Graph Neural Network Based Access Point Selection for Cell-Free Massive MIMO Systems

A graph neural network (GNN) based access point (AP) selection algorithm for cell-free massive multiple-input multiple-output (MIMO) is proposed. Two graphs, a homogeneous graph which includes only AP nodes representing the structure of the APs in the network, and a heterogeneous graph which includes both the AP nodes and user equipment (UE) nodes are constructed to represent a cell-free massive MIMO network. A GNN based on the inductive graph learning framework GraphSAGE is used to obtain the embeddings which are then used to predict the links between the nodes. Numerical results show that compared to proximity-based AP selection algorithms, the proposed GNN based algorithm predicts more potential links with a limited number of reference signal receive power (RSRP) measurements. Unlike the other AP selection algorithms in the literature, the proposed algorithm does not assume the knowledge of RSRP measurements of every AP-UE combination for optimal AP selection. Furthermore, the proposed algorithm is scalable in terms of the number of users in the cell-free system.