Intelligent Over-the-Air Computing Environment

Abstract

A key service of the sixth generation (6G) of wireless networks is envisioned to be native Artificial Intelligence, which calls for radical changes to the way the nodes communicate and perform computations, as well as the role of wireless environment. For this purpose, over-the-air computing (AirComp) is a promising technique for ultra low-latency wireless data aggregation, enabled by the waveform superposition properties of a multiple access channel. In this letter, the synergy of decentralized AirComp, reconfigurable intelligent surfaces (RISs) and machine learning is proposed, to transform the wireless environment to intelligent AirComp environment (IACE), i.e., with inherent and advanced capabilities to perform computations in a fully decentralized way at the physical layer. Specifically, we minimize the AirComp error, i.e., the average mean-square errors of devices with respect to a target function, by jointly optimizing the RIS phase-shift vector and the transmission and reception scaling factors of devices. Also, to solve this challenging problem, we propose an online deep neural network (DNN) optimization approach. Finally, simulation results validate the effectiveness of IACE and the proposed DNN approach.