A Novel Hierarchical Framework for Cooperative Spectrum Sensing using CNN and SVM

Le résumé fourni par la source

Efficient dynamic spectrum access is essential in cognitive radio networks (CRNs) to alleviate spectrum scarcity. Cooperative spectrum sensing (CSS) plays a crucial role in detecting available spectrum by aggregating information from multiple secondary users (SUs). Although deep learning (DL) has significantly advanced CSS, achieving robust and accurate detection remains challenging, especially under low signal-to-noise ratio and severe fading conditions. This paper proposes a novel hierarchical deep learning framework, named CNN-SVM, for CSS that processes raw SU signals using a multi-stage convolutional architecture, followed by a Support Vector Machine (SVM) classifier for final decision-making. First, raw time-series signals from each SU are processed by a 1D-Convolutional Neural Network (1D-CNN) to extract robust temporal and spectral features. These features are then aggregated and passed to a 2D-Convolutional Neural Network (2D-CNN), which captures intricate inter-SU spatial relationships and cooperative patterns. Finally, the high-level features learned by 2D-CNN are classified by SVM to perform accurate channel detection. By integrating feature representation learning of the received signals using CNNs and performing classification with an SVM, this hierarchical approach achieves superior detection performance at low SNRs and in diverse fading environments. This work contributes to more reliable and efficient spectrum utilization in future wireless communication systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.