ai empowers 5g intelligent operation
TRANSCRIPT
©TM Forum | 1
AI Empowers 5G Intelligent Operation
Champion Participants
©TM Forum | 2
Wireless network: 4G/5G col laboration, 5 networks
with 11 bands, and M-MIMO (Massive Multiple Input Multiple Output)Core network: virtualized, microservice-oriented, and edge sinkingBearer network: SDN (Software Defined Network) and
cloud-network synergy
Dense deployment: complex configurations and high costs of site visitsEnergy consumption: high electricity fees and 5G energy consumption
Agility: available once provisioned and self-serviceDeterministic SLA (Service Layer Agreement): delay-sensitive and high reliability
New experience
High bandwidth
Mbit/s-> Gbit/s
Low latency10 ms-> ~1 ms
100,000-> 1 mi l lion
Mass connectivity
Network
complexity
Cost pressure
New 2B
requirements
New experience New challenges
New challenges
At the TM Forum Autonomous Networks Summit 2020 in Beijing, China, Wei Lihong, General Manager of China Mobile's Network Department, delivered a speech titled "AI Enables Autonomous Driving Network and Practices 5G+", which explained the new challenges brought by 5G and called for intelligent network capabilities. This catalyst project is an in-depth exploration into TM Forum AN, with contributions from both China Mobile and industry partners.
New Challenges Brought by 5G
©TM Forum | 3
Call for Intelligent Network Capabilities
eMBB
uRLLC
mMTC
AI Automation
"Soft" capability in the fourth dimension
"Hard" capability of 5G networks
eMBB(enhanced Mobile Broadband)
mMTC(Massive Machine Type Connection)
uRLLC(ultra Reliable Low
Latency Connection)
Large-scale IoT
1 mill ion connections/km2
Ultra-high reliability and low latency
1 ms, 99.999%
Network + AI: lower cost, higher efficiency
Operation + AI: simplified operation and
maintenance
Business + AI: agile support
• IoE's (Internet of Everything) cornerstone is the "hard" pipe capability of 5G networks.
• AI-empowered network automation will become an important "soft" capability of 5G networks, enhancing existing eMBB, URLLC, and mMTC
capabilities while also increasing network dimensionality.
• Autonomous driving network (AN) will bring improvements in three aspects: network + AI (better network performance), operation + AI (simplified
operations and maintenance), and business + AI (deterministic capabilities), meeting the network requirements of vertical industries in different
scenarios and promoting value-oriented network operations.
©TM Forum | 4
Intelligent mining of mass data helpsdetect hidden risks, locate the root causes of alarms, and intelligently
preprocess faults, improving operation and maintenance efficiency.
Intelligent analysis of performance indicators helps proactively identify and
automatically process poor-quality services, improving network quality.
Intelligent modeling based on home network characteristics and customer intents helps accurately identify and
solve customer complaints, improving customer experience.
Operation + AI Network + AI Business + AI
◆ 4G/5G intelligent incident management
◆ Intelligent optimization of 5G wireless networks
◆ NFV(Network Function Virtualization) log analysis
◆ Premium home broadband◆ AI-driven comprehensive improvement of
network satisfaction
AI Drives Higher Operation and Maintenance Efficiency, Better Network Quality, and More Agile Services
©TM Forum | 5
Principles:1. Single-domain, single-
vendor autonomous management and control (small closed loops)
2. Multi-domain, multi -vendor orchestration (large closed loops)
BSS/Application
Plan
Construct Maintai
n
Optimize
Networkautomation
Networkintelligence
Analyzer
Manager
Controller
Network management &control system
OSS (cross-domain orchestration)
Networkautomation
Networkintelligence
Analyzer
Manager Controller
Network management &control system
Autonomous domain X
Autonomous domain Y
Single-domain autonomy
Single-domain autonomy
Operate
5G transport Premium private linePremium broadband
AN framework:3 layers, 4 closed loops
Overall Solution: Autonomous Domains (Small Closed Loops) + Network Automation (Large Closed Loops)
RFS = Resource Facing Services
CFS = Customer Facing Services
©TM Forum | 6
Operation + AI — 4G/5G Intelligent Incident Management :Alarm Aggregation,
RCA, and Accurate Tickets
Mass alarms⚫ 500,000+ alarms a day on a local network⚫ Ticket storm (300+ alarms per second)
ChallengesSolution Benefits
Experience-basedAlarm filtering and RCA (Root Cause Analysis) rely on specialists' experience, which are inefficient and error-prone.
