Low-cost integrated resistor networks can be used to implement a low-parts-count two-op-amp differential amplifier with low common-mode gain. Integrated onto a common substrate and processed together, resistors of integrated networks match well and are an alternative solution to discrete resistors, especially for amplifiers with a wide common-mode input voltage range. The amplifier circuit is shown below.

In some diff-amp applications such as high-side current sensing, where the sense resistor voltage to ground can have a large range, it is important that the op-amp differential amplifier (diff-amp ) reject the floating voltage of the sense resistor. In other words, the common-mode (CM) gain, ACM of the diff-amp must be minimized. The above amplifier circuit is a two-op-amp diff-amp with half of an 8-R resistor network. Two parts (using a quad op-amp) can implement a dual diff-amp circuit with low error caused by CM gain.
Two sources for CM gain are the op-amps used in the diff-amp and the gain-setting resistors around the op-amps. Amplifier CM rejection is usually specified as the CM rejection ratio ,

where Av is the (desired) differential gain,

and vo = amplifier output voltage. The (undesirable) amplifier CM gain is

where the common-mode voltage is

and is the average of the voltages at the inputs. As the CM voltage changes, ideally no change in the output voltage of the diff-amp occurs, and only the difference voltage at the diff-amp inputs is amplified. For the above circuit with equal resistors in the network, the quasistatic gain is Av0 = 2.
The derivation of A¬CM caused by resistor mismatch is rather involved. (See Planet Analog articles, Seemingly Simple Circuits, Part 3: Effect of Resistor Tolerances on Diff-Amp Gains and Seemingly Simple Circuits, Part 4: Diff-Amp Common-Mode Rejection by this author for details.) However, it reduces to a rather simple approximate formula:

where ε = resistor tolerance. A +/-0.2 % resistor network matching tolerance has ε = 0.002.
Thick-film R-networks are low-cost and readily available from passive-components manufacturers. They typically are specified at 1 % to 2 % absolute inaccuracy. What matters for low A-CM is how well the resistors match, and being made in a common process on a common substrate, they match typically to 0.2 % to 0.4 %. Thick-film TCR tracking is typically about 50 ppm/o C, comparable to T2 1 % metal-film resistors.
Suppose that the resistors match to 0.33 %, or ε = 0.0033. Then the CM gain of the given diff-amp is

The CM error resolution is 6.83 bits and ACM ≈ 8.8 mV/V. At a +/-0.2 % match, the error resolves to 1/187.5 = 5.33 mV/V, or about 7.55 bits. For many applications, this CM error is acceptably low. If so, a CM trimpot adjustment (of the bottom R of the string) is eliminated from the circuit. If it is not sufficient but close, such as for an 8- to 10-bit system, then software calibration can remove the CM error. (See Part 2 of the above-mentioned diff-amp article series for that.) Greater precision can be achieved using thin-film R-networks at greater cost – an alternative to discrete +/-0.1 % metal-film resistors.
A 16-pin SMT or DIP R-network has 8 matched resistors, usually of the same value. For a gain of up to 4, the extra resistors can be placed in series or parallel to modify Av . Placing resistors of a given tolerance in either series or parallel results in the same tolerance for the combinations.
Interesting article! As you are aware, this circuit was used by Analog Devices and National Semiconductor some ~30yrs ago. Worked well then, and probably better today with today's better OP amps.
I used this circuit when I designed my high side/low side gate driver input stage. I needed fully differential performance to +/-1200V — using only 30V Vbdss devices with similar gate to substrate breakdowns. It was a simple matter of using a pair of matched 1000:1 attenuation networks (..placed in front of the differential amplifier circuit) created with poly silicon resistors over thick field oxide — 2um thickness (thermal oxide has a breakdown characteristic of ~ 100V/.1um.)
All the Best,
Sam Ochi
sroochi's +/-1200-volt on-chip common-mode-range differential input: Excellent!
I have used this technique to create a product, IX6R11S3, a 6Amp high side/low side gate driver with +/-600V high side to low side gate drive swing capability about 10 yrs ago. I was granted a patent, #6759692, “Gate Driver With Level Shift Circuit,” issued 10/23/03.
Enjoy,
Sam Ochi