The frequency differentiation method would be able to approximate the phase difference by using consecutive I samples. But the demod product would probably be pretty crappy at best.
I think that the transform method should at least be able to simulate a phased Q sample which may be a bit better as a starting point. I messed with this a while ago when I was offloading a synthetic IQ stream to SDRAngel.
it almost worked, but I lost interest in the challenge..?
