Research On Modulation Signals Recognition And Channel Equalization Technology Of Broadband Satellite Communication

With the rapid development of satellite communication technology,people are increasingly demanding the quality of information transmission.The satellite communication system has a long transmission distance,a bad channel environment,and a large fluctuation of the signal-to-noise ratio,which affects the channel equalization effect and reduces the accuracy of the modulation signal pattern recognition.Deep learning technology can automatically extract features in information,fit nonlinear relationships,and is often used to deal with problems with uncertain or difficult to quantify.Introducing deep learning techniques into the modulation signal pattern recognition technique does not require deriving the distinguishing features between the modulated signals.If a new modulation signal is added,the nonlinear characteristics of the time-varying channel can be fitted by simply adding a small amount of data to retrain the network.The use of parallel computing to process large amounts of modulated signals increases computational efficiency,shortens model training duration,and improves the iterative update period of the model,which is critical for time-varying satellite channels.In order to improve the recognition rate of the modulated signal,it is not enough to design an algorithm to distinguish the modulation signal with large signal-to-noise ratio fluctuation.It is also important to design an equalizer that can suppress inter-symbol interference and improve the signal-to-noise ratio.Blind equalization technology is deeply loved by researchers because it does not require training sequences.Deep learning technology changes the tap coefficients through iterative training.Reinforced learning(RL)can guide the model to update direction according to historical data,so the combination of reinforcement learning and deep learning technology can be realized.No training data can dynamically update the tap coefficients to fit the channel environment.This thesis focuses on modulation recognition technology,modulation signal parallel processing technology and channel equalization technology in satellite communication systems,focusing on the design problem of wideband satellite modulation signal recognition model,high-speed parallel data processing problem based on spark framework and channel equalizer design.Research on satellite signal modulation recognition algorithm,distributed image calculation algorithm,convolutional neural network(CNN),long and short memory network(LSTM),RL and blind equalization technology,focusing on broadband satellite modulation signal recognition technology.The main research contents and innovations of this thesis are as follows:(1)Broadband satellite modulation signal pattern recognition algorithm based on Deformable Convolutional Neural Networks(DCNN)networkBased on the research of satellite communication modulation signal pattern recognition technology,this paper proposes DCNN algorithm for signal pattern recognition based on CNN network.This method eliminates the research on the classification characteristics of the modulation signal pattern recognition algorithm in the process of satellite communication.It does not need to perform feature extraction for the new modulated signal,which solves the problem that the on-board modulation signal is difficult to identify and the classification signal needs to be added to the newly added modulation signal.The research results show that the proposed algorithm exhibits better recognition performance and robustness for satellite modulated signals with complex and variable channel conditions.(2)Spark distributed parallel modulation signal image compression storage algorithmBased on the research of modulation signal preprocessing technology,based on the spark framework,the spark distributed parallel modulation signal image compression storage algorithm is proposed.Overcoming the problem of processing large amounts of data when training satellite modulation signal recognition models.The research results show that the algorithm has parallel processing function,improve data processing speed,save memory overhead,solve the problem of large time cost of training DCNN network,and shorten the model update period.(3)LSTME blind equalization algorithm with RL feedbackBased on the reinforcement learning and deep learning techniques,an LSTME blind equalization algorithm with RL feedback is proposed to study the equalization technique which can effectively improve the symbol distortion caused by various noises and interferences of the channel.The research results show that the method can achieve equalization without training sequence and can make the original signal approximate the best estimation value of the received signal,which solves the low signal-to-noise ratio caused by inter-symbol interference and satellite channel environment.The problem can also improve the difficulty of modulation signal pattern recognition.