《GTI:2023年5G无线网络智能化技术需求白皮书(英文版)(70页).pdf》由会员分享,可在线阅读,更多相关《GTI:2023年5G无线网络智能化技术需求白皮书(英文版)(70页).pdf(70页珍藏版)》请在三个皮匠报告上搜索。
1、GTI 5G Intelligent Network Whitepaper V1.0 1 GTI 5G Intelligent Network White Paper GTI 5G Intelligent Network Whitepaper V1.0 2 GTI 5G Intelligent network White Paper Version:V 1.0 Deliverable Type Procedural Document Working Document Confidential Level Open to GTI Operator Members Open to GTI Part
2、ners Open to Public Program 5G eMBB Working Group Network Working Group Project Project 6:Network Intelligence Task Intelligent Network Level(INL)Intelligent Network Architecture(INA)Intelligent Network Elements(INE)Intelligent Network Management (INM)Source members China Mobile,Huawei,CICT,Nokia,ZT
3、E,Ericsson Support members(in alphabetical order)Last Edit Date Approval Date GTI 5G Intelligent Network Whitepaper V1.0 3 Confidentiality:This document may contain information that is confidential and access to this document is restricted to the persons listed in the Confidential Level.This documen
4、t may not be used,disclosed or reproduced,in whole or in part,without the prior written authorization of GTI,and those so authorized may only use this document for the purpose consistent with the authorization.GTI disclaims any liability for the accuracy or completeness or timeliness of the informat
5、ion contained in this document.The information contained in this document may be subject to change without prior notice.Document History Date Meeting#Version#Revision Contents Nov.23,2020 29th GTI Workshop V1.0 The first version of GTI 5G Intelligent Network Whitepaper.The standardization and indust
6、ry status,the practical use cases and corresponding intelligent network level,architecture,function requirements on network elements and network management of 5G intelligent network are presented.GTI 5G Intelligent Network Whitepaper V1.0 4 Table of ContentsTable of Contents GTI 5G Intelligent Netwo
7、rk White Paper.1 GTI 5G Intelligent network White Paper.2 Document History.3 Table of ContentsTable of Contents.4 1 Executive Summary.6 2 Reference.7 3 Abbreviations.9 4 Introduction.11 5 Standardization and Industry Status.13 5.1 Motivation and Overview.13 5.2 Activities in ITU.13 5.3 Activities in
8、 3GPP.14 5.4 Activities in ETSI.14 5.5 Activities in CCSA.15 5.6 Activities in Industry Parties.15 5.7 Summary.16 6 Use Cases.17 6.1 Introduction.17 6.2 Classification for Use Cases.17 6.2.1 Full Life Cycle Dimension.17 6.2.2 Functional Entity Dimension.17 6.3 Use Cases of Network and Service Mainte
9、nance.18 6.3.1 Energy Saving.18 6.3.2 Root Cause Analysis of Alarm.25 6.3.3 Root Cause Analysis of Cell Performance Issue.27 6.3.4 Subscriber Complaint Handling.28 6.4 Use Cases of Network and Service Optimization.30 6.4.1 NR Network UE Throughput Optimization.30 6.4.2 NR Network Coverage Optimizati
10、on.33 6.4.3 ML-Based MU-MIMO Scheduler.36 6.4.4 Link Adaptation.37 6.4.5 Load Balancing Based on the Virtual Grid Technology.40 6.4.6 QoE Optimization.42 6.4.7 Edge QoS.45 6.4.8 Transport Network Optimization.47 7 Intelligent Network Level.50 7.1 Introduction.50 7.1.1 Framework Approach for Classifi
11、cation of Intelligent Network Levels.50 7.2 INL Evaluation of Typical Use Cases.52 7.2.1 Energy Saving.52 7.2.2 NR Network Coverage Optimization.53 7.2.3 NR Network UE Throughput Optimization.54 GTI 5G Intelligent Network Whitepaper V1.0 5 7.2.4 Root Cause Analysis of Cell Performance Issue.56 7.2.5
12、 Root Cause Analysis of Alarm.57 7.3 Summary.58 8 Intelligent Network Architecture.59 8.1 Introduction.59 8.2 Framework Architecture of Intelligent Network.59 8.2.1 Use Cases Classification.60 8.3 Architecture of Typical Use Cases.61 8.3.1 QoE Optimization.61 8.3.2 ML-Based MU-MIMO Scheduler.62 8.3.
13、3 Root Cause Analysis of Alarm.62 8.3.4 Link Adaptation.63 8.3.5 Energy Saving.64 8.4 Summary.65 9 Intelligent Network Elements.66 9.1 Introduction.66 9.2 Function Requirements for Network Elements.66 9.2.1 Multi-Dimensional Data Collection and Reporting.66 9.2.2 Intelligent Data Modelling.66 9.2.3
14、Data Storage and AI Computing Capability.66 9.2.4 Intelligent Feedback and Closed-loop Control.67 9.2.5 AI Model Interactionn.67 10 Intelligent Network Management.68 10.1 Introduction.68 10.2 Function Requirements for Network Management.68 10.2.1 Data Collection and Reporting.68 10.2.2 Data Analysis
15、 and Modelling.68 10.2.3 Network Management and Control.69 11 Conclusion and Recommendation.70 GTI 5G Intelligent Network Whitepaper V1.0 6 1 Executive Summary This is the first version of GTI white paper provides an overview of 5G Intelligent Network.It covers standardization status,practical use c
16、ases,intelligent network level,intelligent network architecture,intelligent network elements and intelligent network management.After a brief review on intelligent network related activities in SDOs and industry parties,the whitepaper introduces a series of practical use cases of intelligent network
17、 for further study.The use cases can be categorized based on the dimension of full life cycle and main functional entities.Potential solutions and application with performance are introduced for each use case.Intelligent network levels are beneficial for the industry to have a clear view on how to i
18、mplement a fully intelligent network step by step.Here we summarize the framework approach for classification of intelligent network levels based on relevant standards.Evaluation on intelligent network levels of typical use cases are analyzed as well.Intelligent network architecture is another impor
19、tant topic,in order to understand the impact of AI/ML on mobile network architecture,the whitepaper analyzes the framework architecture of intelligent network and implementation architecture from the practical uses case perspective.Based on the analysis of use cases,levels and architecture of intell
20、igent network,some general function requirements for intelligent network elements and intelligent network management are highlighted at last,which are derived from the implementation workflow and close-loop of intelligent network.As it comes to 5G era,it is obvious that network has become an indispe
21、nsable part of our lives.And network intelligence is widely considered as an important enabler for the network evolution with purpose of achieving promoted service performance and operational efficiency.This white paper is expected to provide helpful reference for any industry participates who are i
22、nterested in and committed to promoting the development of network intelligence.GTI 5G Intelligent Network Whitepaper V1.0 7 2 Reference The following documents contain provisions which,through reference in this text,constitute provisions of the present document.1 GSMA,“AI in Network Use Case in Chi
23、na”,Oct 2019.2 ITU-T Y.3172:“Architectural framework for machine learning in future networks including IMT-2020”.3 ITU-T Y.3173:“Framework for evaluating intelligence levels of future networks including IMT-2020”.4 ITU-T Y.3174:“Framework for data handling to enable machine learning in future networ
24、ks including IMT-2020”.5 ITU-T Y.ML-IMT2020-RAFR,Architecture framework for AI-based network automation of resource adaptation and failure recovery for future networks including IMT-2020.6 3GPP TR 28.810:“Study on concept,requirements and solutions for levels of autonomous network”.*7+3GPP TS 28.100
25、:“Management and orchestration;Levels of autonomous network”.8 3GPP TR 28.812:“Telecommunication management;Study on scenarios for Intent driven management services for mobile networks”.9 3GPP TS 28.312:“Intent driven management services for mobile networks”.*10+3GPP TS 28.535:“Management and orches
26、tration;Management services for communication service assurance;Requirements”.*11+3GPP TS 28.536:“Management and orchestration;Management services for communication service assurance;Stage 2 and stage 3”.*12+3GPP TR 28.809:“Study on enhancement of Management Data Analytics(MDA)”.13 3GPP TS 28.313:“S
27、elf-Organizing Networks(SON)for 5G networks”.14 3GPP TR 28.861:“Study on the Self-Organizing Networks(SON)for 5G networks”.15 3GPP TR 37.816:“Study on RAN-centric data collection and utilization for LTE and NR”.*16+3GPP TS 38.314:“NR;Layer 2 measurements”.*17+3GPP TS 38.300:“NR;Overall description;S
28、tage-2”.*18+3GPP TS 37.320:“Minimization of Drive Tests(MDT);Overall description;Stage 2”.*19+3GPP TS 38.306:“NR;User Equipment(UE)radio access capabilities”.*20+3GPP TS 38.331:“NR;Radio Resource Control(RRC);Protocol specification”.*21+3GPP TR 23.791:“Study of enablers for Network Automation for 5G
29、”.*22+3GPP TS 23.288:“Architecture enhancements for 5G System to support network data analytics services”.23 3GPP TR 23.700-91:“Study on Enablers for Network Automation for 5G-phase 2”.24 ETSI GS ZSM 002:Zero-touch Network and Service Management(ZSM);Reference Architecture”.25 ETSI GR ZSM 009-3:Zero
30、-touch Network and Service Management(ZSM);Closed-loop automation;Advanced topics”.26 ETSI GS ZSM 009-1:Zero-touch Network and Service Management(ZSM);Closed-loop automation;Enabler”.27 ETSI GS ZSM 009-2:Zero-touch Network and Service Management(ZSM);Closed-loop GTI 5G Intelligent Network Whitepaper
31、 V1.0 8 automation;Solutions”.28 ETSI,ETSI GR ENI 007 V1.1.1(2019-11):“ENI;ENI Definition of Categories for AI Application to Networks”.29 CCSA,“Technical report of telecommunication network planning application based on artificial intelligence”.30 CCSA,“Study on grading method for intelligent capab
32、ility of mobile networks”.31 CCSA,“Technical specification for intelligent level of mobile network management and operation”.32 TMF:A whitepaper of autonomous networks:empowering digital transformation for the telecoms industry,https:/www.tmforum.org/wp-content/uploads/2019/05/22553-Autonomous-Netwo
33、rks-whitepaper.pdf.33 TMF:A whitepaper of autonomous networks:empowering digital transformation for smart societies and industries,https:/inform.tmforum.org/research-reports/autonomous-networks-empowering-digital-transformation-for-smart-societies-and-industries/34Qing Zhang,“AI based intelligent re
34、cognition of 5g base station energy saving scenarios”,Aug 2019.35 Jian Feng Lei,“Research on network traffic prediction based on Neural Network“,May 2008.36 Zhi Rong Zhang,“Research on energy saving technology of 5G base station based on AI“,Oct 2019.37 China Mobile,“5G intelligent network white pap
35、er”,Dec 2018.GTI 5G Intelligent Network Whitepaper V1.0 9 3 Abbreviations Abbreviation Explanation 3D Three-Dimensional 3GPP 3rd Generation Partnership Project AAU Active Antenna Unit AGV Automated Guided Vehicle AI Artificial intelligence API Application Programming Interface BTS Base Transceiver S
36、tation CA Carrier Aggregation CE Cell Edge CM Configuration Management CQI Channel Quality Indication DL DownLink DT Drive Test E2E End to End eMBB enhanced Mobile Broadband EMOS Electric Operation Maintenance System EMS Network Element Management System eNB E-UTRAN NodeB EOMS Electric Operation Mai
37、ntenance System E-RAB Evolved Radio Access Bearer FP Frequent Pattern GANA Generic Autonomic Networking Architecture GTI Global TD-LTE Initiative HO HandOver INE Intelligent Network Elements INL Intelligent Network Level INM Intelligent Network Management KNN K-Nearest Neighbor KPI Key Performance I
38、ndicator LB Load Balance LTE Long Term Evolution MDT Minimization of Drive Tests MEC Multi-Access Edge Computing MIMO Multiple-input multiple-output ML Machine Learning mMTC massive Machine-Type Communications MR Measurement Report MU-MIMO Multi-User Multiple-Input-Multiple-Output NE Network Element
39、 GTI 5G Intelligent Network Whitepaper V1.0 10 NMS Network Management System NR New Radio NWDAF Network Data Analytics Function O&M Operation and Maintenance OPEX Operating Expense PDU Protocol Data Unit PM Performance Management PRB Project Review Board QoE Quality of Experience QoS Quality of Serv
40、ice RAN Radio Access Network RAT Radio Access Technology RCA Root Cause Analysis RF Radio Frequency ROI Return On Investment RRC Radio Resource Control RRM Radio Resource Manage RRU Radio Remote Unit RSRP Reference Signal Receiving Power SCTP Stream Control Transmission Protocol SDO Standards Develo
41、pment Organizations SON Self-Organizing Network TTI Transmission Time Interval UE User Equipment UPF User Plane Function URLLC Ultra-Reliable Low-Latency Communication VNF Virtual Network Function VR Virtual Reality GTI 5G Intelligent Network Whitepaper V1.0 11 4 Introduction This whitepaper mainly
42、focuses on the 5G Intelligent Network.Combined with standardization status,the whitepaper analyses a serious of practical use cases,intelligent network levels,intelligent network architectures of the representative use cases,and general function requirements on intelligent network elements and intel
