Unfortunately I haven't been able to find any fancy numbers on this. The only thing the datasheet tells me (unless I missed it ofc) is that it(the chip) at ~3V "delivers +10dBm (CW) output power into a 50ohm load". Does this also mean that the output impedance of the PA is 50 ohm or only that it is able to deliver that through the matching network?
Dunno. +10dBm isn't much though. A lot of oscillators or DDSs will do more...
It could be unmatched or poorly matched, or it could just be low power. I have no idea...
Have I understood it correctly that I cannot know if the matching network makes sense unless I know the characteristics of the PA?
Yep.
And for example, in your sim -- if you knew it was a perfect voltage source, that would be fine. But a real power amplifier (of that type) never is. An audio amplifier, for example, is usually a pretty reasonable voltage source, compared to its expected loads, over most of its bandwidth -- but that changes at higher frequencies, where the output becomes inductive. Most RF amp devices are open-drain rather than emitter follower style (as audio amps are), so they have a generally CC || capacitive output characteristic.
In any case, the dynamics of the load network aren't separable from the dynamics of the source, not by just one number (like an impedance) anyway. Impedance is a complex valued function of frequency.
Also, you can't really do small-signal testing, because capacitances vary with signal level, and maximum power point (usually also with a modest compromise between low distortion and high efficiency) is not nearly the same as the conjugate small-signal impedance. Sadly, they are determined by separate, unrelated mechanisms, so that doesn't work.
I'm guessing the normal procedure for these is to take the reference design as planned, tweak to satisfaction and be done with it. I'm pretty much set on needing to fiddle around abit with values anyways depending on how the layout and such turns out but would rather have a basic idea on what is reasonable first.
Yeah, tweak it until you get the right output characteristics, is really about the best you can hope for. You'll want a spectrum analyzer to view the fundamental and harmonics, since filtering harmonics is probably a priority as well.
Tim