Research On Automatic Recognition Of Digital Signal Modulation

Digital signal modulation recognition technology is in the case of a non-cooperative communication mode,random intercept a digital communication signal.Under the condition of unknown or a lack of prior information,through the signal processing and feature parameters extraction of intercepted signals,and then the interception of the signal classification and recognition.Since entering the information age,digital communication signal modulation recognition technology has widely used in both military and civil areas.In recent years,no matter the electromagnetic environment or the signal modulation mode is more and more complex,making the digital communication signal modulation mode recognition more difficult.The traditional communication investigation methods has been difficult to satisfy the needs of the modern battlefield,so it is very important to study more practical and effective signal modulation recognition approaches.First of all,this paper gives an overview to the mathematical model and important parameters of the common digital signal modulation mode in communication system,and introduces the common methods of communication signal processing.Secondly,in the study of digital signal feature extraction algorithm,this paper deals with digital signal entropy features,the fractal box dimension feature and higher order cumulant algorithm algorithm research,the simulation and improvement.By means of the SFS feature selection algorithm to select the optimal characteristics of the combination of signal classification and recognition,thus getting choose rid of the redundant features to signal recognition.After that this paper forward a kind of feature extraction algorithm based on the time-frequency image two-dimensional box dimension,to solve the fractal box dimension of the digital signal classification problem of poor effect.Then,the BPA acquisition method of evidence theory has carried on the simple introduction.For distance similarity of BPA for probability limit problems of the method,this paper proposes a BPA access method based on Euclidean distance + Mahalanobis distance.This method combines the Euclidean distance and Mahalanobis distance,to solve the target signal trust probability extremum distribution affected by the number of training signal,which greatly increased the probability values distribution of target signal.Taking advantage of the study of entropy features is put forward a method of BPA acquisition method based on entropy normality,effectively solving the problem of weight distribution of the target signal.This method suppresses the signal characteristics of the gap with the target signal category of weight,thus improve the classification performance of the signal.Finally,this paper conducts an analysis simulation on three different classifier distance measured classifier,neural network classifier and evidence theory classifier algorithm.According to the digital signal characteristics,a classifier based on evidence theory fusion meathod is designed to realized the recognition of digital signal under low SNR,which strongly verified the effectiveness of the proposed classifier.