The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> Circuit Simulators >> simulation of KT/C noise in an Ideal S/H
https://designers-guide.org/forum/YaBB.pl?num=1413365305

Message started by Yashar on Oct 15^th, 2014, 2:28am

Title: simulation of KT/C noise in an Ideal S/H
Post by Yashar on Oct 15^th, 2014, 2:28am

I tried to simulate the KT/C noise by PSS+PNOISE analysis as explained in "Simulating switched-capacitor filters with SpectreRF" (http://www.designers-guide.org/Analysis/sc-filters.pdf).
I used an ideal switch (relay) controlled by square wave clock signal series with a 770ohm resistor, dc input and a 10fF sampling capacitor as seen in attached netlist. The sampling clock was 7GHz and the S/H bandwidth around 20GHz, I lunched a shooting PSS with beat frequency of 7GHz with 100 harmonics and specified 770GHz for maxacfreq (as recommended in article 10xfc+Nmsbxfc). My first interesting range of noise folding in PNOISE analysis was up to 100 sidebands and that is why I choosed the above value for maxacfreq. The PNOISE was defined in range of 0 to 3.5GHz (half of the sampling clock) and I used time domain option and integrate the output noise from 0 to 3.5GHz. The result was 621.2425uV for 100 sideband, while it was 621.8437uV for 500 and 1000 sidebands of pnoise simulation (Just to note that I increased maxacfreq when I increase Nmsb). It seems that the stabilized value is around the 621uV while it has -3.5% error with the calculated analytical value of 643u (sqrt(KT/C)) and noise analysis of shorted switch ideal RC filter sim result. Is there anything wrong in my test bench or simulation setting, cause in the article the error is achieved less than 1% by increasing the number of sidebands?

Title: Re: simulation of KT/C noise in an Ideal S/H
Post by Ken Kundert on Oct 15^th, 2014, 11:55pm

Quote:

I used time domain option

I don't think you want to do that. You want the time-averaged noise, which for unknown reasons, ADE refers to as 'sources'.

-Ken

Title: Re: simulation of KT/C noise in an Ideal S/H
Post by Yashar on Oct 16^th, 2014, 2:24am

Thanks for your reply, I am interested in sampled data noise analysis at specific time point and I think it is the same time averaged noise by time domain feature of PNOISE. It is consistent with the setting of LISTING1 (page 3) and first paragraph of page 13 in reference. Meanwhile integrating the power of noise output from pnoise analysis with 'source' feature of noise type gives 325uV stabilized value.

Title: Re: simulation of KT/C noise in an Ideal S/H
Post by Ken Kundert on Oct 19^th, 2014, 12:30am

Oh my mistake. Yes, you would want to sample the noise. I did that in the original paper using a sample and hold because the work predated the time-domain noise option. I would still take a careful look at the time domain noise option settings. It is poorly designed and does not naturally do what you want. Be sure that you are sampling noise noise at the right time.

You might consider using a sample and hold along with the time-average noise and see how the results compare. When using time-average noise, you have to compensate for the duty cycle of the sample and hold.

-Ken

Title: Re: simulation of KT/C noise in an Ideal S/H
Post by Frank Wiedmann on Oct 20^th, 2014, 1:09am

I suggest that you use pnoise jitter analysis. It is equivalent to pnoise tdnoise but allows you to specify the sampling instant by giving a threshold level for the signal (see also http://www.designers-guide.org/Forum/YaBB.pl?num=1224609785).

I also recommend that you use the fullspectrum option of the pnoise analysis, which is available in APS mode (see http://support.cadence.com/wps/mypoc/cos?uri=deeplinkmin:DocumentViewer;src=pubs;q=/spectreRF/spectreRF14.1/Chap4.html#fullspectrumpnoise). It takes into account all sidebands for white noise, so you only need to specify the number of sidebands required to cover colored noise sources (usually 1/f-noise). If you use the fullspectrum option, be sure to set the value of the maxacfreq parameter of the pss analysis large enough (because it cannot be verified automatically against the specified number of sidebands in the pnoise analysis).