Designer’s Guide Community

dgFL2007

New Member

Offline

Posts: 2

Noise simulation and PNoise simulation on PFD/CP
Mar 27^th, 2007, 10:02am

Hi, all,
I am new to here and I am not sure the same topic being brought up before. If so, please refer me to any similar topic on this forum.

I am designing a PLL right now and try to simulate the PFD/CP for its noise contribution. I had two ways. First method: Runing a small signal noise simulation when the charge pump is constantly on to get the output current noise POWER density (in A²/Hz), scale it with the actual duty cycle of the PFD/CP when the PLL is in lock condition, divided it by the PFD/CP gain which is Icp/2π to convert to PFD input refered inband noise floor. (There should be a scale factor of 2 somewhere to account for Single Side Band noise floor). Second method is from the document "Predicting the phase noise and jitter of PLL based frequency synthesizers, ver 4g" by Ken Kundert on this website. It is to run a pnoise simulation and to get the PFD/CP output current noise floor and directly converting it back to the PFD input refered noise floor by following equation 57 and 58 on page 30 of 51.
My problem is that this two method give me different results, methods 2 seems provides very optimitic results on the noise floor (By 10+dB). After carefully inspecting noise contribution table and checking the component contributing flicker noise, it seems to me that pnoise simulation scale the charge pump output current noise with the duty cycle factor instead of the current noise power as in method one. I myself is confused on this and I am wondering anyone here has an answer on this? (It seems that method one gives me a resutls that matche the measurement)

So my questions is if I am using small signal noise to simulate PFD/CP, should I scale the duty cycle factor with the charge pump output current noise (method 2) or current noise sqared (method 1).

BTW, the duty cycle I mentioned above is that the PFD has a small reset time ΔT (which is to help remove PFD-deadzone) and the duty cycle factor is ΔT/Tref, in which Tref is the reference period at the PFD input.

Thanks,
dgFL2007

Back to top

IP Logged

Frank Wiedmann

Community Fellow

Offline

Posts: 678
Munich, Germany

Re: Noise simulation and PNoise simulation on PFD/
Reply #1 - Mar 28^th, 2007, 12:43am

I believe that you should scale the noise power (current noise squared) with the duty cycle (method 1). The reason for the discrepancies you see lies probably somewhere else. Have you checked all the other points mentioned on page 30 of the article (like using enough bandwidth, sidebands etc.)?

You could try to build a simple example using e.g. an ideal noise source (preferably band-limited, use the noise/freq pairs of a voltage source) and a switch (whose duty cycle you might vary) to better understand the relationship between the results of noise and pnoise simulations. You could also use this example to check if the pnoise result really scales the current noise (and not the noise power) with the duty cycle (which I would doubt).

Back to top

IP Logged

dgFL2007

New Member

Offline

Posts: 2

Re: Noise simulation and PNoise simulation on PFD/
Reply #2 - Mar 29^th, 2007, 10:51am

Hi, Frank,
Thanks for your reply and insight. I did my homework on the cyclostationary stochastic process and found the mathematic answer should be method 1. I also build a behavior model for the charge pump and it showed that pnoise results followed a different equation. (At least for the source type to get SSB noise floor). The sideband I used in my pnoise simulation is 50 which I think is more than enough. I am contacting Cadence on this issue now.

Back to top

IP Logged

Frank Wiedmann

Community Fellow

Offline

Posts: 678
Munich, Germany

Re: Noise simulation and PNoise simulation on PFD/
Reply #3 - Mar 29^th, 2007, 11:55pm

Multiplying a signal (the noise in your case) with a pulse train in the time domain means that in the frequency domain, you convolute the spectrum with a (sin x)/x-shaped series of Dirac pulses. A shorter duty cycle means that there is less total power in the series of Dirac pulses, but at the same time the (sin x)/x shape becomes wider so that the noise at higher frequencies becomes relatively more important. This might lead to counterintuitive results when you look at the power spectral density of the noise. It also makes it important to use enough sidebands so that you take into account all the noise in the convolution (and it is the reason why I recommended that you use a bandlimited noise source in your model). When you look at the total noise (the integral over the power spectral density), however, it should definitely scale with the duty cycle. If that should not be the case in SpectreRF, I would consider it a bug. However, I have used pnoise analysis quite extensively in the past (mostly sampled pnoise) and have found it to be very accurate so far if all parameters are chosen correctly (which may be difficult if you are using this analysis for the first time).

Back to top

IP Logged

Frank Wiedmann

Community Fellow

Offline

Posts: 678
Munich, Germany

Re: Noise simulation and PNoise simulation on PFD/
Reply #4 - Apr 1^st, 2007, 1:27pm

One additional remark: depending on the bandwidth of the noise and on the switching frequency, 50 sidebands may or may not be enough. You also might want to have a look at http://www.designers-guide.org/Analysis/sc-filters.pdf. Although the case treated there is a little different from yours, you might still get some additional insights from this paper.

Back to top

IP Logged