The Future of Wireless Network -AI Inside - ITU
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
The Future of Wireless Network –AI Inside Hua.huang@Huawei.com
Challenge: Low Efficiency and High Cost on Network O&M
Network Planning Network Optimization Network O&M
KPI
Drive- Adjust Monitor
planning Design Deploy
test para KPI Service
Alarm
data
Long period, high labor cost Passive response to performance Difficult to do RCA from massive data
Increasing OPEX
Intensive Manual Work
OPEX : CAPEX ~ 4 : 1
Complicated Wireless Environment Source: Gartner report
More and More sites Cost structure needs change
2
Challenge: Growing Operation Complexity
Contextual Perception
UCNC Network
•
• Ultra Dense Deployment
MM Wave Highway Mall Big Event
2600M Dynamic Scenario/Traffic Model
2100M
1900M C-Band
1800M (~100MHz) Massive Het-net Site
900M 120
800M
700M
100x
Multi-Band/Multi-RAT 80x
60x
40x
20x
0x
2018 2019 2020 2021 2022 2023
3
Challenge: SLA Guaranteed is Entering “Hard Mode”
Service type :10+
FMS, Flexible Manufacturing System
Service type: 1000+，Diverse SLA
Slice Topology Slice Function SLA
Slice Template
Design Define Decomposition
1080 types of personalized BMW7
• Guaranteed SLA requires the adaptability of a
slice
– Like in the Cloud/DC: scale in / scale out
– Dynamic resource assignment
– Dynamic scheduling depending on real-time
resource usage
4
AI Democratization Starts its Journey
Assist Smart Driving On-device AI
IBM Watson Sweeping Robot
Google Alpha go Softbank Pepper
• Machine Leaning
• Deep Learning
• Reinforcement
Learning
Massive Data Hardware Computing Power
Mobike:1TB/day Taobao:7TB/day Web: 500PB/day GPU: Float Operations, 10Tflops; CPU: 1.34Tflops
5
Mobile Network is Feasible move to AI
AI Big Data Algorithm Computing
Wireless Network Inherent 4 Characteristics Make it Fertile Ground for AI
Complicated Computing Diverse
Massive Data
Algorithm capability Service
Real time data Coordination eNB, Cloud with high Blooming service,
from network Resource allocation ability chip diverse SLA
When wireless network meets AI, new potential will be inspired
6
The Industry is on the way to ABCC
IT SDN、NFV、Could、SBA ABC2= AI+BigData+Cloud+Connection
Technologies 、HTTP2、AI...
TOP7 MV A B C C
application APPLE 8890 Siri,chipset App Store, Apple Pay iCloud 3GPP
Google 7235 Deepmind Gmail/Google+/Chr GAE ..
ome/ Andriod
management
Microsoft 6459 ,Zo.. 350 m Azure
Amazon 5491 Echo/Alexa 200 m AWS
Core network
Facebook 5285 AML 2b OCP TIP
Tencent 5236 AI Lab 900 m 腾讯云
Access Network
Alibaba 4889 NASA DT 阿里云 3GPP
2001-2010 ALL IP 2011-2020 ALL IT
The downward trend of IT technology is a basic trend. The major reason is that the carrier
network must adapt to application changes.。
7
5G+ and 6G will be AI inside
eMBB
eMBB Network Performance
eMTC
GAP
Future IMT
mMTC uRLLC AI/Big Data 1G 2G 3G 4G 5G
uRLLC
5G 5G+/6G
• 5G is not only provide connection, and also win the new business in the digital era.
• AI can help to improve the network performance, and simplify the network management.
8
Wireless AI Leapfrog to a New Phase
Creation based
on logic Cognitive
Knowledge based Deep Learning
on data
Result based Machine Learning
on rules
Expert System
IT Industry are Here
We are here
Modeling Modeling Modeling
Self Evolution
Characteristic Characteristic Characteristic
9
Wireless AI Vision and Case list
--Learning Radio Environment ,Make impossible to possible
Network O&M Network Performance New Business
Customer
Values
Simplify O&M Beyond Performance Limits Enable New Business
• Massive MIMO Self-Adaption • Free Measurement multi-carrier • TCP Optimization
• Intelligent Alarm Association
selection • Network Fingerprint Enabled
• LampSite Topology Smart
Typical
Optimization • VoLTE Quality Improvement High Accuracy Location
Cases
• Adaptive KPI Anomaly Detection • Bottom user experience • Using Artificial Intelligence to
• Fault Alarm prediction
improvement Predict Ride Requests
• Scenario Self-Recognition
• PCI-conflict optimization, CCO
…… …… ……
10Case1:Massive MIMO Pattern Self-adaption
MM Solution Bring Challenge to O&M Team ML based Solution lock Best Pattern quickly
Configure Pattern Complexity Explosion ML Boost Fast Optimization
ML Generate Best Route
4.5 G MM Site 5 G MM Site Step0: Initial selection base on
massive experience modeling
3
Pattern Option Pattern Option Step1-5: Automatic Iteration
4
300+ 10000+ 1 2
optimization by ML.
