Tuza2000,
First , there has been a detailed discussion of this issue in the thread,
http://www.designers-guide.org/Forum/?board=ms_design;action=display;num=1118555... 245;start=7#7
For a pipeline ADC, the methodology for simulating an ADC is identical to the
methodology a S/H. In the case of the S/H, the noise is not limited by the
quantization noise of the ADC, or the thermal/shot/... noise of the S/H. So the
FFT noise floor is only limited by the numerical noise of the simulation and the
FFT. The largest noise source in the FFT is usually the interpolation noise. Interpolation
noise occurs because FFTs have very strict requirements about how the signal is
sampled, 2^N equally spaced samples. Since SPICE-type simulators do not usually
stop at the correct time points without help, the FFT noise floor is often limited by the
error in estimating the actual value from the available time points. The easiest
solution is to set the maximum step size to be small to minimize the error. However,
this increases the simulation unacceptably so Spectre provides the strobe option.
It forces a time step at the instant required for the FFT. One other option to consider
is using a Spectre's zvcvs source as an ideal sample and hold. It has the same effect
as strobe and allows the use of Spectre's built-in Fourier integral calculation.
Also, as mentioned in the thread above, the rectangular window is sufficient. Using
windowing functions introduces error. You might want to leave some additional time
in the simulation at the start.Usually there is some delay required to reach sinusiodal
steady-state. For example, common-mode feedback loops usually require time to
settle. So you might want to add an additional 100-200ns to the simulation.
Best Regards,
Sheldon