CKMGAT: An Environment-Aware Dataset and Hybrid-Graph Attention Framework for Channel Knowledge Map Construction
Year: 2026
Type: preprint
Abstract: With the evolution of sixth-generation (6G) wireless systems, the demand of integrated sensing and communication (ISAC) for high-precision environmental perception has become prominent. However, existing channel knowledge map (CKM) datasets still have limitations in characterizing three-dimensional (3D) environmental details, making it difficult to fully reflect the impact of complex urban environments on channel propagation characteristics. To this end, this paper first constructs UrbanCKM, a dataset that places greater emphasis on representing 3D environmental details, explicitly considers the spatial distribution of obstacles such as buildings and vegetation in urban environments, and provides their 3D physical features. Second, to enhance the model’s ability to learn the complex mapping between environmental features and channel parameters, we develop a novel graph attention network (GAT)-based framework named CKMGAT. Unlike existing methods that mainly rely on the spatial correlation among receiver (Rx) nodes, CKMGAT explicitly models obstacles as environmental nodes, constructs an environment-aware hybrid graph structure, and further incorporates key blocking obstacles along non-line-of-sight (NLOS) propagation paths into edge construction as propagation-related nodes, thereby enhancing the graph representation capability for complex propagation environments. In addition, based on the constructed hybrid graph, a dual-stream attention mechanism (DSAM) is designed, consisting of a homogeneous spatial self-attention branch and a heterogeneous environmental cross-attention branch, which jointly capture spatial fading patterns and environment-induced propagation effects. Experimental results demonstrate that CKMGAT exhibits strong adaptability under sparse observation conditions and achieves higher CKM reconstruction accuracy and generalization capability. This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible.
Institution: China University of Mining and Technology
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225783420
Cites:
Cited by:
Related to: