Research On Network Handover Management And Computing Offloading Strategies In Ultra-dense Heterogeneous Networks

Ultra-Dense Heterogeneous Wireless Network(UD-HWN)and Mobile Edge Computing(MEC)have become key technologies for solving various computationally intensive and delay-sensitive application requirements in 5G networks.Among them,UDHWN realizes the integration of multiple wireless technologies(4GLTE,5GNR and Wi-Fi,etc.)by deploying micro base stations within the coverage of macro base stations,which can meet the access of massive mobile devices and the needs of different application services,while MEC By deploying computing,storage,and network services near users,application service latency can be effectively reduced.However,in UD-HWN,due to the dense deployment of various types of micro base stations and the mobility of mobile devices(Mobile Devices,MD),in order to maintain the best network connection state,the traditional network handover decision algorithm is easy to cause frequent network handover due to the relatively single decision factors,which may lead to problems such as ongoing service interruption,increased MD energy consumption,and increased network handover signaling interaction.In order to solve the shortcomings of traditional network handover decision algorithms,relevant scholars have proposed multiattribute handover decision-making methods,but these methods still have some shortcomings.They do not comprehensively consider the influence of various performance characteristics of different types of networks on handover decision and a large number of signaling interactions between MD and the Base Station in the network handover process will resulting in the reduction of available network resources and the problem of MD energy consumption in UD-HWN application scenarios.On the other hand,in UD-HWN,the complexity of the network environment is increased due to the dense deployment of micro base stations,and the dynamic time-varying network state will affect the offloading strategy.The limited computing resources of the MEC server cannot satisfy a large number of task offloading requests at the same time.The mobility of MD may lead to problems such as task service interruption or even computational offload failure.Therefore,task offloading in UDHWN is challenging.So,the problem of computational offloading is one of the problems that needs to be solved urgently.In view of the shortcomings of the existing work,this paper studies the network handover management and task offloading problems in UD-HWN based on the deep reinforcement learning algorithm.The main research contents are as follows:(1)This paper proposes a joint RSS prediction and multi-attribute-based adaptive handover decision scheme to reduce the size of candidate networks and the number of network handovers.First,a screening mechanism based on RSS threshold is set up to filter out qualified networks and add them to the candidate network list to reduce the size of the candidate networks.Second,comprehensively consider five service types and seven network decision attributes,use Analytic Hierarchy Process(AHP)to calculate the decision attribute weight values corresponding to different services,and then re-optimize the attribute weight values based on user preferences.The handover between the MD and the network is jointly restricted to reduce the number of network handovers.Third,the handover process of the network is modeled as a Markov decision process(MDP)model,combined with the deep reinforcement learning A3 C algorithm,the business type and candidate network performance parameters are used as input,and the critical neural network is used to calculate the state value of each candidate network,and select the network with the largest state value as the target handover network to achieve the selection of the optimal target network.Finally,according to the historical information of the MD and the RSS of the target network and the location information corresponding to the MD,based on the Gated Recurrent Unit(GRU)model in deep learning,the RSS of the MD in the future time slot is predicted,and the RSS threshold is updated and optimized and reduce signaling interaction between MD and surrounding network,improve network throughput and reduce energy consumption of MD.(2)This paper proposes an offloading strategy for cooperative multi-point computing based on user location awareness.In order to reduce the task offloading delay and task offloading failure rate,this paper applies the SDN-based centralized computing offloading collaboration architecture to realize the collaborative computing of the MEC server.Firstly,assign task priority to MD according to the characteristics of task types.Then,the task offloading problem is modeled as an MDP model,a task offloading algorithm based on deep reinforcement learning DDPG is proposed,and the appropriate offloading node is selected according to the task type and the load status of the MEC server.Finally,according to the MD historical trajectory,predict the location of the future time slot MD and the connected target Base Station,and return the calculation result to the target Base Station in advance,aiming to minimize the system service delay.