We consider the downlink of a multiuser MIMO channel, corresponding to a single cell with an N
t-antenna base station and K single-antenna mobile terminals (MTs). It is known that when full channel state information (CSI) is available at the transmitter (full CSIT) the capacity of the system scales as N
t log(P/N
t- log K), under a total power constraint P [1],
... [Show full abstract] While, when the transmitter has no CSI, scaling reduces to that of a TDMA system. This paper examines the more realistic case of having an intermediate state of CSI. The key idea is based on a split of the allotted feedback between two stages: A first stage devoted to scheduling followed by a second stage for precoder design for the selected users. Based on an approximation of the achievable sum rate, we introduce a method for determining the splitting of the feedback rate so as to maximize performance and provide intuitions. We illustrate the gains of the 2-stage approach via Monte Carlo simulations.