Dynamic Spectrum Access for Femtocell Networks: A Graph Neural Network Based Learning Approach

Document Type

Conference Proceeding

Date of Original Version



This paper concerns the dynamic spectrum access problem for femtocell networks, where traffic loads are different among cells. We model the interference relationship of the femtocell networks with conflict graphs, and a graphical game is employed as the channel access coordination mechanism. A graph neural network based architecture is proposed, which directly maps traffic loads to the channel access scheme for each femtocell. With our method, each femtocell first estimates the qualities of all available channels based on the information from its neighbors, and then the channels of the highest quality are accessed. A multiagent reinforcement learning framework is designed to train the proposed architecture to make accurate estimations of channel quality.

Publication Title

2020 International Conference on Computing, Networking and Communications, ICNC 2020