Distributed mmWave Massive MIMO:A Performance Comparison with a Centralized Architecture for Various Degrees of Hybridization

mmWave massive MIMO (multiple-input multipleoutput) is envisioned to offer a considerable capacity improvement with respect to sub-6GHz systems. However, it suffers much more from path loss and shadowing. Moreover, the present hardware constraints limit the number of radiofrequency (RF) chains, which, in turn, significantly limits the performance of the system achievable with full digital beamforming. To mitigate these performance limiting factors, we propose to geographically distribute the antennas over the coverage area and cluster them into a number of remote antenna arrays (RAA). This leads to a shorter average distance between antennas and user equipment (UEs), more spatial diversity against shadowing and the opportunity to overcome the limitations on the number of RF chains. We analyze and compare, for a mixed-office environment and a specific geographical distribution of the RAAs, the performance of a centralized architecture and a distributed one with varying numbers of RF chains, encompassing hybrid beamforming and full digital beamforming. We assume low-complexity zeroforcing (ZF) beamforming on the downlink and max-min power allocation. Using Monte Carlo simulation, we obtain the empirical CDF of the per-UE spectral efficiency (SE). The results show that: (a) The distributed architecture has better performance than the centralized one in terms of per-UE SE; (b) When the number of RF chains is no less than twice the number of UEs, hybrid beamforming achieves the same performance as full digital beamforming in terms of per-UE SE.