Civil Airport Communication Emitter Identification With Deep Neural Network

In the field of signal processing,identification of communication emitter is defined to analyze and judge unknown received signal of emitter,to realize the identification of communication emitter and identification of modulation type of signals.The identification of communication emitter can be divided into two directions,which are the identification of modulation type of communication emitter signals and the identification of communication emitter respectively.The research object of this dissertation is the communication emitter based on the airport area.The modulation types and transmitting frequency points of the main interference emitter in the airport terminal area are similar to the normal communication equipment in the airport.One of the key technologies to realize the rapid screening of the interference emitters in the airport area is to realize the identification of modulation types of the communication emitter in the airport area.Therefore,this paper mainly studies the identification of modulation type of communication emitter on airport area.In this paper,firstly,traditional modulation recognition technologies based on feature extraction have been studied.Some traditional algorithm,such as the traditional identification algorithm based on signal decomposition multi-scale permutation entropy support vector machine(SVM)algorithm and higher-order statistics fully connected neural network algorithm have been replicated.The results showed that the identification accuracy of the two algorithms is high in the environment with fewer modulation types and high SNR,but the identification of performance of the two algorithms would decline sharply with the increasing of the modulation types and the declining of the SNR.Then,the identification algorithm of deep neural network was used to recognize the modulation types of communication emitter.The deep neural network included convolutional neural network(CNN)and residual network(Res Net).With the start of architecture design of two kinds of networks,some deep neural networks which are suitable for the identification of modulation type of communication emitter have been designed.The results showed that designed networks have the better performance,the performance with multi-type and low SNR environment is farther better than the traditional classification algorithm of signal feature extraction.In dynamic channel,when using deep neural network,problem about identification algorithm’s performance degradation has been studied in this paper.There are three Transfer Learning(TL)methods were proposed.Three transfer learning algorithms were used in two deep neural networks.The final results showed that the networks have a great performance for the identification of the modulation types in the dynamic channel.And the transfer learning algorithm based on the fine-adjusting of all layer parameters has the best performance for the CNN network.The transfer learning algorithms based on constant full connection layer and fine-adjusting of convolutional layer parameter have better performance than that of the two transfer learning algorithms for Res Net network.By using transfer learning algorithm,retraining of the network can be avoided when the channel has been changed,and identification performance of retraining networks can be achieved by depending on a bit of samples to fine-adjust the weight of the network.Thus,when the channel has changed,the problem about decline of identification performance can be solved well.