The Data-Driven Method Of Communication Signal Recognition

Electromagnetic space is a national strategic space,and electromagnetic spectrum resources are national strategic resources.The ubiquitous perception and precise management of electromagnetic space are the current research hotspots,and communication signal recognition is one of the key technologies.At present,with the rapid growth of electromagnetic space communication signals,the signals are becoming denser and richer in types,and the characteristic differences between signals are getting smaller and smaller.However,the traditional pattern recognition methods based on analytical modeling and feature engineering cannot adapt to the current communication signal recognition task requirements.This work takes the communication signal modulation pattern and individual identification as the research object,establishes a high-quality communication signal data set in a real open environment,develops a data-driven communication signal modulation pattern and individual identification method,explores the intersection of traditional electromagnetic signal analysis,processing and identification technology with big data technology and artificial intelligence technology,aiming to solve the problem of accurate identification of communication signals in complex environments.The specific research work and innovation are as follows:(1)The identification mechanism of the communication signal is analyzed from the perspective of the transmitter of the communication equipment.ADS-B,WiFi and other communication signal sources are selected,and key technologies such as signal acquisition,preprocessing,automatic labeling and storage management are further solved,and a communication signal data set with sufficient categories,large amount of data and high quality in real environment is established.It supports the algorithm research work of this paper.(2)Considering the communication signal as a typical nonlinear time series,the extraction of modulation features of communication signal based on multifractal spectrum is studied,and an adaptive weight grey relational classifier is designed.The experimental results show that the method has a good recognition effect for a small number of simple communication signals,but cannot distinguish high-order phase modulation signals,and is greatly affected by the signalto-noise ratio.(3)Aiming at the problems that traditional feature engineering methods require artificially designed features,the recognition results are greatly affected by the signal-to-noise ratio,and the recognition types are limited,this paper proposes a communication signal modulation recognition method based on adaptive deformable convolutional networks.The method introduces a rectangular density window,calculates the sampling points and energy density distribution on the constellation map,and converts it into an equipotential constellation map with color information.From the perspective of deformable module and deformable organization,an adaptive deformable convolutional network classifier is built.The simulation results show that the method can automatically extract the modulation characteristics of the communication signal and improve the recognition accuracy of the modulation type of the communication signal under low signal-to-noise ratio.(4)In the process of communication signals individual identification,the observed signals are highly dynamic,non-stationary,and unbalanced by noise.This paper proposes a method for communication signals individual identification based on short-time slice transform domain features and ensemble learning.The method divides the communication signal into several short-term stationary signal segments by the sliding window trimming function,and extracts its transform domain features such as power spectrum,bispectrum,wavelet transform and highorder cumulant.The ensemble learning classifier is built by splicing the features of different transform domains.The simulation results show that the method can make full use of the correlation between signal slices,capture the subtle differences of signals,and improve the accuracy of individual identification of communication signals.(5)The individual characteristics of the communication signals selected by humans are not obvious,and the feature extraction method requires a large amount of calculation,which leads to problems such as limited recognition categories,low recognition rate,and poor real-time performance.This paper proposes a communication signal individual recognition method based on deep complex network.This method uses the I/Q complex representation model of the communication signal,and solves the complex batch normalization and complex weight initialization techniques by analyzing the complex network backpropagation and complex network loss function,and designs a deep complex residual unit and residual unit.A stack of difference units is used to build a deep complex residual network that can be used for individual identification of communication signals.The simulation results show that the method can mine the correlation between the I/Q channels of the communication signal,extract the more distinguishable individual characteristics of the communication signal,and improve the identification accuracy of the individual communication signal under the low signal-to-noise ratio.At the same time,the method is an end-to-end recognition model,which can realize the rapid identification of communication signal individuals.