tran vs. pss - reproducing results

Hi,

I built a simple receiver based on a passive mixer. I used pss/pac/pnoise/psp/qpss/qpac et.al. to determine NF, matching, linearity. I relied on it and get good results (e.g. NF=7dB, IIP3=16dBm).

I use a single tone at 1.1 GHz which is downconverted via fLO=1GHz to 100 MHz.

As an example, this shows the frequency domain output of the pss:

vixer2_pss_spectrum.pdf

As can be seen, the spectrum is pretty clean, HD3=100 dB and this matches reasonably well with the results for IIP3 I am getting with qpss/qpac.

Now I need to obtain the same results with tran (I build a new architecture which in the end I must verify using tran), even is simulation takes long. I am shocked that results are nothing but clean and garbage. This shows one DFT of the output signal and a zoomed version. It can be seen that there are tons of spurs:

vixer2_tran_dft.pdf

vixer2_tran_dft_zoom.pdf

I am absolutely aware of caveats of the DFT, hence I made sure that:

• Simulation time is exactly tran_sim_periods/BW where tran_sim_periods=400 and BW=100MHz. Hence the simulation captures 400 periods of the baseband signal precicely.
• To obtain equally-spaced samples for the DFT, I set strobeperiod to 1/(BW*NDFT) where NDFT=2^17. Hence the Nyquist rate is 100MHz*2^17=13 THz, very high!
• I plot the DFT with db20(mag(dft(v("/vod" ?result "tran") 10/BW  tran_sim_periods/BW NDFT "Rectangular" 1 "default" ))). So I skip the first 10 periods such that the circuit can stabilize. The DFT takes 390 periods, hence the frequency resolution is very high!
• As can be seen, the DFT runs over entire periods so no leakage can occur.

Given all these points, tran must give the same results as pss. My initial assumption was that the spurs are higher-order spurs that fold back due to too low Nyquist rate. But apart from the fact that 13 THz is huge, results do not change when I increase either NDFT or tran_sim_periods. Also, my LO has rise and fall time of 10ps, so there can't be infinite frequency content (as it would occur for sign(cos(w*t)) with zero rise/fall time).

Last but not least I cut out the last period of the tran simulation and compare it to the time domain output of pss. I export both signals to MATLAB and would expect them to be identical. Fortunately qualitatively they look similar but not even the amplitude matches:

How can that be?

Any ideas on this?