TY - JOUR
T1 - Wireless Signal Identification for Secure Spectrum Sensing Based on Multi-Scale Fourier Segmented Attention Mechanism
AU - Yang, Ziyi
AU - Lu, Yaojun
AU - Zeng, Liang
AU - Wang, Shuai
AU - An, Jianping
AU - Liu, Zhiquan
N1 - Publisher Copyright:
© 2014 IEEE.
PY - 2025
Y1 - 2025
N2 - The rapid development of the Internet of Things (IoT) has led to exponential growth in wireless network traffic and the number of connected devices, thereby intensifying the demand for scarce spectrum resources. In this context, Wireless signal identification, a key technology in spectrum sensing, is crucial for enhancing spectrum utilization by mitigating interference and ensuring system security. In this study, we treat wireless signal identification as a time series classification task and propose a novel model based on Fourier-segmented attention. In our proposed model, instead of computing point-level attention, we extract sequence dependencies by computing segment-level attention. Moreover, we introduce a method based on the Fourier transform to determine the segment length, ensuring that each segment captures multi-scale features. Experimental results indicate that the proposed method outperforms existing models, achieving an accuracy of approximately 95% on our dataset and representing an improvement of around 1.6% in accuracy over competing approaches. Furthermore, experiments were conducted to evaluate the model’s effectiveness in detecting fake signals and its potential to enhance system security.
AB - The rapid development of the Internet of Things (IoT) has led to exponential growth in wireless network traffic and the number of connected devices, thereby intensifying the demand for scarce spectrum resources. In this context, Wireless signal identification, a key technology in spectrum sensing, is crucial for enhancing spectrum utilization by mitigating interference and ensuring system security. In this study, we treat wireless signal identification as a time series classification task and propose a novel model based on Fourier-segmented attention. In our proposed model, instead of computing point-level attention, we extract sequence dependencies by computing segment-level attention. Moreover, we introduce a method based on the Fourier transform to determine the segment length, ensuring that each segment captures multi-scale features. Experimental results indicate that the proposed method outperforms existing models, achieving an accuracy of approximately 95% on our dataset and representing an improvement of around 1.6% in accuracy over competing approaches. Furthermore, experiments were conducted to evaluate the model’s effectiveness in detecting fake signals and its potential to enhance system security.
KW - deep learning
KW - time series analysis
KW - transformer
KW - Wireless signal identification
UR - http://www.scopus.com/inward/record.url?scp=105003688320&partnerID=8YFLogxK
U2 - 10.1109/JIOT.2025.3564140
DO - 10.1109/JIOT.2025.3564140
M3 - Article
AN - SCOPUS:105003688320
SN - 2327-4662
JO - IEEE Internet of Things Journal
JF - IEEE Internet of Things Journal
ER -