pnoise broken?

Hi

(I posted a similar question in www.edaboard.com/thread349701.html). I realize that NFdsb results from pnoise are fishy.

First, I generate a behavioral model of an amplifier (perfect input match to Rs, gain A, noise factor nfamp). I add output noise so that the amplifier has a certain noise factor nfamp (3dB in my case):

nfamp = 1 + vo^2/(kT*Rs*A^2) --> vo^2 = (nfamp-1)*kT*Rs*A^2

I check this with noise analysis and I get a NF of 3dB. Good.

Not I attach an ideal mixer afterwards and run pss/pnoise (no further noise contributions, tried different settings in terms of #sidebands, shooting vs. hb etc). Setting maximum sidebands to 1 I get 3dB as expected. But increasing the sidebands also increases noise figure. The number converges to something like 3.96 (e.g. 150 sidebands).

Now I set nfamp=1 (no noise at all!) and NFdsb is still 0.856 dB! However, looking into noise summary confirms:

/PORT0    rn    9.77859e-14           100%
/I56/R3   rn    1.81e-23              0%
/I56/R4   rn    1.81e-23              0%

Indeed, there is no other noise source than the input source. NF must be zero dB! What's going on here?