Research On The Mean-field Theory Based Resource Allocation In Ultra-dense Networks

The relentless surge in traffic demands has been continuously putting demand on network capacity.To address such a challenge,the densification of wireless networks is currently viewed as a key enabling solution that meets the one thousand times of capacity argument promised by 5G.However,the network optimization in 5G denser network is facing with the large scale network problems which makes the conventional network management or resource allocation algorithm inadequate to tackle.To address the challenge due to the large amount of devices,a wide variety of solutions have been proposed including graph theory,sparse optimization and grouping/clustering approach,etc.But the heavy channel measure and feedback is still challenging for the large amount of devices.To address these challenges,a promising solution is mean-field(MF)theory.Based on MF theory,interactions among large amount of devices in ultra-dense network(UDN)can be transformed into interaction between a single device and all other devices,so the severe coupling UDN can be decoupled and the complexity of resource allocation can be greatly reduced.This thesis focuses on the application of MF theory in resource allocation in UDN,including two aspects as fellow:Firstly,this thesis proposes a mixed timescale joint power,subcarrier allocation and user equipment(UE)scheduling mechanism based on mean-field approximation(MFA)method and Lyapunov DPP(Drift-Plus-Penalty)method under OFDM transmission based ultra-dense small cell networks.Due to the severe coupling of all base stations(BSs)in interference,MFA method is adopted to decouple the system.The complexity of interference calculating method is reduced.When it comes to the time-correlation of system queue stability constraint,DPP method can transform the original optimization problem into problems each time slot and obtain a UE scheduling mechanism for optimizing network throughput while ensuring queue stability.Simulation results show that MFA method can ensure the rapid convergence of network's mean state.MFA method can obviously bring more throughput than adaptive transmission policy under ultra-dense scenario.Lyapunov DPP method can bring more queue stability than proportional fair UE scheduler.Secondly,this thesis proposes a mixed timescale joint caching and deleting mechanism based on mean-field game(MFG)method and Lyapunov DPP method under ultra-dense caching networks.Considering the complexity of finding the Nash equilibrium(NE)of stochastic differential game(SDG),a MFG based iteration method for solving NE is proposed which can solve BS's optimal caching policy with low complexity.Exploiting DPP method,a virtual queue is constructed for the system caching and deleting stability constraint.Deleting mechanism for minimizing network cost while ensuring caching and deleting stability based on virtual queue is derived.Simulation result shows that using MFG method can both acquire optimal caching strategy through a bit iteration times and save network cost obviously compared to baseline caching policy under ultra-dense scenario.Lyapunov DPP method can ensure network's caching-deleting stability while saving network cost.