Upload
others
View
9
Download
0
Embed Size (px)
Citation preview
1
From ‘Green & Soft’ to ‘Open & Smart’
Dr. Chih-Lin I
CMCC Chief Scientist, Wireless Technologies
CMRI, China Mobile
IEEE 5G World Forum
July 10 2018, Santa Clara, California, USA
2
2
World’s Largest 4G Network
The Largest Scale The Biggest User
672M Subscribers 1.87M Base stations
63% 73%
The Best Experience The Most Popular
~1.3B Pop coverage rate
99%>200M
VOLTE users
2G/3G 2G/3G
VoLTE: 313 cities,
>93% new terminals enabled
as of May 2018
3
3
2016 2017 202020192018
The LARGEST Scale Trial, IOT Commercialization
Beijing, Shanghai, Guangzhou,Ningbo, Suzhou
(3.5GHz, 7 sites/city)
Key Tech Test(Completed)
PoC TestPre-commercial
Trial
CMCC 5G Trials in Sync with IMT-2020 PG Timeline
Trial: Hangzhou, Shanghai,
Guangzhou, Wuhan, Suzhou
Service demo: Beijing, Chengdu,
Shenzhen, …
CMCC makes world’s first
Holographic video call based on
5G SA nework in MWCS 2018
CMCC jointly launched the “5G
Standalone Sailing Action”
together with global partners
(5 cities for trial [500+ sites], 12 cities for
service & application demo [500+ sites])
4
4
CMCC 5G Joint Innovation Center & 5G Device Forerunner Initiative (28 Partners Signed MOU in MWCS18)
Qingdao LabIoT
Beijing Central LabComm Infrastructure, IoT
Shanghai LabIoV, IoT
~Dozens of Projects Initiated with 14 Open Labs (US, SE, HK open labs)
(Intelligent transport, smart home, smart government, smart factory,
connected UAV, IoV……)
Chengdu LabComm Infrastructure
Intelligent manufacturing,
Industrial InternetProjects with Changhong, Huawei, E///
Hangzhou LabIoT
Shenzhen LabRobot/UAV
Yingtan LabIoT
Chongqing LabIoV, IoT
Nanjing/Suzhou LabComm Infrastructure
Industrial internet
China Mobile 5G Joint Innovation Center (since Feb, 2016)
224 Partners (167 of these companies are vertical industries)
… … …
5G Device Forerunner Initiative (MWCS18)
《5GTerminal Guide》released in MWCS18
Objective: 5G terminal to be released in 2019
Guiyang Lab
Wuhan Lab
GTI 5G S-Module Initiative (MWCS18)
Superior universal module: unified size/interface/function…
To be released at the beginning of 2019
5
5
Embracing Verticals, E2E Solutions Demo in MWCS18
Intelligent
Could-Robot Mechanical arm
Power grid and oil pipeline
monitoring
Industrial factory
5G mHealthWater tank monitoring
CMCC IoX Demo in MWC2018 CMCC IoX Demo in MWCS2018
E2E Slicing enabled Smart Grid
5G Remote Touch-Control 5G remote driving based Platooning
E2E latency < 10ms, [w/ MEC]
V2V latency < 5ms,
Throughput > 50Mbps
6
6
Cellular UAV
Intelligent Airport
Ground-to-air communication
First Aerial Internet SUMMIT in
Zhangjiakou, Mar 2018
Traffic Warden
Cloud UAV platform
Secure & Efficient
connected service
“Airplane mode”
ended in Jan
2018 in China
Tech Test
since 2013
Collaboration w/
multiple airports
Aerial Internet: UAV (Unmanned Aerial Vehicle)
Aerial Internet SUMMIT in Shanghai, July 2018
UAV use cases
CMCC UAV cloud platform Safe flight control platform
Fiight parameter
Control
signaling
HD video
Intelligent image
recognition platform
Multi types for terminals 5G NetworkUAV
HD video transmission &
intelligent recognition
Remote real-time
tracking & operationHigh precision real
time flight controlAccurate operation
How to work
UAV flight control system
Demo City:
Shanghai, Hangzhou,
Wuhan, Tianjin, Beijing,
Shenzhen
Location Authentication
CMCC UAV cloud platform
UAV control centerCMCC location service center
CN
Uniform equipment record Human-equipment record
Sales
4G/5G Network
Usage
UAV manufacturers Connected UAVUEUEs
CMCC online service center
7
7
Rethink Fundamentals
Rethink Protocol Stack
Rethink Fronthaul
Rethink Base Station
Rethink Ring & Young
RAN
“Towards Green & Soft” IEEE WCNC Keynote, Apr.8, 2013
“SDX: How Soft is 5G?”, IEEE WCNC Keynote, Mar. 21,2017
“Towards Green & Soft: A 5G Perspective” IEEE Comm. Magazine, Vol.52, Feb.2014
“5G: rethink wireless communication for 2020+”, Philosophical Trans. A. 374(2062), 2015
“On big data analytics for greener and softer RAN,” IEEE Access, vol.3, Mar. 2015.
“New paradigm of 5G wireless internet”, IEEE JSAC, vol.34, no.3, March 2016
“Big Data Enabled Mobile Network Design for 5G & Beyond,” IEEE Comm. Magazine., vol. 55, no. 9, Jul.2017.
