Research On Signal Detection And Recognition Algorithms In Ultra-wideband Electromagnetic Environment

Signal detection and modulation recognition technology is currently one of the core technologies used in the radio monitoring and electronic information warfare.It can not only improve the safety and privacy protection of radio users,but also provide effective and accurate technical support information for the national defense security systems.This thesis mainly studies the rapid signal detection and real-time modulation recognition algorithm in the ultra-wideband electromagnetic environment.The content of research includes broadband multi-target detection and parameter estimation,feature extraction analysis and fusion recognition,and semi-supervised learning decision tree recognition algorithm.The main work and research results obtained in this thesis are as follows.(1)In terms of signal detection and parameter estimation,for practical engineering applications and high-speed sampling data,a multi-signal detection algorithm based on dynamic thresholds is proposed.The algorithm first uses the incoherent accumulation-power spectral density split cancellation method to effectively solve the problem of excessive energy difference between the signal,in an ultra-wide band under low signal-to-noise ratio.In the process of signal detection,the algorithm uses the autocorrelation characteristics of the signal to set the dynamic detection threshold,which effectively realizes the detection of multiple different types of signals within the analysis bandwidth.Finally,the algorithm uses the spectral center of gravity method to estimate the carrier frequency and bandwidth of the signal,then accurately filtering the respective signals separated.(2)In terms of feature extraction analysis and fusion,features such as power spectrum,high order spectrum and high order cumulants are extracted and the values of the features are statistically analyzed,then a modulation recognition algorithm based on feature fusion is proposed.The algorithm fusions the subtle features of the power spectrum,the high order spectrum and the high order cumulants features according to the commonality of each feature.And the optimal classification level is determined according to the uniqueness of the three features,thereby establishing a tree-like classification structure,which better realizes the signal recognition.(3)In terms of decision tree classifier,aiming at the limitation of artificial decision tree's fixed threshold threshold.Then comprehensively considering the accuracy of recognition,the complexity of the algorithm,and the difficulty of implementation.A modulation recognition algorithm based on semi-supervised learning decision tree is proposed.The formula of the C4.5 algorithm is simplified,and the artificial decision tree is fused to optimize and constrain the C4.5 decision tree.Compared with the original C4.5 decision tree,the improved algorithm uses the artificial decision tree as the prior information to update the classification structure,which further improves the accuracy of signal recognition and the generalization ability of the algorithm.