Foreword

Embracing 5.5G
to Advance the
Intelligent World
Corporate Senior Vice President
and President of ICT Sales & Service, Huawei
Li Peng

An opportunity may come only once, so we must seize it Ubiquitous connectivity is pushing the boundaries
before it slips away. of both time and space, allowing information and
value streams to flow faster and more broadly.
Over the past 20 years, informatization and digitalization Networks are delivering stronger capabilities, including
have brought opportunities worth trillions of dollars to the large bandwidth and low latency across different
ICT industry, and today, we're rapidly approaching a fully scenarios. Intelligent autonomous vehicles can operate
intelligent world. According to the Intelligent Economy around the clock; production systems can be controlled
Enabling Sustainable Growth report, the Intelligent Economy in real time; and cloud phones enable consumers to
could be worth US$18.8 trillion by 2030. As intelligence access online games and workplace applications far
continues to redefine information and value streams and beyond the limitations of local device storage and
unleash the power of ubiquitous connectivity, the ICT processing power. In wide-area IoT, connectivity density
industry will embrace new development opportunities. and precision accuracy must increase by over 10 times
to support the needs of any scenario, from smart grids
Intelligence will make the acquisition, presentation, and manufacturing, to storage and logistics.
transfer, storage, and processing of information more
efficient, redefining value streams. Huawei predicts that, The redefined information and value streams, alongside
by 2026, AI will create more than 250 billion high-quality ubiquitous connectivity, will raise the bar for networks,
images and over 70 million short videos, representing and 5.5G will be our path forward in an intelligent
exponential growth in AI-generated content. Additionally, world. In China, the Middle East, and Europe, leading
new types of connected objects and scenarios will emerge, carriers have made nonstop innovations and verified
including AI phones, digital humans, and intelligent vehicles, advanced 5.5G capabilities on commercial networks.
while new models for collaborative storage and computing Their tests cover a wide range of scenarios, including
between cloud, edge, and devices will be created. These smart connections for people, homes, vehicles, and
changes will stimulate the generation of over one trillion stadiums. Furthermore, an increasing number of
gigabytes of data traffic. 10-gigabit smart cities powered by 5.5G are popping
up across the globe.
Innovative applications are driving users to focus on
more dimensions, including uplink, QoS, and latency, Huawei is fully prepared for the first year of
creating new value. New business models centering commercial 5.5G. At MWC Barcelona 2024, we released
on metrics like bundles and SLAs are leading to new a full series of 5.5G products and solutions for a wide
monetization methods, opening up new value streams range of scenarios, along with the industry's first
worth hundreds of billions of dollars to the ICT industry. Telecom Foundation Model. They will help carriers
For example, one carrier launched a 5G livestream package provide better user experience and improve O&M
to provide guaranteed uplink speeds for seamless live efficiency.
streaming. The package's monetization model based on
multiple metrics, like uplink speeds and guaranteed QoS, The poet Kahlil Gibran once said, "Progress lies not in
helped increase the carrier's ARPU by over 70%. Another enhancing what is, but in advancing toward what will
carrier launched 5G New Calling services that provide be." Huawei will continue working with global carriers
capabilities like positioning and real-time interaction. These and partners to innovate and build today's networks
capabilities have made mobility services and car insurance for tomorrow's applications. Together, we can advance
claims more convenient, and have helped the carrier the intelligent world!
successfully monetize the B2B2C market.
 1

CONTENTS
Publisher ICT Strategy & MarketingDept.
Huawei Technologies Co., Ltd.

Presented By Zhou Jun

Consultant Song Xiaodi

Editor-in-Chief

Editor-at-Large
Xing Jingfan

Tang Xiaoqiang
Business Success Intelligent Transformation
Embracing 5.5G to Unleash 08

3
Editors Fu Dongwei
Gary Marcus Maidment Industrial Dividends Accelerating Intelligent ICT with a Four-layer 78
Art Editors Zhou Shumin Liu Kang Framework and Transformation in Three Areas
Xu Chuangliang President of the ICT Marketing
& Solution Sales Dept, Huawei Wang Su
Contributors Wang Xiangtian, Wang Yuhua, Director, Integrated Solution Marketing Dept, Huawei
Zhu Wei, Gong Tao, Liu Haofei,
Qi Fushui, He Hui, Tang Jiang, Ubiquitous Fiber Networks 14
Zhang Liqiang, Zeng Xiaofei,
with Huawei ODN 3.0 EM 2.0: A Road for the Digital Intelligent 86
Wei Yongfu, Zhao Yingnan,
Du Yanchen
Transformation of African Carriers
Zhao Maiqing
CTO, Home Broadband Solutions, Huawei Chris Meng
Director, Northern Africa ICT Marketing & Solution Sales Dept, Huawei
HuaweiTech: 5G & Rail: The Road to the 30 Wang Jie
Trends. National Championship Director, Northern Africa Marketing Dept, Huawei
Liu Jifan
Insights. He Tao CEO, Huawei Ethiopia
Innovation. Senior Consulting Expert and Marketing
Expert, ICT Sales & Service Dept, Huawei
Success. Accelerating Intelligent Transformation with 96
Application-driven Industry Collaboration
Please contact the Li Changwei
Huawei Tech editorial office to
Chief Strategic Marketing Expert, Huawei
request a copy of, contribute to,

Network Evolution
or comment on Huawei Tech.
AI-ready Cloud Drives Intelligent Digital 106
E-mail [email protected] Transformation for Carriers
Tel +86 755 89241326

2
Address G1, Huawei Industrial Base, Wang Xiaobin
Bantian, Longgang, Middle East: Leading the 5.5G 42 Chief Architect, ICT Computing Products and
Shenzhen 518129, China Era and Striding Towards an Solutions (Carrier Domain), Huawei
Publication Registration No.:
Intelligent World
Yue B No. L0230032 Tang Zhentian
Director, Middle East & Central Asia ICT
Printed By Artron Art (Group) Co., Ltd.
Address 19 Shenyun Road, Nanshan
Marketing & Solution Sales Dept, Huawei
District, Shenzhen, China
Printed On May 10, 2024 Future All-optical Network 50
Copies Printed 3000

Innovation for the Future

Architecture and Key
Copyright © Huawei Technologies Co., Ltd. Technologies
2024. All rights reserved.
Tang Xiaojun
No part of this document may be Chief Technology Planner, Optical
reproduced or transmitted in any form or Business Product Line, Huawei
by any means without prior written consent
Jointly Defining ICT Architecture for 5.5G: 116
from Huawei Technologies Co., Ltd. APN6 Enables Innovation in 58 The Key to Unlocking New Opportunities
NO WARRANTY Cloud-Network-Edge-Device Dang Wenshuan
The contents of this document are for Collaboration Chief Strategy Architect, Huawei
information purpose only, and provided "as
is". Except as required by applicable laws, Li Zhenbin
no warranties of any kind, either express Chief Protocol Expert of Huawei and Former AI: The Bridge to 6G 124
or implied, including but not limited to, the Member of the Internet Architecture Board (IAB)
implied warranties of merchantability and
Tong Wen
of the Internet Engineering Task Force (IETF) Huawei Fellow, CTO of Huawei Wireless
fitness for a particular purpose, are made in
relation to the contents of this document. To Zhu Peiying
the maximum extent permitted by applicable AI Data Lake: Breaking Silos 68 Huawei Fellow, Senior Vice President of Wireless Research, Huawei
law, in no case shall Huawei Technologies and Accelerating Intelligence Ma Jianglei
Co., Ltd. be liable for any special, incidental, Technical Vice President of Wireless Research, Huawei
indirect, or consequential damages, or lost Michael Fan Chen Yan
profits, business, revenue, data, goodwill Director, Data Storage Marketing Senior Expert of Wireless Research, Huawei
or anticipated savings arising from or in
Execution Dept, Huawei
connection with any use of this document.

This publication is printed using eco-friendly

paper and ink.
01
Business
Success
Business Success Business Success

5G has achieved significant success since 1-Gbps uplink rate, and a 10-fold im-
its commercial use. By the end of 2023, 294 provement in network capabilities
commercial 5G networks had been deployed
worldwide, serving more than 1.4 billion 5G Evolution towards 5.5G is a continuous process. In
subscribers and establishing 5G as a major engine the first phase, TDD 3CC aggregation above 260
for carriers' revenue growth. In 2023, 90% of MHz is able to achieve a peak rate of at least 5
carriers that invested in 5G saw revenue growth Gbps. This has enabled wireless networks to deliv-
and achieved positive business cycles. 5G has er a deterministic experience for people-connected
helped industries go digital, and has been used at services for the first time. In the second phase, a
scale in more than 10 high-value scenarios, includ- downlink peak rate of 10 Gbps can be achieved.
ing 5G-enabled machine vision and 5G-enabled
remote control. 5G has also become a new engine 5.5G networks that deliver a deterministic expe-
driving the development of the digital economy. rience will significantly improve consumer per-
ception. Voice and 2D videos used to be the main
The industry is undergoing tremendous chang- methods of content interaction, but immersive 3D
es, making it essential to upgrade connectivity videos, XR, and holography are developing into
technologies. The rapid development of gener- the major forms of future content interaction. This

Embracing 5.5G
ative AI presents huge opportunities, driving the upgraded experience will raise the bar for connec-
exponential growth of traffic. tivity. Previously, services have generally required
a latency of 100 ms, but the latency required for

to Unleash In the B2B domain, 5G has been integrated with

core production activities and glasses-free 3D tech-
future services will be as low as 10 to 20 ms. This
tenfold decrease in latency means that 5.5G net-

Industrial Dividends
nology has reached an inflection point in terms of works enabled by 3CC aggregation will be integral
experience. The emergence of new services, con- to network deployment in order to provide a bet-
nections, and experience is bringing unprecedent- ter experience. The high bandwidth delivered by
ed opportunities, driving leading carriers to shift 3CC aggregation will give full play to its strengths
from traffic monetization to the monetization of in transportation hubs, such as high-speed railway
Liu Kang deterministic experience and convergent new ser- stations, subway stations, and airports, and venues

President of the ICT Marketing & Solution Sales Dept, Huawei vices. These changes place higher requirements on like commercial offices, stadiums, tourist attrac-
the capabilities of existing networks. tions, and urban villages.

