A Graph Neural Network Enhanced Deep Q-Learning Method for Time-Domain Interference Coordination in Low-Altitude Integrated Sensing and Communication Networks
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Low-altitude Integrated Sensing and Communication (ISAC) networks face severe inter-site interference problems that significantly degrade sensing performance. This paper proposes a Graph Neural Network Enhanced Deep Q-Learning (GNN-DQN) method to address the time-domain interference coordination problem in ISAC networks. First, we utilize a three-dimensional ray tracing platform to identify strong interference relationships between base stations and model them as a conflict graph. Then, the pulse wave resource allocation problem is formulated as a constrained multi-pulse graph coloring problem. Finally, we propose the GNN-DQN algorithm, which extracts topological features of the conflict graph through graph neural networks, combined with hierarchical action space and coverage prediction-based reward shaping, effectively addressing the action space explosion and sparse reward problems faced by traditional DQN in combinatorial optimization. In simulation experiments with 162 cells in an urban scenario, the proposed GNN-DQN algorithm achieves optimal performance compared to various baseline methods, effectively improving sensing coverage in areas with severe inter-site interference while ensuring communication performance meets engineering requirements.
Fields of Study & Subjects
Radar Systems and Signal Processing
Satellite Communication Systems
Advanced Wireless Network Optimization
Publication details
- Journal / source
- International Conference on Communication Networks and Machine Learning (CNML)
- Pages
- 357–364
- Publication date
- 30 January 2026
- Language
- EN