Saurkt,
    There are a couple of cases to consider. First, if you can simulate the entire PLL in SpectreRF, then it accounts for the variation in the VCO input. The variations have to be periodic of course (otherwise SpectreRF will not work), and the effect of this periodic variation will be sidebands on either side of the oscillation frequency that will be computed by the PSS analysis. If you are interested in both spurs and noise, you will have to manually combine the spurs (computed in PSS) with the noise (computed in PNoise).

If you are breaking the PLL up into it component pieces and characterizing them individually, then you do not need to worry about the variation on the VCO input, it will not be there (well, there may be a small component at the VCO oscillation frequency, but it is not a concern).  Once you create the time-domain models and simulate them, the variation will naturally be produced during the simulation if the models are sufficiently accurate, and the variation will result in spurs in the results. You do not see this in Figure 20 because the reference frequency is 200kHz, and the plot only goes to 100kHz. The spurs you are concerned about would only appear at the reference frequency and its harmonics. If you wanted to view the results at higher frequencies, you would have to take samples more often. In the example, the sample rate is limited by the fact that the divider model is included in the VCO model (this is done for efficiency). To plot to higher frequencies, you would need to pull at least some of the divider out and model it explicitly.

The Jitter Man