Research On Cognitive Covert Communication Technology Based On Deep Learning

Recently,as a key technology to effectively improve the spectral efficiency,realize the intelligent communication and win the electromagnetic spectrum war,cognitive radio has been widely concerned by scholars and institutions.With the extensive application of cognitive radio network,how to realize its highly secure and highly reliable communication has become a hot issue.The traditional encryption and physical layer security technology can not meet the development needs of modern communication.It is urgent to explore new technologies with higher level of information security.As a low detection/intercept-probability communication,covert communication guides the development direction for realizing secure and reliable communication in cognitive radio network.Especially,using deep learning to realize intelligent covert communication will greatly promote the application of cognitive covert communication.Combined with the new requirements of cognitive covert communication network,this paper mainly studies the covert communication in cognitive non-cooperative system or cognitive cooperative system,and resource optimization in cognitive covert communication network.1.For the unmanned aerial vehicle(UAV)jammer assisted covert communication scenario in cognitive non-cooperative system,the optimization problem is first transformed into a dynamic game between the UAV and eavesdropper,a novel model-driven generative adversarial network(MD-GAN)is designed,which makes the unknown channel information and the detection threshold as the learned weights.Especially,the number of layers,the number of neurons,the weights and non-linear transforms of neurons per layer in MD-GAN are explainable.Then a GAN based joint trajectory and power optimization(GAN-JTP)algorithm is developed to alternatively train the generator and discriminator in a competitive manner,which results in the Nash equilibrium to output the feasible scheme.The simulation results show that,the proposed GAN-JTP with a rapid convergence speed can generate UAV’s trajectory and power optimization scheme with partial channel distribution information(CDI)and unknown eavesdropper’s detection threshold,and achieve a much better covert performance over the benchmark algorithm.2.For the intelligent reflecting surface(IRS)enhanced covert communication scenario in cognitive non-cooperative system,with the transfer learning and generative adversarial network,a domain adversarial transfer network composed of source domain generator,target domain generator and discriminator is first designed,and then a domain adversarial transfer network based resource optimization algorithm(DATN-ROA)is proposed,which learns and transfers the hidden features of resource optimization without covert communication,and generates the joint optimization scheme of the legitimate user’s beamforming and IRS’s phase shift matrix through the dynamic game between the legitimate user and eavesdropper.The simulation results show that,the proposed DATN-ROA can generate near-optimal resource optimization solution for the covert communication and achieve rapid convergence with the unknown eavesdropper’s channel state information(CSI)and detection threshold.3.For the cognitive relay covert communication scenario by using the primary user’s authorized channel in cognitive cooperative system,a novel GAN based power allocation algorithm(GAN-PA)is proposed.Under the proposed GAN-PA,the generator imitates the cognitive relay to generate the power allocation scheme for covert communication,and the discriminator imitates the primary user to judge whether the cognitive relay transmits covert information.The generator and discriminator composed of three-layer neural network are trained alternately and finally output the feasible scheme.Numerical results show that the proposed GAN-PA with a rapid convergence speed can attain near-optimal power allocation solution for covert communication with partial network state information and unknown eavesdropper’s detection threshold.4.For the cognitive covert communication network performance optimization scenario,the performance evaluation mechanism is established,which evaluates the cognitive covert communication network by spectrum efficiency,energy efficiency and fairness.A novel modeldriven deep reinforcement learning assisted resource allocation method is investigated.The Alternating Direction Method of Multipliers(ADMM)iterative data stream is first obtained according to the maximum spectrum efficiency,and the deep neural network(DNN)based optimization framework is constructed,which makes the CSI as the learned weights.Then a channel information absent Q-learning resource allocation algorithm(CIAQ)is proposed.By interacting with the network environment,the DNN based optimization framework is trained without a large number of labeling data,and the resource allocation scheme is autonomously obtained with partial CSI.The simulation results show that,the proposed CIAQ algorithm with fast convergence can not only determine the optimization objective by adjusting the discount factor,but also have better optimization performance than the random initialization method and the existing resource allocation algorithms.