Cross-domain intelligent cooperative spectrum sensing algorithm based on Federated Learning and Swin-Transformer neural network

The deployment of 6th Generation (6G) mobile networks will bring revolutionary changes to organizations and users through higher speeds, lower latency, and more intelligent network management. This will require more efficient spectrum sharing to improve network performance. Cognitive Radio (CR) can...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Engineering applications of artificial intelligence Jg. 157; S. 111370
Hauptverfasser: Xu, Lingwei, Gao, Zhihe, Li, Yufang, Gulliver, T. Aaron
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier Ltd 01.10.2025
Schlagworte:
Cross domain processing
Federated learning
Mobile cooperative spectrum sensing
Swin-transformer
Mobile cooperative spectrum sensing
Cross domain processing
Federated learning
Swin-transformer
ISSN:0952-1976
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The deployment of 6th Generation (6G) mobile networks will bring revolutionary changes to organizations and users through higher speeds, lower latency, and more intelligent network management. This will require more efficient spectrum sharing to improve network performance. Cognitive Radio (CR) can be employed to help achieve this goal. However, data collection at the fusion center for cooperative spectrum sensing (CSS) creates a risk of sensing user (SU) data leakage. In this paper, an intelligent CSS algorithm for mobile communication is proposed based on Federated Learning with a Swin-Transformer. First, to achieve the cross-domain process, a Continuous Wavelet Transform (CWT) is used with the normalized time series data to obtain a time-frequency spectrum map with joint time-frequency features. Second, Federated Learning is employed to realize distributed CSS to reduce sensing overhead, improve the security of sensing data, and eliminate the SU data isolation with traditional distributed CSS. Then, a new Swin-Transformer spectrum sensing is proposed for fusion learning of the time-frequency spectrum map. The fusion of Federated Learning and Swin-Transformer neural network provides cooperation with multiple SUs to achieve efficient spectrum sensing using the K-rank fusion criterion. Compared with the Convolutional Neural Network (CNN), ViT-Transformer, Graph Neural Network (GNN) and EfficientNet algorithms, the detection probability of the proposed algorithm has increased by 45 %, and the false alarm probability has decreased by 13 %.
ISSN:0952-1976
DOI:10.1016/j.engappai.2025.111370