43、ligent network management are presented as well.This Whitepaper is expected to help people make a comprehensive understanding of intelligent network and its industry.It is also expected to be helpful on promoting the development of network intelligence.Sincere thanks to all the contributors and the
44、supporters for their hard work in writing this whitepaper.Respectfully,the task leaders and contributors of each chapter are listed as following.Chapter 1 Executive Summary China Mobile Chapter 2 Reference China Mobile,CICT,Ericsson,Huawei,Nokia,ZTE Chapter 3 Abbreviations China Mobile,CICT,Ericsson
45、,Huawei,Nokia,ZTE Chapter 4 Introduction China Mobile,CICT,Ericsson,Huawei,Nokia,ZTE Chapter 5 Standardization and Industry Status China Mobile,Huawei Chapter 6 Use cases CICT,Huawei,Nokia,ZTE,Ericsson,China Mobile Chapter 7 Intelligent Network Level Huawei,CICT,Ericsson,China Mobile Chapter 8 Intel
46、ligent Network Architecture Nokia,Ericsson,Huawei,CICT,ZTE,China Mobile Chapter 9 Intelligent Network Elements ZTE,Nokia,Ericsson,CICT,China Mobile Chapter 10 Intelligent Network Management Huawei,CICT,Ericsson,China Mobile Special thanks to the following contributors for writing the whitepaper.Chin
47、a Mobile Dr.Xi Cao,Ms.Xiaojia Song,Ms.Yanping Liang,Dr.Lin Zhu,Ms Ruoyi Liu,Mr.Peng Zhao,Mr.Shaofan Chen,Ms.Xing Wang,Mr.Li Yu,Mr.Xiangyang Yuan,Dr.Junlan Feng CICT Ms.Xiufeng Wang,Ms.Xuan Zhang Ericsson Ms.Xiaomei Liu,Mr.Bo Guan,Mr.Liyu Liu Huawei GTI 5G Intelligent Network Whitepaper V1.0 12 Ms.Da
48、nni Chen,Mr.Ruiyue Xu,Mr.Junjie Zhang,Mr.Yuanxian Tang,Mr.Chunhong Yuan,Ms.Lan Zou,Mr.Li Lin Nokia Mr.Zhuyan Zhao,Ms.Fangfang Gu ZTE Mr.Xinjian Jiang,Mr.Pan Li,Mr.Wanteng Zhai,Ms.Hong Feng,Mr.Fei Ye,Mr.Honghui Kang,Mr.Xuefeng Lu This is the first version of the whitepaper,it will be continuously upd
49、ated according to the research and development progress.GTI 5G Intelligent Network Whitepaper V1.0 13 5 Standardization and Industry Status 5.1 Motivation and Overview As mobile communication network evolves to 5G,it is obvious that the network has become an indispensable part of our lives.While bei
50、ng the enabler,mobile network itself is evolving into the intelligence era with multiple application scenarios,features,services and operation requirements for the intelligent network.Technologies including artificial intelligence(AI)are expected to be introduced to enable autonomous networks in the
51、 areas of network planning,deployment,operation,optimization,service deployment,assurance,etc.Most of the standards development organizations(SDO),e.g.ITU-T,3GPP,ETSI,CCSA,are taking actions to study and develop standards for autonomous networks,topics related to autonomous networks are discussed in
52、 different working groups.Industry bodies such as GSMA,TM Forum,Global TD-LTE Initiative(GTI)etc.are working to promote autonomous networks.GSMA proposed that the automatic network operation capability will become the indispensable 4th dimension of the 5G era together with eMBB,mMTC,and URLLC,and be
53、come one of the most important driving forces for 5G service innovation and development.1 Figure 5-1 Network automation is the 4th dimension of 5G networks 1 Main activities about autonomous networks in the SDOs and industry parties are introduced briefly in the following.5.2 Activities in ITU In IT
54、U-T,Study Group 13(SG13)has led the ITU standardization work on next generation networks and now caters to the evolution of NGNs,while focusing on future networks and network aspects of mobile telecommunications.A focus group on Machine Learning for Future Networks including 5G(FG-ML5G)had been set
55、up for machine learning for future networks,which includes interfaces,network architectures,protocols,algorithms and data formats.The topic related to autonomous networks came into study since 2017 and recommendations ITU-T Y.3172(Architectural framework for machine learning in future networks inclu
56、ding IMT-2020)2,GTI 5G Intelligent Network Whitepaper V1.0 14 ITU-T Y.3173(Framework for evaluating intelligence levels of future networks including IMT-2020)3,ITU-T Y.3174(Framework for data handling to enable machine learning in future networks including IMT-2020)4,etc.have already published.Recom
57、mendations about AI-based network autonomous,e.g.ITU-T Y.ML-IMT2020-RAFR 5 are in draft stage.5.3 Activities in 3GPP Autonomous network comes into the sight of 3GPP since 4G era,the topics mainly focused on Self-Organizing Network(SON),Minimization of Drive Tests(MDT),etc.In 5G era,3GPP makes more e
58、fforts on the standardization for autonomous networks.3GPP SA WG5 started the study item Study on autonomous network levels 6 from August 2019 and published in September 2020,which output the concept,dimension,framework and typical use cases for classification of autonomous network level.In June 202
59、0,a new work item Autonomous network levels 7 was approved and started.The main objective of this work item is to deliver standard specifications for defining concept and framework for autonomous network level,and providing requirements for corresponding autonomous network enabler related standard f
60、eatures for different autonomous network levels.Since Release 16,3GPP SA WG5 has started a series of standard projects to continuously promote the autonomous network,which cover the entire mobile network lifecycle(including planning,deployment,maintenance,and optimization phases).Examples of these p
61、rojects are Intent driven management service for mobile network 8 9,Closed loop SLS Assurance 10 11,Study on enhancement of Management Data Analytics Service 12,and Self-Organizing Networks(SON)for 5G networks 13 14.3GPP SA WG2 and 3GPP RAN WG3 have also started some works on topics related to netwo
62、rk automation and intelligence.Since Release 16,3GPP RAN WG3 started the RAN-centric Data Collection and Utilization research topic 15 and SON/MDT support for NR standard project 16-20,and researched and defined wireless data collection and application oriented to network automation and intelligence
63、.Since Release 16,3GPP SA WG2 has started the Enablers for Network Automation for 5G standard project 21-23.The objective is to define the introduction of Network Data Analytics Function(NWDAF)on the 5GC network layer to implement control-plane data analysis for 5GC.5.4 Activities in ETSI The study
64、on autonomous network is active in ETSI,there are several groups working on relevant topics of autonomous networks:ENI,NFV,OSM,MEC,F5G,TC INT AFI and ZSM.Generic Autonomic Networking Architecture(GANA)is studying TC INT AFI.ETSI ZSM targeted to deliver E2E solution for Zero-touch network and service
65、 management and orchestration,and ZSM framework is designed for closed-loop automation and optimized for data-driven machine learning and AI algorithms.ZSM published Zero-touch GTI 5G Intelligent Network Whitepaper V1.0 15 network and service management reference architecture specification 24 and st
66、arted discussions on three closed loop topics related projects:closed-loop automation:Advanced topics 25,closed-loop automation:Enablers 26,and closed-loop automation:Solutions for automation of E2E service and network management use cases 27 in June 2019.The goal is to deliver the use case,requirem
67、ents and solutions for the E2E closed loop automation including automation-related policies and intent interfaces to implement interaction between closed loops in E2E management domains and management domains.In November 2019,ETSI published the report ETSI GR ENI 00728:ENI definition of categories f
68、or AI application to networks which defines various categories for the level of application of AI techniques to the management of the network,going from basic limited aspects,to the full use of AI techniques for performing network management.5.5 Activities in CCSA As one of the most influential SDOs
69、 in the field of communication in China,CCSA began the standard works on autonomous networks from 2010s,and the items are mainly set up in TC1,TC5 and TC7,including use cases,architecture,data handling,levels of autonomous network,management requirements etc.CCSA TC1 started the research project 29
70、on telecommunication network planning application based on artificial intelligence from July 2018.The research is focusing on network planning based on AI.The output report will mainly cover network planning needs,network architecture,data training,network evolution.CCSA TC5 started the research pro
71、ject 30 on intelligence levels of mobile networks in November 2018 and finished in August 2019.The output of the research covers the methods for evaluating intelligence levels for mobile networks,and typical use cases for classification of intelligence level and potential relationships between the m
72、obile network architecture and intelligence levels.CCSA TC7 started the standard specification Technical requirements for intelligence levels of mobile network management and operation31 in January 2020.The work target to define the concept,measurable evaluation method,typical use case and requireme
73、nts for intelligence levels of mobile network management and operation.5.6 Activities in Industry Parties Industry bodies such as GSMA,TM Forum are actively taking actions to explore and promote the collaboration of autonomous network topics among the SDOs,operators,vendors and any other industry pa
74、rticipants.In GSMA,AI&Autonomous is one the topic of the“Future Network”.In June 2019,the first GSMA Global AI Challenge was held and the challenge investigated three specific areas:connectivity in rural areas,mobile energy efficiency and enhanced services in urban areas.In June,at the AI in Network
75、 Seminar for Mobile World Congress Shanghai 2019,it is called on the GTI 5G Intelligent Network Whitepaper V1.0 16 entire industry to focus on and contribute to the key applications of AI in the mobile network,and jointly build the 5G era for the intelligent autonomous network in the workshop.In Oct
76、ober 2019,GSMA has published“AI in network use cases in China”1.In TM Forum,several workshops of autonomous have been held since 2019 and Autonomous Networks Project(ANP)was established in August 2019.Since 2019,there are three whitepapers are published officially:AN Whitepaper 1.0,IG1193 Vision&Roa
77、dmap v1.0,IG1218 Business requirement&architecture v1.0.32 This year,AN Whitepaper 2.0 33,business requirements&architecture v1.1,technical architecture,demo of Catalyst projects,user stories/use cased,etc.are now on going.5.7 Summary As the development of technologies and the evolution of networks,
78、intelligent network will be an important enabler for the future networks.In order to promote the industry,SDOs and industry parties are taking activities to form a unified understanding and continuously clarify the concept.GTI 5G Intelligent Network Whitepaper V1.0 17 6 Use Cases 6.1 Introduction Op
79、erators,vendors and third-parties have already begun to explore intelligent networks,and a number of good practices and use cases have emerged.This whitepaper categorizes the intelligent network use cases based on the dimension of full life cycle work and service planning,network and service deploym
80、ent,network and service maintenance,network and service optimization,and the dimension of main functional entities i.e.intelligent network element and intelligent network management.And for each use case,background,solution overview,application and performance are presented.6.2 Classification for Us
81、e Cases 6.2.1 Full Life Cycle Dimension Network and Service Planning:processes of designing and delivering new or enhanced network or service based on the business,market,product and customer service requirements.Network and Service Deployment:processes of allocation,installation,configuration,activ
82、ation and verification of specific network and service.Network and Service Maintenance:processes of monitoring,analyzing and healing of the network and service issue.Network and Service Optimization:processes of monitoring,analyzing and optimization/assurance of the network and service performance.6
83、.2.2 Functional Entity Dimension Intelligent Network Element:the close loop of intelligent function is mainly deployed within and completed by the entity of network element(with network management systems control).Intelligent Network Management:the close loop of intelligent function is mainly deploy
84、ed within and completed by the entity of network management system(with network elements assistance).GTI 5G Intelligent Network Whitepaper V1.0 18 Figure 6-1 Classification of use cases Considering the current development status of the industry,this version of white paper,mainly focuses on the use c
