Research On Time-frequency Domain Prediction Algorithm Based On Spectrum Data

In recent years,with the rapid development of wireless communication technologies and services,the wireless spectrum has become more and more crowded,and the shortage of spectrum resources has become increasingly serious.Together with the disadvantages of the traditional fixed spectrum allocation method,the above phenomenon has become more serious;To solve this problem,some research scholars have proposed the concept of cognitive radio,which means that secondary users can dynamically access some frequency bands through spectrum sensing without affecting the communication of the primary authorized users,so that they can effectively Improve spectrum utilization.The analysis of the electromagnetic spectrum situation is crucial for the sensing process,so that we can provide a factual basis for the secondary users to dynamically access the spectrum,which can improve the accuracy of its access and the effectiveness of spectrum sensing.In this thesis,the spectrum situation prediction algorithm is deeply analyzed and studied from the time dimension and the frequency dimension respectively.Different from the traditional spectrum data analysis method,for the SCR sequence,wavelet decomposition and LSTM network will be introduced to improve the prediction accuracy;for the channel,the spatial frequent mode will be introduced to improve the clustering effect.Here are the main work and contributions of this article:(1)In the time dimension,an SCR sequence prediction algorithm based on WDRCLSTM network is proposed for SCR sequences: This algorithm innovatively introduces wavelet decomposition into LSTM networks based on random connections.Wavelet decomposition can fully detect time series.The associated information in the frequency domain is better than the commonly used time series prediction models of SVR,ARIMA,LSTM,and has better prediction accuracy.At the same time,the WD-RCLSTM network is more flexible than the commonly used LSTM network.By controlling the degree of connectivity,and even if the network is lightweight,it can reduce the training time of the network if the prediction accuracy meets the conditions.(2)Frequency dimension,for the channel in fixed service,according to the shortcomings of traditional MEI-based channel clustering algorithm and density-based channel clustering algorithm,a spectrum channel clustering algorithm based on DBSCAN-SFPM is proposed,which is mainly based on The non-clustering points after the density spectrum channel clustering algorithm are analyzed by the spatial frequent pattern-based spectrum mining algorithm(SFPM)proposed in this paper.Finally,the SFPM algorithm proposed for the non-cluster point utilization is verified by experimental analysis.Effectively improving the channel prediction range,that is,improving the efficiency in the spectrum sensing process.(3)Implemented an intelligent analysis platform for spectrum data based on B/S mode,which is mainly through web page back-end development technology,visualization technology and web front-end development technology,and WD-RCLSTM based on the above mentioned time domain.The SCR sequence prediction algorithm of the network and the DBSCAN-SFPM-based spectrum channel clustering algorithm in the frequency domain are used to analyze the spectrum data,and the analysis results are visually displayed on the browser side,so that the researcher can be more convenient and intuitive.Learn about the spectrum situation.