Research On Spectrum Sensing Technology Based On Deep Learning

Spectrum sensing is the primary problem to be solved in the formation of an actual cognitive radio system.Its significance lies in enabling wireless communication devices to perceive the RF environment in real time and accurately,and to identify and determine “spectral voids” for rational use.This paper starts from two perspectives.On the one hand,aiming at the problem of low probability of detection based on blind spectrum sensing technology and insufficient analysis of the feature vector relationship of sampling covariance matrix at neighboring times,a blind spectrum sensing algorithm based on Hausdorff distance is proposed.The Hausdorff distance is used as a measure of the feature vector relationship of adjacent detection units.The parameters,and based on the stochastic matrix correlation theory and the characteristics of the sampling covariance matrix eigenvalues,the theoretical values of the detection threshold and the detection probability are derived.Through experimental simulation,the improved method is compared with traditional spectrum sensing algorithm,and It can be concluded that when the signal-to-noise ratio is-15 dB,the receiver operating characteristic curve of the blind spectrum sensing algorithm based on the Hausdorff distance is obviously superior to the blind spectrum sensing algorithm based on feature matching.On the other hand,in the process of researching spectrum sensing methods,it has been found that with the arrival of the era of big data and the increase in computing power,traditional spectrum sensing algorithms cannot effectively use the information existing in the channels,nor can they satisfy the intelligence of cognitive radio systems.Problems such as sexual requirements are more prominent.So we apply deep learning methods to cognitive radio spectrum sensing.In machine learning-based spectrum sensing algorithms,shallow-structured machine learning algorithms are prone to overfitting problems in data training.Therefore,we propose a spectrum sensing algorithm based on Convolutional Neural Networks(CNN).In order to enable the CNN method to effectively deal with spectrum sensing issues,we quantified the CNN structure and trained the CNN based on the energy features of the extracted signals and the training set constructed by the cyclic spectral features to establish an appropriate spectrum sensing model.The spectrum of the received signal is further sensed to determine whether the authorized user signal exists or not.Through simulation experiments on the modulated signal,we can know that we can build a proper CNN spectrum sensing model and have better performance than the previous spectrum sensing algorithm.