85、ases of network and service optimization and maintenance.Use cases of network service planning and deployment would be introduced in the next version.6.3 Use Cases of Network and Service Maintenance 6.3.1 Energy Saving 6.3.1.1 Background As operators network energy consumption keeps increasing,reduc
86、ing the energy consumption of main equipment is key to energy saving.Reducing the power consumption of main equipment of wireless sites has become the top priority for all.For a typical carrier,the power consumption of wireless sites accounts for about 45%,and the power consumption of wireless base
87、stations as main equipment accounts for 50%.In the power consumption of a wireless base station,the power consumption of radio remote units(RRUs)accounts for a large proportion,and that of the power amplifiers in the RRUs also accounts for a large proportion.In actual networks,traffic has obvious ti
88、dal effect in most cases.When the traffic is light,the base station is still running,which causes a great waste of energy.GTI 5G Intelligent Network Whitepaper V1.0 19 Figure 6-2 Challenges facing traditional energy saving Reducing unnecessary power consumption is a key measure of energy saving but
89、is faced with many challenges.The network traffic volume varies greatly during peak and off-peak hours.The equipment keeps running,and the power consumption is not dynamically adjusted based on the traffic volume.As a result,a waste of resources is caused.The capability of zero bits,zero watts needs
90、 to be constructed.However,in a typical network,the features of different scenarios vary greatly.How to automatically identify different scenarios and formulate appropriate energy saving policies becomes the key to energy saving.Business district:high requirements on user experience,obvious tidal ef
91、fect,and light traffic at night.Residential area:high requirements on capacity,heavy traffic in a whole day,and no obvious traffic fluctuation.Suburban area:low requirements on capacity,light traffic,sparse sites,and long site coverage distance.To meet the need of environmental-friendly development
92、featured in low-carbon,energy saving and emission reduction and the requirement of cost reduction from telecom operators,the contradiction between the increasing communication data service volume and high energy consumption needs to be resolved while ensuring the development of 5G services.Therefore
93、,energy saving technologies have always been a hot topic in the industry,and equipment energy saving has always been an important direction of research.As 5G technologies become mature,it is urgent to accelerate the commercial application of energy-saving solutions for 5G equipment.6.3.1.2 Solution
94、Overview Based on network-level AI-based intelligent energy saving policy management and site energy saving scheduling control,the mobile network energy saving solution implements network scene adaption,one site one policy,and multi-network collaboration for intelligent base station energy saving ma
95、nagement.This maximizes network energy saving benefits while ensuring stable network performance,and achieves the optimal balance between energy consumption and KPIs.The overall solution is as follows:The system obtains data on the live network,including engineering parameters,MRs,and weather data.B
96、ased on big data analysis,the system uses AI technologies to automatically identify network energy saving scenarios,predict network traffic trends,such as traffic busy/idle hours and areas and traffic/energy consumption trends,and identify multi-cell co-coverage,and automatically generates energy sa
97、ving policies.The system automatically delivers energy saving policies and implements network-level AI-based intelligent energy saving policy management and coordinated management and control of site energy saving scheduling.-The network-level AI-based energy saving algorithm is used to implement au
98、tomatic precise energy saving feature enabling and energy saving parameter optimization with GTI 5G Intelligent Network Whitepaper V1.0 20 lossless performance based on different network scenarios/models,base station configurations,and networking modes(multi-frequency networking and 2G/3G/4G/5G mult
99、i-RAT networking).In addition,the solution implements one site one policy and multi-site collaboration to quickly and efficiently start network-wide energy saving.-Precise energy saving scheduling control(such as carrier shutdown and power adjustment)is implemented for sites under the control of net
100、work AI.Real-time monitoring of impact on network KPIs and energy saving benefits is implemented to achieve manual visualization and management of energy saving benefits on mobile networks.6.3.1.2.1 Energy Saving Scenario Identification For traditional energy saving,due to the diversity of scenes ac
101、ross the network and the large differences in the characteristics of the scenarios,manual operation cannot effectively identify the different scenarios and can only use the same set of energy saving policies for different scenarios.AI-based scenario identification can intelligently identify and labe
102、l the scenario of each cell based on network coverage,users,resources,and other characteristic data.Figure 6-3 Scenario identification for energy saving Scenario identification includes the following two levels:The first level can identify the coverage scenario of a cell,for example,whether it is a
103、high-speed railway,common urban area,rural area,subway,large stadiums,colleges,shopping malls,or office buildings.After scenario identification,for different scenarios can be combined with the characteristics of various energy-saving technologies to preset the recommended energy-saving solutions by
104、scenario,as well as adapting to find the best recommended energy-saving solutions.The specific energy saving technologies include carrier shutdown,channel shutdown,symbol shutdown,cell shutdown and so on.The second level is based on network topology data(for example,engineering parameters and config
105、uration data of the cell),measurement reports,and handover indicators,the system can identify whether a cell is the coverage/capacity-layer cell,overlapping coverage degree of a cell pair,and whether a cell has the same-coverage cell.The result can be used as the input of intra-RAT/inter-RAT collabo
106、rative energy saving.The first step to implement power saving policy for a specific cell is to intelligently identify the scenario of the cell.During scenario identification,each base station can identify the feature data GTI 5G Intelligent Network Whitepaper V1.0 21 such as topology information,upl
107、ink/downlink measurement report,service characteristics,user level information and resource occupation distribution of each cell.The AI algorithm can use such classical machine learning algorithms as K-means clustering algorithm,KNN algorithm,decision tree and logic regression for scenario predictio
108、n and classification.Scenario identification is supported on both the EMS and the NE.When it is implemented on the NE side,more user-level and service-level data is provided to improve the accuracy of identification,but the NE side has the AI data storage and computing capabilities.6.3.1.2.2 Traffic
109、 Prediction Based on historical network data,for example,time,cell traffic statistics,neighbor cell relations,handover data,holidays,and major events,AI modeling is performed at the cell,cell cluster or region level to predict the load flow direction and load level of a cell or cell cluster in the n
110、ext few hours.Based on the load prediction result,it is used to accurately understand the occurrence time and duration of low load.In this way,intelligent energy saving can be performed on cell.In addition,based on historical handover and load migration information,the prediction model can predict t
111、he neighbor cells to which the load of a cell is transferred when the cell enters energy saving state,and then make adaptive settings for the energy saving start and end time of the neighbor cells and the load threshold.34 Load prediction can be supported on both the EMS and the NE sides.However,if
112、it is implemented on the NE side,load prediction can be implemented based on whether each cell is in energy-saving status.In this way,the accuracy of neighbor cell load prediction can be improved.Figure 6-4 Traffic prediction for energy saving The AI approach can accurately predict the effective tim
113、e period for energy saving applications,thereby reducing the impact on performance KPIs caused by irrational energy saving time period configurations in manual configurations.6.3.1.2.3 Intra-RAT/Inter-RAT Collaborative Energy Saving The objective of collaborative energy saving is to select compensab
114、le cells(or groups of cells)simultaneously when selecting energy-saving cells so as to form a group of collaborative cells.Specifically,the gNodeB selects a coordination cell group in accordance with the same-coverage cell,overlapped coverage degree between cells and neighbor cells,coverage scenario
115、,and the real-time load prediction result of each cell.After appropriate energy saving methods are enabled,the handover parameters of the energy-saving cell and neighbor cells can be GTI 5G Intelligent Network Whitepaper V1.0 22 intelligently adjusted to predict the load guarantee capability of the
116、neighbor cells when a cell is in energy saving status.35 Figure 6-5 Intelligent carrier shutdown for energy saving For intra-RAT coordination,the load prediction results of overlapping coverage and candidate supplementary cells are used to determine whether the target cell can be used as a compensat
117、ion cell.For inter-RAT coordination,the support of candidate cells for the QoS level of the energy-saving cell is added.The following describes how to select a compensation cell.Channel shutdown means that when the cell load is low,the shutdown of some transmission channels of the local cell which m
118、ay affect edge coverage and the download rate.For a cell with channel shutdown is enabled,it is necessary to intelligently evaluate whether the coverage of edge users and downlink services of the cell can be serviced by adjacent cells within the radio mode when the cell is in channel shutdown status
119、.Carrier shutdown is applicable to the scenarios where multiple layers of networks are covered,and the traffic tidal effect is obvious,such as high-speed railways,universities,subways,and large stadiums.Basic coverage can be guaranteed by the coverage layer cells during idle times,and the capacity l
120、ayer cells carriers are turned off for energy saving.Therefore,carrier shutdown must ensure that there are two or more carriers covered in the same sector.This action cannot be performed if there is only one carrier covered in a sector,because once it is performed,the entire sector will be without s
121、ignal.Therefore,before carrier shutdown,it is necessary to intelligently analyze the same-coverage situation.Note:To determine energy saving,you need to select energy saving objects and load migration objects on the NE side in real time.Therefore,it is recommended that you select energy saving objec
122、ts on the NE side.6.3.1.2.4 Optimization of Energy Saving Policy Parameters At present,the trigger thresholds for various types of energy saving in the energy saving policy(downlink PRB utilization,number of RRC users,etc.)are mainly set based on manual experience,and the thresholds are not optimize
123、d according to the actual energy saving effect.For energy saving,if the shutdown policy is loose(for example,if the downlink PRB utilization threshold,which triggers entry into the energy-saving state,is set high,it is easier for the cell to enter the energy-saving state,which means that the policy
124、is loose),the energy saving effect is better.However,if the policy is too loose,the cell can easily enter energy saving state,which may reduce the service quality and traffic requirements of users.Otherwise,it is difficult for the cell to enter the energy-saving status,and the energy-saving effect i
125、s poor.Energy saving policy parameter optimization aims at the threshold for triggering energy savings GTI 5G Intelligent Network Whitepaper V1.0 23 (for example,DL PRB usage).The purpose is to iteratively evaluate the energy saving effect and KPI performance of a cell/cell group under various trigg
126、ering thresholds,and obtain the energy saving triggering threshold with the best KPI and energy saving effect.The recommended energy saving policy threshold for a cell/cell group is obtained.That is,the inflection point between the energy saving policy/load threshold and performance/energy saving ef
