If you want infinite bandwidth from such a transformer, and don't mind a flat 6dB insertion loss: split the transmission line in the middle, and add terminations here. You need an R+C across the (now open) ends of these windings, where:
R = Zo
C >= 2.5 * k * (377Ω) * t_half / Zo^2
k = effective dielectric constant (note this is less than the material k itself, when the dielectric is mixed, e.g., microstrip, twisted pair, foam, etc.)
t_half = transmission line electrical length (units of time)
This terminates the half-windings, so the transformer looks like a pair of Zo resistors from P1 to S1 and from P2 to S2. Obviously, this breaks transformer action.
One might rightfully note this isn't a transformer at all anymore, and the circuit can be greatly simplified by removing the transformer altogether; now you have coupling capacitors (the termination resistors are completely optional, unless you liked the insertion loss), and the CMRR is still just as bad, although it's bad at low frequencies too (oh well?
![Tongue :P](https://www.eevblog.com/forum/Smileys/default/tongue.gif)
).
To restore transformer action, connect across the cut ends (primary middle to primary middle, secondary middle to secondary middle) with inductors, of similar value (i.e., L ~ Zo * t_half). Now you get asymptotically zero insertion loss at low frequencies, and at high frequencies, a 6dB shelf instead of the dips and peaks.
Aside of all of this, CMRR is asymptotically bad in the HF limit. To address this, place a CMC on either side of the transformer, as much (equivalent) inductance as you can afford. Add damping R+C across the transformer (P1 to S1, and P2 to S2) to control resonances, if necessary (these will be a large impedance, so will have little impact on HF response, despite their location). The CMC inductance will resonate with the transformer isolation capacitance (and coupling/termination capacitance C, if used, as above), so this prevents that resonance from getting too peaky. An isolation impedance in the low kOhms is very reasonable to achieve this way, and while that doesn't sound terrifically good, understand that we're talking with respect to radio frequencies here, where a thin wire in semi-free space is unlikely to reach half a kohm Zo. This is doing as well as you can, given the limitations of real electromagnetism and not just some inductors and capacitors in a SPICE model.
![Smiley :)](https://www.eevblog.com/forum/Smileys/default/smiley.gif)
Tim