Performance Limits of Compressive Sensing Channel Estimation in Dense Cloud RAN

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.