Research On Congnitive Anti-Jamming Based On Bayesian Compressive Sensing And Reinforcement Learning

Anti-jamming communication has become the focus and difficulty of information-based operations in military field.With the rapid development of battlefield equipment,traditional anti-jamming methods will face the challenges of combating wideband and dynamic jamming.And especially in recent years,algorithms in the field of artificial intelligence have been introduced into the communications field,making jamming intelligent,which needs to explore and research new ideas and methods of communication anti-jamming.Therefore,in response to the demand for real-time acquisition and processing of information in a highly dynamic and complex interference environment,aiming at traditional methods to face the problems of wideband and intelligent jamming with low perception accuracy and slow decision-making speed.According to characterize of jamming signal is different in multiple dimensions.Based on compressive sensing and reinforcement learning theory,a mathematical model of spectrum perception and cognitive decision-making is built,and a method is proposed to realize cognitive anti-jamming from the perspective of multiple domains,such as transform domain,frequency domain and power domain.The proposed methods overcome the key technologies of intelligent anti-jamming solution design,and provide key technologies for China's anti-jamming communications field.The main contents of this paper are summarized as follows:(1)Focusing on the problem that Chirp-like jamming signals have large time and bandwidth characteristics that make it difficult for traditional algorithms to detect them effectively,a multi-antenna wideband spectrum detection algorithm based on compressed sensing and entropy is proposed.First,based on the Bayesian compressed sensing theory,a compressed sensing model is constructed.According to the sparse characteristics of the Chirp signal in the fractional Fourier domain,the signal is sparsely transformed using the fractional Fourier transform.Then,the signal is processed through the Gaussian observation matrix to get data observation.After that,in order to solve the measurement uncertainty due to various factors,the relevance vector machine algorithm based on the probability model is applied to accurately reconstruct the signal.Finally,to solve the noise uncertainty robustness of the traditional energy detection algorithm for poor problems,frequency domain information entropy-based spectrum decision algorithm is applied to improve signal detection performance.Simulation results show that the proposed method improves the reconstructionaccuracy by 25.17% compared with the single-task algorithm,and improves the detection accuracy by 5.63% compared with the Bayesian assisted spectrum sensing algorithm.(2)Focusing on the problem that poor adaptive ability of traditional anti-jamming algorithm with highly dynamic and intelligent interference in the complex electromagnetic communication environment.Taking the unmaned aerial vehicle swarm as a research example,based on the intelligent decision theory,cognitive anti-jamming algorithm combining frequency domain and power domain algorithm is proposed.First,based on the advatange actor-critic theory,the drone is regarded as an agent,and the interference channel is determined by the perceived environmental spectrum state.Then,based on the Stackelberg game theory,the medium interference level channel is suppressed from the power domain to reduce the time overhead of channel switching.Finally,a cluster-head assisted method is introduced to solve the problem of low channel decision success rate due to the weak local spectrum sensing capability of a single agent.Simulation results show that the proposed algorithm can improve network communication security capacity and channel decision success rate.