TY - JOUR

T1 - Resource Allocation in Heterogeneous Network with Supervised GNNs

AU - Sun, Qiushi

AU - Zhang, Yuyi

AU - Wu, Haitao

AU - Petrosian, Ovanes

PY - 2023/7/8

Y1 - 2023/7/8

N2 - Device-to-device (D2D) transmission has become an essential form of wireless communication due to the rise of 5G and Internet of Things (IoT). Unfortunately, most currently available techniques for allocating resources are extremely time-consuming or computationally expensive. Graph neural networks (GNNs) have recently been proposed as a way to improve the efficacy of many network-related tasks. To tackle this issue, we propose a GNN-based method in a supervised manner. We denote heterogeneous networks as directed graphs and model various communication links as nodes and interference between links as edges. The proposed algorithm has two phases: (1) Offline supervised learning obtains the optimal solution through the particle swarm optimization (PSO) algorithm and uses the solution as a sample label to train GNN; (2) Online prediction by well-trained GNN. The simulation results indicate that the proposed method outperforms the state-of-the-art GNN-based benchmarks and achieves substantial speedups over conventional benchmarks.

AB - Device-to-device (D2D) transmission has become an essential form of wireless communication due to the rise of 5G and Internet of Things (IoT). Unfortunately, most currently available techniques for allocating resources are extremely time-consuming or computationally expensive. Graph neural networks (GNNs) have recently been proposed as a way to improve the efficacy of many network-related tasks. To tackle this issue, we propose a GNN-based method in a supervised manner. We denote heterogeneous networks as directed graphs and model various communication links as nodes and interference between links as edges. The proposed algorithm has two phases: (1) Offline supervised learning obtains the optimal solution through the particle swarm optimization (PSO) algorithm and uses the solution as a sample label to train GNN; (2) Online prediction by well-trained GNN. The simulation results indicate that the proposed method outperforms the state-of-the-art GNN-based benchmarks and achieves substantial speedups over conventional benchmarks.

KW - Resource allocation

KW - Heterogeneous networks

KW - Swarm intelligence

KW - Graph neural networks

KW - Graph neural networks

KW - Heterogeneous networks

KW - Resource allocation

KW - Swarm intelligence

UR - https://www.mendeley.com/catalogue/de8b667f-b918-394f-85ae-b5c420148aa0/

U2 - 10.1007/978-3-031-36625-3_28

DO - 10.1007/978-3-031-36625-3_28

M3 - Article

SP - 350

EP - 361

JO - Lecture Notes in Computer Science

JF - Lecture Notes in Computer Science

SN - 0302-9743

IS - 13969

ER -