Research On Deep Feature Modeling Of Electromagnetic Signals And Modulation Classification Based On Constellation Diagram

In a non-cooperative communication system,the signal can be demodulated correctly when the modulation mode of the signal is obtained.Modulation classification is an essential part of non-cooperative communication systems,a key technology to intercept enemy signals and an important step in spectrum detection.It has important theoretical significance and application value in military and civil fields such as UAV cluster operation,electromagnetic spectrum sensing and radio communication.With the rapid development of communication technology,electromagnetic signals have the characteristics of large quantity,high density,and various forms,which makes the communication environment more complex and restricts the improvement of signal modulation classification performance.It is urgent to break through relevant theories and methods.Constellation diagram can intuitively represent the distribution of signals in the in-phase and quadrature directions,which can effectively represent electromagnetic signals.As an important branch of machine learning,deep learning technology can mine the useful information of data which provides new ideas and methods for realizing electromagnetic signal feature modeling and modulation recognition.Based on the constellation diagrams representation of electromagnetic signal and the theory of pattern recognition and deep learning,this project carries out the research on the depth feature modeling of electromagnetic signals and modulation recognition based on constellation diagrams.The specific research contents are as follows:(1)In view of the high computational complexity of the existing constellation construction methods which leads to the slow mapping speed of the constellation diagram,an enhanced representation method of the constellation is proposed,which can effectively realize the modulation recognition of digital signals.Firstly,a fast projection(FP)algorithm for constellation diagrams is introduced,which improves the projection speed of signals to constellation diagrams by presetting the Gaussian weight kernel.Secondly,an enhanced constellation diagram(ECD)construction method is proposed that maps the single channel constellation into three channels,which enriches the constellation diagram information.Finally,a lightweight Convolutional Neural Network(CNN)model is designed which effectively improves the network training speed while ensuring a high recognition rate.Experimental results show that the proposed method has low computational complexity and excellent recognition performance.(2)Aiming at the problem that the existing constellation diagrams have weak ability to represent frequency offset signals,resulting in poor model recognition performance,a signal modulation recognition method based on the characteristics of anti-frequency offset constellation diagrams is proposed.The method works by cumulatively projecting a single signal sample to F constellation diagrams which are extracted convolutional features using CNN.Then the back-end residual unit and Gated Recurrent Unit(GRU)are constructed to realize modulation recognition.The effectiveness of the proposed method is proved on the simulated and measured signals.Experiments show that the proposed method has better recognition performance when the signals have frequency offset.(3)Aiming at the problem that the constellation diagram is easy to lose the useful information of the signal and difficult to represent the electromagnetic signal completely which leads to the poor robustness of the signal representation,a signal modulation classification method based on multi-feature fusion is proposed.The method extracts convolutional features by constructing deep neural networks for I/Q signals and different constellation diagrams mapping representations,useing back-end fusion to realize signals modulation types classification.It is found that this method can effectively improve the performance of signal modulation recognition by constructing a complete electromagnetic signal representation.The research contents of the above three parts complement and support each other which jointly improve the theory of electromagnetic signal mining and modulation recognition,providing method support and reference value for relevant military and civil fields...