Towards a Wireless Network Digital Twin Model: A Heterogeneous Graph Neural Network Approach
- Resource Type
- Conference
- Authors
- Perdomo, Jose; Gutierrez-Estevez, M.A.; Zhou, Chan; Monserrat, Jose F.
- Source
- 2023 IEEE International Conference on Communications Workshops (ICC Workshops) Communications Workshops (ICC Workshops), 2023 IEEE International Conference on. :29-35 May, 2023
- Subject
- Communication, Networking and Broadcast Technologies
Signal Processing and Analysis
Training
Runtime
Wireless networks
Computational modeling
Message passing
Conferences
Downlink
wireless network modeling
heterogeneous graph neural network
beyond 5G
mobility
- Language
- ISSN
- 2694-2941
A digital twin of wireless networks can enable the ability to safely and rapidly recreate what-if scenarios of mobile networks for more efficient and intelligent network optimization and planning. In this work, we present a novel digital twin model of wireless networks based on Heterogeneous Message Passing Graph Neural Networks (HMPGNNs). Our digital twin model represents wireless network nodes and the underlying wireless phenomena between them as nodes and edges of different type in a heterogeneous graph. Heterogeneous graphs are fed as samples into the HMPGNN model so that the model learns to simulate the underlying wireless phenomena. Results using system-level simulations to train and evaluate our proposal, show that our approach accurately and efficiently captures the dynamics of wireless networks to produce accurate reconstruction of downlink data rates of all users while also generalizing to network deployments unseen during training.