An Improved SVM-Based Spatial Spectrum Sensing Scheme via Beamspace at Low SNRs

Most spectrum sensing algorithms mainly use the characteristics of frequency, time, and geographical dimensions to detect spectrum holes. In this paper, we propose a novel spectrum sensing scheme from the space domain by using beamspace transformation and the support vector machine technology. First, a model of beamspace transformation is proposed for the case of complex calculations in a sizeable multi-antenna system. This beamspace transformation has the ability of spatial filtering, which can not only decrease the dimension of the receive matrix but also enhance the signal to noise ratio of the received signal. Then, we employ the support vector machine classification to overcome the problems caused by the inherent threshold of traditional sensing algorithms. We only need to train the historical samples to distinguish between noise and primary user signals effectively. This classification algorithm has self-learning ability, which can adaptively adjust the classification hyperplane according to environmental changes without complex threshold calculation. Finally, simulation results show that the proposed scheme outperforms other related multi-antenna sensing algorithms, especially under low signal to noise ratio and low snapshot.