Hello everyone,

It is a long time that I am desperately looking for a fast, proper and accurate method of measuring the input-reffered noise and dc gain of a regenerative comparator.  :'(
I have recently got familiar with the "periodic ac analysis" option of RFSpectre, yet have not found a detailed discription of using this for a comparator. It was only recently that I came across a set of slides from UC Berkeley and a paper from here, which point to this issue but not very clearly.
http://goo.gl/iYPgQg
http://goo.gl/Hk6lfW

Following them, I built a setup as you see in the photo:
1) An input source with DC=0 and pac=1;
2) An ideal output sampler, which is driven by a pulse signal with a very short duty cycle of 10ps and a delay of Td;
3) A clocked regenerative latch with the clock frequency being 16MHz and a rising of 1ns.

I then set the parameters in pss, pxf and pnoise as follows;
PSS:
Beat frequency=16MHz, Number of harmonics=0
PXF
Output frequency sweep range: 1~8MHz, Maximum sideband=0, Output=VOUT_Sampled
PNOISE
Output frequency seep range: 1~1GHz, Maximum sideband=50, Output+=VOUT+, Output-=VSS, Input=Vin, Reference sideband=1, timedomain: number of points=10

For measuring the dc gain using pxf, I then run a parametric analysis by sweeping the moment of sampling (Td) in one clock period.  The maximum gain, which happens during the rising edge of the clock,  goes to around 180dB! How is it possible?
By running the pnoise analysis, I get the same strange results. What is even more bizzare is that the output integrated noise is dependent on Td, ranging from volts to Mega Volts!! while I measure the noise at VOUT+ not at VOUT_Sampled. Actually, PSS is sort of smart and detects all the periodic signals existing in the schematic and list them in "Fundamental Tones". But does how this effect the results, I have no idea.

This problem is now taking me so much time and energy. If anyone has the same experience and could help, it would be really appreciated.