I am studying how to run PSS+PNoise sim to get total output noise of a Track and Hold circuit.
I am using Virtuoso version ICADVM20.1-64b.500.15 for simulation.
Below is my test bench, in which the switch is driven by a 8GHz 25% duty cycle clock. The closed switch resistance is 0.1Ohm, while the open resistance is 1T Ohm. And the RC time-constant is 15GHz, which is higher than the Nyquist frequency (4GHz).
In document "Simulating Switched-Capacitor Filters with SpectreRF" , it is written that the total noise of T/H circuit is kT/C, which is 41.4nV^2 in this case.
I expect that the noise higher than 4GHz would be folded back into 0~4GHz, so I set the output frequency is from 1~4GHz. And to get a more result, maximum sideband was set as 50 (Because 4GHz *50= 200GHz >> 15GHz).
However, the total noise integrating from 1Hz to 4GHz is only 27nV^2. Moreover, the total noise increases as the duty cycle decreases.
Did I misunderstand anything about the total noise of T/H or do anything wrong in the simulation setup?
Thanks and regards,
You only get all the noise in the frequency range up to half the "Beat Frequency" if you use a sampled pnoise analysis. Take a look at the stop frequencies of unsmpldNoise and smpldNoise in Listing 1 of…
I don't have bandwidth this week to look into your questions or issues in depth, but a quick answer to your points above:
a) If the circuit itself folds noise, then that is handled by time…
Please allow me to try to add to both Andrew's and Frank's very valuable comments...maybe my comments will not but will let you be the judge!
YutaoLiu said: a. it seems that noise folding…
You only get all the noise in the frequency range up to half the "Beat Frequency" if you use a sampled pnoise analysis. Take a look at the stop frequencies of unsmpldNoise and smpldNoise in Listing 1 of the document.
Thanks for your reply.
I have a few questions based on your reply.
a. it seems that noise folding only happens with sampled pnoise sim. I would like to know the reason and more detail on the difference between the "sampled" and "unsampled" pnoise sim. Is there any material about this? Could you point me to that?
b. In the version of tool I am using, there are only "timeaverage" and "sampled (jitter)" for Noise type. I think I should use "sampled (jitter)" to get the sampled Noise. But there is no option for "noisetimepoints" and "numberofpoints", which was used in the Listing1 in the document. How can I do an equivalent setup?
a) If the circuit itself folds noise, then that is handled by timeaverage mode. All that sampled (jitter) does is introduce an ideal sampler which samples at given time steps within the period at the PSS fundamental rate. This is similar to having a stage of circuitry which follows the block you're analysing - so you can think of the analysis being more representative of how the output of the block is used. The time average is more like looking at the output with a spectrum analyser - you'll see the effect of the circuit itself folding noise, but the measurement equipment is not doing any sampling.
b) Within the sampled (jitter) mode, there are three choices - edge crossing, edge delay, and sampled phase. The edge crossing uses the transition of a signal to determine the time which the sampling at another (or the same) signal occurs at; the edge delay measures the delay between a transition of one signal edge and another (could be different signals), and sampled phase allows you to specify the number of sample points, and a list of specific time points to sample at.
BTW, from a very quick glance at your original picture, you almost certainly should be using the fullspectrum choice on the noise form.
YutaoLiu said: a. it seems that noise folding only happens with sampled pnoise sim. I would like to know the reason and more detail on the difference between the "sampled" and "unsampled" pnoise sim. Is there any material about this? Could you point me to that?
Andrew's response is exactly correct. If you need a bit more detailed explanation, Cadence On-line support has a nice Application note that graphically shows why the use of a sampled noise analysis is more appropriate than a linear noise analysis for a switched-capacitor circuit:
There is no "unsampled" pnoise simulation. In all cases, the impact of noise folding is included in a pss analysis. However, for non-autonomous circuits (i.e. driven) the magnitude of the resulting phase noise will be different as the time over which the noise is measured varies between a "timeaverage' (more conventional) and sampled jitter modes. In the former, the noise is averaged over the entire period and in the latter the noise represents the noise at the sample points/times only. As a result, the noise measured in a sampled (jitter) analysis will typically appear larger than that for a more conventional pnoise "timeaveraged" analysis. This is probably more clearly explained than in the few words I am providing at URL:
YutaoLiu said: b. In the version of tool I am using, there are only "timeaverage" and "sampled (jitter)" for Noise type. I think I should use "sampled (jitter)" to get the sampled Noise. But there is no option for "noisetimepoints" and "numberofpoints", which was used in the Listing1 in the document. How can I do an equivalent setup?
Are you studying the most recent version of this Application Note? I am seeing a much newer version (January 2021, page 343 "Characterizing a Switched-Capacitor Filter") and believe you should refer to it; This version of the manual is at Cadence URL:
Do these add any useful intuitive information to Andrew and Frank's comments for you? I hope so!
I need to spend some time to digest Andrew's and your comment.
By the way, the last link in your reply is not valid. It there any other way I can find this updated application note?