Research On Cooperative Spectrum Sensing Algorithm In Cognitive Radio Networks

With the development of wireless communication networks,the contradiction between the rapid increased demand toward spectrum and the real supply ability of spectrum is growing in intensity.The cognitive radio is an effective way to solve this contradiction problem.Spectrum sensing is the core technology of cognitive radio.Its goal is achieving rapid and accurate detection of potential spectrum opportunities in a condition that the primary users are being protected from interruption.However,the actual communication environment is more complicated with diversity of signals.Thus,the traditional spectrum sensing technology cannot adapt to it and led to poor detection performance of the whole network.Therefore,using machine learning to improve the self-learning ability of sensing is an important method to research the spectrum sensing in cognitive radio networks.According to the presence of multiple primary users in the network,two kinds of learning spectrum sensing methods,cooperative spectrum sensing algorithm based on SMO-SVM and algorithm based on extreme learning machine(ELM),are proposed respectively.Based on past experience,these two methods can accurately detect spectrum holes and improve the spectrum utilization of the networks.The main research contents are as follows:(1)On the basis of existing cooperative spectrum sensing algorithm,thesis firstly analyzes the principle and characteristics of spectrum sensing.And then,with the traditional spectrum sensing algorithm,thesis proposed SMO-SVM based spectrum sensing algorithm for single primary user and extreme learning machine based spectrum sensing algorithm for multiple primary users.Finally,two kinds of cooperative spectrum sensing systems model for different number of primary users are established.(2)Base on studying theory of single primary user in spectrum sensing network,the defects of the tradition sensing algorithm in the environment adaptability and detection accuracy,as well as the weakness of support vector machine method in training samples is analyzed.On the results of the above theoretical and analysis,a cooperative spectrum sensing method base on SMO-SVM is proposed.We combine the energy detection and SVM classification algorithm to solve the hyperplane of support vector machines using the SMO minimum sequence method,and introduce the relaxation variables,penalty factors and kernel functions to optimize the SVM.Then,the grid algorithm,particle swarm algorithm and genetic algorithm are used to optimize the unknown parameters in the classification function.The experimental data and simulation results show that SMO-SVM can greatly improve the training speed and classification speed of support vector machine,and the detection probability in the networks is improved.(3)Taking into account the existence of multiple primary users in the network,we first study the existing learning multiple classification methods,and then a cooperative spectrum sensing algorithm based on ELM are proposed.A spectrum sensing model for multiple primary users is established,and the state of multiple primary users in the channel is classified by combining energy and ELM.The algorithm randomly assigned parameter weights to the input signal,and adjust the hidden layer output function until generates a unique optimal solution.Finally,we use three activation functions,Sigmoid,Sine and Hardlim,to remove redundancy in the input data.As simulation and experimental data shows,ELM greatly reduces the training time and classification time,and the detection probability is higher,the detection performance is more stable.