Application Of Machine Learning In Blind Estimation Of Signal Modulation Parameters

Parameter estimation and modulation pattern recognition of communication signals are key issues in communication systems,which directly affect the performance of subsequent signal demodulation.With the application of non-cooperative communication systems in various modern communication environments becoming more and more frequent,blind estimation of signal parameters is playing an increasingly prominent role in military communications,satellite communications and other communication fields.Because the blind estimation of signal parameters is based on the accurate estimation of signal bandwidth,the accurate estimation of bandwidth is the research focus in noncooperative communication.For a long time,the recognition of signal modulation mode is the focus and hotspot in the field of communication.Accurate identification of signal modulation mode is the research direction of researchers.In the actual communication system,the environment is complex and changeable,which may make the spectrum of the signal complex and difficult to process.When no shaping filter is added to the transmitter,side lobes and high noise will appear.At this time,the traditional direct bandwidth estimation method in engineering is difficult to accurately estimate the signal bandwidth,thus affecting the subsequent signal processing steps.So this paper studies the estimation of signal bandwidth parameters based on complex signal spectrum,and explores and studies the identification of signal modulation modes after the estimation of signal parameters.Machine learning is a rising concept in recent years,which contains a large number of excel ent algorithms for various conditions and problems.Many problems in the field of communication and the direction of solving machine learning are similar and overlapping,so this paper devotes to using machine learning algorithm to deal with traditional communication problems and improve the communication algorithm.From the point of view of complex signal spectrum,this paper uses deep learning regression model to estimate the starting and ending position of signal bandwidth.The simulation results show that the accuracy of bandwidth estimation using deep learning scheme on complex signal spectrum data is about 10 to 15 percentage points higher than that of traditional methods.At the same time,the modulation recognition algorithms of five kinds of conventional digital communication signals BPSK,QPSK,8PSK,16 QAM and 32 QAM are simulated.The algorithms based on traditional communication and machine learning are proposed and tested.The simulation results show that the machine learning algorithm can improve the recognition accuracy at the expense of computational complexity,especially in the case of low signal-to-noise ratio.Recognition of signal constellation using image classification algorithm to realize modulation pattern recognition works well under low signal-to-noise ratio(SNR).With the gradual improvement of computational power,machine learning algorithm will also have better performance in practical applications.