Enhanced reconfigurable intelligent surface assisted mmWave communication: A federated learning approach

Abstract

Reconfigurable intelligent surface (RIS) has been proposed as a potential solution to improve the coverage and spectrum efficiency for future wireless communication. However, the privacy of users' data is often ignored in previous works, such as the user's location information during channel estimation. In this paper, we propose a privacy-preserving design paradigm combining federated learning (FL) with RIS in the mmWave communication system. Based on FL, the local models are trained and encrypted using the private data managed on each local device. Following this, a global model is generated by aggregating them at the central server. The optimal model is trained for establishing the mapping function between channel state information (CSI) and RIS' configuration matrix in order to maximize the achievable rate of the received signal. Simulation results demonstrate that the proposed scheme can effectively approach to the theoretical value generated by centralized machine learning (ML), while protecting user' privacy.