127、fect is obtained to maximize the energy saving effect.The comprehensive score of KPI and energy saving effect in the iteration process can be defined in the form of target function,for example,*(1-call drop rate)+*access success rate+*normalized average throughput+*normalized energy consumption,wher
128、e,are the weights of each item.Specific definitions are recommended based on the operators attention to the performance index and the impact of the energy saving action on the network.The system can automatically optimize energy-saving policies and load thresholds based on traffic prediction and rei
129、nforcement learning,and implement online iterative optimization to optimize the shutdown duration without affecting KPIs.Figure 6-6 Energy-saving policy optimization During prediction modeling,the KPIs of key network indicators need to be monitored,and the current prediction models need to be fed ba
130、ck in accordance with the changes of KPIs to achieve the most advantages of energy saving and system performance.Note:The parameters of an energy saving policy can be optimized through rough adjustment of outer-loop parameters on the EMS side,and fine adjustment of inner-loop parameters and individu
131、al parameters can be performed on the NE side to assist in parameter optimization.6.3.1.2.5 Enhanced Symbol Shutdown Based on Intelligent Scheduling of Wireless Resources By intelligently and dynamically adjusting cell-level uplink and downlink resources,the system optimizes the service symbol resou
132、rces allocated to users under the condition that service delay and service level are met.With light network load,the system maximizes the ratio of symbols in energy saving status and improves energy saving efficiency.The AI algorithm predicts service distribution and load in future based on historic
133、al user load/service analysis,current user service type analysis,and service arrival analysis.As the input of symbol resource scheduling,the AI algorithm optimizes symbol resource scheduling without affecting user experience and enables idle symbols to enter energy-saving status in a timely manner.N
134、ote:This module can only be supported on the NE side.GTI 5G Intelligent Network Whitepaper V1.0 24 6.3.1.3 Application and Performance In typical network configurations,the power consumption of base stations can be reduced by 10%15%,and the emission of about 2 million kg carbon dioxide can be avoide
135、d for every 1000 base stations in one year.Operator in China applies AI technologies and automation capabilities to base station energy saving.The RAN element management system(EMS)can automatically identify different scenarios and optimize energy saving policies for different networking modes and l
136、oads,maximizing network energy saving benefits while ensuring KPIs.Energy saving solution is applied more than 11,000 cells in the entire province.The overall energy consumption is reduced by 13.59%.The average shutdown duration is 9.88 hours,which increases by 57%compared with that when the feature
137、 is manually enabled.The tidal effect is obvious in office buildings,business centers,large stadiums,suburban areas,and county-level areas.The average energy consumption is reduced by 16.88%.On the basis of deploying energy saving functions at multiple layers such as base station software and hardwa
138、re,and terminals,AI technologies are used to intelligently deploy scenario-based and cell-level refined energy saving policies,minimizing network energy consumption while ensuring stable KPIs.The AI-based intelligent energy-saving technology collects historical and spatial feature data of each cell
139、on the network to analyze the change rule of radio resource utilization,automatically identifies the coverage characteristics of cells and fully considers the network coverage,UE distribution,and scenario characteristics based on the prediction and evaluation results of coverage scenarios and traffi
140、c variables.In this way,the energy-saving policy can be self-adaptive or selected based on the operators policy.The following figure shows the online energy saving solution for wireless networks based on the layer-and domain-based principle.Figure 6-7 Architecture of the online energy-saving solutio
141、n for wireless networks The online energy-saving solution consists of four sub-functions:Energy-Saving scenario identification GTI 5G Intelligent Network Whitepaper V1.0 25 Traffic prediction Intra-RAT/Inter-RAT coordinated energy saving Optimization of energy-saving policy parameters The AI-based w
142、ireless network solution can implement energy-saving policies such as cell sleeping and carrier shutdown through inter-RAT and intra-RAT coordinated management based on cell scenarios and energy-saving time,cutting energy consumption by 15%.Figure 6-8 Simulation result of online energy-saving soluti
143、on 6.3.2 Root Cause Analysis of Alarm 6.3.2.1 Background With the rapid development of communication network in recent years,its scale has been quite large.In the network,there will be alarm information every day,and the amount of these information data is huge,and there are many sudden failures.Whe
144、n the network equipment fails and causes alarm,the equipment associated with it will also cause corresponding faults,and generate a large number of alarm information in a short time.As a fault often causes multiple alarm events,the equipment and business process related to the fault will send out re
145、levant alarm information.At the same time,the alarm information caused by multiple faults will be superimposed together,which will submerge the real alarm information,which makes fault identification very difficult.The rapid recovery of network fault is the basis to ensure the stable operation of th
146、e network.The traditional fault handling method is completed by manual analysis through a combination of network alarm,operation status and log data for manual analysis,and rely on reliable expert experience to achieve fault analysis and recovery.Analysis efficiency and screening effect were low in
147、time dimension and regional dimension.In 5G the network,combined with big data analysis and machine learning algorithm,the fault analysis experience can be informationized and modeled.Through multi-dimensional analysis of alarm information,network performance,operation log,etc.,the association model
148、 that is difficult to be found manually can be mined out to form a precise root cause analysis system,which helps to improve the efficiency and success rate of fault analysis and recovery in the whole network.At the same time,by accumulating and sharing a large number of case data in the system,the
149、fault prediction based on network operation and maintenance can be realized,and GTI 5G Intelligent Network Whitepaper V1.0 26 timely treatment and prevention can be obtained before the fault occurs,so as to improve the stability of network operation.6.3.2.2 Solution Overview Alarm root cause analysi
150、s is to use machine learning algorithm to train alarm association rules from massive alarm information,and combine with expert rule bases to form alarm diagnosis model bases.The existing network alarm information is diagnosed by matching rules,and the root cause alarm and derivative alarm are analyz
151、ed.The root cause alarm is accurately dispatched to improve the efficiency and success rate of analysis and recovery.36 Figure 6-9 Solution chart Alarm root cause analysis scheme is divided into two stages:alarm rule mining stage,real-time alarm analysis and processing stage.The purpose of alarm rul
152、e mining stage is to analyze the big data based on historical alarm data,and obtain the relationship between alarms(for example,it can be based on Apriori and FP 37 In this stage,offline processing can be used to analyze and mine historical data,and it is not required to be real-time.In the alarm an
153、alysis and processing stage,the purpose is to analyze and process the real-time alarms in the network based on the association rules in the rule database,and identify the source alarms and derived alarms.In this stage,online processing is used to process real-time alarm,which requires real-time perf
154、ormance.6.3.2.3 Application and Performance The current alarm data is monitored in real time in the existing network.When a new alarm is received,it is matched with the alarm association rule base to analyze the alarm root cause and derived alarm.Then,according to the root cause and the derived alar
155、m relationship,the high-efficiency alarm management,such as alarm elimination,alarm merging and associated alarm dispatch,are implemented.It is helpful to improve the efficiency and success rate of fault analysis and recovery in the whole network.It is expected that the efficiency of alarm troublesh
156、ooting can be improved 80%.GTI 5G Intelligent Network Whitepaper V1.0 27 6.3.3 Root Cause Analysis of Cell Performance Issue 6.3.3.1 Background For the increasingly large and complex multi-layer and multi-standard wireless communication network,whether it is network planning,maintenance,or network o
157、ptimization,operators and equipment vendors are facing new opportunities and challenges.To deal with these new challenges,Artificial intelligence is one of our most important solutions for managing modern networks.For increasingly dense multi-layer networks,the traditional way is mainly for network
158、optimization background engineer to check a single KPI through a variety of network optimization tools,and then combine one or more KPIs with the experience of senior network optimization engineers to analyze them,and pre-defined check rules or threshold values,first find out the problem cell,and th
159、en analyze the problem cell with more relevant KPIs and MR information to find out the cause of the problem,and then give the corresponding solution optimization plan for optimized implementation and verification.This traditional way of dealing with the problem of dense and massive communities in la
160、rge modern cities requires more and more network optimization engineers to deal with it.The method of using machine learning algorithms to autonomously discover problems in massive data and quickly identify and classify problems can significantly improve the efficiency of network optimization work.T
161、he transformation and skill upgrade of network optimization personnel and the application of AI modules make fewer network optimizations.The network engineers can handle more and more complex network problems.6.3.3.2 Solution Overview Combined with the actual network cells performance data(KPI),and
162、make full use of the experience and skills of AI experts to complete the exploration and selection algorithms by machine learning on big data platform,take full advantage of the distributed computing power of AI big data platform,complete Iterative development and model training and verification of
163、application core modules.And quickly integrate and docker with the input data of the existing network,and through the deployment of the containerized platform,it can be directly applied to the rapid classification and root cause analysis of the daily network optimization problem cells.Through the se
164、lection of about 100 KPIs from a large number of KPIs,and through multiple iterations of rigorously trained machine learning algorithms,the automatic aggregation and classification of 12 major types of network problems is quickly realized,and the intelligent root cause analysis module is further use
165、d to provide various problems Network root cause.Significantly reduce the workload of on-site network optimization engineers,improve the efficiency of on-site optimization,and quickly maintain the optimal state of the entire network cell performance.GTI 5G Intelligent Network Whitepaper V1.0 28 Figu
166、re 6-10 Root cause analysis system flow chart The intelligent root cause analysis(RCA)solution for automatic network problem cells,on the one hand,realizes the identification of intelligent problem cells in multi-mode networks in one or more cities and gives the root causes of the main problems,and
167、realizes automation through on-site deployment through containerized solutions;On the other hand,it also significantly simplifies the investigation and analysis of daily network optimization engineers problem communities.6.3.3.3 Application and Performance Through the application and on-site verific
168、ation in typical cities(10000+cells),the AI module can quickly,accurately and intelligently identify problem cells and give the root causes of problems in the cell.3 minutes to process more than 10k+cells For the same workload,it takes two weeks of labor The provincial company arranges 4 experts fro
