Spectrum Quality Cognition Driven by GCN-Based with Knowledge Graph Embedding and Knowledge Transfer

Spectrum is an indispensable scarce resource for 5G and beyond mobile networks. Aiming at improving the spectrum utilization, spectrum quality cognition is expected to be a key enabler, which features predicting the transmission performance values and recommending frequency-domain channel for utilization. Nevertheless, the learning of feature representation for entities and relations in the Knowledge Graph (KG) for the electromagnetic spectrum remains largely unexplored. In order to capture the feature of spectrum quality, we construct a KG with transmission performance as node labels in graph model for the communication network and develop a knowledge graph embedding (KGE) model based on convolutional neural network (CNN). In order to address the problem of predicting transmission performance values in the presence of scarce spectrum data, we convert this issue into a node regression problem using a GCNbased graph model. In order to address the problem of sparse feature vector with respect to entities and relationships within the KG, we exploit KGE vectors generated by a CNN-based KGE model to replace the one-hot encoding method in graph model. Moreover, knowledge transfer (KT) can be exploited to extract knowledge from the historical complete transmission performance data of the source domain to the incomplete data of the target domain, so as to make up scarce spectrum data. Simulation results reveal that the algorithm based on GCN-KGE-KT significantly improves the accuracy of predicting transmission performance values compared to the algorithms without KGE.