Multi-view Semi-Supervised Learning for Cooperative Spectrum Sensing

Lusi Li, Laura Slayton, Hepeng Li, Haibo He

Abstract: Cooperative spectrum sensing (CSS) for cognitive radio networks (CRN) emerged to enhance the utilization of the frequency bands. CSS has multiple collaborated secondary users (SUs) to detect the presence of primary users (PUs) by fusing the sensing information and then determining the PU signals. Most of the previous works train detectors with sufficient labeled sensing data. However, collecting labeled and accurate data is time-consuming and expensive. In this paper, we propose a multiview semi-supervised learning framework for CSS (MSCSS). Specifically, MSCSS first combines the complementary sensing information from different SUs to explore a latent representation from both labeled and unlabeled sensing data. Based on the representation, it learns a robust fusion graph, performs the label propagation to unlabeled data, and trains a detector. MSCSS considers the information sensed by each SU as one view of the state of the concerned spectrum channel and fuses the multi-view information to make a decision. Simulation results demonstrate the effectiveness of our method.