“Big Data Driven Intelligent Wireless Network: Architecture, Use Cases, Solutions and Future Trends ,” IEEE VT Magazine, 2017
CT/DT/IT ConvergenceOpen SourceBusiness Model
Green Communication Research
Center established in Oct. 2011,
initiated 5G Key Tech R&D.Green Soft
Rethink Air Interface & Spectrum
Rethink Signaling & control
Rethink Shannon
Air
Interface
To enable wireless signal to “dress for the occasion” via SDAI
To start a green journey of wireless systems, EE/SE
To make network application/load aware
Embracing verticals How it affects the traditional SDOs? What’s Big Data’s role in 5G era
For no more “cells” via C-RAN
To enable Soft RAN via NGFI
To make BS “invisible” via SmarTile
To enable User Centric Cell and flexible AI via MCD
Efficiency Agility
8
8
Soft 5G Arch: SBA [Service based Architecture], UCN [User Centric Network], SDAI
Soft Transport Network
Soft CN
Soft RAN
Serviced based Arch,
Network Slice, NFV
Ultra wide BW, lower latency,
Higher timing precision
UCN: C-RAN, CU/DU design
Flexible/configurable AI
RAN Restructure, CN-RAN Repartition , Turbo Charged Edge , Network Slice as a Service
(S)PTN PON
Unified RAN arch+ Common high layer protocol
(UCN, enabled by C-RAN/NGFI)
SDAI (Software Defined Air Interface)Low Freq.
New RIT
High Freq.
RIT
mMTC
RIT
uRLLC
RIT
Low-latency &
high-reliabilitySeamless wide-area
coverage
Hotspot &
high data rate
Low-power &
m-connections
5G CN
O &
M
NSSF (Network Slice Selection Function),introduced to support flexible deployment,
operation & maintenance of diverse network slices
(Apr., 2017)EC
MEC
DU
DU
CU-U
CU-U CN-U
CU-CCN-U CN-C
CN-C
超低延迟URLLC
延迟敏感eMBB
AR/VR
Industry
DU(s)CU-U CU-C
Video
CN-U延迟不敏感
eMBB/mMTC
WebCN-C
CU-C
V2X
9
9
SA
RAN
5G Services SI 5G SA1 WI
System Arch SI System Arch WI R16 WI
HF Channel Model SI
5G NR
Requirements SI
NR SI R15 NR WI R16 NR WI
2015
R15 LTE-A WI
CMCC activities:
• RANP Vice Chair, RAN2 Vice Chair; Projects in lead: 6, CMCC lead: 3, lead with partners: 3;
• 3GPP submissions: ~600; 5G patent applications:~450; Top journal/conference papers: ~100 (including 5 books )
VerticalsSDAI,UCN,SBA
2016 2017 2018 2019
3GPP Standardization Timeline (SA completed in June 2018)
Now
NSA SA Opt 4/7 (LTE-NR DC)
(NGC, and NR/LTE as anchor)
2020
10
10
A flexible, efficient, scalable and programmable
network towards to “Telecomm 4.0” era
Lead in 5G Network Arch design (Serviced based architecture agreed in 3GPP SA2 #121, May 2017, TR 23.799 )
CMCC delegate as Rapporteur for R-14 SI/R-15 WI
NovoNet: True Convergence of CT and IT
CN Transformation: Service Based Architecture, Telecom 4.0
4G Architecture by equipment: rigid network
Function split &
integrate
11
11
RAN Transformation: a Journey
2012-2013 2014
R&D on C-RAN baseband
pool1, Design, development and test on
front-end accelerator
2, First soft 4G BS PoC based on
COTS platform
3, First field trial in the commercial
networks
20162015
C-RAN PoC development1, OTA test with commercial EPC , RRU
and UE
2, Proposal of NGFI (xHaul) concept
3, Proposal of CU-DU architecture
C-RAN field trials1, Large-scale field trials in over 10 cities
2, PoC field demonstration on virtualized
C-RAN
3, Evaluation of NGFI & design of CU/DU
architecture, anchor CU for reliability
4. Established IEEE 1914 WG
5G C-RAN1, Continuous refinement on design of
CU-DU architecture and the interface
2, In-house PoC development of gNB
with CU-DU, MANO and cloud platform
3, Carrier-grade cloud platform proposal
accepted by Openstack
2017 2018
5G smart RAN1, C-RAN Alliance launched & CU-DU
architecture accepted by 3GPP
2, Proposal of RDA concept for the first
time with AI-based wireless big data
architecture
3, In-hours PoC development on cloud-
based CU-DU with demonstration with
commercial RRU&UE
O-RAN:1、RDA
2、AI
3、MEC
4、……
PCIe CPRICNRT-Linux+
Driver
CU_DU VM
SmarTileFront-
End
12
12
Option2 identified in 3GPP RAN3 #95bis in Apr 2017 [F1 interface]
CU/DU function split: 8 arch options + NG interface definition
Study on various split options and give a preferred recommendation
The proposal of NGFI leads to the CU-DU architecture
PDCPLow-
RLC
High-
MAC
Low-
MAC
High-
PHYLow-PHY
PDCPLow-
RLC
High-
MAC
Low-
MAC
High-
PHYLow-PHY
Option 5Option 4 Option 6 Option 7Option 2Option 1
RRC
RRC
RF
RF
Option 8
Data
Data
High-
RLC
High-
RLC
Option 3
Non-ideal
fronthaul
optimal option
Massive
MIMO
optimal option
Normal
antenna
optimal option
Non-ideal
fronthaul
optimal option
A project under National Science & Technology Program, “Study and demonstration of 5G FH/BH solutions” ongoing,
led by CMCC with partners of Huawei, ZTE, Fiberhome and BUPT
SDAP
Fronthaul: NGFI (xHual),Essential Enabler Element of 5G C-RAN
Function split study, since 2012
White Paper on NGFI (x-Haul) released in June 2015, http://labs.chinamobile.com/cran.
