Low-Rank Subspace Representation for spectrum sensing

Spectrum sensing plays a critical role in cognitive radio. It autonomously detects the presence of primary users. If the primary users are absent, the spectrum holes can be assigned to secondary users. Different methods have been proposed and employed over the years including the covariance-based detection method. Since signal data generally resides in low dimensional spaces, a method based on a new theoretical framework called Robust Principal Component Analysis (RPCA) was recently proposed. It proves to be quite effective to detect the presence of primary signals. However, RPCA assumes that the underlying data is taken from a single low-rank subspace. A better approach, known as Low Rank Subspace Representation (LRSR), models the data from a mixture of multiple subspaces. This paper proposes that LRSR be used in spectrum sensing to produce better detection results. Using simulated and real DTV signals, experimental results show that LRSR delivers higher performance than RPCA especially in the low SNR region.