Research On Key Technologies Of Software-Defined Satellite Networks

Satellite networks with global coverage capability,high transmission capacity,and without limitation of geographic conditions,can effectively complement and improve the terrestrial networks.Especially,they can provide flexible,convenient,and low-cost network access for areas without sufficient terrestrial network infrastructure.Therefore,with the rapid expansion of network scale,the explosive growth of multimedia services,and the vigorous development of various applications,numerous Internet giants and satellite companies have carried out researches on low earth orbit satellite networks with low transmission power,low propagation delay,small free-space attenuation,and global coverage capability.Although satellite communication technology has achieved great progress,its development is far behind the terrestrial networks.The traditional satellite network architecture suffers from the challenges of technology closure,high update cost,and difficulty in flexible control and management.Thus,it cannot efficiently meet the requirements of network management and control,and user Quality of Experience(QoE)improvement brought by the development of personalized,customized and multi-functional network services.To this end,considering the technological advantages of the software-defined networking architecture and solutions,this dissertation conducts research on the following four aspects to improve satellite network openness,flexibility,and user's QoE.1)Designing a novel satellite network architecture to effectively improve network management and control capabilities by introducing "softwarization" and "virtualization" technologies.2)Designing the control plane architecture for the novel satellite network architecture to improve its reliability,scalability,and latency by using the characteristics of satellite networks.3)Establishing a QoE assessment model for satellite network multimedia services to provide the QoE model basis for the QoEaware routing mechanism.4)Designing an intelligent routing mechanism to improve the user's QoE by employing the proposed novel network architecture and QoE assessment model.The main contributions and innovations of this dissertation are as follows.(1)Given the problems of closed technology,high update cost,and lack of interoperability in traditional satellite networks,a software-defined satellite network architecture which is composed of a control plane and a data plane is designed by applying the software-defined network architecture and the ideas of "softwarization" and "virtualization" networking technologies into satellite networks.The control plane is responsible for collecting network status,making and releasing forwarding strategies.The data plane is only responsible for forwarding data flows.By decoupling the data plane and control plane,a centralized management model is formed in this architecture,which simplifies the functions of satellite network forwarding nodes and empowers the nodes with dynamic programming features.Secondly,both southbound interface supporting extended OpenFlow and programmable hardware platform are equipped in satellite network forwarding devices to implement the function virtualization of their network/MAC/physical layers.Thirdly,based on the network virtualization paradigm,hypervisors from high-level to low-level are proposed to create multiple virtual networks on the shared network infrastructure and optimize network utility with Quality of Service(QoS)provision.Finally,combining with typical application scenarios,numerical results validate the feasibility of the proposed network architecture,and its advantages over traditional network architecture in terms of network resource scheduling,rapid function deployment,and CAPEX cost reduction.In addition,the implementation roadmap of software-defined satellite network architecture is presented in aspects of timescale,technology and research focus.(2)Considering the global node distribution,easy link failure,and dynamic topology change in satellite networks,a distributed and multi-layer hierarchical controller placement scheme is proposed to solve the problem of control plane design to improve the network reliability,scalability and delay performance.Firstly,the models of sub-satellite point and satellite-ground station/satellite-satellite accessibility are established,and the metrics of network reliability,link maintenance cost,and link use cost are defined.Based on the above,a controller placement optimization model is established for the software-defined GEO/LEO satellite networks.Secondly,a distributed and hierarchical control architecture is proposed,where the network operations control center,GEO satellites and a part of LEO satellites respectively play the roles of the super controller,domain controllers,and slave controllers.Furthermore,a slave controller selection strategy based on the LEO satellites with lower latitude in the same orbit,and a shortest distance based inter-layer out-of-band control strategy are proposed.The super controller is responsible for information maintenance and policy making/releasing for the entire network;the domain controller is responsible for the network status collection and policy making/releasing in its coverage;the slave controller is responsible for network status collection and control command issue in its orbit by using the Inter-orbital links.In addition,according to the latitude change of the LEO satellites,a time slot division method is proposed to divide the network control topology into a series of time slots.It reduces control link handover and enhances the stability of control links.Finally,extensive simulations demonstrate that the proposed control architecture not only obtains high stability and scalability,but also efficiently saves the maintenance and use cost of control links and reduces intersatellite link interferences.(3)The design of QoE-aware routing in software-defined satellite networks depends on the QoE assessment for multimedia service over satellite networks.Thus,the modular neural network and Deep Belief Networks(DBNs)are combined to design a deep learning-based modular QoE assessment method,which establishes the QoE/QoS correlation model to translate QoS parameters into user's QoE parameter.Firstly,the complex task of QoE assessment for multimedia services is decomposed into various simple sub-tasks by using traffic classification.Then,multiple heterogeneous DBNs are employed to complete each subtask(i.e.,learning the mapping relationships between QoE and QoS for different types of services)simultaneously.An integrative approach based on relative distance is further exploited to select sub-neural network(s)to predict QoE result collaboratively.Secondly,to determine the parameters in QoE assessment model,the dataset composed of QoS parameters and subjective opinion scores is created by combining satellite network simulator with the subjective test.Finally,based on the proposed software-defined satellite network architecture,the QoE assessment application framework is designed.The experimental results demonstrate that the proposed assessment method can not only improve prediction accuracy,computational efficiency,and scalability,but also reduce the impact of traffic classification accuracy on the assessment method performance.Finally,the relationship between adoption rating and user's opinion scores is analyzed with a conclusion that there is a positive correlation between them.(4)To improve the user's QoE of personalized service and self-learning ability of satellite networks,a DRL-based QoE-aware adaptive routing mechanism is proposed for video service over software-defined satellite networks.This routing mechanism combines Deep Reinforcement Learning(DRL)with the established QoE assessment model and makes full use of the promising features of the proposed network architecture,including obtaining global information and dynamically reconfiguring nodes.Firstly,a virtual node strategy is adopted to construct a fixed virtual topology,and then the domains are divided according to the coverage of the domain controller.Thereby,an inter-domain routing strategy based on the request area is proposed.The super controller executes the inter-domain routing strategy to map the crossdomain service requests into different domains.Then,the domain controller aggregates the cross-domain service requests and the intra-domain service requests.The aggregation service flow matrix and the intra-domain distance-based topology are taken as state,the virtual link weights as action,the QoE assessment value as reward,to establish the DRL-based intradomain routing model.The Deep Deterministic Policy Gradient(DDPG)is exploited to learn the strategy of optimizing link weight configuration directly and to determine the transmission path for traffic flows.The parallel execution of multiple domain controllers and the cooperation among different controllers balance the computational load among controllers and improve the computational efficiency of cross-domain requests and intra-domain requests.Theoretical analysis and comparative experiments have shown that the proposed routing mechanism can not only significantly improve the user's QoE and resource utilization,but also reduce packet loss rate and end-to-end delay.