Signal Recognition Algorithm Based on Random Forests for Spectrum Sensing in Cognitive Network

In this paper, a novel approach to signal recognition combining spectral correlation analysis and random forests is introduced to solve the problem of the low accuracy on detection and modulation type recognition of the weak Primary Users (PU) in low signal-to-noise ratio. Three spectral coherence character- istic parameters are chosen via spectral correlation analysis. By utilizing the proposed algorithm, the detecting signals are classified by the trained random forests, which use the Gini index as the classification criteria, to test whether the primary user exists and recognize the modulation type of the signal. The proposed algorithm enhanced the performance of the classification by utilizing the strong classifier synthesizing multiple weak classifiers and the accuracy of spectral correlation analysis method, so it is more suitable for primary user signal detection and recognition under low SNR environment. The performance is evaluated through simulations and compared with ANN and SVM algorithms. The advantages of the proposed algorithm are also shown through simulations.