A Novel Knowledge Graph Driven Automatic Modulation Classification Framework for 6G Wireless Communications

Abstract

Automatic modulation classification (AMC) is a promising technology to realize intelligent wireless communications in the sixth-generation (6G) wireless communication networks. Recently, many data-and-knowledge dual-driven schemes have achieved high accuracy in AMC. However, most of these schemes focus on generating additional prior knowledge of unknown signals, which needs more computation cost in the inference phase. To solve these problems, we propose for the first time a modulation knowledge graph (MKG), and a novel knowledge graph (KG) driven AMC (KGAMC) framework by training the networks under the guidance of MKG domain knowledge. To achieve the best performance by exploiting KGAMC, a KG-driven multi-time-scale network (KG-MTSNet) is proposed to extract the MKG knowledge and the scale and frequency features of the sampled signals. Moreover, to utilize the knowledge, a designed feature aggregation loss is implemented to improve the signal feature presentation obtained by the data-driven model. Simulation results demonstrate that KGAMC significantly boosts the performances of data-driven models, and the KG-MTSNet achieves a superior classification performance compared to other benchmarks. Furthermore, the effectiveness of KGAMC is demonstrated in terms of the interpretability of the feature extraction and the sample shortage situation.