I'm designing an audio amplifier input stage. This amplifier is fully differential. Among other things, one of the design goals is to have good CMRR for UNbalanced (single ended) inputs with RCA to XLR cables (pins 1 and 3 tied to GND at RCA connector).
I mostly succeeded in this goal, except for the high end of the audio spectrum. At 20kHz, the CMRR with unbalanced input drops around 30dB from CMRR at 20Hz. This is in case of a source output impedance of 600 ohm, unbalanced with the low impedance ground of the single ended output.
The cause of the declining CMRR at higher frequencies, is in fact the RF input filters of the amp. It consists of a 100 ohm - 100pF RC networks very close to the input connector, preventing RF signals reaching any electronics where they might be demodulated into audible signals. These 2 RC networks are implemented on pins 2 (hot) and 3 (cold).
If it would be possible to have a selectable impedance at pin 3 (cold input) that could be matched with pin2 output impedance at the source, the CMRR would be much better at higher frequencies. This would mean 3 or 4 settings, preferably freely combined. A similar system is often found in phono preamps, where input loading resistors and capacitors can be combined with dip-switches.
The trouble is, that it must happen in the RF input network, where GHz signals should successfully drain to chassis instead of entering the signal chain. I have no clue if or how this could be implemented without compromising RF draining performance.
Edit: added schematic and simutaltion
The schematic is simple enough as it's a separate passive input connector PCB. The added switches are imaginary for now as I have my doubts they can be layed out in a way that does not compromise RF performance.
Now these are theoretical results and component tolerances significantly reduce CMRR performance, but the large differences between balanced and unbalanced input signal remains.
Edit2: reactions to answers and comments
It has been suggested that RCA to XLR cables are nonsense and should be avoided because they have no added value. Well, I disagree. Consider this type of connection between an unbalanced source and a balanced sink:
In this scenario, the balanced cable can cleary do it's work with regard to noise rejection. Even with the high frequencies compromised, at low frequencies it's still effective and performing better than an unbalanced input. Furthermore, an unbalanced input allows noise currents on to the active PCB's where they can cause havoc with all single ended circuitry. So, IMHO balanced inputs are always better performing, regardless the source that is connected to it.
Another suggestion was capacitive shunting of the differential input (behind the RF RC filter). A typical value would be 1nF. I had thought of this myself and it does indeed work in order to improve high frequency performance of CMRR. But it only works to prevent positive gains above 1Mhz, it does not change anything at 20kHz. Furthermore, this works equally well if implemented in the gain stage further down the chain.