Research On Individual Communication Transmitter Identification Based On Unsupervised Learning

Individual identification of communication radiation source(also known as fingerprint identification of communication radiation source)is the correlation between the signal and the transmitter by measuring the difference reflected in the signal of the transmitter.It plays an important role in electronic countermeasures and is an indispensable means in modern electronic warfare.This paper mainly studies the individual identification of communication radiation sources under unsupervised conditions,and works as follows:(1)The histogram feature of the signal is constructed on the basis of the l bispectrum,so as to improve the ability of feature characterization of signal information and fully reflect the difference between the signals emitted by different transmitters.Unsupervised learning is introduced into the individual recognition of communication radiation sources.First,calculate the histogram features of each signal sample,then calculate the Euclidean distance between each feature sample,then calculate the local density and relative distance of each feature sample to draw a decision graph,and finally select the center point of each cluster and assign labels to the non-center points.The method can achieve a recognition rate of 69% on VHF radio data set,and the experimental results show that the performance of histogram feature is better than that of rectangular integral bisection.(2)In view of the peak density clustering algorithm using truncation nuclear function or gaussian kernel function estimation of the density of sample points and the actual density of sample points the problem of large difference,put forward the kernel density estimation method based on the thermal diffusion equation,the kernel density estimation is regarded as the only solution diffusion partial differential equations,using instead of gaussian kernel said solution of partial differential equation,using the fast Fourier transform can quickly solve,finally to use the improved Sheather-Jones,adaptive optimal bandwidth selection algorithm.The method proposed in this chapter can improve the result of density estimation and make it closer to the real density of data points.The experiment shows that compared with other clustering methods,such as density peak clustering algorithm,the algorithm proposed in this paper performs better on the four data sets,achieving 74% recognition rate on the data set of ultra-short wave radio and 75.9 recognition rate on the data set of krisun radio.(3)In view of the problem that Euclidean distance cannot fully represent the internal structure of data,geodesic distance is introduced in this chapter as the distance measure of two data points;In view of the fact that the relative distance in the density peak algorithm cannot fully reflect the characteristics of the clustering center,this chapter remodels the ratio contest in the second hypothesis of the algorithm and proposes the concept of "comparative distance".Aiming at the problem that the value of parameters in the density peak algorithm depends on the subjective experience of users,local density information entropy is introduced to realize the parameter adaptive.Experimental results show that the performance of the algorithm is obviously better than other clustering methods,and the recognition rate can reach 78.6% on the data set of ultra-short wave radio and 80% on the data set of krisun radio.(4)Aiming at the problem that existing individual identification methods of communication radiation sources can only realize batch processing of data,this chapter proposes the incremental density peak clustering algorithm and builds an incremental model by combining KNN classifier based on chi-square measurement to realize incremental learning of individual identification of communication radiation sources.The incremental model can identify individuals of communication radiation sources in the absence of signal samples with class information,and can effectively use historical data information and limited training samples to solve the problem of identifying observed data of communication radiation sources of unknown class when new data samples are added.The incremental density peak algorithm is used to discover the newly added data,and KNN classifier is used to realize the fine classification of samples.The experimental results show that the recognition rate of VHF radio can reach75% by using fewer data samples.