Research On Anti-Jamming Communication Technology Based On Machine Learning

In modernization and future wars,information and communication technology plays a vital role,making anti-jamming communication one of the core technologies in electronic countermeasures.Frequency hopping technology is widely used in military and civilian fields for advantages,such as easy networking and good anti-jamming ability.To enhance the anti-jamming ability of the frequency hopping system,it is necessary to avoid jamming attacks.Therefore,an interference cognition and avoidance technology are proposed combining the ideas of cognitive radio and frequency hopping technology.Firstly,we exploit digital signal processing technology to estimate the carrier frequency of the interference and build the interference occupation status of the frequency band.Then we derive an optimal anti-jamming policy(optimal sub-band selection policy)using machine learning to learn and predict the knowledge obtained from interference cognition.The intelligent anti-jamming technology can reduce the risk of "collision" with the interference frequency band,and improve the communication anti-jamming utility of the frequency hopping system.1.Interference cognition: The jamming signal is usually a time-domain aliasing signal of multiple narrowband signals,while the traditional parameter estimation algorithm is only applicable to the scenario of a single signal.The cyclic cumulant of the digital signal has the selectivity to the aliasing signal and the excellent noise immune.Therefore,a cyclic cumulant-based carrier frequency estimation algorithm is proposed for multi-signal aliasing,which exploits the 4th order 0 cyclic cumulants of the signal to estimate carrier frequency.Exactly,we do not need the complete cyclic cumulant of the signal.The cyclic cumulant is sparse;therefore,we apply compressed sensing technology to reconstruct the cyclic cumulant of the signal,which reduces the sampling complexity of the ADC compared to the traditional calculation method.Finally,simulation validates the effectiveness of the proposed interference cognition technology.2.Interference avoidance: This paper mainly applies AI algorithms to learn the interference occupation status obtained from interference cognition to realize intelligent interference avoidance in a frequency hopping system.Refer to the common types of interference in wireless communication,such as multitone interference,frequency sweep interference,etc.Markov interference is proposed and defined.For the various types of interference,a 0-1bit quantized state representation is proposed,and then this paper exploits reinforcement learning technology to propose a Q-Learning-based cognition frequency hopping algorithm.Finally,simulation validates the proposed algorithm.3.In some jamming models,the interference frequency band may be related to many moments,and some systems can obtain the interference spectrum.Due to its limitations,the Q-Learning algorithm cannot directly process the spectrum because it has infinite environmental states.Therefore,for this scenario,this paper proposes a spectrum flow model that combines time and spectrum characteristics.Moreover,a DQN-based cognition frequency hopping algorithm is proposed to process the spectrum flow,which combines deep neural network(DNN),convolutional neural network(CNN)and Q-learning algorithm.And it learns interference features directly from the spectrum flow which contains time and frequency domain information,to obtain an effective optimal sub-band selection policy and achieve intelligent anti-jamming.Finally,simulation results validate the algorithm performance and indicate that it can be widely used for combating various jamming models.