Good idea, thanks Andrew.
I have a bit more information to add on this, but sadly the problem has been sticking around over the last few days. I’ve managed to get the lime to run in TDD mode through hacking some SPI register changes into my own custom writeSetting option in the SoapyLMS7 source, but this turns out to have not resolved the issue. I’ve attempted to simplify my tests by changing the antenna for a matched load on the output of the coupler, so now I’m getting:
With this as my test setup:
Looking at the phase graphs, there are discontinuities at about 2.2 GHz and at 2.56 GHz. The discontinuity at 2.2 GHz is not an issue, seeing as it impacts both RX1 and RX2 equally and therefore can be recovered from in the difference. The discontinuity at 2.56 GHz is an issue though, seeing as it is only affecting RX2 and then causing issues in the difference. Over running this test a few times, the discontinuities seem to appear randomly at different frequencies in either RX1 or RX2 or both at once.
I’m not sure if this behaviour is expected by default with SoapyLMS7? If it is, what is the cause of it, and how could I get around it? I think at this point it’s something in the RX PLL/LO chain, so that’s where I’ll be focusing my investigation for the moment, but any suggestions would be more than welcome 
Edit:
I derped out for a second there - It won’t be in the RX PLL settings, because TDD is on and the SXR is powered down.