5.5G is the natural evolution of 5G. 5.5G In the home market, FWA will continue evolving
delivers higher speeds, lower latency, higher towards FWA2, leading to the emergence of three
5.5G is the natural evolution of 5G. With a tenfold increase in
network reliability, wider connectivity, and new application scenarios. First, FWA Pro can
network capabilities, 5.5G is becoming hugely popular among
native intelligence, allowing it to meet the deliver a bandwidth of 1–2 Gbps, meeting the
industries. At the 5G Advanced: Completing the Enterprise
higher requirements of communications demand for high-speed services while providing
Opportunity forum during MWC Barcelona 2024, Huawei
networks in the future. a deterministic experience. This has allowed
proposed that 5.5G will enable industries to accelerate both
home users to enjoy better services, including
digital and intelligent transformation.
5.5G supports multi-carrier aggregation, ultra-HD videos and VR applications. Second, the
delivering a 10-Gbps downlink rate, cost-effective FWA Lite is primarily intended for

8 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 9

 Business Success Business Success

markets evolving from 4G to 5G and new markets based on the telecom foundation model. These • 5.5G marks the first time that integrated 5.5G improves 5G IoT capabilities and
without connectivity. This solution will help carriers facilitate wireless intelligent alarm management, sensing and communications have been enables IoT across various scenarios,
obtain new users, but help release 4G spectrum accurate alarm identification, quick fault location, possible, meaning that 5.5G can deliver both making all things connected possible
resources to 5G as home users migrate from 4G and fault prediction and prevention. With this communications and sensing capabilities. The
to 5G, thus improving network efficiency. Third, solution, carriers can shift from responsive O&M to large-scale deployment of commercial 5.5G 5.5G drives IoT development, as it is capable
FWA Biz is a solution for enterprises across different predictive and preventive O&M, achieving "zero" networks can deliver full-area, grid-based, and of breaking the limits of 5G in some scenarios.
industries. Built on FWA, the solution can achieve network faults. low-cost detection capabilities. Compared with 5.5G also improves network performance and
99.99% network stability and 20-ms low latency. • 5.5G-enabled intelligent network optimization: traditional radars, 5.5G networks have the connectivity and reduces costs, meaning that
This has made it easier for carriers to replace Intelligent NEs have been introduced to following advantages: IOT-related goals that cannot be achieved with
microwave private lines and low-speed copper implement intelligent resource scheduling, • Spectrum sharing: The allocation of 10% 5G become achievable. This makes all things
private lines, accelerating the interconnectivity of optimize the experience and capacity of multi- of spectrum resources for sensing and 90% connected possible, enabling carriers to maximize
small- and medium-sized enterprises. band and multi-site heterogeneous networks, for communications maximizes spectrum the commercial value of 5.5G.
and maximize spectrum efficiency. Features such resources.
With 5.5G, carriers can monetize network capabil- as topology, time sequence, and grids have also • Base station sharing: A wide-area native • 5G RedCap and 5.5G RedCap intrinsically
ities across more dimensions and provide differen- been integrated to create a unified and general sensing network can be built based on support 5G capabilities like large uplink
tiated packages based on deterministic experience. performance prediction model, resulting in a millions of base stations. rates, positioning, and network slicing.
Carriers are now able to meet diversified customer lower number of accumulated errors of multi- • Low cost: Communications can share most These light weight solutions provide
demands and increase their ARPU. objective serial prediction, and the accurate and deployment costs. medium- and high-speed connections that
fast prediction of multiple KPIs. By employing • Network collaboration: The blind spots of meet the high SLA requirements of industry
5.5G enables native intelligence, meaning the solution, carriers can realize optimal network traditional radars can be overcome by multi- unified communications, such as smart
networks will be capable of self-optimi- performance and energy efficiency, meaning site collaboration. power distribution, smart city monitoring,
zation and self-management to better carriers can maximize energy saving results while • 5.5G can be applied in a variety of sensing and industrial sensing, and cut deployment
adapt to different scenarios and service guaranteeing stable network performance. scenarios, including low-altitude unauthorized and usage costs.
requirements • 5.5G-enabled intelligent service operations: drone flight warnings and route supervision; • The RedCap industry ecosystem is rapidly
User-level dynamic simulations help implement offshore sea-surface monitoring; inland wa- maturing. In 2023, eight RedCap modules and
As wireless networks continue to evolve, they will precise network planning based on coverage, terway management; vehicle and pedestrian more than 30 types of RedCap devices (DTUs,
be able to deliver a greater range of services. Net- speed, and latency, and enable fast service sensing; the sensing of high-speed railways cameras, and CPE) were launched worldwide.
works will also inevitably become more complex, provisioning, which can meet differentiated service and related perimeter intrusions; border It is estimated that over 100 types of RedCap
bringing challenges for carriers in terms of offering SLA requirements. Real-time, dynamic resource intrusions; thrown objects; meteorological devices will be launched for commercial use
simplified O&M and lower energy consumption scheduling can be achieved based on prediction monitoring and predictions; the monitoring of in 2024, and the average cost of a RedCap
while delivering diversified services and guaranteed capabilities guaranteeing smooth service micro-deformation of buildings, bridges, and module is expected to drop from US$30 to
service experience. The 5.5G intelligent solution, experience and service-based network adaptation mountains; and crop monitoring. US$20 this year. Furthermore, the average
IntelligentRAN, will help carriers build autonomous when faced with challenging scenarios such as • Based on the native sensing and communications cost of RedCap-based data collection devices
driving networks that feature intelligent and sim- service fluctuations, burst traffic, and large events. capabilities of 5.5G networks, carriers can fully has decreased to just US$150. These costs will
plified O&M, intelligent network optimization, and utilize the strengths of cloud, computing, and continue to fall as millions of RedCap devices
intelligent service operations. 5.5G marks the first time that communi- connectivity to fill in the gaps of industry capa- start to be shipped from China.
cations capabilities have been expanded bilities. They can also provide new and better • 5.5G P-IoT features large bandwidth, continuous
• 5.5G-enabled intelligent and simplified O&M: from connectivity to integrated sensing and key solutions to avoid homogeneous competi- networking, long-distance coverage, and low
The solution provides capabilities like wireless communications, opening up new market tion, greatly expanding carriers' business bound- power consumption, allowing it to support
FME copilots and key service assurance agents opportunities aries and creating new business opportunities. IoT service scenarios that require speeds of

10 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 11

Business Success Business Success

10 Kbps or lower. Such scenarios include Africa, with more than 20 carriers now providing improve efficiency, vehicle-road synergy requires Leaders Summit in December 2023, experts and
warehousing, production and manufacturing, 5G services for mines. 5.5G technologies bring networks with 99% reliability and 20-ms latency. representatives from authoritative organizations
logistics, and remote monitoring. new momentum to the digital upgrade of This is where 5.5G comes in. New E2E technologies and enterprises jointly announced the first
industries, with 5.5G supporting ultra-large uplink have already been verified in Shanghai, where a year of 5.5G deployment. These organizations
5.5G further unleashes industry produc- rate of 1 Gbps and latency as low as 4 ms. These 1-km demonstration route boasts reliable 5.5G include ITU, du and e& (UAE), Ooredoo (Qatar),
tivity and supports more industry sce- features can support enterprises' core production networks with stable latency. With vehicle-road Vodafone Oman, Huawei, Ericsson, and Nokia.
narios, creating greater economic value processes such as high-density quality inspection synergy supported by 5.5G's sensing capabilities, Over the past year, leading carriers from the
and flexible production, greatly enhancing the drivers no longer need to worry about complex Middle East, Europe, and Asia Pacific (including
5.5G is expanding private network services from productivity of enterprises. In the mining industry, traffic situations, while accidents caused by China), have fully verified TDD 3CC on live
industry private lines only to private lines and for example, the length of a fully-mechanized blind spots and other unexpected factors can be networks and started commercial deployment.
ICT services, thus accelerating industry digital mining surface in a coal mine ranges from 100 to prevented. It is forecast that traffic management
transformation. The 5G private network and 300 meters. Previously, it was difficult to clearly efficiency will improve by 30% and average Close industry collaboration and ecosystem
private line market is developing fast and is set see the entire surface. However, 5.5G wireless commute times will be 20% faster. collaboration are also key to unleashing the
to reach US$20 billion by 2025. 5G private lines access technology can be employed to provide value of 5.5G. Huawei stands ready to work with
are extensively used across industries thanks network coverage for underground mining Global carriers have reached an industry-wide industry and ecosystem partners to explore its
to fast, low-cost deployment. 5.5G can further surfaces and areas in which explosions are set off, consensus and are actively deploying 5.5G. extensive applications and provide end-to-end
improve the network performance of private solving issues related to fiber disconnection and During the Global Mobile Broadband Forum intelligent 5.5G solutions. Together, let's usher in
networks, including guaranteed 300 Mbps data collection failures. Low-frequency and large- in October 2023, 13 world-leading carriers a new era of 5.5G-enabled intelligent connectivity
and millisecond-level latency. 5.5G can also be bandwidth resources are used to backhaul videos jointly announced the launch of the Global 5.5G and applications so that all participants will
integrated with networking, computing, cloud, IoT, from more than 100 cameras, and AI is used to Network Pioneers, marking 5.5G's transition benefit from the development of 5.5G and
and other industry requirements to open up more stitch these videos into a panoramic video of the from technical verification to commercial achieve shared success.
possibilities for industry digital transformation. 5.5G mining surface, providing a comprehensive and deployment. At the 17th Telecom Review
supports an uplink transmission rate of 1 Gbps clear picture of underground operations.
for a single user. This can meet the requirements
of interactive, immersive services such as AI 5.5G expands IoV applications from
training data cloudification, cloud photos, and in-vehicle entertainment to vehicle-road
cloud conferencing. In industrial production, 5.5G synergy, accelerating the arrival of
can enable services such as AI quality inspections, smart transportation
security monitoring, and remote control. And in
healthcare, the low latency and high reliability of As the automotive industry becomes more
5.5G can make telemedicine applications possible. autonomous and electric, the number of
connected vehicles worldwide is soaring. By the
With the enhancement of 5.5G capabilities, end of 2023, the number stood at 350 million,
related technical applications are expanding from and 45% of all vehicles on sale today support
enterprises' auxiliary systems to core production autonomous driving at L2 or higher. These trends
systems. 5G serves as an enabler and accelerator will see the traffic of connected vehicles soar
in the digital transformation of various industries. by 100 times, and the data of usage (DOU) in
For example, 5G has been applied in more vehicles to exceed 100 GB. These developments
than 400 mines in countries around the world, will require the support of high-speed 5G
including China, Thailand, Brazil, and South networks. To ensure transportation safety and