UP
RFRRU
BBU
EPC
Backhaul
CPRI
CN
CU
DU
RRU
FH-II
BH
FH-I
UP
CPcore
4G 5G
L1'
L1", L2-RT
L2-NRT, L3
13
13
IEEE 1914 NGFI (xHaul): a Solid Step towards Open Interface
subType flowID length
orderInfo
8 16 24 310
..payload bytes..
DA SA NN-NN subType RoE Payload FCSflowID length orderingInfo
RoE EthType
RoE header
Bit processing
ModulationLayer
mappingPrecoding
Resource mapping
IFFT/ CP
Bit processing
DemodulationCE &
EqualizationPrefiltering
Resource demapping
FFT/ CP
PRACH filter
CorrelationPeak detection
DA
AD
An
alog beamfo
rming
Option 7-1Option 7-2Option 7-3
SRS process
Optional
(for mMIMO)
Bit oriented
IQ oriented
IEEE1914.3
Frequency
domain IQ
IEEE1914.3
Time
domain IQ
eCPRI
Option 8
• P1914.1 TF:
- Use case, Arch and Scenarios
• P1914.3 TF: Radio over Ethernet
encapsulation & mapping
• Sponsor ballot recirculation passed and to
be approved by IEEE RevCom
• Encapsulation & Mapping applicable to any
split option, yet
• Specific objects & parameters defined for
split option 7-1, 7-2
• Enabling OPEN interface- Support I/Q in time & frequency (Option. 7-1 & 7-2)
- Legacy CPRI support
• LS to O-RAN FH WG to seek collaboration
on open FH interface specification, and
adopted by O-RAN Open FH Specification
New project under discussion
Lead in IEEE 1914 WG, the 1st SDO for NGFI Defining mapping & encapsulation of radio over Ethernet
Support open interface
14
14
C-RAN: Revolutionary Evolution of RAN
…
RRU
RRU
RRU
RRU
RRU
RRU
RRU
Virtual BS Pool
Distributed RRU
High bandwidth optical transport
network
Real-time Cloud for centralized processing
…
“CU-DU-RRU” to RAN virtualization/Cloudification
C-RAN has been deemed as a 5G essential enabling element (2011)
F1-C F1-U F1-C F1-U
gNB
Xn/X2
NG/S1-U
gNB-CU
RRC
PDCP-C
SDAP
PDCP-U
gNB-DU gNB-DU
RLC
MAC
PHY
RLC
MAC
PHY
CP UP???
CU/DU based two-level RAN Arch
• CU-DU Arch identified in RANP (Mar 2017)
• E1 SI approved in RANP 76 (June 2017)
• E1 WI approved in RANP 78 (Dec 2017)
Centralized Control and/or Processing
Collaborative Radio, Real-Time Cloud , Clean System
15
15
SDAI: Wireless Signal tailored for Diverse Use Cases
Scenarios&Services Agility & Effectively
5G SDAI:Unified Air Interface Framework
RAT & Parameter
Flexible Configure
Frame structure, Scheduling & HARQ
mMIMO Waveform Duplex SpectrumMAChannel
coding
Enabling Verticals from the PHY Layer, Flexible & Configurable
Consistent progress within 3GPP
Frame structure
Duplex
Self-contained/flexible
Coverage & CLI Interference
Semi-static
Frame structure
Unified Frame structure
Cell specific configure+L1 dynamic configure
Multiple numerologies (15kHz-240kHz])
...
