Research On VHF Spectrum Prediction Technology Based On Data Driven

Based on the scarce spectrum resources and the current static frequency division strategy,the increasing flight traffic in the airport terminal area will face the severe contradiction between the supply and demand of VHF spectrum resources.The dynamic cognitive frequency utilization achieved by cognitive radio technology using spectrum holes will provide a feasible technical solution to alleviate the above-mentioned contradictions,but there are defects such as high spectrum sensing delays and spectrum access conflicts,which threaten the safety of frequency utilization in civil aviation.However,spectrum prediction technology can provide forward-looking basis for spectrum sensing by analyzing historical spectrum laws to predict the future spectrum state,makes up for the above-mentioned shortcomings and promotes the safety of frequency utilization in cognitive communication.Therefore,this research has investigated the current domestic and foreign research status of cognitive radio,spectrum measurement and spectrum prediction,based on algorithm theory,and has launched VHF spectrum prediction technology research based on data-driven.The main contents are as follows:First,the spectrum status of 120.3MHz,129.2MHz and 129.45 MHz in the terminal area of an Airport has been monitored for 28 days,lasting 12 hours a day.A total of 882 G of civil aviation VHF voice communication spectrum data has been obtained through the spectrum data acquisition platform built.A set of standardized data cleaning methods have been proposed for the measured data.According to this method,a second-level channel state data set for frequency occupancy statistics and a millisecond-level compressed spectrum data set for spectrum prediction technology research have been obtained.Then,Through the statistics of the 12-hour frequency occupancy results,it is found that there are a large number of spectrum holes in the measured spectrum,and the dynamic frequency utilization based on the spectrum holes can reduce the frequency occupancy under the solid-state frequency strategy by up to 22%,which are the important research points of this thesis.Secondly,the key technologies of cognitive radio(spectrum sensing,spectrum decision,spectrum sharing and spectrum switching)and the theoretical derivation of the four regression models(Ordinary Least Squares,Support Vector Regression,Autoregressive Integrated Moving Average,and Autoregressive Integrated Moving Average with Kalman filter)have been summarized in detail.The model has been trained and tested using original compressed data and normalized compressed data,and the prediction effect has been evaluated based on the root mean square error,model running time and data correlation.The results show that the regression models can only achieve more accurate time-dimensional predictions of normalized compressed data,which is the first key conclusion of the research.Finally,the architecture and characteristics of recurrent neural networks and convolutional neural networks have been summarized and explained in detail,so the theoretical foundation has been laid.The Conv LSTM network,which combines convolution operation and long-short-term memory neural network,has been applied to spectrum prediction,and the original compressed data and normalized compressed data has been used to train and test the model.The prediction effect has been evaluated based on predefined indicators and compared with the regression model.It is found that the prediction error of the Conv LSTM network is much lower than that of the regression model and the calculation time is not much different.Especially under the condition of not changing the preset signal detection threshold,the Conv LSTM network can directly realize more accurate time-dimensional prediction of the original compressed data,which is the second key conclusion of the research.