Pilot-Efficient Channel Estimation for RIS-Aided Wireless Communication Systems: A Systematization

Passive reconfigurable intelligent surfaces (RIS) can reshape the propagation environment with negligible radio-frequency power, yet they introduce a dominant channel-acquisition bottleneck, because the cascaded channel to be learned is high-dimensional and its naive pilot cost scales with the number of surface elements and base-station antennas, rapidly exhausting realistic coherence intervals. This article systematizes pilot-efficient channel estimation for RIS-aided links. Starting from a structure-free uplink and downlink training model, the identifiability limits are made explicit, including the intrinsic bilinear ambiguity that precludes unique recovery of the two constituent hops without additional information. Least-squares and linear minimum mean-square error estimators serve as neutral baselines that expose pilot-scaling laws, conditioning effects, and the relationship to the Cramér-Rao bound under a linear Gaussian model. The literature is then organized into pilot-based, semi-blind, and learning-based estimators, and each family is tied to the structural priors it exploits to reduce measurements; to compare results across heterogeneous studies, a within-paper normalized mean-square error (NMSE) gap benchmark is introduced. The central finding is that pilot overhead is fundamentally a degrees-of-freedom problem. Without exploitable structure, only the effective cascaded channel is identifiable, so every substantial pilot reduction stems from imposing physically valid priors. Reported gains are shown to track the strength of these priors and the chosen baseline as much as the estimator family, and they erode when structure weakens or the operating regime is mismatched. Guided by this analysis, regime-dependent estimator-selection guidelines and open problems are distilled.