Robust Design of RIS-Assisted Over-the-Air Computation in the Face of Long-Term and Short-Term Imperfections

Xiongfei Zhai,Guojun Han,Yunlong Cai

DOI: https://doi.org/10.1109/tvt.2024.3419807

IF: 6.8

2024-01-01

IEEE Transactions on Vehicular Technology

Abstract:Recently, a promising technique of aggregating massive data from distributed Internet-of-Things Nodes (IoTNs) has been proposed, which is referred to as over-the-air computation (AirComp). Since the computational accuracy degrades as the number of IoTNs increases, reconfigurable intelligent surfaces (RISs) may be considered for improving the computation performance by improving the degrees of freedom. However, the passive beamforming performance of RISs degrades in the face of environmental factors, e.g., snow, dirt, and precipitation. To address this impediment, we investigate the robust design of RIS-assisted AirComp systems under realistic imperfections. We first characterize two models of imperfections in terms of their statistical features, namely by the long-term and the short-term model. To deal with the long-term imperfections, a minimum-mean-squared-error (MMSE) estimator is harnessed first for estimating the imperfections. Then we propose an optimization algorithm to optimize the computational accuracy by alternatively updating the transmit coefficients of the IoTNs, the active receive beamforming of the fusion center (FC), and the passive beamforming of the RIS. In particular, the penalty dual decomposition (PDD) technique is invoked for updating the passive beamforming vector. As for the short-term imperfections, we resort to optimize the computational accuracy in the worst-case scenario owing to imperfect passive beamforming at the RIS. Another alternating optimization algorithm is developed for solving the associated Min-max problem. Our numerical results verify that our developed beamforming designs substantially outperform their non-robust counterparts. Specifically, our algorithm conceived for mitigating the long-term imperfections approaches the performance of idealized perfect beamforming designs relying on unimpaired RISs.