Research On DQN-based Intelligent Anti-jamming Technology Of MAC Layer In MANET

Mobile ad hoc network(MANET)is a distributed,non-centered and multi-hop autonomous network whose nodes have both terminal and routing functions.MANETs have great flexibility and adaptability,which have been used in a wide variety of applications,such as environmental monitoring,rescue and disaster relief,military communications and other fields.However,due to the characteristics of multi-hop transmission,dynamic topology,and resource-limited features,the reliability of mobile ad hoc networks is relatively weak and vulnerable to human jamming.This problem is particularly prominent for military communications applications.Therefore,in order to improve the reliability of MANETs,research on anti-jamming methods is necessary.Communi-cation anti-jamming methods include anti-jamming transmission technology at the physical layer and anti-jamming networking technology at the network layer.This dissertation focuses on the anti-jamming network technology of MANETs,especially the anti-jamming methods of the media access control(MAC)layerThis dissertation focuses on the channel jamming problem of MAC layer in MANETs.In traditional methods,this problem is usually modeled as a Markov decision process and relies on prior statistical information such as the channel state transition matrix.As the number of channels increases,the computational complexity increases dramatically.On the other hand,deep rein-forcement learning algorithm can obtain low-dimensional action output for high-dimensional state input,and does not need to model the environment,which has a great advantage in solving the problem of anti-jamming channel selection.In view of that,this dissertation studies the intelli-gent anti-jamming problem based on Deep Q-Network(DQN)algorithm.By introducing a deep reinforcement learning algorithm,the node channel information is trained,learned,and predicted to select normal channel transmission to achieve the goal of anti-jamming.Specifically,the main contributions of this dissertation are the following aspectsFirstly,in order to solve the problem of MAC layer channel jamming in MANETs,an in-telligent anti-jamming method based on DQN is proposed.The DQN-based anti-jamming model interacts with the channel information of the jamming environment,compares the predicted chan-nel status with the real channel status of the nodes and learns online,analyzes and summarizes the channel decision strategy to obtain the uninterfered channel information,and achieves the purpose of intelligent anti-jamming.After comparing the performance with the traditional method,it can be seen that the DQN-based intelligent anti-jamming method can achieve better anti-jamming per-formance under single-channel and multi-channel condition,and it can not only deal with active jamming sources,but also respond to reactive jamming sources under multi-channel condition.By improving this anti-jamming method,we get an adaptive anti-jamming method based on DQN.This adaptive method can effectively respond to jamming sources with changed jamming modes and complex comprehensive jamming sources.It re-adapts to the new jamming mode through re-training to achieve the purpose of adaptive anti-jamming.Numerical simulation and performance analysis have confirmed its effectiveness and practicality.In addition,in view of the error in the channel spectrum sensing of the nodes'channel j amming information,which leads to the performance degradation of the anti-jamming method,a clustering cooperative sensing anti-jamming method based on DQN is proposed.Firstly,this method performs online training of the DQN anti-jamming model.Ordinary nodes in the subnet sense the channel jamming status and send the results to the cluster head nodes.The cluster head node computes collaboratively sensing results and obtain channel decision results with higher accuracy,and uses these results for DQN anti-jamming model training.Then the DQN anti-jamming model is testing offline.The cluster head node sends the parameters of the trained DQN anti-jamming model to each ordinary node.Each node can realize the local offline independent anti-jamming function by constructing a local DQN anti-jamming model.The method has good anti-jamming performance and strong adaptability on both single-channel and multi-channel conditions.Through numerical simulation and performance analysis,it can be seen that compared with traditional methods,the DQN-based clustered collaborative spectrum sensing anti-jamming method can effectively reduce network message flooding and resource consumption while ensuring anti-jamming performance,which is highly flexible and robust.