I'm trying to predict synthesizer divider noise, but my results seem to be optomistic.

I've run PSS and PNOISE in time domain method to only sample noise on the edge that matters.  I then apply (32) from http://www.designers-guide.com/Analysis/PLLnoise.pdf.  However, when I then do 10log(Sphi(f)), my dBc number seems too optomistic.

Example:
Output noise at 1kHz: 28.8uV/sqrt(Hz)
fo=62.5kHz
SR=35GV/s

Sphi(1kHz) = [2*pi*62.5kHz/(35GV/s)]^2 * (28.8uV/sqrt(Hz))^2
Sphi(1kHz) = 104.4e-21 rad^2/Hz
10*log(104.4e-21) = -190 dBc

My assumptions:
PSS is run at 62.5MHz, not 62.5kHz, to speed simulation.  I don't feel this is a problem, since this is a synchronous counter.

I've simulated with different max sidebands to make sure the noise folding is accounted for.

I'm only using noise at a single point, do I need to integrate the noise over the transistion of the output signal?

Am I correct in thinking that 10*log(Sphi(f)) gives me a dBc, because I'm in radians? I.E., Sphi(f) is the noise at the end of a vector, which is always related normalized?

Thanks.