Machine Learning-Enabled RIS-Assisted mmWave Multi-Hop Communications: Path Scheduling and Beamforming Design

This is the authors' abstract. We don't add key points for this paper.

In this paper, a reconfigurable intelligent surface (RIS)-assisted millimeter-wave (mmWave) multi-hop communication system is considered, where the base station (BS) transmits signals to remote users via multi-hop paths formed by multiple RISs. We aim to maximize the sum achievable rate by jointly optimizing multi-hop path scheduling, active beamforming, and passive beamforming, while satisfying the maximum transmit power of the BS. To tackle the formulated non-convex joint optimization problem, a machine learning (ML)-based two-stage algorithm is proposed. In the first stage, a multi-hop path scheduling algorithm based on graph neural networks (GNN) is investigated. Specifically, the considered system is first modeled as a graph topology, where the BS, RIS, and users are all served as nodes. Then, the weights between adjacent nodes associated with the path gain and the number of RIS reflection units are defined based on the expressions of RIS-assisted equivalent channels. Finally, a GNN-based multi-hop path scheduling algorithm is proposed. In the second stage, according to the obtained optimal multi-hop path, the deep deterministic policy gradient (DDPG) strategy is adopted to optimize the active beamforming of the BS and the passive beamforming of the selected RISs. Simulation results demonstrate that the proposed GNN-based multi-hop path scheduling algorithm maintains an error within 3% compared to the global optimal solution and outperforms the graph-based methods. Additionally, the proposed two-stage algorithm improves the sum achievable rate by approximately 27.6% compared to the alternating optimization (AO) algorithm.