Jim, with all due respect, you need to think about what I wrote in my original post. The sign of the loop gain is opposite for USB and LSB for a given PD null, i.e. if you use an XOR or a diode mixer whose nulls are at quadrature, if the loop wants to lock up at +pi/2 on USB, it'll lock up at -pi/2 on LSB. With a PFD, one sideband will have the right sign of loop gain to lock up at 0, where (as you point out) everything is copacetic.
The point I was making in my original post is that the other sideband will have to try locking up at +- pi, where there's a ruddy great cliff--its PD gain there is like Vdd /2*pi*(f_0*t_PD), i.e. something like 500 times larger than the other null. Of course it's noisy and possibly metastable there, but the point is that the loop gain is _huge_, so no lock will occur there. Thus with a tiny bit of acquisition aiding, e.g. 2 resistors and a cap in positive FB around the loop amplifier, you can make a reliable lock to one sideband and not the other, _without_needing_a_SSB_mixer_.
Cheers,
Phil Hobbs