169、m different vendors to verify,and the accuracy rate 81%Network automation,network intelligent simplification,artificial intelligence,machine learning and applications in 5G and other emerging fields will help improve user perception and satisfaction,accumulate experience for intelligent optimization
170、 and exploration in 5G and other emerging fields,and finally form artificial intelligence results in the network Optimize the scale application in the network service.Improve the overall strategy and service quality of China Mobiles network services through artificial intelligence application resear
171、ch,development and application of results in the field of network services.6.3.4 Subscriber Complaint Handling 6.3.4.1 Background With the network evolving,2/3/4/5G network coexists and thus the problems of multi-RAT network,multiple elements and other issues make the subscriber complaint handling b
172、e completed.Traditionally,it mainly relies on the accumulated experience of experts,and requires high demands on operating labor.Subscriber complaint handling provides an end-to-end self-service solution for network management and operation,including complaint analysis and network fault handling.AI
173、technologies are introduced to replace traditional methods and thus to GTI 5G Intelligent Network Whitepaper V1.0 29 improve the efficiency of fault solving.This solution integrates four AI models and take the results of knowledge demarcation as the final output.Also,it extracts the core features th
174、at affect the results to support the demarcation.In order to facilitate the traceability analysis,it provides a visual display function for the whole process of demarcation.At the late stage of complaint handling,the technology of abstract extraction is used to classify customer complaints automatic
175、ally.It assists the human operator to analyze complaints,find potential problems in time and improve the accuracy of reply.As a result,repeated complaints can be greatly reduced.At the same time,it analyzes the complaint handling process and the results in the receipt,gives accurate and reliable rea
176、sons for complaints cascading to improve the efficiency of manual verification.6.3.4.2 Solution Overview Automatically dock with the EOMS,receive network tickets in real time,realize the automation of submission,delimitation and result confirmation to the EOMS.Also it can automatically return orders
177、,dispatch orders or provide reference suggestions for complaints handlers.Through speech recognition,speaker segmentation,self-supervised learning,natural language processing and other AI technologies,the system can analyze the full number of complaints,extract the summary of complaints,and obtain t
178、he core information of complaints.After analyzing the complaint handling process and the results in the receipt,the accurate and reliable reason cascading is given to improve the efficiency.Automatic docking signaling side wireless side data to define seven categories including Wireless,Core Network
179、,Communication Services,User Terminals,User Terminal Services,User Signing Services and No Exception.Combine curing experts to delimit the knowledge base.Deeply mine the correlation of alerts,use machine learning models to combine the periodic changes of network data with trend items,holidays and ot
180、her influencing factors to fit the development trend of data for anomaly detection.Based on the time sequence characteristics of the signaling data to realize the comprehensive decision-making of the users 24-hour fault situation.Considering the time and space factors to realize the accurate positio
181、ning of poor quality area.Abstract the complaint information and cluster the results for statistical analysis.Based on the knowledge graph and natural language processing technology to establish the alert expert knowledge base and provide the suggestions of alarm processing measures.Use natural lang
182、uage processing technology for receipt quality inspection instead of manual quality inspection personnel.6.3.4.3 Application and Performance At present,Network Self-service Robot system has been put into production.The complaint handling time is shortened from 90 hours to 41.21 hours,and the complai
183、nt location and demarcation time is shortened from 4 hours to 15 minutes.It is estimated that 45000 person GTI 5G Intelligent Network Whitepaper V1.0 30 days can be saved annually.According to the estimation of 700 RMB/person day,about 31.5 million RMB can be saved every year.The system assists oper
184、ation and maintenance personnel to analyze complaints and reduce repeated complaints.Also it greatly improves the quality inspection efficiency.6.4 Use Cases of Network and Service Optimization 6.4.1 NR Network UE Throughput Optimization 6.4.1.1 Background Wireless network parameter configurations a
185、re subject to scenarios.There are thousands of parameters related to the air interface.Different parameters,such as handover,coverage,and power control parameters,have different impact scope on performance counters.The combination of parameters increases exponentially.It is difficult to achieve the
186、optimal combination only through manual commissioning due to many types of parameter optimization,wide value ranges,complex scenario factors,and mutual dependencies between parameters.There are millions of parameter combinations.Figure 6-11 Thousands of parameters related to the air interface In add
187、ition,wireless network scenarios are complex and diversified,and parameter settings need to vary depending on scenarios.Therefore,a large number of experts are required for analysis and processing,and it is difficult to achieve the optimal efficiency and performance.Traditionally,only expert experie
188、nce can be used to analyze problems and optimize parameters.However,the efficiency of manual optimization on the entire network is low.In certain cases,the parameter settings used in a cell may bring negative gains in other cells.Figure 6-12 Parameter settings varies in different scenarios GTI 5G In
189、telligent Network Whitepaper V1.0 31 6.4.1.2 Solution Overview For RAN O&M,multi-parameter optimization is the most basic capability and all optimization tasks are performed based on parameter adjustment.The objective of the multi-parameter optimization solution is to ensure that all parameters of e
190、ach cell can be automatically adjusted without affecting network KPIs.The vigorous development of AI technologies makes automatic multi-parameter optimization possible.With automatic multi-parameter optimization,the RAN Manager:Obtains the parameter optimization area and optimization objective,such
191、as the target network KPI values,from the NMS.Obtain data.The RAN Manager automatically collects live network data(including MR data)based on optimization requirements and preprocesses the data,including data filtering and association.Automatically set scenario-specific parameters.The deep learning
192、AI algorithm is introduced to the RAN manager to perform joint modeling analysis on KPI data of a large number of cells.In addition,the RAN manager automatically identifies networking scenarios based on the collected MR data on the live network and configures initial parameters based on the scenario
193、s.Figure 6-13 Iterative optimizing the model Performs automatic iterative optimization.The RAN manager uses live network data and machine learning AI algorithms to perform fast iterative optimization for multiple times and evaluate the impact of different parameter groups on network performance.This
194、 RAN Manager automatically configures optimal parameter combinations for cells based on different target settings to improve network performance(RRM parameters).In this way,network problems such as load balancing issues can be resolved.GTI 5G Intelligent Network Whitepaper V1.0 32 Figure 6-14 Mappin
195、g different scenario with proper parameters 6.4.1.3 Application and Performance To cope with new network challenges,an operator in middle China has been exploring and innovating network intelligence and automation capabilities since 2019.In the traditional routine optimization process,telecom operat
196、ors need to manually identify network coverage or capacity problems through DTs or network KPI statistics.Based on manual optimization experience,the RAN Manager provides advice on parameter adjustment,such as RF and network configuration,and then issues the network optimization policy.After the AI
197、technology and automation capability are introduced,modeling is performed based on the multi-dimensional characteristics of cells on the live network,such as coverage,networking,traffic,and radio parameter configuration,and the average UE throughput in a cell.The RAN Manager can automatically identi
198、fy low-rate areas on the network and automatically optimize 13 power parameters that are closely related to the single-user throughput of the cell based on this model while ensuring the overall network performance.In the Luoyang city,the average downlink UE throughput(more than 1000)cells increases
199、by 14.5%.Figure 6-15 Modelling and iterative optimization for multiple cell parameters In addition,the RAN Manager can automatically identify the cells subjected to load imbalance based on the live network data.Based on cell configurations and traffic models,the RAN Manager adjusts related parameter
200、s and load balancing policies to achieve the optimal balancing relationship between sectors.In the pilot area the load balance rate of the multi-band and multi-layer network increases by 75%,and the optimization efficiency is greatly improved.GTI 5G Intelligent Network Whitepaper V1.0 33 Figure 6-16
201、 Pilot result of load balance with intelligent multiple parameter optimization 6.4.2 NR Network Coverage Optimization 6.4.2.1 Background Massive MIMO is an evolution form of multiple-antenna technology,and is widely regarded as a key 5G network technology.This technology integrates more RF channels
202、and antennas to implement three-dimensional precise beam forming and multi-stream multi-user multiplexing.Massive MIMO achieves better coverage and larger capacity than traditional technologies.In contrast with 4G massive MIMO that supports more than 200 broadcast beam combinations,5G massive MIMO s
203、upports thousands of broadcast beam combinations.The pattern adjustment scope varies according to AAU types.Pure manual configuration and adjustment of broadcast beam combinations cannot achieve the optimal performance of massive MIMO due to its complexity.When massive MIMO modules are deployed on a
204、 large scale,the adjustment workload is heavy,and it is difficult to complete the adjustment manually.According to the test results of multiple operators on the live network,massive MIMO intelligent optimization can improve the RSRP and UE throughput and maximize operators ROI.6.4.2.2 Solution Overv
205、iew When massive MIMO intelligent optimization is enabled,the RAN Manager:Obtains coverage optimization areas and objectives,such as the proportion of weak coverage areas,from the NMS.Obtains DT data,performance counters,traffic statistics,engineering parameters,configuration parameters,and other ba
206、sic information,including electronic maps,antenna patterns,frequency bands,and AAU types.GTI 5G Intelligent Network Whitepaper V1.0 34 Figure 6-17 Multiple dimension of data collection Creates grids for DT/MR data,identifies problematic grids,and converges them into problematic areas.Then,the RAN Ma
207、nager selects the best scenario-based beam,azimuth,and down tilt configurations for problematic cells.In this step,antenna hardware must meet the corresponding configuration requirements.Figure 6-18 Locate the problematic area and find the optimal configuration Performs iterative reinforcement AI le
208、arning based on the preset optimization objectives to obtain the optimal optimization advice.Automatically delivers the massive MIMO pattern parameter combination,down tilt,and azimuth parameters of problematic cells and their neighboring cells base on Massive MIMO pattern common AI model.Evaluates
209、and verifies the optimization advice based on user experience after issuing the optimization advice.If the KPIs do not meet the target requirements,the RAN manager rolls back the optimization advice.Figure 6-19 Evaluate and verify the optimization advice 6.4.2.3 Application and Performance In a typi
210、cal operator application scenario,the RAN Manager interconnects with the NMS through an open API.The NMS delivers the network coverage optimization objectives and areas to be optimized to the RAN Manager.The RAN manger sends the final optimization result and GTI 5G Intelligent Network Whitepaper V1.
