Research On Digital Modulation Jamming Method Based On Deep Learning

In today’s society,digital communication has gradually replaced analog communication and becomes the main method in the field of communication.The rapid growth of the number of communication equipment and the increasing complexity of the electromagnetic spectrum environment have brought great pressure to the management of the electromagnetic spectrum.Because the inherent openness of wireless communication system,jamming attack has become an important method of electronic warfare and electromagnetic spectrum control.As a non-partner,under the premise of complex electromagnetic environment and rapid transformation of target parameters,how to make intelligent jamming decision according to the enemy signal characteristics is an urgent problem to be solved.Therefore,based on the deep learning,this paper studies the jamming waveform design of digital modulation signals according to the loop of "perceiving the environment-adapting to the environment-making decisions-taking actions".The main contents of this paper are as followed:Aiming at the low recognition accuracy of actual electromagnetic signals in perceptual environment,we propose an auto modulation classification(AMC)algorithm based on permutation entropy,high-order cumulant and Res Net-Complex Net.Firstly,the parameters of each layer of convolutional neural network are optimized,and the classification performance of VGGNet,Res Net,CLDNN,Dense Net and Res Net-Complex Net is compared.It is proved that the Res Net-Complex Net is feasible as the backbone network.Aiming at the serious confusion between MPSK and MQAM signals,the features extracted by neural network are combined with artificial features such as permutation entropy,high-order cumulant and instantaneous frequency,which are jointly recognized by K-Nearest Neighbor Classifier in the Radio ML 2016.10 a.The simulation results show that the algorithm we proposed can effectively solve the problem of confusion between MPSK and MQAM signals and improve signal recognition accuracy.Aiming at the scene that the target communication signal can be identified,a method of using deep learning to identify the signal and jamming according to the signal category is proposed.The optimal jamming pattern of common signals under additive gaussian channel is deduced,and the optimal jamming mapping relationship is obtained on the premise that the carrier frequency,symbol interval of jamming signal and communication signal are the same.A communication system based on GNU Radio is constructed in the simulation to generate signal and test the jamming performance.The signals are identified by the above method,and the corresponding jamming waveform is sent according to the identification result.Simulation results show that the bit error rate of the proposed jamming method is higher than that of gaussian jamming under high SNR.Aiming at the situation that the characteristics of the target signal are unknown or the dynamic changes of parameters are difficult to deal with,a jamming waveform generation technology based on convolutional auto encoder(CAE)is proposed.The potential characteristics of the target signal are extracted by the encoder,and a jamming signal similar to the communication signal is generated.The simulation results show that the jamming waveform generated by the proposed algorithm can approach the theoretically efficient jamming effect,and provide a new solution for the generation of communication jamming waveform.