Low troubleshooting efficiency
Alarm compression rate: 99.9%; tickets: reduced by 35%
35%43%
23%
9%
0
100
200
300
400
2019.1 2019.12 2020.1 2020.2
Tickets for or iginal alarms (daily)
Tickets for intel ligent incidents (daily)
Decrease rate
Recall: 95%; precision: 90%
Recall Precision
95% 90%
OSS (Operations Support Subsystem)
Ticketing
Single-domain autonomy
Alarm reporting
Intelligent incident API(Open API)
Network management and control system (single-domain, single-vendor intelligent incidents)
MBB FBB
Ala
rm n
oise
red
ucti
on
Ala
rm
aggr
egat
ion
RCA
Fault 1
Fault 2...
Fault 1
Alarm Monitoring
• Ticket storm handling• Ticket correlation/recall• Ticket delay analysis
• Multi-domain, multi-vendor analysis
• Multi-domain RCA• Handling Policy delivered
to the root fault domain
Delivery policies to the root fault domain.
(Open API)
Multi-domain collaboration
Troubleshooting⚫ Tickets are dispatched repeatedly, not
dispatched as expected, or overstocked.⚫ Low alarm & ticket accuracy, difficult to
locate and rectify faults across layers and vendors
RCA = Root Cause Analysis MBB = Mobile BroadbandAPI = Application Programming Interface FBB = Fixed Broadband
©TM Forum | 7
⚫Massive MIMO service complexity: 1000+ patterns, involving horizontal and vertical beamwidths, downtiltangles, and azimuth
⚫Complex and diversified scenarios: wide coverage, high-rise buildings...
⚫Highly dependent on drive tests: overshoot, overlapping, or weak coverage
ChallengesSolution Benefits
5G network in XX city (1426 cells)
Coverage improved by 15.8% on 91% of roads
15.59
18.07
14
16
18
20
优化前 优化后
Average SINR (dB)
2.38
% 0.98
%
0.00%
2.00%
4.00%
优化前 优化后
Percentage of SINR < 0
Network management & control system
LTE NR(New Radio)
MR(Measurement report)
MDT(Minimizationof Drive Tests)
DT
Initial network construction phase:
Automatically improve coverage by importing drive test (DT) data.
Routine operation and maintenance phase:
Continuously improve coverage and user experience based on AI and online data
Pattern adjustment
CHR/MR/...Before After
Before After
Online learning, iterative optimizationIterative optimization
Tilt angle Optimal point
Start point
Azimuth
Performance
Optimal solution
Network + AI — 5G Intelligent Optimization :Coverage Improvement with Massive
MIMO(multiple-input multiple-output )
Network management &
control system
SINR = Signal Interference Noise Ratio
CHR = Call History Record
©TM Forum | 8
ChallengesSolution Benefits
Risk monitoring scope ↑⚫ More comprehensive device risk
monitoring, preventing major risks from being missed
⚫ Auxiliary measures for alarms and traffic statistics
Risk detection speed ↑1-30 minutes earlier than alarm generation
Device security ↑⚫ Prevent faults in advance⚫ Keep network devices secure
Risks on NFV devices
⚫ Open and multi-layer NFV architecture⚫ Risks on any nodes can cause
accidents.
Manual, inefficient log analysis
⚫ NFV logs are large in quantity. In a region of country C, more than 10 billion logs are generated every day.
⚫ Manual analysis relies heavily on specialists' experience, efficiency, and status.
Log-based risk detection
⚫ IT device logs record 90% of information and are therefore very important for O&M.
⚫ Logs are more sensitive than alarms.