211、0 35 optimization advice of each round to the NMS of the operator.Figure 6-20 Network framework in a typical operator application scenario In 5G network deployment scenarios,the number of 5G UEs is small.One operator in middle China applies AI technologies and automation to optimize massive MIMO bro
212、adcast beams.The RAN Manager can automatically identify problems found during drive tests,such as weak coverage,poor SINR,overlapping coverage,overshoot coverage,and frequent handovers.Based on experience rules,coverage prediction,and 5G weight parameter optimization,the RAN Manager provides paramet
213、er adjustment advice on mechanical tilts,azimuths,and broadcast beam weight.In this way,5G coverage and performance can be quickly improved to ensure better experience.This solution increases the average coverage of 5G massive MIMO cells by 15.8%and the road coverage by 91%.In addition,the optimizat
214、ion efficiency is significantly improved in contrast with traditional optimization methods.Figure 6-21 Pilot result of NR network coverage optimization GTI 5G Intelligent Network Whitepaper V1.0 36 6.4.3 ML-Based MU-MIMO Scheduler 6.4.3.1 Background Massive MIMO with massive number of antennas is on
215、e of the key enhancements of 5G.Narrow beams can be used in the regions of high user density whereas wider beams could be used in the regions of low user density.It brings opportunities to further enhance cellar network,in terms of spectral efficiency and user throughput.By utilizing the radio resou
216、rces more efficiently,the next generation 5G promises to bring much better services to consumers,to open more business opportunities and revenues to operators.However,it comes along with a great potential challenge.Massive MIMO means there are large number of beams and user layers needed to be manag
217、ed.In order to realize the full potential of multi-user massive MIMO,the scheduler is needed to solve very complicated problems,which include to figure out the best set of beams.Simply it figures out that combinatorial complexity of selecting 4 beams from 32 is more than 100,000 choices.The challeng
218、e here is to be able to design advanced scheduler that can optimize the spectral efficacy within practical compute complexity.Based on the channel sensing measurements,an optimal beam-former configuration might be derived,which can improve the user throughput.6.4.3.2 Solution Overview The basic idea
219、 on how to model this problem is to decompose the multi-step selection problem,without using programming,into simple sub-problems in a recursive manner.For every TTI,the objective is to find the set of beams that maximize the Q value,in this case,is max Sum UE-PF.(UE Proportional fair),which is key
220、system performance indicator,also commonly used to evaluated scheduler performance.Figure 6-22 Deep Q-learning or reinforcement learning Then the Q value function is the state as the function pair.The action in simply,which is the beam selection.The state is designed to capture the beam UE_PF as the
221、 function of its channel GTI 5G Intelligent Network Whitepaper V1.0 37 condition and into beam interference.And the reward is the net benefit,i.e.beam on the sum UE_PF when adding a new beam UE.It is important here to use MU_PF rather than SU_PF,which means multi-user proportional fair.To use MU_PF
222、here is to take the inter-beam inference into account.Due to the recursive property of such a dynamic programming,the value function can be represented as immediate reward as future reward possible which is called Bellman Equation.Using this property,Q Value can be obtained via value iteration.Howev
223、er,such value iteration is hard to store conventionally as the state space is huge.So rather a deep neural network is used to approximate the value function here.Such model free dynamic program is also called Deep Q-Learning or Reinforcement learning.6.4.3.3 Application and Performance Figure 6-23 S
224、imulation results of ML-based MU-MIMO scheduler This diagram shows the simulation evaluation of ML algorithm,compared with the traditional greedy algorithm in the current existing product and the best theoretical possible in this case.The theoretical max can be obtained via exhausted search.The high
225、light the result is:The ML scheduler(algorithm)can achieve close to theoretical max.The ML scheduler(algorithm)can achieve very good performance gains 17%in geomean(GM)UE throughput and 31%in cell edge(CE)user Tput comparing to traditional algorithm.6.4.4 Link Adaptation 6.4.4.1 Background 5G spectr
226、um is limited and frequency efficiency is very important for operators.In 5G network,DL spectral efficiency and throughput may be affected by inter-cell interference.The current DL Link Adaptation is optimized towards slowly varying and stationary channel variations.The network may show suboptimal p
227、erformance when adjusting to interference created by burst traffic,such as low spectral efficiency,low throughput.It will impact the end user experience.17%19%31%35%0%10%20%30%40%MLTheory MaxGMCEGTI 5G Intelligent Network Whitepaper V1.0 38 Figure 6-24 Interference created by burst traffic It is a b
228、ig challenge how network adapt the interference to get better performance.New technology AI can be used into network to recognize burst interference and optimize link adaptation.6.4.4.2 Solution Overview Using neural network on RAN,introduces a Machine Learning algorithm for steering of the existing
229、 Link Adaptation.The ML algorithm is trained to recognize refined interference scenarios based on the history of the neighbor cell activities and UE signal quality.Figure 6-25 Intelligent DL Link Adaptation It can optimize radio link performance using pattern recognition for each radio link.Collect
230、adjacent cell data in real time,use that together with mobile in order to make smart link adaptation to better fit air quality.Proper tuning the link adaptation by a Machine Learning algorithm is built on the history of neighbor cell activity and serving UE measurement.The ML algorithm tune the link
231、 adaptation dynamically in time and individually for each UE for a short period of time(sub-seconds).This replaces the module based constant homogenous parameters setting.6.4.4.3 Application and Performance Intelligent link adaptation can improve end user performance and spectrum efficiency.It will
232、be no UE dependency and can get more optimistic scheduling which can allow user to schedule higher data rate.The biggest gain can get from overlap cells outdoor.For indoor scenario,only single cell and no overlap cells and it will not have any gain since intelligent link adaptation takes into consid
233、er other cell interference.If there are no other cells around the cell,there is no gain.There will have GTI 5G Intelligent Network Whitepaper V1.0 39 big value for multi cell in city center and urban area with intelligent link adaptation.Based on simulation result,the cell edge downlink throughput c
234、an be up to 50%gains.Figure 6-26 Simulation result of intelligent link adaption With simulations,it shows that in areas with high cell overlap and medium to high loaded,the spectral efficiency can be improved up to 15%.The gain is present in light to high load,not overload sites.70%cell has 50%PRB l
235、oad.This function can be used in 70%cell.Figure 6-27 Simulation result of intelligent link adaption According with 4G test result,the cell edge downlink throughput can be up to 50%gains.In areas with high cell overlap and medium to high loaded,the spectral efficiency can be improved up to 15%.Consid
236、ering 5G scenario,the gain should be the same as 4G.Average PRB usage per cell network view Off peak:90%cells 50%PRB usage Peak:70%cells 50%PRB usage GTI 5G Intelligent Network Whitepaper V1.0 40 6.4.5 Load Balancing Based on the Virtual Grid Technology 6.4.5.1 Background With the continuous develop
237、ment of communication technologies,brings more subscribers with higher data consumption.As a result,the cell load surges and the load among multi-carrier cells is unbalanced.Therefore,the load distribution among cells needs to be adjusted based on the load balancing policy to improve user experience
238、.In the current balancing policy,the selection of the balancing UE and the balancing target cell is based on capabilities and some cell-level information.Relatively blindly,the load of the target cell may be low.However,the selected UE is located in the area with poor or no coverage of the target ce
239、ll,resulting in invalid measurement or blind handover failures.By introducing virtual grids,you can predict the radio characteristics of the neighbor cells where UEs are located,and rapidly and accurately implement load balancing between cells on the multi-frequency layer,improving resource utilizat
240、ion and user experience.A virtual grid is a space division method based on geographical location information to obtain the signals of multiple intra-frequency cells under the current environment of a UE,and then divides areas based on the cell and signal quality.By collecting statistics on the radio
241、 features of each grid(such as inter-frequency adjacent cell coverage),we can deploy more streamline network strategies.6.4.5.2 Solution Overview The virtual grid-based balancing solution includes the following aspects:Grid library building:Based on the historical measurement reports and handover in
242、formation of UEs,the AI algorithm is used to construct virtual grids and obtain the relationship between the grid-level UE and the wireless coverage of surrounding cells.Grid database update and evaluation:-Writes UE information into virtual grids and updates them in accordance with the real-time in
243、tra-frequency measurement information of UEs.-This feature collects statistics of real-time intra-frequency and inter-frequency measurement information and handover information of UEs,evaluates neighbor cell information in the grid database,and updates neighbor cell information as required.Virtual G
244、rid Application:The system monitors the load of each cell in real time.If the load unbalance conditions between cells are met,the system deduces the UEs that can be balanced and their target cells based on the cell load,cell characteristics,UE characteristics,and relations between UEs and surroundin
245、g cells(virtual grid information),and provides execution suggestions.While executing load balancing based on virtual grids,perform availability assessment on GTI 5G Intelligent Network Whitepaper V1.0 41 virtual grids.When availability is low,initiate a request for updating virtual grids.Figure 6-28
246、 Basic flow chart 6.4.5.3 Application and Performance Application scenario:Load balancing in a multi-frequency network architecture.Expected application effect:In a traditional load balancing cell,the same policy is applied to almost all UEs,but the coverage and performance of neighbor cells in diff
247、erent areas are different.That is,the target cells where UEs can be balanced may be different.The grid information can be used to determine whether a UE is suitable for balance and the target cell that can be balanced,thus improving the overall performance.No balance cell is available for the grid w
248、here the UE is located:The UE does not perform balance to reduce handover failures.The grid where the UE is located has cells that can be balanced:The UE performs blind balancing to reduce inter-frequency measurement,or performs targeted measurement only for the cells that can be balanced to improve
249、 the measurement efficiency,improve the UE migration speed,shorten the load unbalance time,improve the resource usage,and improve the overall throughput of the region.GTI 5G Intelligent Network Whitepaper V1.0 42 Table 6-1 The actual application effect in China Mobile QuanZhou(23 cells of 8 eNBs for
250、 a week)Parameter Normal LB RF Fingerprint LB(non-measurement)Improvement High Load Time of Cell(s)1542918 1338440 13.25%Number of Cell Load Balance Occurred 93853 79339 15.46%Number of Load Balance intra-system Measurement Report 1022960 169424 83.44%Number of Load Balance intra-system Measurement
251、Configuration UE(Due to Enhanced Load Balance)909375 415774 54.28%Handover success rate based on load balance stable Basic KPI (such as RRC Establishment Success Rate,E-RAB Setup Success Rate,E-RAB Drop Rate)stable The result shows that high load time,load balance times,MR reports and measurement co
