Research On Approaches Of Cooperative Spectrum Sensing For Cognitive Radio

With the rapid development of mobile internet and internet of things,the demands of spectrum resource are increasing quickly,leading to the shortage of spectrum resource,which is a bottleneck of wireless communication development.The cognitive radio(CR)has been proved an effective way to resolve this bottleneck.The CR's priority is to detect spectrum state accurately and efficiently.The primary methods of spectrum sensing mainly include single-node spectrum sensing technology and cooperative spectrum sensing(CSS)technology.The single-node spectrum sensing technology is with the merit of fast efficiency,and has a number of challenges to be solved including the hidden terminal problem,the poor performance of network,shadows,fading,the multi-path effect,and so on.CSS has been one of the key technologies to break through the above challenges of the single-node spectrum sensing technology.A series of CSS methods have been successfully proposed and applied one after another in recent years.The CSS technologies mainly cover hard fusion and soft fusion models.The hard fusion models are with a number of core superiorities including simple disposition of SUs,low cost,low communication costs and high transmission efficiency.However,the hard fusion models have the deficiency of low detection reliability and the challenge of parameter optimization problem.By comparison with hard fusion models,the soft fusion models can get better detection reliability but to face the challenges including quantization the uncertain sensing data,spectrum sensing data falsification,high complexity of existing sensing models and design of soft fusion rules.To overcome the mentioned challenges of CSS,this thesis investigates the topic of CSS for CR via deep analysis of sensing conditions,from a multi-disciplinary perspective,by integrating wireless communication with the recent advances in statistical learning,fuzzy decision making,uncertain inference theory and machine learning.The main work of this thesis is summarized as follows:1.Analyze and summarize both the single-node spectrum sensing technologies and the CSS technologies.Firstly,we recall three classical single-node spectrum sensing technologies including energy detector,matched filter and cyclostationarity detector.Meanwhile,we present their merits and limitations.Secondly,we review and analyze three hard fusion models(AND,OR and K-N)and their characteristics.Meanwhile,we discuss the key challenge of K-N-based CSS,i.e.,optimizing parameters.Finally,we recall threetypical fusion models including energy detection method based on Neyman-Pearson rule,CSS methods based on D-S theory,and machine-learning-based methods.At the same time,we investigate the challenges of the latter two methods.2.Propose a parameter-optimization-based K-N sensing method in the presence of primary user emulation attack.Firstly,we introduce a four hypothesis testing model based on the presence or absence of the PU.Secondly,we design an optimization model of parameters of the proposed K-N spectrum sensing algorithm and present the analytic expression formula to determine the optimal solution for choosing K through defining both the attack strength and attack probability.Finally,several examples are employed to analyze the sensing effect derived from different parameters.The simulation results indicate that the optimal solution of relevant parameters is helpful to enhance the sensing capability of the proposed K-N spectrum sensing algorithm.3.Propose a CSS method using minimum-spanning-tree clustering from the perspective of fuzzy logic and graph theory to accurately describe the uncertain sensing information and to reduce the computational complexity compared with existing sensing methods.Firstly,we employ energy of sensing data and intuitionistic fuzzy to quantify the local sensing information from all SUs.Secondly,we divide local sensing data from fusion center into several clusters through the minimum-spanning-tree clustering method and select certain cluster having the maximum number of sensing nodes for spectrum decision.Finally,we make a decision on the basis of the selected cluster from a standpoint of TOPSIS.The simulation indicates that this method not only improves the sensing performance but also reduce the computational complexity than existing models.4.Propose evidential-reasoning-theory-based CSS method to overcome the challenges including spectrum sensing data falsification,high complexity of existing sensing models and design of soft fusion rules.Firstly,we propose an evidential-reasoningtheory-based CSS with efficient quantization method to solve the problems of fusion evidence in conflict and Zadeh's paradox from D-S theory which may induce unreasonable or even wrong sensing results.Secondly,we propose a CSS algorithm based on adaptive reputation and evidential reasoning theory to detect the malicious SU and to improve the sensing performance in the presence of spectrum sensing data falsification.Concerning this algorithm,an adaptive reputation is introduced to detect malicious SUs and a model by integrating the adaptive reputation and evidential reasoning theory is designed to enhance the sensing reliability and the sensing stability.Finally,we propose a grid-based distributed cooperative spectrum sensing algorithm based on the analytic hierarchy pro-cess(AHP)and evidential reasoning theory by considering the increasing SUs and the consequent uncertain sensing information.This algorithm contains three key steps,i.e.,dividing sensing area of all SUs into a number of grids depending on AHP method,local sensing information fusion in every gird using the evidential-reasoning-theory-based CSS with efficient quantization algorithm,determining the weights of every grid,making a comprehensive fusion of sensing reports from grids which is same as the process of fusion in every grid.A lot of simulation examples show that the evidential-reasoning-theorybased CSS method is helpful to solve evidence in conflict,to reduce the computational complexity,to detect malicious SUs and to improve sensing performance by comparison with conventional sensing models.