Research On Spectrum Sensing And Anti-SSDF Attack Based On Machine Learning

With the continuous development of wireless communication technology,the contradiction between the explosive growth of the number of wireless devices and the limited spectrum resources is becoming increasingly serious.Cognitive Radio(CR)Networks is committed to improving the utilization rate of spectrum resources.As a key technology in CR networks,spectrum sensing is used to detect the main user signals in the target frequency band.Since spectrum awareness performance is susceptible to the adverse effects of complex wireless environment and CR network security problems,it is of great significance to study high-performance and secure spectrum sensing technology.In view of the excellent performance of machine learning,this thesis studies the cognitive radio spectrum sensing and anti-attack technology based on machine learning.In order to improve the performance of collaborative spectrum sensing in low signal to noise ratio,a collaborative spectrum sensing method based on eigenvalues and cascade clustering is proposed in this thesis by using fuzzy C-means(FCM)and Gaussian mixture model(GMM).The eigenvectors are constructed by extracting the eigenvalues from the covariance matrix of the received signals,and the classification model of channel availability is obtained through the clustering in the three-dimensional space.This process does not need to obtain the prior information of the primary user signal and the noise power,which further avoids the complex threshold calculation on the basis of preserving the advantages of completely blind spectrum sensing.FCM clustering is used to optimize the initial parameters of GMM clustering,which effectively solves the problem that GMM is prone to fall into the local minimum value in low signal to noise ratio.Simulation results show that this method both reduces the convergence time of GMM clustering and improves the accuracy of model classification.Compared with other mainstream methods,it can effectively improve the performance of spectrum sensing.Aiming at spectrum sensing data falsification(SSDF)attacks in CR networks,this thesis gives consideration to both malicious secondary user identification and data fusion,and proposes a secure collaborative spectrum sensing strategy against SSDF attacks.A malicious secondary user detection method based on HHO-SVM is proposed by using the Harris hawks optimization(HHO)algorithm and support vector machine(SVM),and the honesty of the cooperative spectrum sensing nodes are improved by eliminating the identified malicious secondary users.The fusion center extracts the key attributes which are used to distinguish normal secondary users and malicious secondary users from the report information matrix and constructs eigenvectors,and the training set of SVM model is simulated by combining the historical accumulated report information.The HHO algorithm is used to optimize the kernel parameters of SVM,so that the model can further adapt to the complex sensing environment and improve the accuracy of the identification of malicious users.In order to improve the anti-aggression of collaborative spectrum sensing,a data fusion algorithm based on credibility weighting is proposed.The credibility of secondary users is obtained by SVM model calculation,after detecting malicious secondary users,the fusion center marks them and deletes the reported data.By assigning reasonable weights to the secondary users which are remaining,the interference of outliers and missed malicious secondary users on spectrum sensing that can be reduced to the greatest extent.Simulation results show that this strategy can effectively resist SSDF attacks in a variety of scenarios and improve the security of spectrum sensing system.