Spectrum Sensing for Cognitive Networks Based on Dimensionality Reduction and Random Forest

In this paper, we study spectrum sensing based on dimensionality reduction and random forest (RF) in low signal-to-noise ratio environments. Classifications of three digital modulation types, including BPSK, OFDM and 2FSK, are investigated. From the received radio signal, a set of cyclic spectrum features are first calculated, and the principal component analysis (PCA) is applied to extract the most discriminant feature vector for classification. Furthermore, the detecting signal is classified by the trained random forest, which uses the Gini index as the classification criteria, to test whether the primary user exists. The performance of our proposed PCA combining with RF algorithm is evaluated through simulations and compared with MME, SVM, RF. Experimental results show that with dimensionality reduction, the performance of classification is much better with fewer features than that of without dimensionality reduction.