Spatial CSIT Allocation Policies for Network MIMO Channels

TL;DR

This paper investigates optimal CSI sharing among distributed transmitters in network MIMO, proposing a finite neighborhood CSI allocation policy that achieves maximal degrees-of-freedom without unbounded CSI requirements.

Contribution

It introduces a network geometry-aware CSI allocation policy that limits cooperation to finite neighborhoods, overcoming limitations of traditional uniform CSI dissemination.

Findings

CSI requirements are finite and depend on network geometry.

Finite neighborhood cooperation achieves maximal degrees-of-freedom.

Proposed policy overcomes edge interference and unbounded CSI growth.

Abstract

In this work, we study the problem of the optimal dissemination of channel state information (CSI) among K spatially distributed transmitters (TXs) jointly cooperating to serve K receivers (RXs). One of the particularities of this work lies in the fact that the CSI is distributed in the sense that each TX obtains its own estimate of the global multi-user MIMO channel with no further exchange of information being allowed between the TXs. Although this is well suited to model the cooperation between non-colocated TXs, e.g., in cellular Coordinated Multipoint (CoMP) schemes, this type of setting has received little attention so far in the information theoretic society. We study in this work what are the CSI requirements at every TX, as a function of the network geometry, to ensure that the maximal number of degrees-of-freedom (DoF) is achieved, i.e., the same DoF as obtained under perfect CSI at all TXs. We advocate the use of the generalized DoF to take into account the geometry of the network in the analysis. Consistent with the intuition, the derived DoF maximizing CSI allocation policy suggests that TX cooperation should be limited to a specific finite neighborhood around each TX. This is in sharp contrast with the conventional (uniform) CSI dissemination policy which induces CSI requirements that grow unbounded with the network size. The proposed CSI allocation policy suggests an alternative to clustering which overcomes fundamental limitations such as (i) edge interference and (ii) unbounded increase of the CSIT requirements with the cluster size. Finally, we show how finite neighborhood CSIT exchange translates into finite neighborhood message exchange so that finally global interference management is possible with only local cooperation