12 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 13

Business Success Business Success

Ubiquitous Fiber Two key transformative technologies

in ODN 3.0
1. Uneven optical splitters: Efficient
connections and reduced costs

Networks with In the earliest FTTH solution, ODN 1.0 optical

splitting was used for optical splitters, while
An uneven optical splitter (as shown in Figure
1) unevenly splits 100% of optical power signals

Huawei ODN 3.0

fusion splicing or mechanical splicing was from COs, with 70% of output allocated to
reserved for fiber connections. In 2015, some remote ends and 30% retained for local users,
vendors implemented drop cable pre-connection and then evenly splits the portion of optical
by connecting fiber drop cables to fiber access power between local users. Although only 30%
terminals (FATs), forming the ODN 2.0 solution. of optical power is retained for local users, it is

Zhao Maiqing Since 2018, based on ODN 2.0, Huawei has sufficient to meet local users' needs.
gradually realized pre-connection between
CTO, Home Broadband Solutions, Huawei
distribution optical cables and level-2 optical
Remote end
splitters, uneven optical splitting of level-2 optical CO
70% optical power

1:2
splitter FATs, and pre-connection between fiber 100% optical power
30% optical power

feeder cables and level-1 optical splitters. This

With Huawei's core concept for ODN construction has resulted in a comprehensive solution that
1:8
centering on full and dense coverage coupled with implements full pre-connection, cascading, and
short and easy access, Huawei's ODN 3.0 solution uneven optical splitting technologies, culminating
uses two transformative technologies to support To local-user ONTs
in the ODN 3.0 solution.
five typical network scenarios.
Figure 1: Uneven optical splitter

FAT 1 FAT 2 FAT 3 FAT 4

70% optical power 70% optical power 70% optical power 100% optical power
CO
1:2 1:2 1:2
30% optical power 30% optical power 30% optical power
100% optical power

1:8 1:8 1:8 1:8

To local-user ONTs To local-user ONTs To local-user ONTs To local-user ONTs

Figure 2: Four-level cascade with uneven optical splitting

14 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 15

Business Success Business Success

Several uneven optical splitters are usually Generally, in an ODN project, costs related to Another way is mechanical splicing, which Pre-connection (Figure 4) means fiber sockets
cascaded together. For example, for four-level optical splitter material and installation account is simpler than fiber splicing. However, poor are reserved outside an FAT, and a fiber plug
cascading, as shown in Figure 2, the first three for less than 15% of total investment, while 85% connection quality can result in faults or is reserved in the factory, so that the plug can
splitters perform uneven optical splitting, and of investment relates to optical fibers. excessive signal attenuation within just three easily be fixed to the socket at the construction
the fourth performs even optical splitting. This years. Furthermore, both methods require device site. Pre-connection ensures that FATs never
means 100% of the optical power from the CO Therefore, uneven optical splitting is an cover plates to be removed from boxes, meaning need to be uncovered. Of course, reliable pre-
is retained for FAT 4 users. Although the third advanced and revolutionary development for exposure to the weather and compromised connection and adequate optical attenuation
and fourth splitters receive only a relatively FTTH network solutions. protection of components over time. Most ODN requires firm conjunctions, meaning both
small amount of CO optical power, it is still faults actually occur at these junctions. A typical the sockets on devices like FATs and the pre-
sufficient to meet the needs of the local users. 2. Pre-connection: Reduced faults for ODNs example is tail-end optical splitters inside FATs. connection plugs of optical fibers must strictly
Frequent removal of FAT cover plates during be manufactured to a high standard.
Uneven optical splitters are advantageous in that An FTTH ODN is a network of optical fibers onsite installation and repairs often leads to
they significantly reduce the number of optical connected to different devices, such as optical faults over time from compromised protection.
cables required. Only a single-core optical cable splitters, FATs, and optical cable junction boxes.
is needed between the four optical splitters and One way to connect these fibers is using
Level-2 optical splitters
between the optical splitters and the CO. By fiber fusion splicers onsite, but fiber splicing User
FAT FAT terminal
contrast, an even optical splitting solution would requires highly skilled technicians and can be CO
Drop cable
ONT
require four optical cables, as shown in Figure 3. very difficult if operation sites are unfavorable. FDT Level-1 optical
splitter
Distribution
cable 1
OLT ODF FDT
FAT 1 FAT 2 FAT 3 FAT 4

Distribution
cable 2

1:8 1:8 1:8 1:8 Feeder section Distribution section FDrop section

HP (ODN) HC (drop)
1
2
3 Figure 5: FTTH network
4

Figure 3: Four-level cascade with even optical splitting

Single-core FAT MPO HUBBOX Blade Nano

pre-connection cable pre-connection cable MA5800 F01M500A MA5800-X2 F01M50
OLT OLT

Figure 4: Pre-connection devices Figure 6: Different types of OLTs

16 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 17

Business Success Business Success

OLT options in FTTH networks optical network terminal (ONT), should be a 3. Compact OLT mean sparser coverage and longer access lines,
maximum of 6 kilometers. leading to repeated construction that wastes
An FTTH network (Figure 5) consists of three • Blade OLT: OLTs of this kind are compact both time and investment. Generally, a 30%
parts: OLT, ODN, and drop cable segment. 2. Outdoor OLT and weigh about 13 kilograms. These OLTs take-up rate within three years after an FTTH
Generally, the ODN part is referred to as home can be installed on a pole or in an open project is completed is already considered fair
pass (HP), and the drop part is referred to When no indoor equipment room is available, space within a base station. They can carry performance. For a large-scale, multi-area FTTH
as home connection (HC). HP consists of the an outdoor OLT cabinet is typically installed in 1,000 FTTH users each, or 2,000 FTTH users network construction project, a total take-up
feeder section and distribution cable section. an open outdoor space, preferably where base when two units are installed back to back rate of over 50% within five years of project
Several types of OLT can be used depending on stations are located. A dedicated concrete base and share two uplink optical fibers to the CO. completion would be considered impressive
the actual scenario (Figure 6). will be built in advance for such OLTs to keep • M A 5 8 0 0 -X 2 : T h i s OLT m o de l ca n be performance. Some carriers require their local
water out. Each outdoor cabinet houses only installed inside a mini outdoor cabinet which business units to achieve a take-up rate of more
1. Indoor OLT one OLT, typically covering 4,000 to 6,000 users. is then fixed at a base station or street cabi- than 70%. This is incredibly difficult to achieve
WDM and IP equipment can also be installed in net to support up to 2,000 users per unit. when they are developing the business at scale.
Applicable to COs or indoor base stations, each outdoor cabinets as required. Power supply and • Nano OLT: This type of OLT provides only
OLT can typically carry 4,000 to 6,000 users storage batteries are required in the cabinets, four GPON ports and can carry 256 Household Number of FAT Ports
(in an area)
and each OLT rack can house two or three while natural air cooling, fan-based air cooling, broadband users. It is smaller and lighter Penetration =
Number of Households
OLTs. Generally, the coverage radius, that is, or dedicated air-conditioning can be selected than other OLTs, and can be installed inside Rate of Ports (in an area)
the distance between an OLT and the farthest based on local temperatures. a mini outdoor cabinet which is then fixed
outdoors. Nano OLTs are particularly suitable Household penetration rate of ports means the
for rural areas, as they are an extraordinarily number of optical fiber ports to be deployed
low-cost solution for areas with small for every 100 households in an area. Carriers
populations and low household density. that emphasize lower CAPEX will reduce the

Uneven optical splitting is an advanced number of ports planned for the first phase of

and revolutionary development for Core concepts in ODN design construction projects, resulting in low household
port penetration rate.
FTTH network solutions.
1. Another look at two carrier KPIs
The CAPEX of an FTTH project typically compris-
Number of Subscribers es construction labor (60%), optical cable (25%),
(in an area)
Port Take-up Rate = devices like optical splitters (15%), and minor
Number of FAT Ports
(in an area) costs such as design and supervision. Labor cost
clearly accounts for the majority of the total
Traditionally, carrier performance is evaluated cost. A low household penetration rate of ports
according to take-up rate. This is problematic will mean two, three, or even more phases of
because carriers tend to require high port- construction are required over just one or two
take-up rate soon after project acceptance. years, with each phase resulting in basically
Setting a high port-take-up rate as a KPI for the same labor cost. Three more phases of
the first two years after project completion will construction mean three times the required
inevitably result in fewer ports being planned investment compared with everything being
for the initial phase of the project. This will done at once.