one slot
CP symbol15kHz
30kHz
60kHz
Sub1GHz: 15kHz, 3.5GHz: 30kHz
mmW: 60kHz/120kHz
High Mobility: 30KHz NCP or 60kHz +ECP
Prototype of full duplex
Self-interference cancellation
capability:112dB
Bottleneck: networking solution
Dynamic TDD, small cell
BS-BS crosslink interference
UE-UE crosslink interference
Remote BS crosslink interference
D
LD
L
D
LU
LD
LD
L
D
L
D
L
D
LU
L
Optimization on Frame structure
Inter-cell coordination
Crosslink interference
measurement & management
16
16
5G DU
5G CN
AAU/RRU
NMS MANO
Open Interface &
ArchitectureMEC
Big Data
Analytics & AI
5G CU
RAN NFVI
CU-C
NGFI-I
NGFI-II
E2 E1
CU-U
Vision of O-RAN
Intelligence &
Standardization
Open Source &
Virtualization
White box
&Reference design
O-RAN: Open & Smart Ecosystem for 5G RAN (Feb 27, 2018, MWC18)
• E2, E3 Interface Standardization
• Open Interface of protocol stack
• Open Capability of Edge Computing
• Open Interface (NGFI-I/NGFI-II)
•Open-source Software,
white-box reference design
CU
DU
AAU/RRU
Intelligent
Management
• Big Data-based RRM
• Intelligent computing-
based apps
17
17
A1: btw RIC near-RT and RIC non-RT, ONAP
CU-UPCU-CP
SDAP
PDCP-U
RRC
PDCP-C
E1Multi-RAT
CU Protocol Stack
F1
NGFI-I
Orchestration & Automation (e.g. ONAP): MANO, NMS
RAN DU: RLC/MAC/PHY-high
RAN RRU: PHY-low/RF
NFVI Platform: Virtualization layer & COTS platform
Design Inventory Policy Configuration RAN Intelligent Controller (RIC)
non-RT
E
2
Radio-Network Information Base
Applications Layer
RAN Intelligent Controller (RIC) near-RT
E2 :btw RIC near-RT and CU/DU
3rd party
APP
Radio Connection Mgmt Mobility
Mgmt
QoS
MgmtInterference
Mgmt
Trained
Model
O-RAN Working Group (WG) Structure (O-RAN Founding Meeting in MWCS18)
MWCS18, Jun 27, 2018
WG1: Use cases & Overall architecture
WG2: RIC(non-RT) & A1 interface
WG3: RIC(near-RT) & E2 Interface
WG4: Open FH Interface
WG7: White-Box Hardware
WG5: Stack Reference Design & E1 & F1/V1 Interfaces
WG6: Cloudification& MANO Enhance
TSC Co-Chair
CMCC & AT&T
ORANGE & DOCOMO
AT&T & ORANGE
DT & CMCC
DOCOMO
CMCC & AT&T
•12 Board members [Operators]
- Founding Members:
AT&T/CMCC/DT/DCM/ORANGE
- New members:
Bharti Airtel/China Telecom/KT/Singtel/
SKT/Telefonica/Telstra
Board(EC inside)
Technical Steering Committee
WG 1 WG 2 WG n
18
18
Open Fronthaul Interface Spec to be released
• Target: to enable true open FH interface & Multi-vendor interoperability
• Subject: low-layer split (LLS):
• Version 1 released in April
- Ethernet/IP-based
- split option 7-2x (i.e. b/w RE mapping and beanforming)
- C/U/S-plane specified;
- eCPRI as transport encapsulation method
• Version 2 to be released soon, which would feature
- M-plane specification
- 1914.3RoE adopted as a second transport option
• Future work
- continue work on e.g. PoC, trials, test specification, certification etc.
under the auspicious of O-RAN
Source: xRAN presentation in NGMN
Methodology
19
19
Reference design is opened step by step
HW
reference
design
Low-level
driverFPGA
Code
Key Alg.
RealizedBBU
software
HW White-Box: The scale effect of the
reference design lowers cost.
White-Box Hardware to Reduce the Cost
ADC
DAC
PATx
LNA
ADC
DAC RX
PA
LNA
Digital Transceivers PAs Filter Antenna
Digital
Processor
RX
Tx
To specify and release a complete reference design of a highperformance, spectral and energy efficient white box basestation. Within the scope any kinds of design material are notprecluded, such as documentation of reference hardwareand software architectures, detailed design of schematic,POC hardware, test cases for verification & certification forall BS types and usage scenarios and so on.
White-box Small BS Demo in MWCS18
Targeting white-box small BSs trial, from end of 2018 to Q2 2019,
in Guangdong, Jiangsu, Anhui
20
20
1.14
0.64 0.60
0.35
0.21
0.39
0.06
0.37 0.28
0.34 0.32
0.00
1.00
2.00
3.00
4.00
5.00
6.000.00
0.20
0.40
0.60
0.80
1.00
1.20
Silicone_v1 Silicone_v2 Silicone_v3 CarbonFibre_v1
CarbonFibre_v2
Graphene_v1 Graphene_v2 Graphene_v3 Graphene_v4 Graphene_v5 Graphene_v6
For extreme heat conductivity Graphenes (Graphene v2), Primary Chips temperature is nearly <10%
than latest silicone Materials
For Graphene_v6, Primary Chips temperature is nearly <4% than latest silicone Materials
Exploring new Material for 5G Heat Dissipation Technology
芯片
Heat Dispersing
Thermal Conductive Pads(bottle neck)
Heat Dissipating shell
Material Typesilicone Material
(Current Use)
Graphenes
(New research)
Coefficient of heat
conductivity(W/(m K))6W/[m*K] >15W/[m*K]
Thermal Resistance Performance Comparison Test:
More tradeoff between heat conductivity and electrical Characteristics
RRU
Test Pont 1 Test Point2
RRU System Test:
Th
erm
al R
es
ista
nc
e
Bre
ak
do
wn
Vo
ltag
e
5kv 5kv 5kv
1.1kv1kv
0kv 0kv
2.5kv
1.3KV
1KV 1KV
Use in 2012 Current use Latest LatestConductor
Deformation
Breakdown
Volt.
Conductor
Deformation
Breakdown
Volt.
