What for?
The transformation from a single ended filter is trivial: mirror it so you have a symmetrical network around ground. Then remove the ground -- all the capacitors get connected in series, so their values go down by half. And the filters are wired in series, so the system impedance is double.
So for a 100 ohm differential filter, use a 50 ohm design, halve the capacitor values, and double the *count* (but not value) of inductors (because there's one in each line).
Which is one reason why they suck: you need double the inductors!
They're also not that great because the common mode impedance is poorly controlled, which can make things quite a bit worse for whatever your signal is. You're typically better off filtering each signal independently, using matched filters so that the common mode is attenuated without causing a difference in phase shifts.
If the common mode impedance is coming from unshielded cable, then it's probably a somewhat high impedance, like 150 ohms or so. You could, in principle, design a properly matched CM+D filter, where instead of a matched pair of filters (e.g., 50 ohms normal mode per filter, 100 ohm differential), or a fully differential filter (100 ohm differential, ??? normal mode), you use capacitors in a delta/wye configuration, and coupled inductors, everything dimensioned so that the cutoff frequencies in both modes are equal, and the impedances are appropriate (e.g., 100 ohm diff,
150 ohm CM).
But that's a whole lot of bother, for no value (who cares about CM signal quality!...as long as you can filter it effectively, that is).
So, this is further reason to be gentle in the first place:
- Use low order, Bessel or Butterworth type networks, which have looser tolerances.
- Design the rest of your system around this limitation, say by sampling an ADC at a relatively excessive rate, or using more RF conversion steps to pare down the bandwidth more gradually.
- Get rid of the common mode signal at the earliest convenience, say by using a transformer or differential op-amp. (The xfmr/amp has to be very well balanced itself, of course.)
- Or filter it out separately, which can be done reasonably well (without impacting the original signal) using coupled inductors and whatnot.
Tim