18 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 19

Business Success Business Success

Therefore, a 70% or higher household Dense coverage means a 60% or higher users at scale. High-quality network 3. Drawbacks of sparse coverage and long
penetration rate of ports is recommended to household penetration rate of ports. This construction will mean efficient device access
be achieved following the very first phase of reflects carriers' expectations for and installation, low fault rate, good user
FTTH construction. confidence in the market. experience, and a good service brand. Sparse coverage and long access are the
main reasons for poor ROI for global carriers
2. Full and dense coverage, short and easy Dense coverage can improve ROI, shorten • Easy access: Quick installation in FTTH deployment over the past decade or
access project duration, and ensure high network and troubleshooting so. Repeated construction in the same area is
operations quality, while enhancing con- The end-to- end pre- connected FAT the primary reason why some carriers fail to
• Full coverage: Key to one-time fidence in market development, reducing solution eliminates the need for high- succeed in FTTH, as this wastes investment,
holistic planning management costs, and reducing competi- altitude fiber splicing while decreasing time, and market opportunities.
One-time holistic planning should be tors' market expectations. the failure rate caused by fiber clamping.
performed for small- and medium-sized When a FAT is faulty, it can simply be • Sparse coverage means wasted investment
cities to achieve full coverage in one de- A 60%–100% household penetration replaced. This ensures quick installation for secondary construction
ployment, without requiring secondary rate of ports is recommended, as the and troubleshooting for FATs. Sparse coverage in the initial phase
design and construction. The FTTH con- number of FAT ports in an FTTH network
struction project of a city should have as is a rational standard for measuring
many areas as possible contiguously and a carrier's network capabilities, which
fully covered, again in one deployment. represents medium- and long-term market
This is especially true for traditional fixed development potential. This is intended
network carriers, because they need to mi- to support development over at least the
Huawei's core concept for ODN construction
grate existing xDSL users across all areas next five years, and carriers should not
centers on full and dense coverage coupled
to FTTH. expect a high take-up rate in the first year
of project completion. with short and easy access.
Full coverage is required for user volume
to ramp up across an area or city and for • Short access: Improving FTTH
carriers to build their brand influence, installation efficiency
because existing users will pass on their For areas with average resident density,
experience to potential new users through 8-port FATs are recommended. For areas
word of mouth. with low resident density, 4-port or 2-port
FATs can be used. The average access
By contrast, interspersed coverage may distance between an FAT and a household
re s u l t i n c o n s u m e r c o m p l a i n t s a n d ranges from 30 meters to 80 meters.
negative market effects in non-covered
areas. This may encourage competitors to Shorter access ensures efficient installation.
fill coverage vacancies through repeated Broadband installations per capita per
construction and compete for users. day is a major factor affecting the scale of
carriers' user development and total HC
• Dense coverage: Improving return on installation costs. FTTH networks are a
investment and market confidence key foundation for developing broadband

20 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 21

Business Success Business Success

of construction results in a huge waste • Long access leads to high installation costs total installation cost resulting from high FATs should be located as close to consumers as
of investment for a second phase of Figure 8 shows a typical case of long access. labor costs of HC installation. The criteria possible. For example, an 8-port FAT, which can
construction. Due to a sparse coverage approach, users in for evaluating FTTH network construction connect to eight households, should be located
the area are covered by only FAT 1, deployed involves how many broadband lines an with the shortest average distance to the eight
Figure 7 shows a typical case of secondary during the first phase of construction. When engineer can install per day. Therefore, it households it connects to.
construction. The total investment users 4, 10, and 11 subscribe to the broad- is recommended that the average distance
required to achieve a 70%–90% household band service, their access to an FAT can only between an FAT and a household be a The capacity of a hub box should first be iden-
penetration rate of ports over two phases of be delivered over a long distance. maximum of 100 meters. tified to determine the maximum number of
construction is twice that required to achieve FATs that can be connected to it. For example,
this over a single phase. Some carriers en- As FAT 1 ports become almost fully loaded, Three keys to ODN design a hub box that can cover eight groups of FATs,
counter a shortage of available ports and a FAT 2 is installed through second-phase with four FATs in each group, should be locat-
budget crunch despite constant construction construction to address additional user 1. Locations of level-1 splitters (hub boxes) ed with the shortest average distance to the
projects in virtually the same area year af- needs. However, users 4, 10, and 11 who and level-2 splitters (FATs) eight groups of FATs it connects to.
ter year. Typically, this situation is primarily were connected to FAT 1 cannot be migrated
caused by the dogmatic practice of evaluat- to FAT 2.
ing performance according to take-up rate,
FAT 1 and FAT 2 separately deployed over two construction phases (sparse coverage, long access)
resulting in a shortage of upfront investment When FAT 1 is fully loaded and no idle port
and subsequent repeated construction. is available, new users near FAT 1 (e.g., user
User 4
Generally, during the second and third 7) can only be connected to FAT 2. In this User 1
User 2
User 5 User 3
phases of construction, identifying idle fibers case, the HC installation distance is very long, Phase-2
construction
in fiber distribution terminals (FDTs) is more significantly increasing installation costs.
FAT 1 User 6 Flexible drop cable FAT 2 User 9
time-consuming than laying new optical Phase-1
construction
cables. Some carriers end up having as many Deploying the two FATs at the same time
as three to five phases of construction in the can prevent the long-access problem caused User 8
User 7 User 10 User 11
Flexible drop cable
same areas, wasting huge investment. by sparse coverage and address the high

Two FATs deployed at one time (dense coverage, short access)

Cost of traditional ODN User 4

solution: User 2
User 1 User 5
Optical splitters (15%) User 3 Optimized solution:
1 2 1 1 2 2 1 Optical cables (25%) One-off construction
Feeder
FDT 1 Labor (60%)
cable
Level-2 splitters Optimized solution: User 6 User 9
FAT 0 FAT 0
Level-1 splitters FAT (level-2 splitter) One-off construction
deployment

1 Phase-1 construction
1 2 2 1 2 2 1 Flexible drop cable User 8 User 10 Flexible drop cable
2 Phase-2 construction User 7 User 11
FDT 2

Figure 7: Investment wasted due to secondary construction Figure 8: Long access increases the HC installation cost

22 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 23

Business Success Business Success

Of course, the exact locations of FATs and hub required in the feeder section, distribution cable pipes requires enormous investment and makes Overview of the ODN 3.0 solution
boxes are the result of a comprehensive analysis section, and HC section. future maintenance more difficult.
of onsite conditions. 1. Solution for high-density, low-rise,
Neither putting level-1 splitters inside COs nor The feeder cables between COs and level-1 overhead scenarios
This is the only way to ensure the minimum installing multiple level-2 splitters in a street splitters are considered scarce resources, so using
number of optical cables used between cabinet is advisable, as both approaches would the minimal number is recommended. As shown in Figure 9, this solution consists of
hub boxes and a CO, as well as the shortest result in unsuitably long distances between the three types of boxes (hub boxes, sub-boxes, and
distances between hub boxes and FATs and FATs and consumers' homes — far more than Given the fixed number of optical cables end boxes) and three types of optical cables
between FATs and consumer homes. Ensuring 100 meters. Laying a large number of drop between level-1 and level-2 splitters and (MPO cables, single-core distribution cables, and
this minimizes the optical cable investment cables inside street cabinets and underground between level-2 splitters and ONTs, their lengths drop cables). Each hub box performs 1:2 optical
should be minimized. splitting on the fiber cores of the four GPON
ports and outputs eight links. Sub-boxes use
2. Leveraging existing network resources uneven optical splitting and can consist of four
HUBBOX 1
2
1 cascaded FATs, with each connected to eight
FAT 1 FAT 2 FAT 3 FAT 4
Single-core ONT 8
distribution cable 1 COs, wireless base stations, and existing FDTs households. For areas that are far away from an
≤2Km Single-

…
core
… Drop cable
already have the power supply and optical cable OLT, a regular optical cable and an MPO X-box
…

Single-core ONT 1
2
1
12-core distribution cable 8 resources required for FTTH networks, such as can be used. X-boxes can be directly connected
MPO cable Single- 3
OLT
core
outgoing optical cables from COs and redundant to both 12-core pre-connected optical cables
HUBBOX 2
optical cables between base stations or optical and regular optical cables for output. This
Single-core
distribution
FAT 1 FAT 2 FAT 3 FAT 4
ONT 8
nodes and COs. Deploying OLTs where these solution uses end-to-end pre-connection, and
cable 1 Single-
…
core existing resources can be leveraged will help is suitable for overhead-deployment scenarios,

