A Simultaneous Spectrum Sensing and Anomaly Detection Deep Learning Framework for Dynamic Spectrum Sharing Networks

Spectrum sensing and anomaly detection are two fundamental techniques in dynamic spectrum sharing networks. However, traditionally, those two functions are separately realized, which results in high implementation complexity and non-real-time. To tackle this problem, a novel deep learning frame-work is proposed for simultaneous spectrum sensing and anomaly detection. Moreover, a feature anchor clustering loss function is designed to improve the performance of both functions. Simulation results demonstrate that our proposed framework reduces the time complexity by at least 16.91% and the model complexity by at least 16.94% while improving the detection performance.