Design And Implementation Of DCOP Based Management Method For Large-scale Small Cells

Massive small cells are very important in increasing cellular network frequency nowadays,and they are mostly applied in filling up the signal holes and expanding the network capacity.A great challenge is the automatic management of the power of largescale small cells.Traditional centralized methods are not suitable for massive small cells scenarios with high dynamics due to the long time of management routine,high difficulty of central optimization and high delay of data relay.Recently,decentralized power management algorithms solve some of the problems that the centralized method can not solve based on partial observation and limited communication,which makes the management of large-scale small cells possible.However,these algorithms have not solevd the connection problem between the local optimization and the global optimization,thus making them less effective than centralized methods.This paper designs an infrastructure based on X2 interface for dynamic distributed management of large-scall small cells.Through the information from measure report of UEs,constraints between basestations can be constructed and turned into the Distributed Constraint Optimization Problem(DCOP).A limited solving window algorithm is designed for solving the DCOP.This algorithm can find the optimal solution with infinite time and find the best solution within given time,providing a dynamic,scale-free distributed global optimization algorithm for the power management problem of largescale small cells.For optimizing different scale problems,the group based hierarchical algorithm restricts the dynamics within one group or one layer,making the tradeoff between different scenarios.Also,the algorithm is based on the global pseudo-tree and search for the global goal without biases,which allows us to tradeoff between dynamics,quality,and concurrency.To check the optimization effect on scenarios with dynamic non-uniform signal coverage,this paper procedurally generates sim cities that match certain distribution conditions and simulate the citizens' life cycle,and uses multi-agent system to simulate the signal change of basestations causing by the movement of citizens carrying user equipments.This paper's distributed optimization algorithm is deployed in such scenarios and several experiments are running to verify the effectiveness and scalability of the algorithm.