Towards Optimal CSI Allocation in Multicell MIMO Channels

TL;DR

This paper proposes an efficient CSI feedback allocation method for multicell MIMO channels that reduces the feedback needed and enables scalable joint transmission, especially in large networks, by exploiting channel structure.

Contribution

It introduces a novel CSI allocation strategy based on the Wyner model and pathloss structure, improving scalability and reducing feedback in distributed MIMO systems.

Findings

Achieves good performance with feedback bits per TX independent of the number of TXs.

Reduces cooperation to a local scale, simplifying implementation.

Extends to exponentially decaying channels modeling 1D TX cooperation.

Abstract

In this work, we consider the joint precoding across K transmitters (TXs), sharing the knowledge of the user's data symbols to be transmitted towards K single-antenna receivers (RXs). We consider a distributed channel state information (DCSI) configuration where each TX has its own local estimate of the overall multiuser MIMO channel. The focus of this work is on the optimization of the allocation of the CSI feedback subject to a constraint on the total sharing through the backhaul network. Building upon the Wyner model, we derive a new approach to allocate the CSI feedback while making efficient use of the pathloss structure to reduce the amount of feedback necessary. We show that the proposed CSI allocation achieves good performance with only a number of CSI bits per TX which does not scale with the number of cooperating TXs, thus making the joint transmission from a large number of TXs more practical than previously thought. Indeed, the proposed CSI allocation reduces the cooperation to a local scale, which allows also for a reduced allocation of the user's data symbols. We further show that the approach can be extended to a more general class of channel: the exponentially decaying channels, which model accuratly the cooperation of TXs located on a one dimensional space. Finally, we verify by simulations that the proposed CSI allocation leads to very little performance losses.