Computing the NF of the three blocks combined is straight forward, however computing P1dB or IP3 is not possible because SpectreRF currently does not allow an autonomous circuit, the VCO, to be driven by a large tone, the input signal. However, this is generally not a problem. One does not normally include the VCO when using simulation to predict P1db, iIP3, and NF of the LNA & mixer because the VCO does not substantially affect these performance metrics.

However, having said that, these metrics are affected by the phase noise of the VCO when a large blocker in present. In this case, one would want to co-simulate the LNA, mixer, and VCO. In this situation, the best approach is to simulate the VCO alone, determine its phase noise, and then replace the VCO with a behavioral model that produces the same phase noise but is not autonomous. This procedure is described in Introduction to RF simulation and its application, which can be found at http://www.designers-guide.com/Analysis. Look in 6.3.3 on page 38.

-Ken

To compute noise figure you need a port at the input of the LNA, which would be set to produce a constant-valued (DC) input. You would then perform a PSS analysis to determine the large signal operating point. In this case, it is the oscillator that is setting that operating point. The PSS analysis starts with a transient analysis, you will need to use initial conditions or an input pulse to start the oscillator. Once it is started, set the PSS tstab parameter to give the oscillator sufficient time to settle down. That will allows the PSS analysis to converge without difficulty. Then perform a PNoise analysis. With the port defining the input to the circuit, SpectreRF will naturally compute the noise figure.

-Ken

I would like to make one very minor side comment about noise figure and Spectre/SpectreRF. The default port temperature is something like 16.8 degrees C. However, the default temperature of the rest of the circuit is 27 degrees C. The temperature difference affects the noise figure measurement only by a few tenths of a dB so you probably would not notice it. I stumbled upon the difference because I wanted to check my noise figure test bench with behavioral models and the test bench did not produce the exact noise figure I specified. Changing the port temperature to 27 degrees produced the exact noise figure  I specified in the behavioral model.

(The correct thing to do may very well be to add a temperature parameter to the behavioral model to account for the circuit temperature for which the noise figure is defined, but that is a modeling issue that probably does not belong in this thread.)

Frank,
Thanks for the link. That's one of the best definitions of noise figure I've seen, especially since it addresses the image frequency. This helps justify the way my group distinguished between noise figures of direct conversion receivers and super het receivers. It seems the correct approach to modeling noise in a behavioral model is indeed to account for the difference between circuit temperature and noise source temperature.
  -Eugene

I have never tried setting the port resistance to zero but I am not sure it would be a good idea. The port would only deliver the desired signal level if the port drove a short. For any finite load, you would see double the desired signal.

But why would you want to eliminate the port noise? The port noise is necessary for computing noise figure. If you still want to simulate noise only from the circuit, without any contribution from the port, and you still want to use the port for gain etc., I would eliminate the port noise by setting the port's noise temperature to -273.15 degrees C (zero Kelvin).

Oh, one more thing, anybody tried to simulate  Mixer + LNA + divide-by-2 circuit.  Could you tell me if you know how to do that

Thank you

Thank you,

Actually i put the divider circuit on the LO side, i would like to see if i can simulate the NF with divider also