…
HUBBOX 3
ONT 1 reduce the need to deploy new optical cables with high installation efficiency and low fault
in the feeder section, which is typically long rates.
Single-core FAT 1
distribution
FAT 2 FAT 3 FAT 4
ONT 8 and comes with a large number of fiber cores,
cable 1 Single-
… core and will help reduce additional investment in 2. Solution for overhead deployment in low-
…

MPO XBOX
ONT 1 deploying the power supply. density areas
HUBBOX 4

FAT 1 FAT 2 FAT 3 FAT 4

2~5Km ONT 8 3. Replacing traditional ODNs with ODN 3.0 In this solution, as shown in Figure 10, each hub
Single-
core
box has a 1:2 optical splitter built in. FAT 1–FAT
…

48-core/96-core regular …
optical cable ONT 1
The technical design and construction of ODN 3 use uneven optical splitting with eight ports,
12-core
FAT 1 FAT 2 FAT 3 FAT 4
MPO
ONT 8 3.0 are simple and efficient, and the ODNs while FAT 4 uses even optical splitting. The three
cable 1
Single-
core can run stably with low fault rates, reducing types of optical cables involved are 4-core pre-
…

ONT 1
ODN maintenance costs. If the fault rate of connected cables, single-core distribution cables,
FAT 1 FAT 2 FAT 3 FAT 4
ONT 8
carrier ODNs that are built earlier is too high and single-core pre-connected drop cables. If
Regular
Single-
optical
core to keep maintenance costs down, then carriers the covered area is close to the CO, a 4-core
…

cable 1
ONT 1 can consider replacing legacy ODNs through pre-connected cable can connect the four GPON
maintenance investment. ports of the OLT with four hub boxes, thus
Figure 9: Solution for overhead deployment in high-density, low-rise facilities supporting up to 256 households in total.

24 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 25

Business Success Business Success

3. Solution for multistory buildings with with four cascaded FATs, covering up to 4. Solution for underground deployment FATs can be used depending on the household
32 households in a building unit. With eight density in the area. Pre-connection is used
This solution, as shown in Figure 11, uses a building units covered, where each unit has four In this scenario, it is recommended that level-1 between FATs and households, and OneTB's
traditional feeder cable connected to a hub box stories and each story has eight households, up optical splitters (hub boxes) are installed in (underground) are deployed outside household
in the building. Each hub box has four built-in to 256 households can be supported in total. street cabinets. If possible, the coverage area doorways. Pre-connected cables used between
1:2 optical splitters to output eight links, each of level-1 splitters should be narrowed so that FATs and OneTB's can be vertically deployed
hub boxes are as close to the covered FATs as in the same trenches as distribution cables. In
possible to shorten the distance between them. addition, protection pipes that lead all the way
to household doorways can be added for these
2
HUBBOX 1
≤2Km
1
Single-core
FAT 1 FAT 2 FAT 3 FAT 4 3
ONT 8
The FATs shown in Figure 12 are level-2 splitters. cables, where a trench is dug horizontally, and
distribution cable 1 Single-
core Drop cable
It is recommended that these FATs are installed the cables are connected to OneTB's.

…
ONT 1 on walls inside manholes. 8-port or 4-port
Single-core 2
4-core pre- distribution cable 2
1 Single-
connected 3
cable 1 core
OLT
Unit 1 Unit 8
Single-core FAT 1 FAT 2 FAT 3 FAT 4
4-core pre- ONT 8
connected
HUBBOX 2 distribution
cable 1 Single- Indoor unevenly Indoor unevenly
cable core distributed FATs distributed FATs

…
Transparent optical cable Transparent optical cable
ONT 1
4-core pre-
Single-core FAT 1 FAT 2 FAT 3 FAT 4
Floor 4 1:8 1:8
connected
HUBBOX 3 distribution ONT 8
cable
cable 1 Single-
core

…
ONT 1

Single-core FAT 1 FAT 2 FAT 3 FAT 4

HUBBOX 4 distribution ONT 8 Transparent optical cable Transparent optical cable
cable 1 Single-
core

…
Floor 3 1:8 1:8
ONT 1

MPO XBOX
4-core pre- FAT 1 FAT 4
connected HUBBOX 1 ONT 8
cable
…

2~5Km
ONT 1
Transparent optical cable
… Transparent optical cable

48-core/96-core regular Floor 2 1:8 1:8

optical cable FAT 1 FAT 4
HUBBOX 2 ONT 8
4-core pre-
…

connected
cable ONT 1

FAT 1 FAT 4 Transparent optical cable Transparent optical cable

HUBBOX 3 ONT 8

Floor 1 1:8 1:8

…

4-core pre- ONT 1

connected
cable
FAT 1 FAT 4
HUBBOX 4 ONT 8
…

Spliced 4-core fiber hub box Basement

ONT 1
Regular optical cable 1
Traditional feeder cable

Figure 10: Solution for overhead deployment in low-density areas Figure 11: Solution for multistory buildings

26 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 27

 Business Success Business Success

Optical cables between hub boxes and FATs and 5. Leveraging existing backbone cable
between different FATs can be pulled through conduits resources
or blown through microducts. The greatest benefit FAT 1
4-core pre- 1:8 even optical splitting
of this solution is that it does not require secondary One solution is to reuse existing FDTs to connected

PON board

…
cable Single-core drop cable 1
…
ground excavation at household doorways. Of course, expand the network (Figure 13).
it also has other significant benefits, such as shortened Single-core drop cable 8

HC length and distribution optical cable length, as The benefit of this solution is that it allows 1:8 even optical
4-core pre- FAT 2
connected
splitting cable
mentioned previously. the utilization of existing feeder optical cables 1:8 even optical splitting

…
Single-core drop cable 1
and FDTs, with pre-connection used on …

sections beyond the FDT, removing the need Single-core drop cable 8
4-core pre-
for fiber splicing. Fiber splicing needs to be connected FAT 3
cable
performed only once at the FDT to connect 1:8 even optical splitting

…
Single-core drop cable 1
the traditional optical cable with the 4-core …

pre-connected optical cable that is connected Single-core drop cable 8

4-core pre-
Single-core OneTB
to FATs. Each FAT performs 1:8 splitting on an connected FAT 4
distribution cable
cable optical fiber to connect to eight households. 1:8 even optical splitting

…
… Single-core drop cable 1
Single-core
drop cable Another solution is to reuse existing backbone
Single-core drop cable 8
OneTB
… optical cables to develop B2B services and

FAT fiber to the site services (Figure 14).

…
The biggest problem with traditional ODNs Figure 13: Reusing existing FDTs to expand the network

Manhole
OneTB is low optical cable utilization due to multi-
core cables laid over long distances. Along the
HUBBOX 1 HUBBOX 2 HUBBOX 3
routes from the CO to optical splitters, there (Level-1 optical
splitter)
(Level-1 optical
splitter)
(Level-1 optical
splitter)
FAT1 FAT2 FAT3 FAT4

… 1
1 1
are multiple optical cable splicing points such
1:8 1:8 1:8 1:8

as FDTs and optical closures. These cause high MPO 8 8 8

12F
Home broadband
fault rates and optical attenuation, difficulties
with maintenance and troubleshooting, and Rack-mounted
(< 2 km from CO with OLT)
ODF
OLT
… complex resource management. GPON port
X-box shared by HBB and P2P services
1
(> 2 km from CO with OLT) HUB HUB HUB 8 single-
FAT XBOX BOX BOX BOX … core unevenly
The P2P box used in the ODN 3.0 solution can 48-core regular optical cable 1-1 1-2 1-3
8
distributed links
…
WDM
directly output dual-core optical cables to P2P P2P private
line or BSC
To P2P private line users P2P
Manhole private line users and mobile base stations. equipment
(dual-core) private
Therefore, this solution is particularly suited to lines
CO
supporting integrated carriers with deploying To new mobile base
stations (dual-core) New base
networks and fully leveraging their backbone P2P box stations
(dual-core)
optical cable resources.
Figure 12: Solution for
underground deployment
Figure 14: Reusing existing backbone optical cables

28 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 29

Business Success Business Success

Wuhan Metro's 5G-powered Smart Urban Rail project

recently won a national award for being one of China's
most exciting 5G applications in 2023. The project now
serves as a benchmark for helping industry customers to
implement dedicated public 5G network projects.

On October 17, 2023, China's Ministry of Industry This 5G-powered Smart Urban Rail project, while
and Information Technology (MIIT) kicked off the a success, had not progressed without its own
final of their annual 5G application competition challenges and setbacks. And so, the summary
in Shanghai. This year's 6th "Blooming Cup" presented at the Blooming Cup interested many

5G & Rail: The Road to the

Competition challenged participants to think as it answered two important questions: How
about how "5G Drives Digital and Real Economy did China Mobile persuade a customer from the
Development" and find new solutions that rail transit industry to put its services on a 5G

National Championship
would deeply integrate 5G into industrial digital network? And, what implications did this project's
transformation. success have for future 5GtoB services?

Of the 45,728 projects submitted from across The road to a national

China this year, Wuhan Metro's 5G-powered Smart championship
He Tao Urban Rail project stood out and took first prize.
The project's application of public 5G networks China Mobile initially kicked off this project
Senior Consulting Expert and Marketing Expert,
ICT Sales & Service Dept, Huawei in rail transit attracted significant praise from to figure out how to market connectivity to
state agencies, China's carriers, and enterprises in the rail transit industry. As they dug deeper
related fields, as it presented a novel, and more into how the industry's complex production
importantly, feasible, business model for public 5G systems worked, however, it became clear the
networks.

