Machine Learning-Based Infrastructure Sharing and Shared Operations for Intelligent Reflecting Surface-Aided Communications.

This study investigates the sharing of an intelligent reflecting surface (IRS) among multi-mobile network operators (MMNOs). IRSs are an energy-efficient option to manipulate electromagnetic waves; however, constraints on their installation can result in competition among MMNOs, increasing redundancy and energy consumption. A promising solution is to share the IRS among MMNOs; however, negotiating among MMNOs is challenging; thus, reaching a consensus on different IRS control requirements of each MNO is challenging. We propose a machine learning-based method that realizes IRS sharing among MMNOs to resolve this problem. In the proposed method, each MNO trains a neural network to obtain knowledge of the relationship between the IRS reflective coefficient and received power at each location. Subsequently, the IRS controller uses the models trained by each MNO to design a passive beamforming strategy that maximizes the total transmission capacity of all MNOs. The effectiveness of the proposed method in the MMNO scenario is demonstrated through simulations. Particularly, the proposed method outperforms the time-division method, which divides time to configure the optimal phase shift for each MNO at each divided time slot. Based on the results, our proposed method determines the best reflection coefficient to reach a consensus among MNOs in IRSs shared scenario.