Thanks, Jim. That doesn't correspond to the data sheet, though. Fig 20 and eqns 3, 4, and 5 show how to calculate U', which I want to be 1/37 = .027. (From equation 1 this should provide a gain on the XY product of 37.) For a U' of .027 and U = 1.05, k = .974 from eqn 4. If R1 in figure 20 = 100 that makes R2 = 3740.
If you solve for k and U' from the resistors in Fig 21, you get k = .755, U' = .257, which provides the gain of 4 (12 dB), which the example is showing, also.
Personally, I find the whole k / U' thing confusing. If you write the equation for the output voltage of the block diagram of Fig 19 with feedback from W to Z as in Fig 20, you have vo = XY/1.05 + vo*(R2/(R1+R2)). Gain Av from XY to vo is vo/XY, so divide the whole equation by XY and solve for Av = (R1+R2)/(1.05*R1). Using this equation gives you the correct gain on XY for the examples in Fig 20 and 21, and also for my circuit.
Gerrit
EDIT 2 Jan: I tried swapping the resistors (against my better judgment) and it didn't work. The offset is indeed removed, but so is the gain (aka scaling with U' < 1). The gain is now a max of ~1, as the equation Av = (R1+R2)/(1.05*R1) predicts.