Research On Cognitive Radio Spectrum Prediction Based On LSTM Neural Network

Spectrum perception,spectrum decision,spectrum sharing and spectrum management are four functions of cognitive radio technology.Spectrum perception is used to detect the spectrum hole,and then cognitive users can make secondary use of the mined spectrum hole through three functions of spectrum decision,spectrum sharing and spectrum management.However,the above four functions usually cause large time delays and energy consumption in cognitive radio systems.Spectrum prediction technology is an effective method to solve the above problems.Many spectrum prediction algorithms have good performance.With the popularity of deep learning,it will be a good and innovative research in the application of spectrum prediction.Therefore,this paper mainly focuses on the study of spectrum prediction algorithm which based on deep learning in cognitive radio.The main research works are described as follows:Firstly,the key technologies of cognitive radio network are investigated and summarized.Through literature review,the paper proposes an improved scheme based on the spectrum prediction technology aiming at the defects of four key technologies in cognitive radio network.Meanwhile,a comprehensive understanding of cognitive radio is obtained by elaborating and analyzing the spectrum prediction methods studied extensively.Secondly,a prediction model based on Long Short Term Memory(LSTM)is designed to tackle the issue of vanishing gradient in Recurrent Neural Network(RNN)which deals with time series problems.According to the characteristics of cognitive radio,the evaluation indexes for the performance of the spectrum prediction model in cognitive radio are defined.Then,the training and testing data sets used in this paper were preprocessed and statistically analyzed.In a single channel,the proposed prediction algorithm was simulated experimentally and compared with the Multilayer Perceptron(MLP)prediction method and Support Vector Machines(SVM)prediction method.Thirdly,on account of the potential relationship between the various channels in the spectrum can not be well explored under the single channel spectrum prediction,We have made a detailed introduction to the theory of convolutional neural network(CNN)which frequently used in deep learning to lay a theoretical foundation for the subsequent establishment of prediction model.Then in this paper,CNN and LSTM models are used as the basic network structure to carry out the combined design,and the cnn-lstm structure prediction model is obtained,which can be used for spectrum prediction under multiple channels.Finally,the real spectrum data which is collected by Agilent spectrum analyzer is used as experimental data to train and test the spectrum prediction algorithm designed under single channel and multiple channels,so as to verify its performance in the proposed evaluation metrics.