Enhancing Channel Estimation in RIS-aided Systems via Observation Matrix Design
Zijian Zhang, Mingyao Cui
TL;DR
The paper tackles high pilot overhead in RIS-aided dense array systems by formulating a Bayesian observation-matrix design that maximizes the mutual information between received pilots and the cascaded channel. It develops an alternating Riemannian manifold optimization (ARMO) algorithm to jointly optimize the BS combiners and RIS phase shifts, guided by the channel kernel $\boldsymbol{\Sigma}_{\mathbf{h}}$. An adaptive kernel-training scheme is introduced to refine $\boldsymbol{\Sigma}_{\mathbf{h}}$ frame-by-frame without extra pilots, enabling simultaneous channel estimation and kernel learning. Across simulations, ARMO achieves substantial NMSE gains and improved pilot efficiency, approaching ideal-kernel performance when kernel training is employed.
Abstract
Reconfigurable intelligent surfaces (RISs) have emerged as a promising technology for enhancing wireless communications through dense antenna arrays. Accurate channel estimation is critical to unlocking their full performance potential. To enhance RIS channel estimators, this paper proposes a novel observation matrix design scheme. Bayesian optimization framework is adopted to generate observation matrices that maximize the mutual information between received pilot signals and RIS channels. To solve the formulated problem efficiently, we develop an alternating Riemannian manifold optimization (ARMO) algorithm to alternately update the receiver combiners and RIS phase-shift matrices. An adaptive kernel training strategy is further introduced to iteratively refine the channel covariance matrix without requiring additional pilot resources. Simulation results demonstrate that the proposed ARMO-enhanced estimator achieves substantial gains in estimation accuracy over state-of-the-art methods.