problem with spectreRF

Mohamed,

The reason is that it's looking at the difference between the 1st order signal and the 3rd order product in the IP3 tree (i.e. when you have a response that looks like .II. (the "I" are the desired 1st order tones, and the "." are the IM3 products) it's half the difference between the desired and the IM3, plus the input power (which comes from the number you enter on the form, or the variable you've swept). It would only be infinite if there was no 3rd order IM3 appearing at the input port. That would only happen if you have no reflection at all and (say) an ideal voltage source. If there's an impedance there, you'll always end up with some IM3. If you plot the spectrum at the input port, you should be able to see this.

I'm guessing a bit here, because I can't quite see what you're doing (pictures would have helped).

Regards,

Andrew.

1st thanks very much.
I will explain my original problem,
I design a differential common gate LNA working at 2GHz with LC tank at output(QL = 10 low which corresponds to higher BW)
In order to account for the loading of the following mixers, I modeled its capacitance with some value.
Then, The loading mixers are 3 mixers, direct conversion Active mixers.
3 mixers due to some specs. in the system design.
To simulate the IIP3 with the mixers, I switch off the clock, i.e. clk = vdd and clk_bar=gnd, such that I get their loading on the LNA stage.
Doing this hurts the LNA linearity so much(from 4dBm to -4dBm).
After some debuging, I saw the balun issue i mentioned previuosly.

How do I measure the IIP3?
by PSS and PAC:
f1 is the PSS freq. and f2 is the PAC freq.
These freq.s are inserted in the input port which is connected to an ideal balun that transform impedance from 50 ohm to 100 ohm differential, i.e. sine wave and pac magnitude with same magnitude prf dBm respectively.
then, in PAC direct plot window, i choose ipn curves and 1st harmonic is f2
and 3rd order harmonic is 2*f2-f1.
and I choose differential nets as output and select the LNA output nets.

Hi Mohamed,

I think the best way forward here would be for you to contact customer support via http://support.cadence.com - it will almost certainly be simpler to see what you're doing (with a testcase ideally) and then provide advice. You could probably simulate the whole circuit (with the mixer present) using harmonic balance (either hb analysis, or hb-based QPSS, possibly replacing one of the tones with an hbac/qpac). RapidIP3 is another possibility - but probably I'd ensure that you're getting reasonable results first.

As I said before, looking at the spectrum at a particular input power would be a good idea to figure out what is going on.

Best Regards,

Andrew.

Dear Andrew,

Thanks for your replays.

At page 40 of the following LNA workshop:

http://www.cdnusers.org/community/virtuoso/resources/SpectreRF_LNA533AN.pdf

For the IIP3 graph:

the 3dB/dB is wrong spcially for the -50dBm input. Why the curve fails at such low input value?

Is it a simulation issue or circuit issue?

Best regards,

Mohamed

Mohamed,

It's due to numerical noise in the simulator. Tightening the tolerances, or using Harmonic Balance, would have increased the dynamic range and allowed it to more accurately resolve the signals. One of the challenges of IP3 simulations (particularly with a mixer) is that you have large signals (the LO), small signals (the IF) and then very small signals (the IM3 signal) - that's quite high dynamic range to resolve accurately. 

So it's a simulator issue. That's a very old app note too... so I wouldn't spend too long looking at that.

Better to look at <MMSIMinstDir>/tools/spectre/examples/SpectreRF_workshop/rfworkshop.tar.Z

Andrew.

Dear Andrew,

Thanks again. I checked the documents of the LNA. When I saw the IIP3 curves, the input power is started at the x-axis from -40dBm in order not to show the previous problem. But I think it still exists. About tightening the tolerance, I use conservative and further I go to analog sisimulator options and tighten all the tolreances by a factor of 1000. But the problem exists.

Best Regards,

Mohamed

Mohamed,

I have examples where starting from -50dBm is fine - it depends on the circuit, and simulation setup. 

Tightening tolerances by a factor of 1000 is probably not sensible - that might actually make things worse.

As I suggested earlier, you should contact customer support - it's very hard to know what you're actually doing without seeing it. In the case of the app note, it doesn't matter that the curve is inaccurate at -50dBm because the IP3 is being measured at  -40dBm. However, the results look pretty poorly controlled in that app note - so that's why I suggested you look at something more recent.

So please contact customer support.

Andrew.