Research On Adaptive Spectrum Sensing Based On Convolutional Neural Network

Cognitive radio technology can realize uninterrupted detection of free spectrum in the channel,and use dynamic spectrum access technology to realize spectrum sharing strategy,effectively alleviating the shortage of spectrum resources and meeting the rapid growth of business demand.As an important part of the cognitive radio system,spectrum sensing can enable cognitive users to detect idle frequency bands in wireless channels,and access the available frequency bands without interfering with authorized users,greatly improving the utilization of spectrum resources.The signal-to-noise ratio estimation is a pre-prior condition for spectrum sensing,which can sense the accurate spectrum status information of authorized users,and ensure the smooth operation of spectrum analysis and spectrum management in cognitive radio.This thesis focuses on the research of signal-to-noise ratio estimation and spectrum sensing technology in wireless communication.In order to improve the adaptability and high availability of the system model,the research on the adaptive spectrum sensing model based on signal-to-noise ratio estimation is carried out.The main work and innovations of the thesis are as follows:Firstly,in view of the sharp deterioration of traditional detection algorithms under low signal-to-noise ratio and the high computational complexity of artificial neural networks in processing large-scale data,this thesis proposes an adaptive spectrum sensing model based on convolutional neural networks.First,map the simulated signal sequence to an RGB picture,and input the picture into the convolutional neural network to train the model.Since the trained neural network model is based on a single signal-to-noise ratio,the actual channel is time-varying,so use The signal-to-noise ratio estimation algorithm matches the signal to be detected with the neural network model and performs spectrum sensing.The simulation experiment shows that under the low signal-to-noise ratio,when the false alarm probability is the same,the detection probability of the convolutional neural network model is greatly improved than that of the traditional energy detection method.At the same time,with the increase of signal data sampling points,the detection performance is gradually improved.Secondly,the signal-to-noise ratio estimation algorithm studied in this thesis is to fit the adaptive spectrum sensing model,that is,it needs to use a common data set.The traditional signal-to-noise ratio estimation algorithm does not have this ability,so the signal based on HOG and SVM is proposed.The noise ratio estimation algorithm optimizes the adaptive spectrum sensing model.Simulation experiments show that: compared with the traditional maximum likelihood estimation method,the algorithm proposed in this thesis has better performance.In the experiment to verify the adaptive process of the system,it is found that the performance of the matched spectrum sensing model is far superior to that of the differentiated model.