Research On Environment Information Sensing Based Communication And Computation Resource Allocation Method In Vehicular Networks

Vehicular network is one of the typical application areas for the deep integration of the Fifth Generation Mobile Communication System(5G)and vertical industries,which is expected to become a new round of economic growth by integrating various advanced technologies such as cloud computing,edge computing,information sensing and artificial intelligence.However,the various emerging services under the vehicular network put higher demands on the performance of wireless networks.In order to meet these extreme performance demands,communication and computation resources can be managed based on the environment information sensing such as network load,transmission environment and information generation status,however,there are still many challenges.First,the existing vehicular network mainly conducts discrete analysis and optimization of sensing,communication and computing processes,and does not consider joint analysis of the performance of the overall sensingcommunication-computing process which is unable to grasp the balance between the three performance constraints.In addition,limited by the communication environment and computation resources,the performance of continuous services can hardly be fully guaranteed to meet the requirements of delay-sensitive tasks.Finally,the related characteristics among tasks of vehicle users(e.g.,distributed learning)limit the transmission and computation performance,thus affecting the overall task processing efficiency.In view of this,this paper conducts a research on the environment information sensing based computation resource allocation method in vehicular networks.The main content and innovation points can be summarized as follows:First,the space-time domain performance analysis method of information freshness is studied for the communication and computation process for vehicle sensing data,and the modeling and analysis of the joint sensing-communication-computation process is realized based on the stochastic geometry theory and queue theory.Specifically,considering the linear distribution characteristics of users and road side unit facilities in vehicular networks,a closed-form solution of uplink coverage probability and expected data rate is derived using Cox random point process modeling by modeling interference as a combination of standard path loss and smallscale fast fading.After that,the closed-form solution for the average age of information is derived by proposing a tandem queue model for the timing relationship between communication and computation.Finally,the correctness of the mathematical derivation is verified based on Monte Carlo simulations.The theoretical and simulation results show that improving the communication or computation capacity alone will cause waste of resources,and the existence of optimal communication and computation resource allocation methods can ensure the information freshness while using the resources efficiently.Second,the resource allocation method for optimizing the average age of information performance is studied,and an algorithm for adaptively adjusting the communication and computation capacity of the wireless link is designed to achieve efficient utilization of resources.Specifically,firstly,starting from the communication power,the transmit power corresponding to the minimum transmit energy consumption is determined using convex optimization method,and this transmit power subinterval is divided using the in-out method.Then a one-dimensional search algorithm is used to search for the local optimal solution within the sub interval.Finally,the global optimal solution is obtained by comparing the local optimal solutions of the subintervals.The simulation results illustrate that the proposed algorithm obtains a performance close to the theoretical optimum by balancing the communication and computation capacities.Third,an environment sensing based distributed learning communication and computation resource allocation method for joint vehicles and roadside units is proposed to improve the efficiency of overall task processing.Specifically,a three-tier network deployment scheme for distributed learning is designed,where the free computing power of the edge servers of the roadside units is used for model training,and an optimization problem for minimizing the long-term weighted latency and training loss sum is constructed on this basis.After that a time-scale division and service management method is proposed so as to decouple the original problem into a short time-scale communication resource allocation subproblem and a long time-scale computation resource allocation subproblem.An iterative algorithm based on fractional programming(FP)is used to solve the straggler problem of distributed learning;a multi-agent reinforcement learning algorithm(MARL)is used to dynamically achieve access selection and data offloading ratio control.Finally,the convergence and robustness of the algorithm are verified by building a simulation platform based on the protocol specification related to Sidelink of the 3rd Generation Partnership Project(3GPP).The simulation results illustrate the performance advantages of environment sensing distributed learning based on computation offloading over distributed learning based on baseline federated computing and edge computing.In summary,this paper is investigating environment information sensing based communication and computation resource allocation method in vehicular networks,which provides the theoretical basis and algorithmic practice for future intelligent management of self-driving vehicles and wireless networks.