Deep-Learning for Cooperative Spectrum Sensing Optimization in Cognitive Internet of Things

Abstract

Spectrum sensing is a critical component of Cognitive Internet of Things. It allows Secondary Users(SUs) to access underutilized frequency bands licensed to Primary Users (PUs) opportunistically without causing harmful interference to them. How-ever, accurate individual spectrum sensing solutions are complex to deploy. Thus, Cooperative Spectrum Sensing (CSS) techniques have flourished. These techniques combine individual sensing through a weighting mechanism at a fusion center to assess the channel status. The fusion process depends heavily on the indi-vidual detection thresholds at each SU and the weights attributed to their sensing results by the Fusion Center. In this paper, we propose to use Deep Neural Net-work to compute the optimal energy detection thresh-old and fusion weights. Our goal is to develop a solution that optimally adapts to the time-varying wireless channel conditions. Furthermore, our DNN-based so-lution eliminates the need to solve hard optimization problems, thus significantly reducing computational complexity, especially in large networks.