Regularized ZF in cooperative broadcast channels under distributed CSIT: A large system analysis

Obtaining accurate Channel State Information (CSI) at multiple Transmitters (TXs) is critical to the performance of many cooperative transmission schemes, including joint precoding in the context of network MIMO. Practical CSI feedback and limited backhaul-based sharing creates degradations of CSI which are specific to each TX, giving rise to a Distributed (D-) CSI configuration. In the D-CSI broadcast channel setting, each TX implements separate elements of the joint multi-user precoder based on its own individual CSI estimate. In this work, we presents a first finite-SNR regime rate analysis for a network-MIMO (broadcast channel) under a distributed CSI setting. Of particular importance is the notion of a “price of distributedness” which penalizes the D-CSI setting over the conventional centralized one with the same overall feedback quality. To tackle this problem, we apply tools from the field of Random Matrix Theory (RMT) to derive deterministic equivalents of the Signal to Interference plus Noise Ratio (SINR) for a popular class of precoders. Our key finding lies in the notion that the price of distributedness converges to a predictable value, bounded away from zero, as the number of antennas grows.