Well I can see how that bit works. If the TX and RX signal paths are short enough (remember EM energy doesn't move infinitely fast), you can subtract the TX's signal from the RX's signal and appropriately gain adjust it, then you will be left with the RX's signal only.

However the problem I have with this is that if your signal propagates 100m then is reflected back another 100m, how does the RX then discriminate between that signal and the signal from another TX on the basis that the signal is so far out of phase (670ns P-P approx assuming vacuum) having travelled such a distance and back.

You need buffers, software and all sorts of magic there as the RF carrier has gone a few cycles then.

Good point. It comes down to how well the analog cancellation will attenuate the in-band blocker (cf. IIP3 spec of the RF LNA).