30 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 31

Business Success Business Success

The second step was to establish a defined that improved the industry's confidence.
industry ecosystem to better aggregate Carriers' existing expertise in 5G cybersecurity
industry resources. 5G application generally from other industries has also improved the
It became clear the project needed to refocus
needs significant support from not only the rail transit industry's trust in the reliability of
on broader industry digitalization, as true
carrier industry, but also the target industry. these services. Commercially though, many
success could not be achieved by targeting The more robust the industry ecosystem, the of China's carriers had years of experience in
an individual metro company – they needed more collaborative resources they had to leasing, instead of building, networks that can

to target the industry as a whole. achieve shared success. best serve another industry. Under this model,
customers pay carriers annually to use carrier
Industry adoption of a new technology is 5G networks for around the same total cost
never just a technical matter. It has to be as building their own traditional network.
tackled from both the technological and the
commercial ends. Technologically, carriers By combining the technological and com-
project needed to refocus on broader industry block (PRB) usage of carrier 5G networks along must ensure that industry customers are mercial elements, China Mobile was able to
digitalization, as true success could not be most metro lines is only at about 20%. confident about 5G. 5G network slicing had resolve 14 of the key problems facing the rail
achieved by targeting an individual metro already been piloted with Nanjing Metro, transit industry within just 3 years. This finally
company – they needed to target the industry At the same time, metro operators traditionally using dedicated resources to carry rail transit paid off in the Wuhan Metro project.
as a whole. build their own private networks, which can services. That pilot project set a benchmark
cost over 1 million Chinese yuan per kilometer
China Mobile had previous experience in of track. Metro wireless network bandwidth is
5GtoB services. They knew that industry also typically not large enough to carry more
customers often need patient persuasion data services, and the network can only run in · Network: Metro systems do not necessarily need to build their own networks;
carrier 5G construction costs are about 70% less for dedicated wireless networks;
before they decide to adopt 5G. They needed a best-effort manner. Service data that cannot carriers can carry out network construction and O&M at significantly lower costs.

to take a chain-like approach to bridge gaps be transmitted needs to be manually copied, · Application: Innovative 5G applications improve management capabilities to
reduce costs, while improving business operation capabilities to find new business
in the industry, going slowly and steadily in which increases OPEX. Idle carrier network growth areas for increased revenue.

the early stages to pave the way for a boom resources can be leveraged to address this
in adoption at the later stages. They broke private-network bandwidth problem—one of
this strategy down into three parts: the biggest headaches of metro operators. · ToC and toB service flows To metro
are highly complementary, operators · 5GtoC networks provide
Fully transitioning to carrier networks, which increasing network passengers with better
utilization rate increases mobile communications
experience in metros.
First, they had to identify a correct overall could simultaneously serve multiple purposes, from 23% to over 50%
(a more than 100%
direction at the beginning of the project. Once presented a win-win situation. increase). · 5GtoB networks and
To To applications facilitate
better metro applications
that direction was defined, all further efforts · Metro 5G creates both
carriers passengers (e.g., 5G + facial
toC and toB revenue,
recognition-based access
Value
improving network
had to strictly stay the course. China Mobile A smooth and workable business model resource utilization and control) and better travel
increasing 5G network experience.
chose the rail transit industry based on the required thorough analysis to determine ROI.

logic presented in Figure 1. Many carriers are the base logic. Once the general model was
installing 5G networks in metros, but those decided, China Mobile was able to call upon
China Association of Metros – There will be two types of wireless networks for rail transit in
network resources are often not being fully resources from across the industry to address
the future: LTE-M core production network and dedicated public 5G network.
utilized due to the tidal effect in the number the specific challenges that arose during
of passengers at different time periods. Traffic implementation.
analysis showed that the physical resource Figure 1: The value of dedicated public 5G networks for rail transit

32 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 33

Business Success Business Success

The third step was to develop scenario-based to make the collected data a key part of The 6th "Blooming Cup" Competition has 1. C ( Construction cost ): The full cost of
solutions to resolve key implementation production. Industry customers who actually helped promote 5G applications and explore ensuring full coverage of new unmanned
issues. 5G networks are designed to essentially use these solutions are most concerned about new business models. Dedicated public 5G sections like train depots and turn-back
amplify connectivity for terminal devices, but the practical value that they can create. The networks have been recognized across the rail tracks
data connections alone do not create real development of dedicated public 5G networks transit industry as a feasible solution. 2. B ( Basic sharing ): The cost of deploying
value. To effectively support metro operation in Wuhan Metro therefore underwent three p u b l i c 5 G n e t w o r k s t h a t a re u n d e r
services, technologies like AI, big data, and phases: reshaping connectivity, reshaping the Planning is key construction
integrated communications were also needed platform, and building intelligence (Figure 2). 3. X : The percentage of the public 5G
The bidding for the 5G 800 MHz trunking network's uplink resource blocks used by
communications on Wuhan Metro's Line 19 metro operations within a security baseline
was completed in early 2023. Since then, 4. A ( Annual fees ): The fees charged for
E2E facial many other metro operators in China initiated assurance services, such as slicing service,
Operations map
recognition
similar projects. For example, Shanghai Metro net work opt imizat ion, and net work
Smart Smart passenger Urban rail Smart Business
Smart O&M plans to have all of its 23 lines covered by a maintenance over a 10- to 15-year period
construction transport IOC mobility operations
• Self-service
dedicated public 5G network by the end of 5. T: Off-peak usage duration
• Urban rail • Multi-metric • Running gear • In-station
construction map passenger traffic • Decision- ticketing
• Intelligent display
monitoring and
making • Differentiated
store 2024. Guangzhou Metro also plans to expand
diagnosis optimization
identification of • Passenger traffic support security Enabling
construction site prediction and
• Safe driver
• Collaborative checks
• Digital the scope of its existing coverage from a The cost of using dedicated public 5G
operations billboards intelligence
risks warnings command • Facial
• 5G train-ground • Smart
• Track-area security • Operation data transfer • Index system recognition
business
single line to their entire metro system. Other networks can then be calculated as:
dispatch simulation and turnstiles
areas
optimization • Collaborative cities have also produced blueprints for future C + B × X + A (by bandwidth)
mobility
rollout, including Nanning, Wuxi, Tianjin, and or C + B × T ÷ 24 + A (by time)
Suzhou. The clear industry demand presents
Data Application Integration Development
enablement enablement enablement enablement large opportunities for dedicated public 5G Given this, the cost of using wireless 5G
Huawei
Horizon Integrated network projects, but carriers still struggle network resources on demand roughly equals
AI IoT Big data Video GIS/BIM
Operations

for communications
Security

Reshaping with implementation speed and customer that of building the metro operator's own
O&M

urban
rail platforms confidence. wireless private network. Leaving resources
Computing Storage Network Security
resource pool resource pool resource pool resource pool
unused outside metro service hours wastes
Urban
rail cloud Infrastructure Notably, this project has provided carriers with 5G coverage along the line. Metro operators
platform
clear principles for future planning. can also apply unmanned inspection robots
to tunnel inspection, which use 5G networks

NOCC network
First, business models: The business model for data backhaul, achieving unmanned and
design must be based on actual service remote operations.
Line transport Line transport
Rail yard access
In-station access
network Signal DCS network
network
scenarios to keep the price of 5G reasonable
network Reshaping
Dedicated public Dedicated public connectivity for the target customer industry. In metro For metro ground services, 5G can provide
5G network 5G network
Journal temperature, rail, a complete dedicated public 5G network the following alternative private network
LTE-M LTE-M
power supply, and
wheel hub project covers two parts: main lines and capabilities (Table 1).
PSD Turnstile SCADA Camera Dispatch and Phone
command rail yards. For carriers, the cost of providing
private network services mainly includes the The second principle relates to the network
following items: solution. To support train-ground backhaul
Figure 2: Blueprint architecture for smart urban rail

34 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 35

Business Success Business Success

Quasi-real-time services: Video backhaul,

Network Built by
Metro Service 5G Network big data upload, and PIS update
Metro Operator

13(thousand people) Inbound

CCTV service (partial) Guaranteed 12-Mbps uplink traffic
Wi-Fi (target network
Guaranteed 16-Mbps downlink for service 8
migration) PIS service
migration

3
TETRA 800 (target Guaranteed 4-Mbps uplink & downlink
Narrowband trunking
network migration) for broadband video communications 0 01:00 04:00 07:00 10:00 13:00 16:00 19:00 22:00

TCMS 500 Kbps Guaranteed 5 Mbps for TCMS 2

LTE1800 reuse Off-peak services:

(adding links) CBTC redundant networks 7 Rail inspection, pantograph
monitoring, and intelligent O&M
Train trunking
Integrated 5G communications
communications 12

Outbound
Basic 5G network packages provide private uplink lines with 26 Mbps bandwidth for real-time services, 80 17 traffic
Mbps bandwidth for quasi-real-time services hours, and 200 Mbps bandwidth for off-peak services, with a
99.9% network availability over a 10-year contract period. 5G inherits the services previously carried on Wi-Fi
and TETRA networks, while improving TCMS services and providing backup for LTE-M network services. Real-time services: CCTV video backhaul and broadband trunking