Conductor
Deformation
Deformation Breakdown voltage
Trade off
21
21
Open Source: LNF, Linux Foundation Networking Fund
LNF established at beginning of 2018
CMCC selected as VChair of the Board in ONS, Mar 2018
Co-Lead , on Networking slicing, O&M etc
Merger of OPEN-O and Open Source ECOMP
CMCC OPNFV Test Lab
Lead in C-RAN project On SDN Controller Platform
Big data based network data analytics
On deployment servicesModularized & scalable IO framework
Streaming Network analytics System
On L2/L3 data mining & analytics
22
22
Timeline of Open Source in O-RAN
Open-sourced CU+DU (2018)
• Modularized L1/L2/L3 function
• Shared Framework to reduce the R&D cost
• Common platform based
• Pooling gain to support the tidal effect
Open-sourced Near RT RIC (2019)
• Open-sourced RRM framework to implement embedded AI-based function block
• Unified abstraction of the function block to adapt to the complex environment
Open-sourced Non RT RIC (2020)
• Data set with unified structure to share
• Machine learning and prediction function optimization for radio network
Big data & Intelligent tools: Hadoop, Tensorflow, Torch, etc.
Enhanced NFVI in existed community: KVM, Containor, GPU, FPGA, accelerator.
2018 Q2 2018 Q3 2018 Q4 2019 Q1 2019 Q2
1. Define the overall architecture and
function splits
2. First release of O-RAN
whitepaper
1. Identify & finalize use cases &
POCs for each WG
2. First Release of FH interface,
including C/U/S/M-plan
3. Define key capabilities of NFVI
and VIM
4. Launch the open source
community for RAN network
1. Initial results of field trial/demo of
each WG.
2. First Release of A1 interface specs.
3. First Release of E2 interface,
including AI/data analytics support.
4. First release of specification of VIM
and orchestration interfaces
5. First internal release of TD-LTE BBU
framework software including
CU/DU
1. Barcelona MWC demos as well
as POCs for AI-based RIC key
use cases and NFVI
2. First Release of F1/X2
enhancement specs
3. Second Release of FH specs,
including C/U/S/M-plan
4. Finalize 2019 planning for
architecture refinements and new
use cases
5. Embedded AI based LB demo by
using open source framework
1. First Release of E1/W1
enhancement specs
2. First Release of white-
box hardware including
Component Selection
and Design Certification
23
23
Data Set
App-
lication
RT/near RT data collection
(log, buffer, event, procedure)
RRM Optimization
Data Layer : data extraction, data cleaning, data processing and storage
Mobility
management
Load balance
strategy
Multi connection
management
QoS guarantee
Beam & PC
optimization
Time-frequency
resource scheduling
& Multi-user pairing
Link adaptation
Data Link
Optimization
Network Management
& Optimization
Energy saving
Reasonable diagnosis
of small BS operation
Cell splitting
configuration
Fundamental
EnvironmentSystem-level
simulation platform
AI algrithm
(Tensorflow…)
Big data
processing
platform
(Hadoop)
Protocal stack
Optimization
E
2
E
PHY Optimization RF Optimiziation Channel Modeling
User service characteristics;
User trajectories analysis;
User value grade;
User level
Service feature
recognition;
Wireless QoS
prediction;
Service level
Cell load prediction;
User spatio-temporal distribution;
Cell KPI prediction;
Cell energy consumption analysis;
Cell level
Network coverage model
analysis;
Network parameter analysis;
Network energy consumption;
Network level
Analysis
&
Prediction
Wireless channel
environmental fingerprint
library(CQI,MCS,TA);
Interference pattern
model among UEs;
Wireless Environment
API
Cluster
ManagementAPI
MRO,MDT… XDR, BOSS…PM, NRM, Enginnering
Parameter…
PA nonlinear
optimization
Memory effect
modeling
PA joint
optimization
DPD flexible
deployment
Network slicing scheduling and related
RRM optimization(PRB allocation,etc)Parameter configuration/control in
control plane
Parameter control and optimization in
user plane(QoS,DRB,etc)
Transport layer
(TCP,QUIC,etc) send
window adjustment
Services Codec Rate
Adjustment based on
predictable network status
Special services (VR/AR,
payment etc) Scheduling
Optimization based on QoE
Network Slicing
Optimization
Cross Layer
Optimization
Retransmission
Optimization
Procedure
optimization
Architecture
optimizationData cache
optimization
SDAP QoS to
DRB mapping
Multi connection
traffic control
Space-time
correlation
modeling
3D environment
reconstruction
Scatterers to
channel
parameters
mapping
Scheduling
Optimization
Link/system-level
Autoencoder
Ideal/average/non
-ideal channel
design
Specification
impact analysis
QoE Modeling (Video, VR, Naked
eye 3D, payment,game, etc)
CMCC Research Framework of AI for 5G RAN
24
24
Wireless AI Alliance (WAIA), Aug 2017
Relying on the cooperation platform of industry-university-research, to realize the intelligent
guidance and in-depth integration of wireless big data and AI, and promote the
development of green, efficient and intelligent communication.
