다중 사용자 MIMO 시스템에서 그래프신경망 기반 확장 가능한 빔포밍 최적화

Abstract

This paper addresses a beamforming optimization inmulti-user multiple input multiple output (MIMO) systemapplying a graph neural network (GNN) that can bescaled to accommodate varying numbers of base station(BS) antennas, users, and user antennas. The intractablebeamforming optimization problem for multi-user MIMOsystem is simplified by adopting a multi-user multipleinput single output (MISO) system-induced approximation. Additionally, we propose an efficient optimization methodwith low complexity by applying a GNN-based inferenceprocedure. The effectiveness of the proposed method isdemonstrated through the numerical results, showingconsiderable sum rate performance and system scalabilityin the dynamic wireless environments.