An SVM-Based Feature Detection Scheme for Spatial Spectrum Sensing

In cognitive radio, most spectrum sensing algorithms detect spectrum holes in the time or spectrum domain. In this letter, we propose a novel spatial spectrum sensing scheme that can detect the signal of the primary user (PU) and offer the angle of arrival (AoA) information. First, two new effective spatial features are introduced to distinguish the PU signal from the noise at low signal-to-noise ratio (SNR), namely the maximum value of the spatial spectrum (MVSS) and the angle of arrival difference (AAD). Then the support vector machine (SVM) algorithm is utilized for the feature classification to adapt to the varying environment, rather than using inherent thresholds as in traditional spectrum sensing methods. Finally, simulation results show that the proposed scheme outperforms the state-of-the-art multi-antenna sensing methods, especially at low SNR.