Alliance goals
Founders
Members
Sponsors
Requirements
WG
Architecture
WG
Tech & Field
Trial WG
Platform
WG
Standards
WG
Working Groups of WAIA
25
25
Wireless AI Alliance (WAIA)
2017.11 2018. 2 2018. 11 2019. 02
2018 MWC WP v1.1
Arch Concept Demo
2019 MWC Demo
(2nd version )
Release
Field Test Results
2st version WP
2017.08/10 2018. 05
TR on Use Case
& Requirements1st version WP
MBBF
Demo
29th Aug Future Forum
19th Oct WWRF
3GPP RAN3 SI approved
on
“RAN-centric Data
Collection & Utilization”
2018. 06
26
26
Table of Contents
Executive Summary
1 Introduction
2 Wireless Big Data
3 Use Cases for Wireless Big Data
3.1 Smart Operation & Services
3.2 Automatic Network Planning & Operation
3.3 Intelligent Network Design & Optimization
3.3.1 Network Slicing Optimization
3.3.2 Service type Recognition for Traffic
3.3.3 Customized Mobility Management
3.3.4 Context Aware Cross-Layer Opt
3.3.5 Proactive Network Resource Magnt
3.3.6 Coverage and Capacity Optimization
3.3.7 Virtual Grid Enabled Network Opt
3.3.8 User Portraits Enabled User Experience
3.4 Emerging Physical Layer Technology
4 WBD Enabled Network Architecture
5 Wireless Big Data Platform
6 Impact on the Standardization
7 Summary Data Source
Current
network data
Laboratory
simulation data
Third party
provides data
Data
Preprocessing
Data
cleaning
Feature
extractionUniform feature
representation
Off-line
transmission
Online
transmission
HDFS DBServiceZookeeper
Kafka YARN
MapReduce Spark
FTP-Server
Safety
management
System
managementPlugin API
……
Big Data
Managerment
Platform
Big Data Platform
API REST/SNMP/Syslog
Problem
Modeling
Result Show
Classic machine
learning
Deep
Learning
Reinforcement
Learning……
Data saved Online transmission
Model
display
Business
statistics show
Model
derived
……
……
……
Analysis Result ModelApplication/Algorithm
platform Layer
Big Data computing
system platform layer
Data acquisition and
pre-processing layer
Wireless Big Data Enabled Network Architecture Framework of Wireless Big Data Platform
-110dB-96dB
-98dB
-110dB
G rid-level K P I is good
G rid-level K P I is norm al
G rid-level K P I is badCA CANo-CA CA
Low quality CA Coverage
Area
High quality CA Coverage
Area
CA
Frequency 1
Frequency 2
Signalling event?
drop/HO/re-establish?
YN
Same issue of other users in
the same location?
Y
Coverage
optimiation
N
Specicial UE
related?
Y
UE issue
N
Dev
checking
YN
Bad radio condition?
Simul connected user
number excceed
threshold?Y
Load
balancing
worked?N
Product
issue
Enough data
in buffer? Y
N
N
Specific app server?
Y
App server
issue
N
Y
Y
Product
issue
N
Dev
checking
Low tput
Scheduling/Power
control issue?
CU/
MEC
Use Cases
Architecture Platform
QoE issue debugging Network Energy Saving Customized Mobility Management
Cross Layer Optimization Coverage and Capacity Optimization Virtual Grid
WP《Wireless Big Data For Smart 5G》v1.0-2017.11
27
27
WP《Mobile AI For Smart 5G-empowered by WBD》v1.1-2018.02 MWC
Understanding of Wireless Big Data
CN
OSS/
NMS/
MANO
with Big Data
AnalyticsCUDA
DUDA
SDAP
PDCP-U
RRC
PDCP-C
RLC
MAC
PHY
PCFNWDA
SMF AMF
gNB
RRC SDAP
PDCP
RLC
MAC
PHY
DUDA
UPF
SMF
UPF
gNB-CU
gNB-DU
MEC
CUDA
RDA RDA
Wireless Big Data Collection &
Feature/Model Distribution
Control Plane User Plane
China Mobile Chih-Lin I, Chunfeng Cui, Qi Sun, Zhiming Liu, Siming Zhang
Huawei Hua Huang, Yan Wang, Wei Zhou, Yixu Xu, Qinghua Chi
Alibaba Chunhui Zhu
USTC Jinkang Zhu, Sihai Zhang
BUAA Chenyang Yang, Tingting Liu
ZJU Honggang Zhang, Rongpeng Li
BUPT Wenbo Wang, Jiaxin Zhang
28
28
Wireless AI Use Cases
29
29
AI-empowered EE inmprovement
By the full integration with BD platform and taking full advantage of multidimensional data, MCES is developed
to find the low traffic cells and deactivate/activate them appropriately without any performance deterioration
• By the middle of 2018 MCES is deployed for more than 11
provinces and 300,000+ cells.
• The total energy saving is over 12 million KHW.
Deep integration with BD platform Collecting
varied data
XDR/MR/PM/CM …
Using ML to forecast the user trajectory and
service profile
Near Real time interaction with RAN by MML to achieve
precise energy saving
Data Set
MCES1.0:
• MR/PM/CM for half
year;Real time PM
every 15min
• Distributed deployment
MCES2.0:• MR/PM/CM for half
year; Real time PM
every 15min
• Cloud deployment
MCES3.0:• MR/PM/CM for half
year; Real time PM
every 15min
• Real time XDR from
S1-U for user location
and service profile
• Cloud deployment
30
30
AI-empowered Crosslayer Optimization
Application server
Transmission
layer
BS
Application Optimization
Transmission Optimization
Air interface status
information collection
Air interface status
forecast report Traffic characteristics recognition
(type,rate,delay, etc.)
QoE Modeling
QoE detection and analysis
Air interface scheduling
optimization
Core
NetworkAI engine
AI engine
1. BS sends air interface status information to facilitate
application/transmission adjustment
2. Application sends traffic characteristics to facilitate air
interface optimization
QoE
Resolution viewpoints
For example, it is
a QoE model of
naked eye 3D at
a fixed bandwidth.
