Graph Learning-Based Cooperative Spectrum Sensing With Corrupted RSSs in Spectrum-Heterogeneous Cognitive Radio Networks

Spatiotemporal spectrum sensing of multiple primary users (PUs) sharing the same channels with unknown and irregular coverage presents significant challenges. The receive signal strength (RSS) levels at secondary users (SUs) vary greatly due to path propagation loss and shadowing. Moreover, due to security concerns and limited energy at SUs, the reported RSS measurements to a fusion center are noisy and incomplete. These challenges seriously impact the performance of existing cooperative spectrum sensing (CSS) techniques. In this work, to address these issues, we propose a robust graph learning based CSS (RoGL-CSS) detector, after revealing the low rank property of the expectation of RSS matrix and formulating a graph learning problem. By solving the graph learning problem with the alternating direction method of multipliers (ADMM), a probability matrix (graph) representing the correlations among SUs is acquired, which is applied to recover the expectation of RSSs from corrupted measurements and select SUs for CSS. Specifically, a robust RSS recovery algorithm with a learned probability matrix is adopted, and the SUs with recovered RSSs of high correlations are collected for implementing CSS. Numerical results are provided to demonstrate the superiority of the proposed detector compared to state-of-the-art detectors. At a false-alarm probability of 10%, with measurement missing rate 20% and outlier rate 30%, RoGL-CSS achieves performance improvement of at least 16% and 9% in detection probability, compared to other detectors for scenarios of two and five PUs, respectively.