Table 1: 5G private network capabilities for metro services Figure 3: Traffic peaks and troughs of the three types of metro backhaul services

services, 5G network transport must meet the tidal effect in metro services. This solution passenger traffic, network resource usage parameters for each of the three domains
SLA requirements. For individual users, consists of three steps: service analysis, also sees peaks and troughs. By accounting – wireless network, transport network, and
disconnections and dropped calls are not existing network assessment and optimization, for these traffic flows, they can use carrier core network – they can guarantee SLAs for
likely to cause serious loss. However, carrier and slicing customization and provisioning. networks for quasi-real-time metro services industry applications. The dedicated public
5G networks must ensure the services they during off-peak hours. At night , when 5G network slicing solution is customized
carry are always available. Metro operators During service analysis, plans are based on the trains are not in service and tunnel for metro service scenarios to maximize
need these high-quality network capabilities the actual needs of metro backhaul services. n e t w o r k s a re c o m p l e t e l y i d l e , c a r r i e r resource utilization. This solution monitors
to ensure continuous 5G access. They are Depending on how services are used, services networks can be utilized for the backhaul the resource availability of each cell at
also key for further application development. can be classified into three types: real-time of off-peak services. This model maximizes different periods, and provides time- and
During the Wuhan Metro project's bidding, the services, quasi-real-time services, and off-peak the utilization of 5G network resources and space-domain service policies. It also sets up
key factor that drove the customer to choose services (Figure 3). Real-time services are reduces costs, and can be quickly replicated automatic parameter configuration for the
China Mobile was not price, but network identified and basic bandwidth is provided to by the rest of the industry. core and wireless networks to implement a
reliability, availability, system security, and guarantee these services. site-specific service design.
data redundancy design. Finally, during slicing customization and
During existing network assessment, they provisioning, they determine time- and The third principle from this project relates
China Mobile uses a customizable five-domain analyze the 5G networks already deployed space-based 5GtoB service policies through to value-added innovation. An example of
5G network slicing solution that addresses in metro facilities. Due to the tidal effect in net work analysis (Figure 4). By setting such innovation can be seen in the figures

36 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 37

Business Success Business Success

b e l o w, w h i c h s h o w a 5 G - b a s e d l e a k y positioning allows metro operators to save the

cable positioning solution that aims to cost of building their own UWB networks.
support positioning in rail transit (Figure 5).
Operation Control Center (OCC) Depot Control Center (DCC)
Fiber
Customized leaky cable and wireless UTDOA The success carriers are finding in rail transit is
private line Intelligent O&M
(existing) terminal platform
features can help position people and vehicles an important step in realizing the full value of
(new)
Magic Window
content editing and
MEC edge compute
platform
Trunking
communication
Intelligent O&M
server
in rail areas. People positioning is mainly used 5G. The dedicated public 5G network model is
broadcasting platform (new) dispatching terminal (new)
(new) (new)
by maintenance personnel to accurately and rapidly becoming the new standard for the rail
Fiber private line (existing)

OCC fiber
quickly locate faults during track inspections. transit industry, and one that can potentially
private line 5G smart station
(new)
Vehicle positioning is used by the pantograph be applied to more scenarios, including civil
monitoring and O&M system to locate airports, high-speed railway stations, and
Firewall
Fiber 5G trunking
(new) PIS content CCTV platform Smart station private line communication Smart station
faults during train operations. Many metro healthcare campuses.
management platform (existing) server (existing) handheld terminal application terminal
(existing) (new) (new) (new) systems currently use axle counters and

5G transport network connection

UWB base stations for positioning, but this More technological advancements and the
is costly, requiring about 2 million Chinese promotion of other successful applications will
Rail transit private 5G network
yuan in investment per line (for a standard help 5G continue creating value in more fields
30-kilometer line). In contrast, utilizing the and help industries go digital and thrive.

UDM+PCF
5G networks already deployed in tunnels for

5G core IMS
network P/PE CE AR UPF
AMF+SMF

UPF
Tunnel positioning Current worker Current rail transit positioning
5G signal requirements positioning solution solution
Tunnel leaky cable
source

POI between cables Worker positioning Positioning with wall tags: high error rate;
real-time monitoring unfeasible
Inspector positioning and real-time fault
5G train-ground communications location reporting

5G metro cars
5G antenna 5G antenna
(new) In-car data gateway In-car data gateway (new)
5G TAU (new) Magic Window (new) 5G TAU
(new) controller (new)
(new)
In-car ERM In-car ERM
In-car switch In-car switch
(existing) (existing)
(new) (new)
High UWB construction (CNY60,000–
In-car NVR 80,000/km) and maintenance costs
In-car NVR PIS host PIS host Vehicle pantograph monitoring, locating,
(existing) (existing) (existing) (existing) and data backhaul

Magic Window
In-car camera PIS screen (new) PIS screen In-car camera
(existing) (existing) (existing) (existing)
Axle-counter-based positioning error:
20m/km

Figure 4: Networking architecture of the dedicated public 5G network solution for rail transit Figure 5: Rail transit positioning requirements and solutions

38 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 39

02
Network
Evolution
Network Evolution Network Evolution

Middle East:
The Middle East has always played a unique role perfect opportunity to reduce their economic

in history as a place of connections, bridging great reliance on oil. The United Arab Emirates and

geographic, economic, and cultural divides. Saudi Arabia, for example, have set in motion

Leading the 5.5G Era

plans to become digital hubs for their surrounding

During the Middle Ages, Arab merchants used regions. Multinational carriers in the region, like

camels and ships to traverse the Silk Road, stc, e&, and Zain have jumped at the chance to

and Striding Towards bringing silk and porcelain from China to India
and the Malay Islands, where they would pick
become the new merchants of the digital era by
supporting digital upgrades across the Middle

an Intelligent World
up spices, minerals, and cane sugar to trade for East and Africa, and developing new strategies for

gemstones in Central Asia and ivory and gold sand cloud, DCs, and toB integration.

from East Africa. Their powerful networks then

went on to connect Northern Europe with honey, All have turned their attention towards building

fur, and wood. This mercantile influence spanned world-class ICT infrastructure.

the world and they played an important role in

Tang Zhentian the Arab Empire's prosperity, in turn influencing The first year of commercial 5.5G in
the world's economy, science, technology, culture, the Middle East
Director, Middle East & Central Asia ICT Marketing
& Solution Sales Dept, Huawei and art.
During the 2023 Global Mobile Broadband Forum

Today, the six nations of the Gulf Cooperation (MBBF), six Middle Eastern carriers kicked off the

Council (GCC) are returning to this model region's commercial launch of 5.5G . Shortly after,

through economic restructuring. Digital and those same carriers successfully completed 10-

intelligent transformation are presenting the Gbps testing and more than 20 scenario-specific
The advancements in 5G and 5.5G that carriers have made in the
Middle East's Gulf Cooperation Council (GCC) countries have set
new benchmarks in scenario-based technology verification and
growth opportunities, setting the stage for region-wide digital Qatar has already put a roadmap in place to transition to 5.5G
and intelligent upgrades. in partnership with key stakeholders, like vendors, regulators,
and strategic partners. By planning for upcoming events and
digital transformation, we've positioned ourselves at the
forefront of this technological revolution. [1]

— Ramy Boctor, Chief Technology Officer, Vodafone Qatar

42 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 43

Network Evolution Network Evolution

pilots of 5.5G technologies like RedCap, Passive IoT,

and glasses-free 3D.

Exploring new 5.5G services for

sustained 5G business success

So far, these carriers have been pragmatic in their

approach to 5.5G. Their existing success in 5G has
been the core driver of their continued investment.

By the end of 2023, 5G had a user penetration

rate of over 40%, and over 40% of mobile data
traffic came from 5G connections. Around 2018,
many carriers had begun experiencing a decline 5G is going to deliver an experience that will be close to or even
in revenue as 4G reached maturity, but 5G has match the fiber experience. We believe that by moving from 5G
reversed this trend. Leading carriers have realized to 5G SA, and to 5G-Advanced, we are halfway to 6G. [3]
20% to 45% revenue growth thanks to 5G, and — Karim Benkirane, Chief Commercial Officer, du

FWA services are rapidly transforming the industry

landscape in the Middle East.

