Spectrum Hole Geolocation via Deep Learning-Assisted Non-Cooperative Sensing

Accurate spectrum hole geolocation (SHG) is essential for efficient dynamic spectrum access. NCSS enables direct mapping between sensing decisions and sensor locations, but sparse observations limit spatial reconstruction. This paper proposes a deep learning-assisted non-cooperative spectrum sensing (NCSS) framework for SHG in Internet of Things (IoT) environments. The approach combines Voronoi-based mapping with an Attention U-Net architecture to reconstruct spectrum hole distributions from dispersed binary decisions. The proposed method improves reconstruction accuracy and consistency, approaching cooperative sensing performance without requiring information exchange among sensors.