I think the reciprocity-based beamforming will fail in both asymptotic regimes. It is more clear to be affected at the low-snr, as the wideband slope is inversely proportional to the second order terms of the mutual information in the low-snr Taylor expansion. This in effect explains how the minimal energy per bit, is inversely proportional to SNR^2 at low-snr, while the tradeoff is that the SINR increases with M. Space or degrees of freedoms dimension increase tradeoffs the power dimension, which in essence breaks reciprocity-based beamforming even on the high-snr.
Many more can be done on the Massive MIMO research, if all of you here are interetsted in some joint collaboration!

]]>