Here's why I think it's better to use a balanced attenuator on balanced inputs and outputs. Consider the simple equivalent circuit of the
ideal impedances of a balanced output driving a balanced input:
Rs is the balanced source impedance, Rdiff is the differential input impedance, and Rcom is the common-mode input impedance. The attenuation in both "legs" of the balanced line is equal.
It may have been more accurate, in an absolute sense, to model Rcom as a single resistor to ground off a centertap of Rdiff, but the result is going to be the same for this particular analysis, anyway.
Here's what happens when you introduce an unbalanced attenuator (such as a pot or an unbalanced bridged-T):
The attenuation in the two legs is now unequal and CMRR suffers. If Rcom is very high, then it may not suffer enough to be very noticeable. That's why high common-mode input impedance is important to CMRR, and transformers tend to excel in this regard. But we can't always guarantee that we're always going to hook up devices with a high common-mode input Z and excellent CMRR to the output of our device; we may very well need all the help we can get.
Here's the case with a balanced attenuator, shown here as a simple "u" pad:
Balance and CMRR are preserved.
If you have an output which needs to see something not-too-much-more-or-less-than 600 ohms, and a high-Z input that is not sensitive to variations in source impedance,
and if you can tolerate a 3dB minimum loss, then this simple variable "u" attenuator might be the ticket:
I've had an idea for a "universal" 600-ohm bridged-H attenuator, 0 to -50dB in 2dB steps plus "off", that could be build cheaply in a little Bud box. It looks like it's time to flesh out that idea. I'll post a schematic soon, depending on how much goof-off time my day job allows this week