Research On Intelligent Routing And Embedding Of Sliced Ad-hoc Network

As a kind of decentralized,self-organizing and self-optimizing peer-to-peer multi hop communication network,Ad-hoc has been widely used in military and civil communication networks for its simple deployment and strong invulnerability.With the advancement of technology changeing and industrial reforming globally,military reform has also entered a new phase of development.As one of the key technologies in the intelligent era,cloud computing was first applied in the military field by the United States: the concept of “combat cloud” was proposed.“Battle cloud” platform is expected to build an overall mesh network which can be used for data distribution and information sharing in the battle space.Each authorized user,platform,or node can send and receive basic information transparently,and can use this information in the whole range of military operations to change the current old military network structure from the aspect of network.As an important part of military network,Ad-hoc can only provides services for a small number of traffics whose service requirements are simple due to its simple network structure and scarce communication resources,and can not support the on-demand service and collaborative organization required by “battle cloud”.Therefore,this paper applies network slicing technology in 5G to Ad-hoc network and proposes sliced Ad-hoc network architecture enables Ad-hoc network to use limited network resources,support more complex and diverse business types,and provide a solid network platform for future new application scenarios such as “battle cloud”.This paper proposes the architecture of sliced self-organizing network according to the technical characteristics of network slicing,and studies the slice embedding and service routing in sliced self-organizing network.The main research contents are as follow.Firstly,this paper studies the sliced network architecture.Sliced Ad-hoc network consists of arrangement management slice,connectivity management slice,slice control slice,and service slice.Different sections are flexibly configured and deployed according to their functionalities and service requests.Under this network architecture,the physical resources in the substrate network are divided into different sections for management,which realizes the flexible configuration of Ad-hoc network resources and improves the performance of Ad-hoc network.Then we study the slice embedding problem in sliced self-organizing network.Due to the dynamic change of network topology,the self-organizing network needs to adjust the embedding mode of virtual link after slicing,and ensure the slicing performance by re-embedding.Frequent re-embedding will lead to huge network overhead.In order to reduce network overhead,this paper proposes a slice re-embedding algorithm based on AC(Actor-Critic)algorithm,and introduces graph neural network technology to learn the topology of network to enhance the generalization of the algorithm.The simulation results show that the algorithm can achieve lower re-embedding cost and higher embedding success rate in dynamic topology environment compared with traditional algorithms.Next,we study the OLSR(Optimized Link State Routing)routing protocol in sliced self-organizing network.As a common routing protocol in wireless Ad-hoc networks,OLSR still works for sliced Ad-hoc networks.In this paper,we optimize the selection of MPR nodes in this protocol.MPR selection problem is NP-complete.Because of the high dynamic of self-organizing network,many MPR node selection algorithms will meet probelms such as redundancy of node selection and unfair node selection when running.In order to solve these problems existing in traditional algorithms,we propose a new MPR node selection algorithm based on DQN(Deep-Q Network)technology.The simulation results show that compared with traditional algorithms,our algorithm reduces redundancy and unfairness in the process of node selection,and the application of graph neural network module also makes the algorithm have better performance in high dynamic network environment.Finally,we study QoS(Quality of Service)routing problem in sliced self-organizing network.The QoS routing problem with multi performance index is NP-complete.We propose an algorithm named QLRA(Q-learning Routing Algorithm)based on Q-Learning technology to approximate the QoS routing problem,and design a corresponding routing optimization algorithm.The simulation results show that compared with the AODV(Ad hoc On-Demand Distance Vector Routing)algorithm,the algorithm can meet the service performance each QoS requirement and get higher routing success rate.