Reararch On Modulation Recognition And Signal Detection Methods For Communication Signals Based On Graph Domain Mapping

Communication signal modulation recognition and detection is the key technology in electromagnetic(EM)spectrum warfare in complex EM environments.It is widely used in dynamic spectrum access and supervision systems,target detection and guidance,military electronic countermeasures,and EM anti-interference fields.The technology is an important foundation for wireless receivers to demodulate and analyze target signals,spectrum detection,and other related signal processing work.With the evolution of EM space combat forms,military wireless communication has become an important carrier for transmitting combat instructions and evaluating war situations.Strong EM interference and high-precision electronic countermeasures have led to complex signal systems and waveforms in modern EM warfare spaces,increasingly congested and aliased spectra.The communication signals of multiple types of EM equipment are densely distributed and dynamically overlapped in the time,frequency,spatial,and energy domains,resulting in extremely complex EM environments.In addition,traditional radio receiver signal recognition and detection methods have degraded or even failed in this situation.Therefore,based on a comprehensive review of existing methods,this thesis focuses on the low signal-to-noise ratio(SNR)and high dynamic complex EM environment and conducts automatic modulation recognition(AMR)of communication signals under Gaussian noise and non-Gaussian Alpha stable distribution noise,as well as communication signal detection under non-Gaussian Alpha stable distribution noise,and conduct in-depth and systematic research on key technologies for detecting abnormal radiation source signals.The main research work and achievements completed in this thesis are as follows:1、To address the problem of feature extraction-based AMR methods in low SNR Gaussian noise environments ignoring the complementarity between different modal features,an AMR model based on dual-model feature fusion neural network is proposed to improve model recognition accuracy.Based on the principle of graph domain mapping,the Gram angle field(GAF)coding theory is first introduced into the field of AMR.Using deep CNN as the framework,the noisy signals received by the radio receiver are mapped into GAF images and one-dimensional in-phase orthogonal data.Based on the data characteristics,GAF-Res Net50 and CV-CNN are designed to extract features from different modal data.To improve recognition accuracy,a dual-model feature fusion mechanism is proposed to fuse the features extracted from the two networks and train them as model inputs,ultimately achieving the AMR task of multiple types of communication signals.Simulation and practical experiments indicate that the proposed model has good recognition performance in low SNR Gaussian noise EM environments.2、Aiming at the problem of severe performance degradation of traditional AMR methods designed with Gaussian noise channels in low GSNR non-Gaussian Alpha stable distribution noise environments,an AMR method based on log-domain 3D Constellation Diagrams and Multiscale Dual-Convolutional 3DCNN is proposed to improve model recognition accuracy.In the signal preprocessing stage,nonlinear logarithmic functions are used to suppress non-Gaussian noise.Based on the principle of graph domain mapping,log-domain 3D constellation diagrams are proposed to characterize multiple types of signals on the basis of a2 D constellation image.A new MDC-3DCNN is designed based on the features of the log-domain 3D constellation diagrams.The network quantity and scale of the dual-convolutional structure are used as the standard components and hyperparameters of the3 DCNN,effectively avoiding the disadvantage of repeated parameter tuning when the dataset is different,and enhancing the scalability of the network model,and achieving multi-dimensional feature extraction in both plane and space of the log-domain 3D constellation diagrams.Experiments indicate that the proposed method can effectively achieve high-precision and robust AMR tasks for communication signals in non-Gaussian noise channels.3、A signal detection algorithm based on Information Geometry Difference Mapping(IGDM)is proposed to address the problem of low GSNR non-Gaussian Alpha stable distribution noise without a closed probability density function(PDF),and the difficulty in modeling statistical moments,which leads to a decrease in the performance of signal detection methods based on statistical theory.The signal detection problem is equivalent to measuring the differences between different statistical moment models in manifold space.This algorithm introduces a multi-scale Gaussian mixture model to approximate the explicit expression of the PDF for noisy observation signals.Using an improved EM algorithm to estimate the parameters of the PDF,complete the statistical moment modeling of the observed signal.Furthermore,based on the information geometry theory,the observed signal’s PDF is mapped to coordinate points in the manifold space.Combined with the Kullback Leibler divergence function,a dual-threshold detection method based on information geometry is proposed to measure the differences between coordinate points,achieving signal detection under non-Gaussian Alpha stable distribution noise.The experiment shows that the proposed method has a good improvement in the accuracy of communication signal detection in low GSNR conditions.4、Aiming at the problem of difficult detection caused by the stacking of abnormal radiation source signals with unknown parameters,high density,and multiple systems in complex EM environments,a method for detecting abnormal radiation source signals based on image-like situation difference mapping is proposed.By utilizing the advantages of graph domain representation of electromagnetic spatial information,the problem of radiation source signal detection is equivalent to the measurement of differences between different image situations,and a data-driven detection framework is proposed.The data collection part adopts a distributed sensor network combined with a spatial autocorrelation interpolation algorithm to solve the perception blind spots and incomplete data caused by environmental attributes in complex EM scenes.The Gaussian mixture model is used to fit and solve the PDF of signal and noise data.Based on information geometry theory,the solution results are constructed into class image situations through continuous mapping.Furthermore,the differential mapping distance function is proposed to detect abnormal radiation source signals by measuring the degree of difference between different image-like situations.Simulation and practical experiments indicate that the proposed method can effectively detect abnormal radiation source signals,and provide a new approach for radiation source detection in complex EM environments.