Enhanced Spectrum Sensing by Combining Feature Selection and Optimal Margin Distribution Machine

The accuracy of spectrum sensing is crucial to improve the utilization of spectrum resources. By fully utilizing the information of received signals, this paper proposes a spectrum sensing algorithm to improve the sensing performance, which is based on feature selection and optimal margin distribution machine (ODM). Firstly, a variety of features such as energy, correlation, and dispersion are extracted from the sample covariance matrix to generate a feature fusion vector that can describe the signal information more comprehensively; Secondly, a feature selection method based on random forest and Pearson correlation coefficient is proposed to obtain optimal features that are highly correlated with the class labels and have minimum redundancy, which avoids overfitting and reduces the computational complexity; Finally, the ODM is introduced to enhance classification capability of primary signal and noise. Simulation results show that the detection performance of the proposed algorithm is better than that of the SVM-based and the classical model-driven based spectrum sensing algorithms.