AI identification and root cause location of 5G wireless problems help improve network quality

Author: Zhang Zhe and Chen Juanjuan, unit: Hebei Mobile

Labs Guide

Since the Ministry of Industry and Information Technology officially issued 5G commercial licenses, my country has entered the 5G era. Hebei Mobile has actively invested in network construction and promoted the development of 5G commercial networks. The large-scale investment in 5G base stations is accompanied by a series of wireless network quality and customer satisfaction issues. We are currently in the stage of stabilizing 4G and expanding 5G, so how to reduce the burden, efficiently and accurately solve problems has become a key challenge. Against the backdrop of increasing complexity of 5G networks and complex scenario-based pattern combinations, the requirements for optimization personnel's capabilities are gradually increasing. This requires the accumulation and iteration of experts' optimization experience and the gradual realization of tool intelligence. Hebei Mobile has established a VoNR evaluation system by governing wireless performance, alarms and MR data, and used AI prediction capabilities to accurately identify network problems. It has also built a rule engine module to determine the wireless root cause of problems, realizing the ability to identify wireless problems and intelligently analyze them, effectively improving the accuracy of 5G network problem identification and problem handling efficiency, helping to improve the quality of 5G networks and create a high-quality image and reputation for 5G networks.

1. Technical Solution Introduction

1.1 Data Governance

Data governance uses asset catalogs, model stratification, permission management, and model management to precipitate and summarize performance, alarm, and MR data on the O-domain platform, allowing for quick viewing and call-up when in use.

1.2 VoNR Evaluation System

After data governance, key VoNR indicators can be effectively identified, supporting the discovery of network problems in the early stage of VoNR commercialization, achieving end-to-end perception improvement of VoNR and rapid problem demarcation, and building a full-process end-to-end VoNR experience guarantee system based on key interface association. Automatic demarcation based on expert experience library + big data analysis, automatic analysis and positioning capabilities of wireless quality-poor cells based on rule engine & customizable configuration, and root cause recommendation are established to support VoNR commercialization guarantee.

A VoNR perception evaluation system is built around the dimensions of registration, access, retention, and voice quality. Key indicator trend insights, problem demarcation capabilities, and abnormal problem traceback capabilities are built based on VoNR characteristics to support network quality monitoring in VoNR commercial use and rapid discovery and demarcation of abnormal problems.

1.3 Identification of poor wireless quality issues

Wireless poor quality identification sets the TOPN threshold for the registration, access and retention indicators of the VoNR evaluation system, and automatically identifies poor quality cells using the multi-data source automatic association, multi-window linkage analysis, and deep analysis based on AI algorithms of intelligent KPI anomaly detection. It also supports manual custom configuration to identify poor quality cells, which is applicable to a variety of scenarios and makes problem discovery faster and more efficient.

1.4 Root cause diagnosis and analysis

1.4.1 Building rule engine capabilities

The delimitation rule engine provides wireless side rule management functions, and the wireless side problem delimitation and location rules can be configured flexibly and autonomously in the rule arrangement interface. The rule engine automatically integrates wireless MR, alarms, operation logs and performance indicators, can accurately locate the cause of the problem period, and supports custom arrangement rules, including creation, deletion, modification, etc., and can flexibly set rule parameter thresholds. The system provides 700+ preset rules based on expert experience.

1.4.2 Intelligent Location of Wireless Root Causes

The rule engine associates operation logs, alarms, neighboring cells, parameters, coverage, capacity, interference, transmission information, etc. to facilitate manual analysis. Based on the TOPN-identified poor wireless quality cells, the rule engine can be actively called, and the root causes can be sorted, and the key reasons for the poor wireless quality cells can be output to guide the rapid processing and closed loop of the problem.

2Technological innovations

2.1 KPI Anomaly Detection

Wireless KPIs have characteristics such as periodicity, volatility, and segmentation. It is difficult to solidify the monitoring threshold. If it is too low, a large number of false alarms may occur, and if it is too high, anomalies may not be detected. The static/rule thresholds used in traditional wireless network performance are difficult to fully adapt to the irregularity and diversity of performance characteristics. In addition, the network (equipment) structure in the IT and CT fields is becoming more and more complex, which leads to more and more monitoring indicators. The traditional method of setting static thresholds based on expert experience is becoming increasingly difficult to complete comprehensive monitoring of the network (equipment) and is difficult to adapt to the dynamic changes of indicators. AI-based KPI anomaly detection intelligently generates dynamic thresholds and realizes simultaneous monitoring of large-scale indicators based on the funnel principle.

Through KPI anomaly detection, the anomaly perception time can be shortened from 2 hours to 30 minutes, improving the efficiency of VoNR anomaly perception detection.

2.2 Alarm Correlation and Aggregation

Through KPI alarm aggregation and merging, VoNR problem clusters can be quickly identified, problem levels can be upgraded, and problems can be quickly identified and closed.

Principles of KPI alarm aggregation:

• Happening simultaneously in time.
• Proximity in the spatial dimension: Proximity in space must be considered, and aggregation can be achieved through important neighboring areas, co-stations, proximity, topology, etc.
• Business related: It is necessary to consider that the cause values ​​corresponding to the degradation of the main indicators are the same.

Principle of KPI alarm merging: Depth-first search DFS

• All cells and their neighboring cell relationships can be viewed as a graph structure, with cells as vertices and neighboring cell relationships as edges;
• Starting from the root node, traverse the nodes of the tree (graph) along the width of the tree (graph) until all nodes starting from vt are visited. If two alarms are detected, aggregation is performed.

Based on the above analysis results in the time dimension (occurring simultaneously), space dimension (adjacent/topological), and business dimension (business-related), related alarms are merged into the same event to improve the correlation accuracy.

2.3 Card Management

It supports users to flexibly and intuitively design and customize dashboard display layout, indicators and granularity, and set and save personalized dashboard templates for reuse.

3 Application Effects

Positioning accuracy: The automatic diagnosis of the root cause of the problem has an accuracy rate of over 80%, of which the SA connection rate is 86.52%.

Production efficiency: The processing efficiency of VoNR top cells with poor wireless quality has been increased from 13 per person-day to 20 per day, a 50% increase in efficiency. The analysis cycle of all top cells has been increased from weeks to days.

Improved workflow efficiency: The workflow efficiency has been significantly improved, with the end-to-end efficiency of problem handling increasing by approximately 35.83%.

The analysis workflow has changed. Compared with the traditional analysis process, data collection and VoNR top N identification processes are omitted, and root cause analysis and optimization suggestions are output more quickly.