The datasheet does not mention anything about differential input (e.g., Figure 1 on page 1). Even Table 6 does not even mention J15 or J16 (which I see on the actual board and in the schematic).
From the schematic, it seems straightforward --- to set up input A as differential, just disconnect S25 and connect S15. However: what about S9 and S19? My signal will be band-limited to 1 or 2MHz, so I don't see the impedance matching to be an issue (I suspect I'm probably wrong on this?), and I don't like presenting a 50-ohm load to the signal source's output stage. Is it ok if I disconnect S9 and S19?
A related detail: why is the AD8138's feedback network non-symmetric? The non-inverting input (the one that connects to Va) has Rf = 499ohm and Rg = 499ohm. The inverting input (the one that connects to J15) has Rf = 499ohm and Rg = 523 ohm. 523 looks like 499 + 50/2 --- the Thevenin equivalent of the input stage is a source with a 50-ohm / 2 series resistor, so the 523 looks like attempting to match that. However, both inputs have the same Thevenin equivalent input, so it looks like one of them would end up with 499 + 50/2 and the other one 523 + 50/2.
I see that Table 3 reports S9 to be open (and it is reported to have a "N/A" function --- ???) whereas S19 has a factory default setting of closed (to terminate the input through a 50 ohm resistor). Is this the reason for the discrepancy? If so, then closing S9 will cause the imbalance --- but if I don't close it, then I get "matched" impedance on one input and non-matched on the other.
Bottom line: what should I do? I absolutely want to try to feed the board with a differential signal, since my setup involves 5 different boards, so ground loops seem to be a problem.