Hi!

Have been working somewhat with simulations of phasenoise of frequency dividers.
This post will first contain some questions and then some advise for you guys doing phase noise simulations.

Question 1:
In the paper, which I tend to forget but I guess it's title is "Predicting jitter and..."
He says that [Spectral phaseNoise density] = (2*pi*f0/SLOPE)^2*[SpectralDensityNoiseV^2/Hz]
The unit of [Spectral phaseNoise density] is it in radians? So 10*log([Spectral phaseNoise density]) would be dBr/Hz.
Getting rid of the 2*pi would give me dBc/Hz?


Question 2a:
In his paper he says that f0 is the VCO frequency, fvco. Is that really so? Not fvco/divisionNbr? Don't have all my equations right here but I found it to be fout_dividerX instead of fvco etc. Finally I got: if I have chain of dividers that they all add equally much phase noise to the output (one have to relate the phase noise from each divider to the input frequency).
Question 2b:
The Slope I assumed to be slope of the output of the frequency divider. Am I all wrong?

Question 3:
Perhaps the key question. I have found two ways of simulating the phase noise. One is according to the paper "Predicting jitter and..." and one by performing PSS and a pnoise and instead of jitter/timedomain using source. Under "Main form" there is a magic button saying "Phase Noise". Can anyone explain what that PhaseNoise button does? It gives me some what equal result  ( 2db off) compared to the jitter/timedomain simulations before I apply any equations to it.



Thank you in advance,

Andreas

Edit:
Promised some advice:

When setting PSS: Fundamental tone: InputFrequency/DivisionNbr otherwise it will not converge.
When setting pNoise do not set start frequency to 0 Hz since the flicker noise will be infinity causing a warning and in the result no flicker noise will be shown (even at f > 0 Hz).
When using timedomain: pNoise starts over? Set skip counts = 0 then the simulator will only simulate your specified point.
When using jitter you will not be able to print noise summary, use time-domain instead.

One thing that you need to change is the fundamental beat tone. It should be the lowest frequency appearing in your circuit. (i.e input frequency / N)

I don't know if that will help you.

Further on I ran a PSS, noted when my positive transition occurred. Typed in the time in strobed PNoise simulations.
What you get i the spectral noise density at that time. Divide it by the slope^2 of the signal and you will get time jitter spectral density. (s^2/Hz)

Divide the time jitter spectral density with the (period time of the output)^2 and you will get some kind of phase noise error spectral density. (the number is between 0 and 1) At a 2*pi in you calculations and you will get in radius.


This is at least how I calculated it. I have not done any measurements to see if it is correct. Does anyone have any oppinion?