Modulation Type Recognition Of Communication Signal Under α-stable Distribution Noise

As a key technology in electronic warfare,modulation type identification of communication signal has an important application in electronic reconnaissance.In many literatures,modulation recognition algorithms assume that noise environment obeys gaussian distribution,but non-Gaussian distribution is widely used in various natural noise,electromagnetic noise and other practical communication environment.It is found that the αstable distribution has the characteristics of probability distribution stability and spike pulse,and it can accurately describe the statistical characteristics of noise in the actual communication environment.Therefore,non-Gaussian noise is usually modeled as an α-stable distribution model.Because of the variability of electronic reconnaissance environment,wireless communication channel environment also has more complex noise composition,which leads to the degradation of many traditional modulation recognition algorithms under gaussian noise,or even failure.Therefore,it is of great practical value and theoretical significance to study the modulation type recognition technology of communication signal in α-stable distributed noise environment.The existing problems and corresponding research contents of this thesis are as follows:First of all,most of the existing recognition methods do not fully consider the modulation types of signals,and cannot effectively solve the recognition problem when the modulation types of multi-angle and multi-base signals from frequency,amplitude and phase modulation exist simultaneously.The modulation recognition of five kinds of signals(including 14 digital modulation types)is realized by using the classification recognition architecture.The cross-section of fractional low order cyclic spectrum(FLOCS)had excellent differentiation of frequency modulated signals,while the enhancement constellation of amplitude phase modulated signals had remarkable characteristics.Therefore,frequency modulation signals are firstly distinguished by section diagram,and amplitude-phase modulation signals are identified by enhanced constellation diagram.Secondly,due to the high complexity of manual extraction of deep dimensional features and the incomplete representation of signal feature information,FLOCS section diagram,enhanced constellation diagram and IQ data were directly input into the Convolutional Neural Network(CNN).CNN was used to automatically learn signal features to avoid the error caused by manual extraction of features.By employing FLOCS sectional diagram and enhanced constellation diagram to assist CNN training,the complexity and difficulty of network training can be reduced.At the same time,IQ data provides the most complete characteristic information for the network.Then,the signal recognition rate is low in the α stable distributed noise environment.CNN was used as a classifier to extract FLOCS section characteristics automatically by single channel CNN network and classify frequency modulation signals.Then IQ data and enhanced constellation map were input into two-channel CNN network,and amplitudephase modulation signals were identified through feature combination.When the Generalized Signal-to-Noise Ratio(GSNR)was higher than-4 d B,the recognition rate of M-FSK modulated signals by FLOCS single-channel CNN network was more than 99%.However,the two-channel CNN network based on IQ data and enhanced constellation diagram can correctly classify and recognize all the M-ASK,M-PSK,M-QAM and MAPSK modulated signals when the GSNR is close to 20 d B.Finally,based on the above research content,this thesis combines the characteristics of fractional low-order cyclic spectrum and constellation diagram with CNN to form a modulation type classification and recognition algorithm suitable for communication signal under α stable distributed noise.