Research On Anti-SSDF Attack Algorithm In Collaborative Spectrum Sensing

In the contemporary era of ubiquitous interconnectivity,the Internet of Things（Io T）technology has garnered significant attention.However,the dwindling supply of spectrum resources has emerged as a major impediment to the development of Io T applications.In this context,Cognitive Radio（CR）technology has been proposed to facilitate unlicensed Secondary Users（SUs）in opportunistically detecting and utilizing spectrum holes that are not occupied by licensed Primary Users（PUs）to enhance overall spectrum utilization.Nonetheless,due to the open nature of wireless networks,Cooperative Spectrum Sensing（CSS）,a core component of CR,is susceptible to Spectrum Sensing Data Falsification（SSDF）attacks.Therefore,this study aims to investigate the SSDF attacks and propose a mechanism for ensuring the security of spectrum sensing data in cognitive radio networks by eliminating the SSDF attackers while maintaining system reliability and security.The specific content is as follows.1.The Soft Byzantine Attack with Variable Attack Probability（VAP）is proposed as an attack model in this thesis to address the existing SSDF attack model.The VAP model involves mutual cooperation among attackers based on continuously changing credibility values and is designed for CR networks’spectrum sensing.The VAP model distinguishes attackers into two categories:Primary Attacker（PA）and Cooperative Attacker（CA）,where the CA passively receives reports from the PA,minimizing the attack cost.Simulation results demonstrate that the VAP model,compared to traditional non-cooperative attack models,exhibits stronger attack potential,higher channel disruption,and faster credibility attainment,rendering traditional credibility algorithms ineffective.2.In the VAP attack model,the attackers collaborate with each other and the attack parameters are constantly changing,which is a disastrous blow to the existing large number of reputation defense algorithms against the SSDF attacks.The Frequency Change Rate（FCR）defense algorithm,which is an improved algorithm based on trust value frequency changes on top of credibility algorithms,is proposed in this thesis.All users are classified by the FCR algorithm by comparing the frequency changes between each SU and the minimum change threshold₁,which eliminates not only PA and CA but also ensures that a large number of honest users with sensing errors are not strictly excluded by the defense algorithm.The simulation results show that PA and CA can be effectively distinguished by the proposed the FCR defense algorithm,and the attackers’credibility values can be reduced compared to traditional credibility algorithms,resulting in an increase in detection probability by around 40%.Furthermore,the detection probability increases by about 20%compared to the absence of defense algorithms.3.Because the minimum threshold₁ in the above the FCR defense algorithm is a static value,and the classification effect depends entirely on the₁ value selection effect.Therefore,in this thesis,an improved algorithm called-Mean-Shift++algorithm is proposed based on Mean-Shift（MS）algorithm,which uses unsupervised machine learning in an adaptive way to automatically complete the classification without considering the effect brought by static values.The algorithm first uses the raster method to divide the data by converting the energy value part of the spectrum report received by FC into an energy vector.The offset vectors of the energy vectors within the same raster can be considered equal,thus ensuring faster processing of the data with large samples.However,the raster method has a good differentiation in the part of the data where the energy vectors are widely distributed.However,when the energy vectors are densely distributed,it is easy to cause too many data to be divided into the same grid and thus fall into a pseudo-optimal.Therefore,this thesis introduces an acceleration factorbased on the raster method,and correlates the size of the raster with this variable.The advantage of this move is that a larger grid is selected when the energy vector is widely distributed to accelerate the determination of the center of mass,and a smaller grid is selected when the energy vector is densely distributed to ensure the accuracy of the center of mass.The simulation results show that the clustering effect of this algorithm improves 11%and 8%compared with K-means and Mean-Shift algorithms,respectively,when the number of samples is larger,and the detection probability improves 28%,14%and 50%compared with K-means,Mean-Shift and no clustering algorithms,respectively,because more attackers are eliminated.