5 Avoid (Bad) and (Normal)
0 , fast approach (Good) Area
1 Broadcast beam 8 Broadcast beam Optima
Powerful Strategy Library Accelerate Optimization
Diverse Scenarios, Fluctuated Traffic Joint Trial Test Result in Japan
Optimization Time
2persons Assumption:
20 Day 100MM Cells
mAOS
7 Day
Traditional Automatic
Way Way
11Case2: Alarm Processing
Compress Alarms, Reduce Dispatch Orders Analysis Root Alarm, Improve Efficiency of
After Real-time Compressing
Troubleshooting
Alarm A … ... A Alarm2
Alarm1 Root Alarm1
Alarm2
Alarm B … ... B
Compressed by Alarm A Alarm B Alarm3 Alarm4
Alarm4
Alarm5 Alarm5 Alarm3
Time
Adaptive Rule
Alarms AI Correlative Alarms
• Adaptive Rule Will Be Setup Correctly by 100%
• Intermittent Alarm Will Be Reduced by 99%
Troubleshooting Efficiency
964
Before 44K 11K
After Compressed Rate
Alarm Groups
10
Alarms
Intermittent Alarms/Day/NE
Based on LTE Network of Changsha in 1 Day Based on LTE Network of Shanghai in 1 week
12Case3:Adaptive KPI Anomaly Detection
Call Drop Rate(%)
1.4
Training Period Threshold Prediction Adaptive Threshold
1.2 Period Based on prediction by ML
1
• Find Concealed Anomaly KPI
0.8 • Avoid tremendous fault alarm
X X • 95% Accuracy
0.6
Historical KPI
• Setting at Cell/Cluster Level
0.4
0.2
0
92.4% 99.4% 86.9%
10:00
15:00
20:00
11:00
16:00
21:00
12:00
17:00
22:00
13:00
18:00
23:00
14:00
19:00
10:00
15:00
20:00
11:00
16:00
21:00
0:00
5:00
1:00
6:00
2:00
7:00
3:00
8:00
4:00
9:00
0:00
5:00
1:00
6:00
KPI Trend Concealed Anomaly
X KPI
Fixed Anomaly Adaptive Anomaly
Threshold Threshold
Adaptive KPI Anomaly Detection
13Case4:Smart CA with Virtual Grid
Always on the Best Carriers: 60+% Improvement
基于虚拟栅格的数据模型化
3.5G
2.6G
2.1G
1.8G
900M
800M
Recognition Prediction
14AI Standard in 3GPP
S2-173192
TR 23.799| 2016.11 S2-164035
Network Data Analytics (NWDA)
1:N 1:N 1:N
AP
Edge DC Core DC
Terminal AN(1..n) TN(1..n) CN(1..n) IMS/3rd APP
S2-164691
• SA2/RAN3
The focus of this key issue is to highlight the need of a mechanism which can discover,
• SA5 reason and predict a situation by efficiently turning raw measurements into well-defined
knowledge (referred here as context ). Solutions to this Key Issue will:
1: S2-173192/173193 Discussion about Big Data Driven - Determine which information from the UE, external applications, Network Functions,
Network Architecture and RAN can be combined and how, to create richer session/network context
2: S2-164035 Analytics-based Policy (Motorola Mobility, Lenovo) information that can optimize decision making.
3: S2-164691 New Key Issue on Context Awareness (Telenor, - Investigate which kind of analysis can be applied;
NEC, Orange, Deutche Telekom AG)
4: S5-173364 New SID Study on utilizing artificial intelligence - Investigate which reference points or communication models should be used to
in mobile network management enable monitored information to flow among NFs (e.g., UP and/or CP functions) and
3rd party applications
155G+ AI forecast
2020 2023 2027
L2 L3 L5
Partial No Complete
intelligence configuration intelligence
My dream: we establish such a network, like a large machine, he seems to have life, in a variety of complex
environment, breathing with the environment changes, the flow of resources, the antenna of the base
station wagging. He knows all the state of the network, also predict all changes that will happen in the
network, including possible failures, even changes in the environment, and timely adjustment, and
continuous evolution, and the maximum efficiency of information transmission and service. And no more
managers are needed.
16Huawei Wireless AI DA-Lab
Wireless Intelligence DA-Lab 4 Key ability for DA-Lab
Data training Pre-research for AI Algorithm
• Data storage • Import industry AI
• Data analysis algorithm
• Data modeling • Machine Learning
• Deep Learning
Verification for AI Algorithm Incubate valuable model
Data Analytics Laboratory (DA LAB) • 3-rd party • High accuracy
•
Established in 2016.9, Core ability, •
Self-study
Off-line verify •
position
Scenario
Data storage and process: 0.5PB, recognition
0.5PB equivalent to 1,000,000 GUL cell，
Plan to 2PB in 2019.
17Wireless AI Alliance
18Alliance goals and participating units
Alliance goals
Relying on the cooperation platform of industry-university-research, to realize the intelligent
guidance and in-depth integration of wireless big data, and promote the
development of green, efficient and intelligent communication.
Organizational Units
Participating Units
China University of Beijing University of Zhejiang Beijing University of
Science and Aeronautics and University Posts and
Technology Astronautics Telecommunications
Cooperative Units
19Thank You. Copyright©2017 Huawei Technologies Co., Ltd. All Rights Reserved. The information in this document may contain predictive statements including, without limitation, statements regarding the future financial and operating results, future product portfolio, new technology, etc. There are a number of factors that could cause actual results and developments to differ materially from those expressed or implied in the predictive statements. Therefore, such information is provided for reference purpose only and constitutes neither an offer nor an acceptance. Huawei may change the information at any time without notice.
You can also read
