway. Still, I look to know if I can trust it to keep it in my arsenal as a sanity check.

I'm not going to append a "defence" of Spectre, because much of what has been said is not generally correct. It may be correct for specific circumstances (I'm not doubting Panchoo Hideboo's honesty), but is not correct in general. Otherwise all the large numbers of SpectreRF users would be getting incorrect results which don't match with silicon, and that isn't the case. We've got a lot of very vocal customers, and believe me, they tell us! Similarly if the simulator was vastly slower, Cadence would hear about that too.

Quick rebuttal - I don't have the patience to go into all of the points in detail:

I don't think you should need to use oversampling most of the time. maximorder/selectharm is OK providing you use them sensibly. Shooting QPSS is appropriate for appropriate circuits - but large, high frequency, weakly non-linear, multitone simulations is not its strength. 3-tone HB should be OK, but 2-tone HB+PAC should be OK too - again, which is appropriate depends on your needs. Similarly you could use PSS+RapidIP3 - it depends on your needs.

Regards,

Andrew.

eskiyerli wrote on Mar 30^th, 2009, 2:04am:
Generally reliable valid input level for small signal analysis subjected to master large signal analysis is under 10dB from 1dB Gain Compression point.

Again you should plot both Fundamental and IM3 powers vs input power and compare results between 3tones-HB and 2tones-HB/QPAC.
See http://www.designers-guide.org/Forum/YaBB.pl?num=1205223090
Show me your results.

If you have other simulators such as Nexxim, Aplac, ADSsim, GoldenGate, eldoRF, HSPICE-RF, SmartSpice-RF, etc.
show me results of comparison with Cadence Spectre.

http://www.ansoft.com/products/hf/nexxim/datasheet.cfm?f=Nexxim_Flysheet.pdf