252、nfiguration are all decreased after applying RF Fingerprint LB(non-measurement),which indicate that some UEs can move to the other cells more easily without measurement,therefore load balancing efficiency is improved,and some inter-frequency measurements are saved.Note:since cell load is not high en
253、ough,the improvement of throughput is not observed.6.4.6 QoE Optimization 6.4.6.1 Background With the deployment of 5g network,many 5g native applications,such as cloud VR,8K video are booming.Cloud VR service needs high transmission bandwidth and is sensitive to delay.Compared with the traditional
254、audio and video services,the user experience(QoE)of cloud VR is more vulnerable to the fluctuations of wireless transmission,resulting in stuck,mosaic and vertigo.Traditional semi-static QoS framework cant efficiently satisfy diversified QoE requirements of different applications.At the same time,Qo
255、E estimation through users interactive information in the application server usually results in a large delay,cant prevent the decline of user experience.The“QoE Optimization”use case usually involves a network element function of gNB and Local server/MEC which collects service requirements of the V
256、R application from MEC/local server and radio status of UEs the BTS.Then,using data analytics and ML inference,predict the UEs radio status e.g bandwidth in next 10 millisecond,and coordinate VR streaming encoding rate and radio resource scheduled to optimize the user experience and prevent QoE degr
257、adation(video stream jitter,mosaic etc.)as the fluctuations of wireless transmission.It is expected that QoE optimization via the intelligent collaboration between the application server and RAN can help deal with wireless transmission uncertainty and improve the efficiency of radio resources,and GT
258、I 5G Intelligent Network Whitepaper V1.0 43 eventually improve user experience.6.4.6.2 Solution Overview As shown in Figure 6-29,an example of intelligent management function is introduced.The function of intelligent management function is to deploy and manage intelligent applications and provide th
259、e data and management channels to application server and BTS.Through Intelligent management function,three artificial intelligence application modules are deployed for QoE optimization:AI based Application recognition,AI based QoE evaluation and AI based wireless bandwidth prediction.AI based Applic
260、ation recognition module is to identify VR applications by using the transmission pattern via ML,AI based QoE evaluation is to evaluate the score of user experience of the VR applications according the jitter,mosaic etc which can be recognized by the traffic pattern by AI,AI based wireless bandwidth
261、 prediction is to forecast radio channel quality in next 10-20ms.A close loop QoE optimization is realized via interaction among 3 AI modules.Using QoE optimization procedure of a VR application as example,firstly,the application recognition module identifies the VR application among the number of b
262、roadband applications which are transmitted by the BTS.The QoE evaluation module monitors the VR applications user experience online.When the users signal-to-noise ratio becomes worse,the wireless bandwidth predicts that the transmission rate will decline,and the QoE starts to deteriorate,it can inf
263、orm/suggest BTS and application server to act and prevent QoE decline.e.g.the VR application server is informed/suggested to reduce the coding rate and keep the stream smooth,and BTS is informed/suggested a min reserved PRB which satisfy the required bandwidth for the service with lower coding rate.
264、Since the action is taken according to QoE predication,the corresponding code rate and schedule rule adjustment have been completed before the wireless bandwidth changes.When the fluctuations of wireless transmission happen,the users experience will be affected little.Figure 6-29 Intelligent managem
265、ent function used in VR APP for QoE optimization(example)6.4.6.3 Application and Performance In the case trial,the VR application server and intelligent management function were introduced,GTI 5G Intelligent Network Whitepaper V1.0 44 which is located in an aggregation transport equipment room,where
266、 is approximately 2 km from the 5G BTS.An example of VR cloud gaming over 5G network were used to test user experience score with/without QoE optimization.As the test results,the estimation accuracy of Application recognition evaluation Wireless bandwidth prediction is more than 90%.Network adjustme
267、nt latency,which is the VR encoding rate adjustment latency after radio quality change,reduce from 20s to 1s,as action can be taken earlier via wireless bandwidth predication.User QoE Score increase from 40%(bad user experience)to 90%,which means a fluently VR stream.Where user-specific 3D game vide
268、o is rendered in the Edge App server:Figure 6-30 AI powered Cloud VR experience trial on China Mobile live 5G network Figure 6-31 trial results:VR game QoE evaluation without QoE optimization VR game QoE score detected by AI reduce to 31%,which indicate low QoE,when congestion happens.GTI 5G Intelli
269、gent Network Whitepaper V1.0 45 Figure 6-32 trial results:VR game QoE score with QoE optimization VR QoE score detected by AI reduce increase to 100%,which indicate high QoE,by AI command VR server decrease the encoding rate.VR QoE increased in short time(0.1s),and User didnt notice the VR QoE chang
270、e.The intelligent management function is deployed to manage intelligent applications and provide the data and management channels to application server and BTS.AI/ML-based QoE optimization via the intelligent collaboration between the application server and RAN improve the user experience.The 5G tri
271、al results shown with AI/ML-based QoE optimization,network adjustment latency for the fluctuations of wireless transmission reduce 10 times,reduce from 20s to 1s,as action can be taken earlier via wireless bandwidth predication.And User QoE Score increase 3 times,from 40%(bad user experience)to 90%(
272、VR stream play fluently).6.4.7 Edge QoS 6.4.7.1 Background In the 5G era,more than ever before,the access network forms and contents of services are enriched.With its position in the network,MEC makes it possible to provide near-real-time services for latency-sensitive,user-sensitive,and service-sen
273、sitive applications.In the campus application scenario,the MEC can obtain the service requirements of the application and the UEs connected to the application service from the app.Therefore,a QoS control policy can be delivered to the base station in accordance with the service requirements of the a
274、pp,and the service QoS of some UEs can be adjusted to meet the service requirements of the app.GTI 5G Intelligent Network Whitepaper V1.0 46 Figure 6-33 Edge Qos application scenario 6.4.7.2 Solution Overview The wireless network information,UE context information and measurement information reporte
275、d will provide the location perception and wireless environment perception for the Apps on the MEC.Meanwhile,in the UPF/MEC co-deployment scenario,the MEC can obtain the UE ID to provide the UE ID perception for the Apps.Apps obtains the near real-time perceived data,generates the near real-time con
276、trol policy according to the application requirements and radio capability optimization algorithm,and then delivers it to the BTS for executing the scheduling policy action.The gNodeB needs to abstract a service model in accordance with the function type that provides wireless network information.A
277、function is mapped to a service model.When establishing a connection,the gNodeB notifies the MEC of the functions it supports,and the MEC determines which functions need to subscribe to RAN wireless network information.The MEC generates a control policy according to the service identification,applic
278、ation requirements and algorithm optimization.It selects some UEs or a certain QoS Flow under a UEs PDU Session,delivers the control policy to the BTS,and the BTS executes the QoS Flow-based scheduling optimization.6.4.7.3 Application and Performance In the uplink direction,the gNodeB needs to repor
279、t the radio network information,UE context information,and measurement information to the MEC.In the downlink direction,the MEC delivers the service control policy and QoS parameters to the gNodeB in accordance with the application requirements.In this case,the AI technology is used for service mode
280、l identification and scheduling model optimization output.For example,based on various sensing data,the ML/DL algorithm is used to output a best scheduling model and deliver it to the base station for QoS guarantee.GTI 5G Intelligent Network Whitepaper V1.0 47 Figure 6-34 Edge Qos remote control gua
281、rantee application 6.4.8 Transport Network Optimization 6.4.8.1 Background Nodes of transport network can be logically divided into three types:Access Layer Node,Convergent Layer Node and Backbone Layer Node.Access Layer Nodes are generally deployed as user access points.Convergent Layer Nodes are u
282、sually deployed at sites with convenient optical routing and a large number of optical cables.Multiple Access Layer Nodes converge at the Convergent Layer Node,and multiple Convergent Layer Nodes converge at the Backbone Layer Node.The topology of the traditional transport network is chain architect
283、ure,which is easy for extension.However,this topology will split transport network into multiple isolated areas,which causes seriously affects on the capacity and performance of transport network when a single node or link fails.In view of this disadvantage,current transport network transforms into
284、ring architecture.This type of transport network has high robustness and reliability where failure of single node or link will not lead to network partitions.Based on the characteristics of ring architecture,in the traditional capacity management approach,the virtual network function(VNF)among the t
285、ransport ring will be expanded to ensure ring capacity requirement of business transport,when the capacity usage rate of transport ring reaches a certain threshold.However,this management mode lacks information linkage between transport rings,which means different transport rings cannot perceive the
286、 service load among each other.This mode will lead the redundant expansion operation,resulting in low utilization rate of the overall capacity of the transport network,and increasing the construction cost.In order to utilize the network resources more efficient,operators put forward related manual-d
287、ecision optimization plans.However,the formulation of the final optimization plan is highly dependent on the manual analysis and decision based on expert experience in related fields,leading the process tedious and time-consuming.GTI 5G Intelligent Network Whitepaper V1.0 48 With the advent of 5G an
288、d saturation of telecom industry market,higher requirements are put forward for the guarantee of transport network capacity,that the guarantee of capacity is bound to be carried out with lower cost and higher efficiency.6.4.8.2 Solution Overview In order to ensure the optimum capacity utilization of
289、 transport network and avoid unnecessary capacity expansion operations,traditional optimization methods include:Keep the link connection between VNFs unchanged,and expand the capacity of targeted VNF(s)among transport ring with excessive capacity usage rate.Keep the link connection between VNFs and
290、the capacity of each VNF unchanged,and add VNFs to form new transport ring.Keep the overall topology of transport network and capacity of each VNF unchanged,and adjust the service load on each transport ring.However,when the capacity usage rate of a ring exceeds the pre-determined health threshold,t
291、he other rings are usually in a state of idle or low usage rate.Due to the lack of information linkage between transport rings,above methods cannot accomplish intelligent optimization on the transport network.Therefore,in order to maximize the optimization effect and reduce the labor cost as much as
292、 possible,the following method is preferred:Keep the capacity of each VNF unchanged,and then adjust the link connection between VNFs to achieve the globally optimum utilization.If the capacity usage rate of the transport ring is still over specific threshold,then carry out capacity expansion operati
