My perception for capacity in this case follows:

According to TRP constraints and Pattern-Agnostic estimates of that for real time, the available power at the ports of the transmitter can be found. Then, capacity can be determined using the well-known formula of Shannon for an arbitrary noise. The discussion is about the proper choice of noise that should correspond to that vector on the ports of the transmitter. I suggest that it should be the power sum of all thermal noises on the ports.