The phenomena:
Use PNOISE with noiseType "timedomain" to analyze Mixer. Increasing the number of "Maxsideband" will increase the output noise power almost linearly (during simulation I set "Maxsideband" from 10 to 50).

I am surprised why Spectre gives that result, since the noise transfer function has certain limit bandwidth. Very High order harmonics almost have no effect on the frequency of interest because conversion gain for that harmonic is very small.

Frank:

The fundamental frequency from PSS is 2GHz.  I feel surprised because there is no obvious increment (within 0.5dB) when I sim by PNOISE with noiseType "sources" and sweep "Maxsideband" from 10 to 50. However if by PNOISE with noiseType "timedomain", there is a large increment.

I don't know why there is so big difference between "sources" and "timedomain". Is it a usual phenomena?

In my opinion  "sources" and "timedomain" should show the same trend.

-JC

Frank:
Yes. The case you described will cause that behavior. However it seems not the same case in my circuit. I post the circuit Fig. 3 and the simulation results Fig. 1and Fig. 2.

I probe the output noise  current using dc voltage source with Vdc=0.

Fig. 1 is the time domain output noise current of 2GH compnoent for "maxsideband" 10, 20 and 50.

Fig. 2 is the output noise current of PNOISE with noiseType "sources" for "maxsideband" 10, 20 and 50.

As can seen there is a large increment in time domain noise while the increment is very small of PNOISE sources analysis .

By the way I use behavior model for the MOSFET that has no capacitance and no noise effect. So I add a noise current source in parallel with M2.

Frank,
Thanks for your reply.
Frank Wiedmann wrote on Jul 9^th, 2007, 11:49pm:

This is because the max noise power is limited by the noise source with the PSD 4kTR in the pnoise sources analysis.  

But what surprises me is that why I can't see that limit in the pnoise tdnoise analysis, in which the power is increasing almost linear with the number of sidebands. Or maybe, as you have said, it is caused by the ideal behavior FET model without capacitive effect.

And do you have any ideal about which simulation setting will probably cause this strange phenomenon?  
-JC

Geoffrey_Coram wrote on Jul 27^th, 2007, 7:14am:

Thanks.
Yes. I agree with you. I think it should be the reason.
But now I have another question: What's the relationship between value from time domain pnoise analysis and that from sources pnoise analysis? For example by sources pnoise analysis I can obtain the time-averaged PSD value at frequency 2GHz and by time domain pnoise analysis I can obtain the waveform of frequency component of 2GHz. Since in my case one has limited value and the other unlimited, it seems strange and What does the value from time domain pnoise represent? .

-JC