Several pieces of research model this self-interference (SI) using different random distributions like Rician, Nakagami-m, Rayleigh, etc. The cancellation techniques allow the SI suppression up to a certain level and only a very small amount of residual SI remains. According to my understanding, due to the presence of transmitting and receiving antennas at the same node the LOS component will dominate that can be canceled using cancellation techniques. But due to multipath effects, the residual SI component is very small.