Performance Limits of Compressive Sensing Channel Estimation in Dense Cloud RAN

TL;DR

This paper analyzes the fundamental performance limits of compressive sensing-based channel estimation in dense Cloud RANs, providing insights into the conditions where channel sparsification is valid and how training length impacts estimation accuracy.

Contribution

It offers a rigorous, analytical performance characterization of oracle estimators in CS-based CRAN channel estimation using stochastic geometry, highlighting operational conditions for sparsification validity.

Findings

Performance bounds for practical CS algorithms derived

Training sequence length significantly influences estimation accuracy

Channel sparsification valid only under high path loss conditions

Abstract

Towards reducing the training signaling overhead in large scale and dense cloud radio access networks (CRAN), various approaches have been proposed based on the channel sparsification assumption, namely, only a small subset of the deployed remote radio heads (RRHs) are of significance to any user in the system. Motivated by the potential of compressive sensing (CS) techniques in this setting, this paper provides a rigorous description of the performance limits of many practical CS algorithms by considering the performance of the, so called, oracle estimator, which knows a priori which RRHs are of significance but not their corresponding channel values. By using tools from stochastic geometry, a closed form analytical expression of the oracle estimator performance is obtained, averaged over distribution of RRH positions and channel statistics. Apart from a bound on practical CS algorithms, the analysis provides important design insights, e.g., on how the training sequence length affects performance, and identifies the operational conditions where the channel sparsification assumption is valid. It is shown that the latter is true only in operational conditions with sufficiently large path loss exponents.