A Spectrum Sensing Algorithm Based on Linear Discriminant Analysis and Extreme Gradient Boosting in Cognitive Radio

Abstract

Spectrum sensing is one of the key technologies in cognitive radio, which can greatly improve the utilization rate of spectrum resources by detecting idle spectrum in real time. In this paper, a new spectrum sensing algorithm based on LDA (Linear discriminant analysis) and XGBoost (Extreme gradient boosting) is proposed. Firstly, the three-dimensional eigenvectors are extracted according to the cyclic spectrum of the signal, and the anti-noise performance can be improved by using the cyclic spectrum features. Then the LDA algorithm is used to reduce the feature dimension, which reduces the similar feature redundancy and improves the training speed of the subsequent algorithm. Finally, based on the transformed samples, the spectrum sensing model is trained by the XGBoost algorithm. The experimental results show that for different primary user signals, the proposed algorithm has better detection performance and faster training speed than SVM and AdaBoost algorithm at low SNR.