Summary: I would like to build a differential amplifier with differential output, but shift the common mode to a different level from the original.
My current knowledge takes me this far: Take the traditional 3-opamp instrumentation amplifier, like the one in this image:
Now if you take the left two opamps without the third, those already give you almost what I want, that is, amplify the differential input and give a differential output. The only problem is it preserves the common-mode of the input. Adding the 3rd opamp on the right, it is easy to shift the CM by biasing its ground (in fact this is what most single-chip instr. amplifiers do when they provide a Vbias pin), but the output of the circuit is now single-ended.
So what is the best way of keeping both the differential output and the CM-shift? One way is, I guess, to take only the left two opamps of the above instrumentation amplifier, and shift the ground of each separately.
Another option that comes to my mind is to take only the left two opamps again, and (using an example when I want to halve the CM) use twice the gain as needed, and then divide each output by 2.
Unfortunately both of these solutions require more (in quantity) highly matched resistors with low TCR (I'm trying to keep temperature drift of the circuit very low), and those are bloody expensive.
So how would you go about this problem? Maybe taking an instrumentation amplifier is the wrong start? Is one of my above solutions the "standard" way of doing this, or are there better circuits for this purpose?
EDIT: Clarification on matching resistors: What I mean is to match them in TCR, because I am aiming to minimize temperature drift. This means I need to match resistors in TCR, not in absolute value, so that when they drift due to temperature, they will keep their original ratios. Actually I am uninterested in matching absolute values (almost, I still need a little bit of matching to maintain CMRR), for two reasons: 1) a mismatch in absolute value causes offset and gain errors, both of which are easy to calibrate out at system level. Measuring and correcting temperature drift is a lot more difficult. 2) Most of the offset errors will be non-existent anyway without even calibrating, because this will be a frontend to a sensor, and offset errors will cancel out due to AC excitation of the sensor. Anyway: closely TCR-matched resistors are expensive (easily more expensive than precision opamps), so I want to minimize their use.