To me the most interesting part of the paper was how they synchronize the streams from multiple receivers (they inject a common noise signal to all receivers and then align the individual streams using cross-correlation).

> It was also observed that the relative phases of the signal streams drift slowly during the operation.

This is an interesting observation. If they are running all RF PLLs from a common reference clock, and the PLLs remain locked, there shouldn't be any long-term drift.