Precise Spectrum Sensing Technology In Cognitive Radio Network

With the large-scale commercialization of the fifth-generation mobile communication system,the exponential growth of wireless terminals and devices has resulted in the increasing shortage of spectrum resources.The frequency efficiency has become an important factor restricting the development of wireless communication technology.By periodically monitoring the spectrum usage of the target frequency band,the spectrum sensing technology provides a guarantee for cognitive users' opportunistic access to the target spectrum and can effectively improve the frequency utilization rate.However,due to the hidden terminal problem,the current local sensing scheme has low sensing accuracy and poor sensing robustness.Although cooperative spectrum sensing(CSS)technology can effectively alleviate the problem of hidden terminals,the problem still exists for the poor sensing performance under the condition of low SNR or insufficient sampling points.And the high computational complexity in the case of heterogeneous and related nodes will affect the sensing efficiency as the same.Therefore,the study of accurate spectrum sensing technology under complex environments such as strong noise,undersampling and heterogeneity nodes will help to improve the sensing accuracy and efficiency,consummate the basic spectrum sensing theory,and then promote the rapid development of wireless communication technology.For this purpose,this dissertation studies the precise spectrum sensing technology in cognitive wireless network.Semi-adaptive sampling method,artificial intelligence technology and N-out-of-K approximation model are used to improve the spectrum sensing accuracy and reduce the complexity of collaborative spectrum sensing.Firstly,a spectrum sensing method for semi-adaptive sampling of received signals is proposed,and the analytical solution of the optimal sampling point for single-channel spectrum sensing is given.Meanwhile,an environmental signal-to-noise ratio evaluation framework based on convolutional neural network short-term memory network is designed,which effectively improves the spectrum sensing accuracy in strong noise environment.Secondly,a variety of Deep Neural Network(DNN)architectures suitable for spectrum sensing technology are proposed to achieve high precision spectrum sensing in the target frequency band under complex scenes.Finally,aiming at the problem of high computational complexity of collaborative spectrum sensing decision threshold in heterogeneous networks,an N-out-of-K approximation model was proposed,which greatly reduced the complexity of perception while ensuring high perception accuracy.The main work and innovations are organized as follows:1.A spectrum sensing method based on semi-adaptive sampling of received signals is proposed,which provides a closed solution of the optimal sampling point for single-channel detection,thereby reducing the sum of false alarm probability and missed alarm probability.Meanwhile,an environmental SNR evaluation architecture based on convolutional neural nets-LSTN is proposed,which integrates the depth feature and time dimension feature,and discusses the optimal CNN and LSTM layers.The results show that the network with two layers OF CNN and two layers of LSTM can obtain high accuracy of SNR estimation.At the same time,compared with the classical spectrum sensing scheme,the detection accuracy is improved by about 1dB.2.For spectrum sensing of a single node,a two-dimensional data set for receiving observed signals at different SNR is constructed,and an improved deep convolutional generative adversal network(DCGAN)is proposed to expand the training set and solve the problem of data shortage.On the basis of the enhanced 2d data set,the classical LeNet,AlexNet,VGG-16 neural networks were trained and optimized,and a novel neural network architecture CNN-1 was proposed to take into account both the perceptual performance and the perceptual complexity.3.A two-dimensional covariance matrix dataset based on local sensing energy vector for CSS is constructed in the fusion center.CSS schemes of three basic convolutional neural networks(LeNet,AlexNet and VGG-16)are trained and analyzed on the dataset.In addition,the perceptual performance of the proposed scheme based on convolutional neural network(CNN)is compared basing on AND,OR,Majority voting.4.For heterogeneous cognitive radio networks with limited nodes and correlation,an effective approximate model of cooperative spectrum sensing scheme based on N-out-of-K rule is established,and the approximate closed expression of optimal sensing threshold at the fusion center is given.Finally,the effect of approximate error on threshold can be controlled effectively under the condition of finite nodes.Finally,a numerical simulation for a special case is presented and compared with the exact algorithm for the closed-form solution,which proves the effectiveness of the proposed scheme.The simulation experiments under various scenarios are designed to verify the effectiveness of the multiproposal scheme.