293、on on the relevant VNFs among this transport ring.This method adjusts the network topology to improve capacity utilization and service load-balance effect.After topology optimization,if the capacity of transport ring is still unable to meet the service requirement,then the relevant VNFs can be expan
294、ded.In order to achieve intelligent optimization method,Transport Network Optimization system is established,which consists of following stages:Stage 1:Data Processing.In this stage,the detailed information of transport network(work topology,coordinates of VNF,VNF capacity,peak-hour data flow on tra
295、nsmission ring,etc.)are collected.Moreover,parameters used for optimization policy management will be also input.These data are transformed into a common format based on normalization algorithms and will be further processed by following stages.Stage 2:Data Analysis.In this stage,based on the existe
296、d network topology and pre-processed data,Transport Network Optimization system analyses the relation amongst capacity of transmission ring,topology of transmission ring and the capacity of VNF.Besides,this system will calculate the global utilization rate of capacity.Stage 3:Decision and Output.In
297、this stage,according to the pre-determined optimization policy,Transport Network Optimization system will decide the optimization plan,including optimized network topology and VNF capacity requirement among each transmission ring,based on the result of Data Analysis and AI/ML algorithm.This plan wil
298、l be used by the operator to optimize the transport network reaching globally optimum capacity utilization GTI 5G Intelligent Network Whitepaper V1.0 49 rate.Figure 6-35 Original and Optimized Transport Network Link Physical Architecture 6.4.8.3 Application and Performance Nowadays,Transport Network
299、 Optimization system has been put into production and provides the following effects:The time taken to formulate the final optimization plan has been shortened from 48 hours to 2 hours with approximate 1000 nodes;The capacity expansion cost has been reduced from 50-million RMB per year to 40-million
300、 RMB per year by applying the intelligent optimization system;The labor cost has decreased to 960 person-hour,saving 1500 person-hour compared to the traditional approach.Transport Network Optimization system also assists operation and maintenance personnel to monitor,analyze transport network servi
301、ce-load and formulate the optimization plan,greatly improving the network performance.GTI 5G Intelligent Network Whitepaper V1.0 50 7 Intelligent Network Level 7.1 Introduction Intelligent network level describes the level of application of intelligence capabilities in the network management and con
302、trol workflow.The participation of the human and telecom system in the network management and control workflow are important factors to evaluate the intelligent network level.For each intelligent network level,which tasks can be performed by telecom system,which tasks can be performed by human,and w
303、hich tasks can be performed by cooperation of human and telecom system needs to be clarified.For example,in the highest intelligent network level,all tasks are performed by telecom system.The industry benefit from common method for evaluating intelligent network level,which provides evaluation basis
304、 for measuring the level of an intelligent network along with its components and workflows,reference for gaps and priorities analysis for standardization works on intelligent network level and guidance to operators,vendors and other participants of telecommunications industry for roadmap planning.7.
305、1.1 Framework Approach for Classification of Intelligent Network Levels According to the potential categorization of the tasks in a general network management and control workflow(including intent management,collection,analysis,decision and execution),a framework approach for classification of intel
306、ligent network level is introduced as following,which is used for evaluating the intelligence capabilities of telecom system.Note:the following framework approach for classification of intelligent network level is based on the framework approach for classification of autonomous network level defined
307、 in 3GPP TR 28.8107.And the framework approach for classification of mobile network management and operation intelligence levels defined CCSA TC7 34.Table 7-1 Framework approach for classification of autonomous network level GTI 5G Intelligent Network Whitepaper V1.0 51 Network autonomy level Task c
308、ategories Execution Awareness Analysis Decision Intent management L0 Manual operating network Human Human Human Human Human L1 Assisted operating network Human&Telecom system Human&Telecom system Human Human Human L2 Preliminary intelligent network Telecom system Human&Telecom system Human&Telecom s
309、ystem Human Human L3 Intermediate intelligent network Telecom system Telecom system Human&Telecom system Human&Telecom system Human L4 Advanced intelligent network Telecom system Telecom system Telecom system Telecom system Human&Telecom system L5 Full intelligent network Telecom system Telecom syst
310、em Telecom system Telecom system Telecom system Note:Human reviewed decision have the highest authority in each level if there is any confliction between human reviewed decision and telecom system generated decision.Level 0 manual operating network:No categorization of the tasks is accomplished by t
311、elecom system itself.Level 1 assisted operating network:A part of the execution and awareness tasks are accomplished automatically by telecom system itself based on human defined rules.At this level,telecom system can assist human to improve the execution and awareness efficiency.Level 2 preliminary
312、 intelligent network:All the execution tasks are accomplished automatically by telecom system itself.A part of the awareness and analysis tasks are accomplished automatically by telecom system itself based on human defined policies.At this level,telecom system can assist human to achieve the closed
313、loop based on human defined policies.Level 3 intermediate intelligent network:All the execution and awareness tasks are accomplished automatically by telecom system itself.A part of the analysis and decision tasks are accomplished automatically by telecom system itself based on human defined policie
314、s.At this level,the telecom system can achieve the closed loop automation based on the human defined closed loop automation policies.Level 4 advanced intelligent network:All the execution,awareness,analysis and decision tasks are accomplished automatically by telecom system itself.And intent managem
315、ent tasks can be partly accomplished automatically by telecom system itself based on human defined intent translation policies.At this level,telecom system can achieve the intent driven closed loop automation based on human defined intent management policies,which means the telecom system can transl
316、ate the intent to the detailed closed loop automation policies and evaluate intent fulfillment information(e.g.the intent is satisfied or not)based GTI 5G Intelligent Network Whitepaper V1.0 52 on human defined intent management policies.Level 5 fully intelligent network:The entire intelligent netwo
317、rk management and control workflow is accomplished automatically by telecom system without human intervention.At this level,telecom system can achieve the whole entire intelligent network management and control workflow to satisfy the received intent.Note:Above framework approach for classification
318、of intelligent network level are applicable for evaluating the intelligent network level from both applicable scope(including NE,domain,cross domain)and applicable scenario perspective.The overall intelligent network level of the whole telecom system is a comprehensive reflection of intelligent netw
319、ork level of the individual applicable scope and applicable scenarios,which means in fully intelligent network level,the telecom system can achieve the whole intelligent for all applicable scopes and applicable scenarios.7.2 INL Evaluation of Typical Use Cases 7.2.1 Energy Saving Based on the framew
320、ork approach for classification of intelligent network level,the use case of online energy saving wireless networks defined in clause 6.3.1 achieves intelligent capabilities of level 3:Intent management(Human):Performed by human Awareness(Telecom System):The RAN Manager can automatically collect net
321、work data and generate traffic distribution information based on the collection results.Analysis(Human&Telecom System):Based on the manually specified areas and energy saving objectives,the RAN Manager can automatically generate energy saving solutions specific for network scenarios and traffic cond
322、itions.Decision(Human&Telecom System):After the RAN Manager generates an energy saving solution,the network automatically evaluates the energy saving solution,continuously performs iterative optimization,and finally provides the optimal solution.Execution(Telecom System):The RAN Manager automaticall
323、y delivers the energy saving solution and automatically configures the entire network and sites.Table 7-2 Level evaluation of Energy Saving GTI 5G Intelligent Network Whitepaper V1.0 53 To evolve to a level 4 intelligent network,the RAN Manager needs to provide the intelligent capability of predicti
324、ng network traffic and generating energy saving policies automatically based on capabilities of level 3.7.2.2 NR Network Coverage Optimization Based on the framework approach for classification of intelligent network level,the use case of NR network coverage optimization defined in clause 6.6.4 achi
325、eves intelligent capabilities of level 3:Intent management(Human):performed by human Awareness(Telecom System):The RAN Manager can automatically collect coverage-related performance data and generate coverage geographic distribution information(for example,RSRP distribution information based on geog
326、raphic grids)based on the collection results.Analysis(Human&Telecom System):The RAN Manager can automatically analyze and determine the root causes of coverage problems,such as identifying top N abnormal cells,based on the specified coverage problem analysis policies.The RAN Manager generates a cove
327、rage optimization and adjustment solution(such as the optimal massive MIMO coverage scenario,azimuth,and tilts of top N abnormal cells and their neighboring cells)based on the manually specified coverage optimization and adjustment policies,such as range of the massive MIMO pattern,azimuth,and tilts
328、.Decision(Human&Telecom System):For an automatically generated optimization solution,GTI 5G Intelligent Network Whitepaper V1.0 54 the RAN Manager can evaluate the coverage optimization solution and its impact on network performance based on the manually specified coverage optimization and adjustmen
329、t policy.Finally,the RAN Manger automatically determines the coverage optimization and adjustment solution to be executed based on the manually specified coverage optimization and adjustment policies.Execution(Telecom System):Based on the automatically determined optimization and adjustment solution
330、,the RAN Manager automatically adjusts related network parameters.Table 7-3 Level evaluation of NR network coverage optimization Level-3 policy driven closed-loop coverage optimization automation is achieved in the use case.As the next-step evolution of intelligent networks,intent driven closed-loop
331、 coverage optimization automation based on specific service assurance intents for certain scenarios needs to be implemented.That is,monitoring rule determination,optimization requirement and policy determination,network/service assurance intent evaluation,and performance deterioration prediction nee
332、d to be automated.7.2.3 NR Network UE Throughput Optimization Based on the framework approach for classification of intelligent network level,the use case of NR network UE throughput optimization defined in clause 6.6.1 achieves intelligent capabilities of level 3:GTI 5G Intelligent Network Whitepap
333、er V1.0 55 Intent management(Human):performed by human Awareness(Human&Telecom System):The RAN Manager can automatically collect performance data and generate performance geographic distribution information(for example,grid-based traffic distribution information)based on the collection results.Analysis(Human&Telecom System):Based on the manually specified adjustment objectives of locating and opti