The wireless network can understand the user's QoE based on
the resolution, viewpoints and current wireless bandwidth.
31
31
AI-empowered Cell Splitting and Merging of Indoor System
Requirements of Indoor Scenarios:
Static topology can not meet the requirements of high throughput and
tidal effect
Manual modification leads to high cost and low efficiency … …
CU
DU/AAU
Centralized baseband
processing pool support
dynamic topology
AI-based cell splitting and
merging solution:
• Splitting and merging pattern
design based on load
prediction
• Power parameters
optimization
DU/AAU
Cell merging
DU/AAU
DU/AAU
DU/AAU
CU
DU/AAU3
DU/AAU5
CUDU/AAU1
DU/AAU2
DU/AAU4
DU/AAU6
Field trial in
Ningbo:
•27 DUs/AAUs
•2 Cells
•Internal data
(ms/s) is collected
for data analysis
Cell Splitting
32
32
AI-empowered Network Load Balancing, tested in LTE network
Future
Requirements
D1
A
BC
D2
D3
FDD 1800
F1
F2
FDD 900
CapacityLayer
Coverage Layer
Enhancing Coverage
Layer
The effect and efficiency highly depend on the
engineer experience.
Poor portability of the traditional cell parameter
optimization
Traditional
method
Data cleaning analyzing and
feature selection
Clustering the user distribution
of cells
Generating solution using static training
result for different
cluster cells
Revise the solution
dynamically using field test
data
With dramatic growth of unlimited data user plan the
DOU doubled in 2017 compared to 2016 of China
Mobile. Up to 7 carriers/cell on different frequency
band will be collocated in a single site including FDD
LTE. How to steer the traffic in a balanced
distribution among different site confirmations
becomes a big challenge to the operation
33
33
Massive MIMO Multi-user scheduling
•Desired outcome:
Efficient UE pairing, better capacity
Multi-cell 3D-MIMO optimization
• Desired outcome:
Improved coverage, reduced interference, better spectral
efficiency
AI-empowered Beamforming
Smart UE Pairing
Interference
Data Collection UE distribution
statistics Inference
Intelligent and Cooperative BF
parameters prediction Evaluation
Adaptation
UE-specific DOA, PL, Interference (HII), CSI, RSRP(serving & neighbour cell)…
UE location information
Parameters: Az and El angles, H- and V-plane beamwidth, beam number, antenna tilt…
Cost function: outage% or gap to optimal capacity
Algorithm: RL+NN
Monitoring KPIs: No. of active RRC connections, traffic volume, spectral efficiency…
Data Collection
UE clustering using
unsupervised learning
Accuracy Enhancement
using RL
Mobility enhancement
UE’s channel response matrix, UE-specific BF matrix
UE radio characteristics
Features: Correlation between different users’ BF matrices, e.g. chordal distance
Reduced correlation computation by leveraging historic pairing information
Algorithm: DQN, for more accurate correlation profile and MIMO mode selection
Feed-forward architecture
Mobility pattern prediction enabling correlation prediction and channel prediction
UE grouping &Optimal UE pairing
34
34
Limited
application
Poor generality
Large feedback
overhead
Simple PA model
Respective modeling for different
manufactures & PA types
Large bandwidth
High cost
Local deployment Local deployment & training
Limited resources & data
Traditional
Linearization
One DA/AD to multiple PAs
in hybrid beamforming arch
Multiband & Ultra wide bandwidth
Digital and hybrid BF, Lens antenna
Big Data & AI
White-box
RRU
Generalized model:different PA types & manufactures
Cloud deployment
Flexible API encapsulation
Applicable to
multiple architectures
Low overhead & cost
AI-DPDData
collection & preprocessing
Model training &
optimization
Model deployment &
distribution
PA types
Memory properties
Nonlinear characteristic
Temperature
ML: #neurons & #NN layers
Feature selection & Time delay & nonlinear order
Training & optimization & testing
Cloud deployment
Distribution to RRUs
• Generalized DPD model
• Reduce cost
• White-box RRU
AI-DPD
Future
Requirements
35
35
SA
RAN
5G Services SI 5G SA1 WI
System Arch SI System Arch WI R16 WI
HF Channel Model SI
5G NR
Requirements SI
NR SI R15 NR WI R16 NR WI
2015
R15 LTE-A WI
CMCC activities:
• RANP Vice Chair, RAN2 Vice Chair; Projects in lead: 6, CMCC lead: 3, lead with partners: 3;
• 3GPP submissions: ~600; 5G patent applications:~450; Top journal/conference papers: ~100 (including 5 books )
VerticalsSDAI,UCN,SBA
2016 2017 2018 2019
3GPP Standardization Timeline: SI on WBD established in RAN
Now
NSA SA Opt 4/7 (LTE-NR DC)
(NGC, and NR/LTE as anchor)
2020
RAN-Centric
Data Collection & Utilization
36
36
3GPP: Towards Intelligent Network
“Study on RAN Centric Data Collection and Utilization for NR” SI approved (3GPP RAN3, Jun 2018) 【CMCC】
“Study of Enablers for Network Automation for 5G”SI approved(3GPP SA2, May 2017)【Huawei】
•Study the use cases and benefits of RAN centric Data
utilization
•Identify necessary standard impact on data collection and
utilization for the defined use cases and scenarios
•If necessary, investigate the benefits and feasibility of
introducing a logical entity/function for RAN centric data
collection and utilization
•RRM measurement
•L2 measurement quantities
•L1 measurement quantities
•Sensor data for UE orientation/altitude
Definition
•Procedure for collection from UE, L1/L2 RAN node
•Signaling procedure for distributed and centralized analysis
Collection
•SON
•RRM enhancement
•Edge computing
•Radio network information exposure
•URLLC, LTE-V2X
Utilization
NWDAF
Data Repositories
NF
Data Access
NF NF
NF
NF
NF NF
NF NF
NF
OAM
Delivery of analytics
data
AF
Delivery of activity
data
AF
13 use cases are discussed, 11 key Issues are studied
Key issue 5: NWDAF-Assisted QoS Profile Provisioning
(huawei/intel, ….)
