Frank_Heart wrote on Nov 19th, 2017, 10:18pm: Any insights why CMRR is limited by R1/R2 matching?
The ideally same value (except from sign) of the gain from V1 and V2 is achieved with nominally same valued, but still different resistors. If these resistors differs (R1_1 from R1_2) than the gain from V1 and V2 will differ, thus there will be a common mode component at the output.
Frank_Heart wrote on Nov 19th, 2017, 10:18pm: Is it also ture for the conventional 3-amp instrumentation amplifier?
Yes. The same principle.
Frank_Heart wrote on Nov 19th, 2017, 10:18pm: What's the advantage of this topology, other than saving one opamp?
Saving the opamp. I just read a book chapter just for fun, so this is my first encounter with this circuit and I did no detailed comparison. This circuit is asymmetrical. Enough reason to be skeptical about it if high performance is required. The path from the two input voltages will see different amount of offset, noise, PSR, coupling noise, and who know what else. I would go for this topology only if power and area is important while the other specifications are relaxed.