Research On MAC Cognitive Jamming Technology Based On Reinforcement Learning Theory

In communication jamming,jamming techniques utilizing medium access control(MAC)protocol of target improve jamming efficiency.However,achieving MAC protocol jamming is difficult in practice,mainly in two aspects.The first one is identification of target MAC protocol type.In MAC protocols of different types,such as non-contention MAC protocol and contention based MAC protocol,communication behavior patterns vary largely,so different jamming strategies should be adopted accordingly.Therefore,it is essential to study MAC protocol type identification methods.Second,it is difficult to solve the optimal jamming parameters directly.On the one hand,the optimal jamming parameter value relate to the MAC protocol parameters of the target network.But in the non-cooperative conditions,especially when the communication system utilizes encryption technology,it is difficult to obtain the relevant parameter information for the jammer.On the other hand,due to the complexity of the MAC protocol process,it is hard to analyze the jamming effect theoretically under different parameters,and to establish the theoretical model of the jamming effect.Due to technical constraints,traditional jamming methods cannot solve above problems effectively.Cognitive radio gives wireless communication equipment tremendous ability of sensing and learning.The application of cognitive radio in wireless communication jamming,which is called cognitive jamming,is expected to improve the combat effectiveness of jamming system.Cognitive jamming system is a dynamic closed-loop system composed of observation,decision-making and evaluation,and adopts machine learning method to enhance the system's sensing and learning ability.In this thesis,the reinforcement learning theory in machine learning domain is applied to improve observation and decision-making ability respectively,to solve the problem of protocol type identification and jamming parameter optimization in MAC protocol jamming.The main contributions and innovations of this work are summarized as follows:1)Aiming at the problem of MAC protocol type identification,a MAC type identification method based on power detection and ensemble selection algorithm.The proposed method can identify non-contention and contention-based MAC.On the basis of the existing characteristics,collision probability estimation and the periodic estimation feature are proposed to improve identification performance.Selective ensemble algorithm based on reinforcement learning is utilized to improve the generalization ability of the classification system.Experiments using data from OPNET simulation show that the proposed features improve identification performance,and that the proposed method can effectively improve the identification performance when parameters of the target are different from the training samples.2)Aiming at the problem of jamming power allocation on different channels in none-contention MAC protocol,cognitive jamming power allocation algorithm based on reinforcement learning is proposed.The algorithm learns optimal jamming power allocation online with targets' positions unknown.Multi-arm bandit model in reinforcement learning is used to analyze the jamming power allocation problem.The jamming power allocation schemes are modeled as the individual actions,while the jamming effect evaluation is modeled as reward,and the upper bound of the confidence interval algorithm is applied as the online learning method.The simulation results show that in the case of target positions unknown,the method converged to the optimal jamming power allocation method through online learning.3)In order to solve the problem of jamming frame type selection and jamming probability optimization in contention based MAC jamming,cognitive contention based MAC jamming method based on reinforcement learning is proposed.The method optimizes jamming frame type and jamming probability through online learning.Because the jammer cannot get frame type directly under encryption,the method extracts time interval feature and utilizes clustering algorithm to identify frame type.Using multi-armed bandit model,jamming effectiveness evaluation is modeled as rewards,jamming frame type and jamming probability are modeled as actions respectively,and upper confidence bound algorithm is adopted as online learning algorithm.Experiments based on OPNET simulation show that when jamming a single network,the proposed method has similar jamming effectiveness compared with traditional jamming,and saves jamming energy.When jamming multiple networks,the proposed method is more effective than traditional methods after some time of learning.