Research On Radar Anti-Jamming Technology Based On Deep Reinforcement Learning And Game Theory

As the main equipment for detecting and tracking targets,radar plays an important role in modern electronic warfare.In order to reduce the detection performance of radar,opponents usually use jamming technology to attack radar.Nowadays,the diversified and intelligent jamming means make the traditional radar anti-jamming methods difficult to cope with,which brings great challenges to the radar working environment.With the rapid development of the field of artificial intelligence,radar intelligent anti-jamming technology combined with machine learning has been widely studied.At the same time,machine game theory is also used in the study of electronic countermeasures game.In this thesis,an intelligent anti-jamming strategy learning method is proposed by combining the deep reinforcement learning algorithm in machine learning with adaptive frequency hopping and pulse width allocation.And for the intelligent jammer with self-learning ability,the game confrontation between radar and jammer is studied.The main research work of this thesis is summarized as follows:(1)Aiming at the problem of radar coping with common jammers,this paper proposes a radar antijamming strategy learning method based on deep reinforcement learning,which combines adaptive frequency hopping and pulse width allocation.It not only effectively deals with external malicious interference,but also improves the integration efficiency and Doppler frequency resolution of radar echo processing.Firstly,the framework model of radar anti-jamming system is constructed,and the radar state is taken as the input of the strategic neural network of deep reinforcement learning.The action selection is carried out according to the dynamic greedy algorithm of DQN algorithm,and the anti-jamming strategy is realized by selecting the optimized pulse transmitting frequency and pulse width.The experience replay pool is then used to store the agent’s previous states,actions,rewards,and next states.During the training process,the agent randomly selects a batch of data from the experience playback pool to update the parameters of the neural network.Through iterative training,the radar anti-jamming intelligent decision method based on deep reinforcement learning is finally realized.The experimental results show that the radar anti-jamming performance of this method is better than random frequency hopping,Q learning and other anti-jamming methods.(2)Aiming at the problem of radar and intelligent jammer confrontation,a radar anti-jamming strategy learning method based on deep counterfactual regret minimization algorithm with regret value discount is proposed.The radar can change its carrier frequency according to the strategy.The competition between the frequency agility radar and the intelligent jammer can be regarded as a noncooperative game between the two parties,in which there are multiple rounds of interaction in the case of imperfect information.The competition process between radar and intelligent jammer is modeled in the extensive-form game,and considering that both sides have the same intelligence level,radar and jammer are described in the form of game tree.Because the calculation amount increases exponentially with the number of radar pulses in the game,the traditional counterfactual regret minimization algorithm is difficult to obtain Nash equilibrium strategy.In order to accelerate the convergence of the algorithm,deep neural network is used to approximate the behavior of the counterfactual regret minimization algorithm by combining deep reinforcement learning and adding regret discount.Finally,the effectiveness of the algorithm in finding Nash equilibrium and obtaining the optimal strategy is verified by the analysis of experimental results.