That should capture the noise folding, although of course the ideal nature of the controlled sources will mean that they are not capturing any device noise of the equivalent blocks in the design - but apart from that, it's fine.
You cannot use an ac analysis, because this is not a time-invariant linear system. You can't use pss, because if it is a fractional pll (a sigma delta), then there's no periodic solution; so it's a non-periodic time-varying system. Secondly, the z-domain controlled sources also suffer from "hidden states" and so cannot be used with pss.
So transient is your only solution here, I think (from the level of detail that you've provided, at least).
You can use vsource in analogLib and set the type to "prbs". See "spectre -h vsource" for more details. Alternatively there's a veriloga model in (I think) ahdlLib called something like rand_bit_stream.
Where can I find this document?
Application Notes on Direct Time-Domain Noise Analysis using Virtuoso Spectre
Not sure where you saw a reference to that document, but I think that the app note you're requesting has been replaced by this one.