Research On Cognitive Radio Spectrum Sensing Algorithm Based On Deep Learning

The continuous innovation of wireless communication technology makes the communication modes diversified,but it also brings pressure to the utilization of wireless resources.Cognitive radio,as a new technology to improve the utilization of wireless resources,has developed rapidly.Spectrum sensing technology is an important technology in cognitive radio.It assumes that when authorized users occupy the frequency band,cognitive users do not use the frequency band.When the authorized user does not occupy the frequency band,the user can be quickly connected to realize the secondary utilization of the free frequency band.This paper discusses the application of deep learning to spectrum sensing technology.Firstly,the research background and the concept and research status of cognitive radio and spectrum sensing technology are introduced.Then,the theoretical analysis of the four traditional spectrum sensing technologies is carried out,and the advantages and disadvantages of the four methods are deeply analyzed through comparative experiments.In addition,different detection algorithms are used in different communication environments and ways to provide basic theoretical support for the proposed new algorithm.For deep learning of artificial,loop,convolution neural network to make a simple analysis,observation of activation function in the neural network's influence on the ability to learn,put forward an application to the convolutional neural network combined Relu-tanh activation function of image classification,with the help of single layer activation function in the different network performance advantages.In the public Mnist and cifar-10 data sets,experimental results show that the proposed method has better convergence and classification performance than the convolutional neural network model with single activation function.Good signal detection performance is an important part of improving the cognitive radio spectrum perception.Aiming at the problem of signal detection in OFDM system,based on the analysis of OFDM system,an OFDM signal spectrum sensing algorithm based on cyclic neural network is proposed.Firstly,the framework of OFDM system was built and the time series data set for model learning was generated by OFDM system framework.Then,the LSTM model of cyclic neural network is used to realize end-to-end spectrum signal detection based on the learning ability of time series,so as to deal with nonlinear distortion and interference in wireless channels and improve the accuracy of signal transmission.Simulation results show that this algorithm has some advantages over traditional detection algorithms.Aiming at the carrier extraction of OFDM signals is subject to the interference of wireless transmission environment and electromagnetic field.A method for estimating the carrier frequency of OFDM signals based on FAM algorithm is proposed.Simulation results show that this method has better performance of carrier frequency estimation than the traditional power spectrum method under low SNR.Based on the study of cyclic autocorrelation and cyclic spectrum of OFDM signals,a frequency spectrum sensing algorithm for OFDM signals based on convolutional neural network is proposed.In this algorithm,the three-dimensional spectral map with certain cyclic characteristics is converted into grayscale image through normalization and grayscale processing,and the grayscale value is randomly assigned to the fixed position of the grayscale image to improve the robustness of the data set.Then,a series of preprocessed data sets are input into the improved convolutional neural network to complete the learning,and the spectrum perception model is constructed to transform the spectrum perception problem into an image classification problem.The simulation results show that the proposed algorithm can not only achieve spectrum sensing,but also significantly improve the performance of other traditional spectrum sensing algorithms and the spectrum sensing algorithms based on machine learning at low SNR.