Open AccessStructure learning‐based unsupervised root cause diagnosis for radio access networks
Author(s) -
Li Qi,
Zhang Yong,
Hua Xiaolei,
Zhu Lin,
Lei Zhe
Publication year - 2022
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
eISSN - 1350-911X
pISSN - 0013-5194
DOI - 10.1049/ell2.12481
Subject(s) - computer science , directed acyclic graph , directed graph , radio access network , ranking (information retrieval) , root cause analysis , data mining , performance indicator , key (lock) , c ran , base station , graph , ran , artificial intelligence , machine learning , computer network , theoretical computer science , reliability engineering , algorithm , engineering , mobile station , computer security , management , economics
The next generation radio access networks (RAN) put forward higher requirements for the reliability of base stations. Telecom operators are facing a series of challenges such as network complexity and diversified user needs, which puts forward higher requirements for the performance of RAN. In order to provide users with high‐quality experience, causes of performance problems must be quickly identified. Previous work on unsupervised RAN performance diagnosis mainly used correlation information of the data. This work introduces causality into performance diagnosis of RAN. A score‐based directed acyclic graph (DAG) structure learning network is used to extract causal relations between key performance indicators (KPIs). In this study, an unsupervised root cause analysis (RCA) scheme, DAG‐RCA, is proposed to achieve rapid diagnosis of abnormal performance on the edge nodes of the RAN. DAG‐RCA decomposes the work into four phases: abnormal ranking, DAG generation, propagation graph construction and random walk diagnosis. Using the KPI data collected from China Mobile's RAN, DAG‐RCA achieves83.75 % $83.75\%$ top‐1 precision on the RCA task.