Unsupervised Learning for Weighted Resource Allocation in RIS-Assisted mmWave MIMO Systems
Reconfigurable intelligent surface (RIS) has emerged as a promising technology for next-generation wireless networks due to its ability to intelligently manipulate the propagation environment. In RIS-assisted millimeter-wave multiantenna MIMO communication networks, the joint optimization of RIS phase configuration and resource allocation under heterogeneous user priorities remains challenging. This paper proposes a deep learning-based framework that incorporates user priority weights into both channel estimation and resource allocation through and unsupervised learning. We formulate the joint optimization problem of RIS phase shifts, base station beamforming, and user priority scheduling under α-fairness criteria. A neural network architecture is designed to learn the mapping from channel state information and user weights to optimal resource allocation policies. Simulation results demonstrate that the proposed approach achieves significant performance improvements of 6.5–13.8% in throughput compared to baseline schemes across multiple metrics. The devised framework attains enhanced performance metrics with lower computational burden, which renders it far more expandable than iterative optimization approaches.