Log f iles
Log distribution featuremodel training
Anomaly detection
Log clustering analysis
Faultscenario matching
Manual analysis and labeling
Log templates
Fault log feature database
Anomalykey logs Fault symptom description
Un
kno
wn
fau
lt m
on
ito
rin
g
Kn
ow
n f
ault
mo
nit
ori
ng
1. Data mining-enabled unknown and known fault monitoring based on logs
2. Real-time display of device status
Network + AI — AI Driven Log Analysis :Intelligent Risk Warning based on NFV Logs
©TM Forum | 9
Challenges Solution Benefits
Router of vendor A
Antenna: 2 x 2Single/dual-band: 2.4GStandard: Wi-Fi 5
High-dimensional mapping of neural
network models and features
Collect three types of data
• ONT/AP/STA capabilities:frequency bands, network ports, and antennas
• Network connectivity:network cables
• Wi-Fi capabilities:Wi-Fi strength, interference...
Device capability learning and bottleneck
identification
200 Mbit/s 20- 100 M bit /s
1
32
45
6Subscribed bandwidth
Actual bandwidth
For 76% of users, subscribed bandwidth ≠ actual bandwidth
6%
15%
35%
38.50%
40%
100BAS E-T cable
100BAS E-T port
Si ng le-ba nd ON T
Si ng le-ba nd router
Si ng le-ba nd termina l
The home network is complex, in which user experience is affected by
multiple factors.
According to data of 27,000 broadband users in XX city
Marketing conversion rate increased 5-7 times
Future
Now
Precis ion marketing based on data analysis
Ground promotion
User Device Capability Wi-Fi CapabilityPotential
Customer Level
User ASingle-band optical modem
Poor coverage High
User BSingle-band optical modem + single-band router
Poor coverage and high interference
High
User C Dual-band optical modemGood coverage and low interference
Low
...
?
Subscribe to 200 Mbit/s but receive
only 50 Mbit/s
End users: Subscribe to 200 Mbit/s and receive 200 Mbit/s.
A p air o f M AC
ad d re sse s
Digi tal twin
Sh are p arame te rs
M AC ad dress in vector re p re se n tatio n
M AC ad dress in vector re p re se n tatio n
Business + AI — Premium Broadband : Accurate Identification of Home Network Bottlenecks to Meet Bandwidth Experience Commitments
©TM Forum | 10
Business + AI : AI-Driven User Satisfaction Improvement
Challenges Solution Benefits
Time consuming
◼ Cal center In-time answer rate < 50%
◼ 80% of Ticket handling duration > 4hours
Poor solving
◼ Root cause accuracy < 50%
Purely Reactive
◼ Lack of evaluation on overall usersatisfaction
◼ Not able to prevent potentialcomplaint
Definition of network characteristics
XDR, MR, KPI, FM, CM
Dimension Reduction
u1
u2
u3
⋮un
G1
G2
⋮Gk
Clustering
Initial portrait of user perception based on network data
Accurate portrait of user perception based on AI algorithm
❶ ❷
❸❹
Root cause analysis and pinpoint the user complaint on map
Efficiency Improvement↑
- Call center In-time answer rate 70%
- resolution is available within 3 seconds。
Accuracy Improvement↑
- Root cause accuracy > 85%
- Locating accuracy > 95%
Proactive Enhancement↑
- Overall user satisfaction evaluation is proven with 92% accuracy
- Prevent complaints from high potential users
5
XDR= X Detail Record FM= Fault ManagementMR = Measurement Report CM= Configuration ManagementKPI = Key Performance Indicator
©TM Forum | 11
September 2019
Research
December 2019
Topicdeclaration
Team
building
January 2020
June 2020
Achievements
R&Ddeployment
April 2020
September 2020
October 2020
Catalyst phase-1achievements
Attain L3
Exploratory phase Systematic phase
Optimization and
expansion
Summary and Outlook
• In this catalyst, China Mobile has been progressed on fully implementation on it
“5G+” strategy to promote 5G+AICDE(AI,IoT, Cloud computing, Big Data, Edge
Computing) innovations. Based on the TM Forum Autonomous Network concept,
this catalyst project focuses on 5G+AI explorations.
• With Huawei, Nokia, AsiaInfo and BOCO, we deeply explored network operations
and maintenance with five use cases in this catalyst.