Doesn't matter very much.
Consider the power supply network impedances. Both will have bypass capacitors in their respective domains (analog and digital), so we can assume whatever noise is present on one acts as a low impedance source.
The only thing different is which side the regulator is connected to. So let's consider the regulator's impedance. A 7805 or similar has fractional ohms impedance at signal frequencies, rising slightly at higher frequencies (roughly corresponding to the internal emitter resistance of the pass device). See Fig. 32, p.36:
http://www.st.com/web/en/resource/technical/document/datasheet/CD00000444.pdfAt such frequencies, bypass caps take over (~20m to 20 ohms depending on type).
So the question is, which side should get slightly more bypass -- because, the regulator itself acts as a dynamic bypass, as can be seen from the impedance curve. Not a great one: a pretty good voltage source at low frequencies, but inductive at higher frequencies.
Which is why you'll get a "blippy" sort of step response when paralleled with a small capacitor, but it needn't be unstable (i.e., an oscillator). Sufficiently large, and optimally lossy, caps can be chosen to effectively bring that resonance down to a frequency where it just goes "thud" instead, and then you have a well bypassed supply. (Roughly, 50mohm and 3kHz would suggest 1000uF, quite a lot for a linear regulator, but hey, if you're desperate, it should be optimal. Just beware of its peak current capacity, eh?)
In short, not very much difference. One way, the regulator shunts digital noise directly, then the ferrite bead and bypassing filters it.
The other way, the bypassing shunts digital noise directly, then the ferrite bead, and regulator and bypassing, filters it.
BTW, ferrite beads are probably useless here: they saturate at extremely low currents (compare the recent video on ceramic capacitors, but just switch everywhere "voltage" with "current"). Look for real inductors, and design the supply network so it remains well damped. (FWIW, a ferrite bead is roughly 0.1 to 2uH depending on value and size.) Usually, a bunch of small, low ESR caps, with one or a few bulk caps with optimal ESR to dampen any resonances (between the small caps and trace inductances, and between filter inductances and whatnot).
Tim