Low Latency Random Access And Resource Scheduling In Large-Scale Wireless Network Based On Mean Field Theory

Nowadays,with the wide application of the Internet of Things(Io T)in various fields of society and life,the number of devices and computing application tasks have explosive growth.Therefore,an efficient distributed communication algorithm is needed to achieve random access and resource scheduling in large-scale wireless network,thereby reducing the delay.However,traditional distributed communication algorithms suffer from access conflicts and severe network overload in large-scale wireless network.In this scenario,the mean field theory is utilized to make the distributed decisions in large-scale wireless network,which can reduce the complexity of large-scale interactive systems.The thesis focuses on the problems of the limited communication resources and complex interaction among devices in the large-scale wireless network.The main goal of the paper is the low latency.Based on the main theory,i.e.,mean field theory,a series of low latency distributed communication algorithms for large-scale wireless networks are proposed.The thesis designs efficient distributed algorithms for random access and resource scheduling problems in large-scale network interactive scenarios,the main work and innovation of the thesis are as follows.For the large-scale homogeneous devices random access problem,the couple energy and data queues and the influence from large-scale devices on random access are studied and a delay-optimal random access algorithm based on mean field theory are proposed.For coupled energy and data queues of homogeneous devices,a two-dimensional Markov decision process(MDP)model is proposed and then derive the Hamilton-Jacobi-Bellman(HJB)equation is derived to solve the optimal access decision of each device considering the influence from other devices.In view of the situation that the detail information of each device in the large-scale wireless network cannot be obtained,the influence from other devices is used to converted to mean field,and then derive the Fokker-Planck-Kolmogorov(FPK)equation to describe the evolution of the system state distribution.The delay-optimal random access algorithm is obtained by iterative solution,which can reduce the delay and improve the efficiency of random access based on the consideration of the influence from other devices.For the large-scale heterogeneous devices random access problem,in this thesis,we consider that the heterogeneous devices adopt different access strategies in the same data and energy state,making it difficult to estimate the influence from other devices accurately during the random access process.Then,a delay-optimal random access strategy is proposed based on the information,i.e.,mean field term,exchange among neighbor devices.For the coupled energy and data queues,a two-dimensional MDP model is established and the Bellman equation is derived.To reduce the computational complexity of the solution,the system Q-factor is decomposed into the perdevice Q-factor,and the distributed random access algorithm is obtained by stochastic learning scheme.Considering the deviation of the estimation of the mean field caused by the heterogeneous devices,a stochastic approximation method is used to achieve the consensus of the estimation by exchanging the mean field with neighbor devices,and the convergence of the algorithm is proved.The simulation results show that the proposed algorithm has a good effect on reducing access conflict and delay.For the resource scheduling problem in large-scale LEO satellite network,considering the complex interactive problem between large-scale satellites and massive users,a resource scheduling strategy based on the Stackelberg mean field game is proposed.To formulate the interaction between the users and satellites,a Stackelberg game model is established at first,where the satellites are the leaders,determining the price for data service.The users are the followers,making the power scheduling decision to transmit the data according to its state information,the influence from other users and the fee paid to the satellites.Since the detailed information of users and satellites cannot be obtained,we adopt the mean field game algorithm to transform the influence from other users and satellites into mean field term and reformulate the optimization problem as a Stackelberg mean field game.Then,we can obtain the optimal resource power scheduling strategy for each user.The simulation results show that the power scheduling algorithm can effectively balance the data service resources of satellites and reduce the energy consumption of users.For the resource scheduling problem in large-scale multi-access edge computing(MEC),considering dual properties of the task,i.e.,computation amount and data size,and the multidimensional influence from other devices and MEC servers,a task scheduling scheme based on the generative adversarial network(GAN)structure is proposed.For the dual properties of the task,we first model the delay-optimal edge computing problem into an MDP and derive the HJB equation with the unknown task proportion distribution,where the multi-dimensional influence from other devices and servers need to be estimated.To obtain the multi-dimensional influence,the FPK equation is derived and we transform the multi-dimensional influence into the mean field term.Considering the non-calculability and insufficient data of the task proportion distribution,we propose a algorithm based on the GAN structure to solve the large-scale MEC problem.For the generator,it mainly trains the unknown proportion distribution.For the discriminator,it mainly trains the value function and obtains the optimal task scheduling policy according to the waterfilling algorithm.The simulation results show that the proposed algorithm decides the priority of task scheduling according to the characteristics of the task and reduces the delay.The research of this paper provides a certain theoretical support for making efficient distributed decision-making for massive devices in future large-scale networks such as the Internet of Things.At the same time,the research results can be extended to solve the decision-making problems in complex physical and social systems.