Rethinking network MIMO: Cost of CSIT, performance analysis, and architecture comparisons

We compare the downlink throughput of various cellular architectures with multi-antenna base stations and multiple single-antenna users per cell, by considering a number of inherent physical layer issues such as path-loss and time and frequency selective fading. In particular, we focus on Multiuser MIMO (MU-MIMO) downlink techniques that require channel state information at the transmitter (CSIT). Our analysis takes explicit account of the cost of CSIT estimation and illuminates the tradeoffs between CSIT, estimation error, and system resource dedicated to training. This tradeoff shows that the number of antennas that can be jointly coordinated (either on the same base station or across multiple base stations) is intrinsically limited not just by ¿external factors¿, such as complexity and rate of the backbone wired network, but by the inherent time and frequency variability of the fading channels. Our analysis, in agreement with a number of recent simulation results, shows that conventional MU-MIMO cellular architectures may outperform schemes based on coordinated transmission from base stations (referred to as Network MIMO schemes, NW-MIMO), at the negligible cost of a few extra antennas per station. In light of these results, it appears that the inherent bottleneck of NW-MIMO systems is not the backbone network (which here is assumed ideal with infinite capacity) but the intrinsic dimensional limitation of estimating the channels.