Key issues 9: Customizing mobility management based on
NWDAF output
Key Issues 12: NWDA-Assisted predictable network performance
(CMCC, Alibaba)
general framework for 5G network automation
37
37
ITU-T Focus Group on “Machine Learning for 5G and Future Networks”
Background
Working Groups of ML5G
FG-ML5G established by ITU-T SG 13, 6-17 November 2017
Mission: The Focus Group will draft technical reports and
specifications for machine learning for future networks, including
interfaces, network arch, protocols, algorithms & data formats.
CMCC Contributions
WG 1: Input 9 Use Cases
WG2 and WG3
Actively involved in the technical discussion
regarding ML algorithms, data formats and impact
on network (esp. RAN) architecture.
WG1: Use cases, services and requirements (CMCC co-editor)
Specify important use cases, technical requirements and standardization gap
WG2: Data formats & ML technologies (CMCC co-chair)
Analyze ML technology and data formats for communication networks, with
special focus on the uses cases of WG1
WG3: ML-aware network architecture
Analyse comm. network arch from viewpoint of ML & standardization gap
1 Personalized Mobile Edge Caching
2 RAN-assisted Transmission Control Protocol (TCP) Window
Optimization
3 Machine Learning based Radio Network Planning and Radio
Resource Management for Network Slicing
4 cell splitting and merging in indoor distribution system for ML5G
5 Load balance among cells for ML5G
6 User Profile Prediction to Improve the Energy-Efficiency of
Radio Access Network
7 Machine Learning based Handover Optimization
8 Machine Learning based Link Adaptation Optimization
9 Big-data-aided channel modelling and prediction
38
38
Can WBD & AI simplify future mobile standards?
Increasing Complicated Network
Consistence High Quality
Experience
Introducing the unified flexible IT style interface to meet the diversified control and management requirements and simply the
traditional case by case interface design with enhanced flexibility
Introducing the loosely-coupled IT framework based on the unified interface to allow diversified data driven network optimization
implementation to simplify the dedicated case by case specification works
Diversified vertical services
Data Driven
Machine Learning
IT+CT+DT
Standards
AI Embedded
Efficient
Information
Model for Data
Collection
Unified flexible
IT style
Interface
Autonomous
Algorithm
upgrade
Intent Driven
Functionality
Orchestration
Open Source
De-facto Standards Standardized Framework
39
39
The Influence of AI on the Protocol Stack Standardization
PDCP: Header compression
Integrity protection
Encryption/decryption
RLC : Segmentation/Concatenation
ARQ
Reordering
MAC : Scheduling
HARQ
mandatory features:
Header compression
Integrity protection
Segmentation /Concatenation
Reordering
AI function features::Scheduling
Retransmission (optional)
Encryption and decryption (optional)
Protocol Stack of existing LTE system AI enabled higher layer architecture
Key Features for the AI empowered Protocol Stack
Flattened
Protocol LayersMerged &
Simplified
Functionality
Simplified
Signaling Flow &
data processing
More powerful
and accurate
Decision making
40
40
Source
Encoding
Channel
codingModulation MIMO OFDM
Channel
Source
decodingChannel
decoding
De-
ModulationChannel Estimation &
EqualizationDe-OFDM
Mu
ltip
le D
en
se
La
ye
r
Norm
aliz
atio
n
La
ye
r
0
...
0
1
0
...
0
f(s)
s
Mu
ltip
le D
en
se
La
ye
r
Norm
aliz
atio
n
La
ye
r
g(y)
x y
0.01
...
0.1
0.95
0.02
...
0.01
S
’
Fa
din
g &
no
ise
laye
r
channelTransmitter Receiver
( | )p y x
(c) Auto-encoder based communication system
(a) Conventional building blocks based communication system
Transmitter
Receiver
Machine Learning Module
(b) AI enabled building blocks optimization
AI-Enabled PHY Design: Facilitating Software Upgrade of Protocols
41
41
Standards Open Source
WBD enabled AI into the picture
42
42
Summary: ICDT Deep Convergence
• ‘Green & Soft’:
• 5G E2E SDX based Arch embracing Verticals
- SDX: SBA(Telecom 4.0, NFV/SDN), UCN/C-RAN, SDAI
• New Frontier: ‘Open & Smart’
• O-RAN, ONAP, LNF...
• WAIA, Wireless Big Data (Zhejiang Mobile alone ~30PB/day)
• WBD/AI: App to M to C to…
• Rethink Standards: NWDA, RDA, …
• Open Source!
• WBD/AI Impact! Simplification?
• Paradigm shift of protocol based wireless communication?