To compute the noise of a sample and hold, you would apply a periodic clock and run a PSS analysis to find the periodic operating point, then run a PNoise analysis to compute the output noise. I sounds like this is what you did, but you claim that you cannot see the output noise. I'm not sure I know what this means. Is it that you do not know how to plot the output noise, or is it that when you plot the noise you find it to be zero? If it is the latter, please describe your sample-and-hold circuit. Is it constructed from transistors, or is it a behavioral model?

Jitter Man

Once you run the PNoise analysis, you can use Results->Print->Noise Summary to calculate the total integrated noise.  Be sure to select integrated noise. You may have to scroll down to see the total output noise.

You will probably want to take some steps to assure that the noise you measure is the noise you want. In particular, you probably only want the noise present when the output has settled, and not when the output is still settling or the noise during the inactive phase. This is most easily accomplised by using the time-domain noise feature (Noise Type = timedomain) on the PNoise form.

Jitter Man

The input signal should be turned off (it should be constant valued), but you must apply the clock. Perform the PSS analysis with only the clock applied, with the period of the PSS analysis matching that of the clock period.

Then perform the PNoise analysis.

Jitter Man

You are simulating a periodically clocked sample-and-hold, so you will see a sin(x)/x transfer characteristic. However, when computing total noise, it is generally not necessary to follow the sample-and-hold with another sampler (ZOH). The power in the hold mode noise generally dominates over the power in the track mode noise. If you are looking to compute the noise in the discrete-time output signal (as opposed to the continuous time signal that is output by default by SpectreRF), you should instead use the time-domain noise feature (noisetype=timedomain). That way you do not have to normalize out the sin(x)/x variations. Once you have the noise in the discrete time sequence (the time-domain noise), then you integrate from 0 to half of the clock frequency to calculate total noise.

Jitter Man

SpectreRF's PNoise analysis reports on the differential noise if you specify the output to PNoise using two nodes.

You do not need to specify the output nodes to the PSS analysis. You specify nodes to the PSS analysis only when simulating oscillators.

How you specify the output to the PNoise analysis is unaffected by the noisetype parameter.

Jitter Man