Research On Wireless Network Cell Outage Detection And Compensation Based On OpenAirInterface Platform

With the explosive growth of smart phones and differentiated traffic demands,enhancing network performance while reducing expenditure has put great pressure on mobile network operators in recent years.To reduce the pressure of network management and automate operation of wireless networks,self-organizing networks(SONs)are proposed.As an essential part of SONs,self-healing aims to detect and diagnose cell outages and make corresponding compensations.However,traditional approaches to cell outage detection often relies on historical network data with the cell outage state labeled,while compensation algorithms are usually centralized,which needs to acquire global network information.To overcome these issues,this thesis proposes an unsupervised cell outage detection method and a multi-agent deep reinforcement learning(DRL)based performance compensation method,respectively.Specifically,the main work and contributions are as follows.Firstly,facing the difficulties of obtaining the labeled data with cell outage states,an unsupervised cell outage detection method integrating Self Organizing Maps(SOM)and unsupervised clustering is designed and implemented.The method firstly extracts the characteristics of the key performance indicators(KPIs)of the cell by training SOM.Then,the neurons of the SOM model are clustered and the cell state corresponding to each cluster is identified by analyzing the statistical characteristics of the cluster.When new KPI data of a cell comes,the similarity of the input data is compared with the weight of each neuron,and the outage state is output based on the cluster to which the closest neuron belongs.By the verification on an OpenAirInterface(OAI)based platform,it can be observed that the proposal can achieve higher detection accuracy in a near real-time fashion.Secondly,to avoid the overhead led by global network information collection of centralized compensation approaches,a cell outage compensation method based on distributed DRL is proposed.Via the interaction with the network environments,each compensation base station independently uses a DRL model to adjust its antenna downtilt and user power according to local cell information and meanwhile updates the adjustment strategy according to the resulted performance of the cell.Compared with a compensation scheme based on genetic algorithm,it is found that the proposed distributed method can effectively restore the performance of the users in the outage cells and reach a near optimal system capacity.Thirdly,aiming at the problem that the outage detection method is difficult to verify performance in real networks,an outage detection demonstration platform is designed based on OAI open source platform.The platform consists of server-side software,base station-side software,and user-side software.The server-side software is responsible for continuously collecting KPI data of the cells and using the data to train the outage detection method.The software on the base station side and user side are used to receive server instructions and visually display the performance indicators of the cell and users.After training is finished,the script is used to make the cells intermittently in different outage states,so that the online detection accuracy and detection time of the outage detection method can be effectively evaluated.