Using pss option "high order" should drop numerical noise floor enough to do your 100dBc measurement.
It invokes Multi-Inverval-Chebyshev analysis after pss has nominally converged, to improve noise floor. According spectre -h pss it perports to improve it to noise floor to -150 dB using "conservative", but i was told before by a Cadence guy that 200dBc distortion should be achieveable... unfortunately my circuits are never that linear so i haven;t needed to test it  ;)

In fact, I did try flexible balance for measuring IIP2 of a current commutating active mixer. The resulting graph does not make sense! I have attached the graphs I got. The upper left corner graph is using shooting method (which is good) and the rest are harmonic balance but with different number of harmonics.
For some other type of mixer, both methods fail and IM1/IM2 vs input power graph looks like harmonic balance ones in the attached figure.

I have had this problem with IIP2 simulation for a long time and nobody has been able to help. Shooting method works sometimes, I have not been able to make flexible balance work. I can't figure out what the trick is since everything makes sense to me the way I run the simulation. Interestingly, with the same simulation, IM3 vs input is linear and looks very good!

I would be grateful if you can advise.

Thank you I will contact Cadence. However, this is another observation of mine. Imagine the same current commutating active mixer with three large signal tones, LO=1GHz, RF1=1.1GHz and RF2=1.16GHz. After downconversion, I expect to see IF1=100MHz, IF2=160MHz, IM2=60MHz, IM3_1=40MHz and IM3_2=220MHz and probably a lot of spurious tones which are very small compared to these main tones (except maybe for IM2).  This is exactly what shooting method gives (Right hand side of the firgure below).

When I change the PSS method to flexible balance, I get some resutls which don't make sense. I see a lot of spurious tones at integer multiples of the beat frequency (20MHz) which are relatively large. The next largest tone after the fundamentals is only 32dB lower than the IF1 and it is at 80MHz! (Left hand side of the firgure below). Besides that, magnitude of the desired downconverted tones and the upper IM3 tone match in both shooting and flexible balance methods however, that of the lower IM3 tone and IM2 tones don't match!

I should say that I am using 200 harmonics in both simulations.

What is wrong here? I expected to get similar results with both methods.

Thank you in advance

No, I did not select highorder in shooting method. The results look to be good and valid and the noise floor is low enough even without highorder option.

My stimuli are sinusoid.

Results of shooting and flexible balance with QPSS match well.

Does HB simulation have a fundamental limit when there are multiple inputs? or the accuracy can be improved some way?

schehrazi wrote on Mar 16^th, 2007, 11:02am:


I don't believe Cadence HB has a limit to the number of input but the more inputs you have, the more memory you will require to handle all the mixing products etc. If you have a large circuit you could quickly run out of memory with a large number of input tones. I generally have not had to use more than three inputs

The accuracy of the HB simulation depends on a number of factors
1. The order of harmonics of your input tones - the more the better but the slower the sim
     - if you have squarish looking signal (say on an LO port of a mixer) you will need more harmonics to get good accuarcy
2. The order of the intermod products - again the more the better but the slower the sim
     - cadence default is to calculate all intermods but you can select diamond rule and limit the number of intermods
3. Oversampling ratio can affect accuracy some as well - usually an OSR of 2 or 4 will suffice but will slow down the sim compared to 1

Unfortunately there is no golden rule about what order to use for your harmonics and intermod. It highly depends on your circuit, signals, and which tone you are interested in. (ie if you are only interested in the fundamental output you probably don't need that much accuracy but if you are interested in the intermods (IP2or3) then you need higher accuarcy.)