On the degrees-of-freedom of the K-user distributed broadcast channel

In this paper, we consider the network MIMO channel under the so-called distributed Channel State Information (CSI) configuration. In that setting, each Transmitter (TX) shares the common knowledge of all user's data symbols but has only its own imperfect estimate of the CSI, based on which it designs its transmission coefficients in a decentralized manner. This setting could for example be encountered when the user's data symbols are cached beforehand at the TXs or in a Cloud-RAN scenario with partial centralization where the CSI sharing between TXs suffers from the delay due to the sharing through the backhaul network. Our first contribution consists in rigorously computing an upper-bound for the DoF through the analysis of a genie-aided setting. Building on this upper-bound, we present a new robust decentralized transmission scheme that achieves the optimal DoF over several important CSI configurations. In strong contrast to the conventional zero-forcing scheme which is known to be limited by the worst estimate across TXs, the proposed scheme is able to leverage the more accurate CSI at few TXs to improve the DoF. Surprisingly, we show that there is a CSI regime in which the optimal DoF only depends on the CSI quality at the best TX, such that sharing the channel estimate to the other TXs does not improve the DoF.