It's a fine point and not really worth arguing about but, by definition, the signal on a balanced pair is differential and has no intrinsic relationship to ground (that's a property of the common-mode or CM component of the signal). But, in the real world, it's not desirable to have an infinite CM impedance at both ends - for one thing, such "floating" requires extreme shielding to avoid picking up high common voltages from relatively weak electric fields - it's why open telephone lines had a center-tapped transformer at at least one end. So having a low CM impedance at one end of a balanced line is a good thing. And generally, because the line driver must deal with supplying high currents (to drive cable capacitance at high frequencies), it's simplest to make the driver low impedance for both differential/signal mode and CM. Mics with balanced lines are usually an exception. A dynamic mic has very high CM impedance - much higher than the typical mic preamp input. So the interface can work well either way, it's strictly a practical consideration. But the signal in a balanced system doesn't require ground-referenced signals, it just happens when the line driver is a pair of voltage sources, whether symmetrical or not. Since the op-amp driving the high line is ground referenced in most inexpensive designs, in order to maintain impedance balance, an equal impedance must be used in the low line. Grounded voltage sources (op-amps) are the simplest and cheapest line driver designs. Of course, a balanced driver that "floats" for CM will theoretically be able to overcome the ground-voltage-difference (GVD) issues that plague unbalanced interfaces (provided the cable's low side is grounded only at the destination - an adapter at the driver output defeats the purpose). An undriven low side op-amp or matching grounded resistor obviously can't do that. It can only prevent high (and usually severly distorted) audio current from flowing into a remote ground point (potentially creating crosstalk) or permanently damaging itself. But, as a balanced output, it's as good as any other at driving a balanced input. A transformer, whether real or simulated with something like THAT's 1646, is the preferred solution. Cost, of course, will dictate the choice for many.