Middle Eastern carriers are therefore eager to searches for "glasses-free 3D" have surged over And a fourth carrier has decided to first explore 3D pressing challenges for carriers. First, per-site user
capitalize on 5.5G. Many have already identified a the past half year according to Google Trends, with screen protector services before turning to 3D content. volume growth is increasing network congestion,
range of high-value 5.5G use cases that will help over half of those searches being made in countries which in turn harms user experience and increases
improve user experience and foster innovative new like Saudi Arabia, the UAE, and Oman. This is a Many carriers are already designing premium complaints. This has made FWA user satisfaction a
applications for individuals, homes, and industries. prime example of how carriers are exploring new service packages that will increase monetization and key area of concern. Second, fiber downlink rates
approaches to 5G monetization. One carrier found are directly working on developing a 3D content are growing by over 40% every year. Mainstream
Individual services: Immersive glasses-free more success with 3D tablets than with glasses- ecosystem. Many of these glasses-free 3D services are fiber-based packages already deliver 300 Mbps
3D experience stimulates 10-fold traffic free 3D phone screen protectors thanks to the expected to launch in the first half of 2024. to 500 Mbps rates, so, how can FWA maintain a
better visual experience they deliver, deciding to competitive edge?
Glasses-free 3D started making waves back bundle 3D tablets with FWA and phone screen Home services: Fiber-like to air-fiber FWA
in 2009 with the release of the sci-fi movie protectors with toC services. These sales plans are experiences Over the past year, multiple carriers have worked
Avatar . Now, more than 10 years later, the core set to launch in the near future. A second carrier with Huawei to explore these issues. What we've
technologies behind glasses-free 3D are beginning plans to introduce glasses-free 3D use cases to Another top area of focus for carriers is sustainable found is that there is still tremendous untapped
to see large-scale commercialization as related their 5.5G Innovation Center and work with OSN (a FWA growth. potential in FWA if we upgrade offerings from
technologies become increasingly reliable, with local OTT service provider) and YouTube to build a unguaranteed 100-Mbps experience packages to
better experiences delivered, 3D content easier to 3D content ecosystem. A third carrier is planning The household penetration of FWA services in the reliable 300-Mbps experience packages. Our surveys
produce, and more affordable devices available. to launch its own 3D livestreaming services and Middle East already exceeds 21%, and the service have found that 70% of user packages can be
working with OTT service providers to establish is growing rapidly, even in markets with extensive upgraded, and 67% of users are willing to pay over
Shopping malls in the UAE and Oman are already a dedicated 3D service section and also plans to optical deployment, such as the UAE, Qatar, and 20% more for better experiences. This means that a
using glasses-free 3D screens for advertising, bundle 3D services into 5G subscription packages. Bahrain. However, this rapid growth has created two reliable user experience is required throughout the

44 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 45

Network Evolution Network Evolution

Ports Authority launched the Smart Ports so the next milestone will be achieving 10-Gbps
Initiative to automate port operations through user-perceived downlink and sub-10-ms latency
different scenario-based applications including by 2030.
remote crane operation, 3D visualization of
We are proud to launch our white paper, which outlines
cranes, license plate recognition, and facial A consensus has already been reached by the
our commitment to technological excellence and leadership
in the advancement of 5G technology in the UAE. It reflects recognition. etisalat by e& also released region's carriers that TDD 3CC aggregation (300
our proactive approach to adopting advanced technologies a 5G-Advanced white paper on multiple MHz band) will be one of the fundamental
and our dedication to pioneering innovation. [7] 5G products, including 5G mobile edge, differences between 5.5G and 5G. Because of this,

— Khalid Murshed, Chief Technology and Information Officer, etisalat by e&

5G campus, and 5G networks for cars to they are working with Huawei on joint innovation
promote the application of 5G and 5.5G in in this area. For example, the region has already
the government, transportation, healthcare, taken the lead in fostering dual-band META
[6]
aviation, and oil and gas sectors . (2.6/2.3 GHz + C-band, dual-band 3CC) products
user journey. Carriers will need to assess available being implemented as the nation plans new • Digital twins: Carriers in the Middle East that support simplified 5.5G deployment, as well
network resources, coverage, and interference urban areas. The enhanced network capabilities are also actively testing and applying 5.5G as 3CC small cell products, to improve indoor
before high-speed FWA services can be provisioned. brought by 5G and 5.5G will enable carriers to technologies such as passive IoT and RedCap. user experience. The UAE's du is also already
Provisioning will also often need higher resource provide ubiquitous basic connectivity while also In the UAE, passive IoT's long standby times proceeding with large-scale TDD 3CC deployment.
scheduling priorities and higher performing CPEs. helping them build global leadership in integrated and low costs are enabling a number of smart The extremely large antenna array (ELAA)
In some cases, onsite assisted installation will be national ICT solutions. retail applications, connecting a wide range technology has increased the number of dipoles
needed to ensure the optimal installation position. of goods to digital twins. This has increased from 192 to 384, enabling a peak rate of 5 Gbps,
Once services go live, carriers will need to be able • A low-carbon "cognitive city": Saudi Arabia's tracking and stock allocation efficiency. In 30% higher user-perceived rates and 30% lower
[8]
to monitor user experience KPIs in real time for NEOM aims to use 5G to connect people Saudi Arabia, low-cost 5G RedCap modules energy consumption .
preventive optimization, and provide dedicated VIP and things to build a "cognitive city". The are also being used to explore new CCTV and
service channels for consumers. tourism developer Red Sea Global then smart security applications. Saudi Arabia's stc is similarly deploying 4CC,
worked with Zain to build zero-carbon 5G which includes C-band and low and medium
The UAE's du is an FWA pioneer. During MWC networks for a massive future city project Building 10G networks using leading bands, to improve network performance and user
Barcelona 2024, du announced that it was ready that would deploy 5G-powered IoT solutions technologies experience, and speed up 5.5G evolution[9].
to launch high-speed, low-latency packages for to monitor important ecological sites, like
[4]
gamers and bundle more TV privileges with FWA coral reefs and sea turtle nesting sites . Leading carriers in the region have already At MWC Barcelona 2024, Huawei signed three
[2]
packages . Carriers are now verifying multi-scenario completed 5.5G target network planning. MoUs with:
IoT applications using 5G smart lamp poles. • du to transform the UAE into a 5.5G country
FWA's potential remains huge as optical Some of these applications include intelligent User-perceived speeds are expected to double • Kuwait's Communication and Information
technologies evolve. security, electronic screens, EV charging piles, in the next three years, with the average 5G Technology Regulatory Authority (CITRA)
environmental monitoring, and autonomous downlink rate expected to jump from 100 Mbps to build 5.5G smart cities
Industry services: 10 Gigabit City infrastruc- driving. to 200 Mbps by 2026, and the user-perceived FWA • Zain KSA on "All in 5.5G".
ture facilitates industry digitalization • Digital and intelligent industries: Industries rate spiking from 165 Mbps to 500 Mbps.
are now increasingly receptive of 5G. Many carriers are working with vendors like
In 2022, Huawei signed an MoU with Saudi According to a survey by the UAE government, Once this is achieved, carriers are likely to Huawei to seek smooth 5.5G network evolution,
Arabia's Ministry of Communications and about 40% of enterprises plan to use some shift their focus towards pursuing ubiquitous 5G super packages (which use new services
Information Technology on building a 10-Gbps form of 5G application within the next 10G connectivity. By 2026, peak user rates will like New Calling, 3D, and MR to build the 5.5G
Society. The country's 10 gigabit vision is now year. For example, stc Group and the Saudi already be at 5 Gbps with under 20 ms latency, brand), and rate-based FWA pricing evolution.

46 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 47

Network Evolution Network Evolution

These demands have become an important factor Success in 5G has helped many of these carriers UAE 2031 and stimulating new digital vitality [6] https://www.eand.com/content/dam/eand/
assets/docs/general/5G-advanced_the_future_
driving innovation at Huawei. improve brand value. From 2018 to 2023, stc's by attracting global talent and encouraging of%20_wireless_communications_in_uae.pdf
brand value increased by 85%, helping them more innovation. This growing digital oasis will
[7] etisalat by e& launches white paper on UAE's
High-quality 5G construction is move up 19 places in global rankings and making undoubtably create a new benchmark for the transition to 5G-advanced and commitment to
creating a digital oasis them the 11th top global carrier. e& also moved world in digital transformation. global standards.
https://wam.ae/article/b0ys2bv-etisalat-
up to 14th place. Middle Eastern carriers are launches-white-paper-uaes-transition
More carriers and governments are pursuing pursuing industry leadership with their 5.5G Jan 03 2024
References :
digital, intelligent, and low-carbon transformation. evolution. 2024 will be the first year of large- [1] https://www.marhaba.qa/vodafone-qatar- [8] du Spearheads Commercial Deployment
scale 5.5G construction in the Middle East, with hosts-groundbreaking-discussion-on-5-5g- of 5G-Advanced.
technology/ https://www.telecomreview.com/articles/
The GCC nations have all launched national visions leading countries like the UAE expected to achieve telecom-operators/7818-du-spearheads-
that will significantly increase the prominence of nationwide 5.5G deployment. [2] du, Huawei sign strategic cooperation to boost commercial-deployment-of-5g-advanced
5G-A in the UAE. Feb 07, 2024
their digital economies and help them become https://www.rcrwireless.com/20240228/
global leaders, such as Saudi Arabia's Vision This has made the Gulf region significantly more 5g/du-huawei-sign-strategic-cooperation- [9] stc Achieves Substantial Progress in
boost-5g-uae Feb 28, 2024 Advancing 5G.
2030 and the UAE's "We the UAE 2031". These economically dynamic. In 2023, Dubai Chambers https://www.telecomreview.com/articles/
governments are encouraging 5G network saw a 43% increase in the number of registered [3] Du to bolster FWA 5G play to challenge rival. telecom-operators/7719-stc-achieves-substan-
https://developingtelecoms.com/telecom- tial-progress-in-advancing-5g Jan 02 2024
deployment through policy and funding support, new enterprises, with the growth rate of Abu
business/operator-news/16333-du-to-bolster-
which is successfully driving improvements in Dhabi's startup ecosystem ranking sixth globally. fwa-5g-play-to-challenge-rival.html [10] du Commits to Building UAE as 5G-Advanced
Feb 29, 2024 Country.
mobile network quality. Currently, four out of The UAE plans to incubate 20 unicorn companies https://www.telecomreview.com/articles/tele-
the world's 10 leading countries according to by 2031 and Saudi Arabia saw a 33% increase in [4] An Inspiring Journey: Zain KSA and Red Sea Global com-operators/7882-mwc24-du-commits-to-
Pioneering World’s First Zero-Emission 5G Network. building-uae-as-5g-advanced-country
Speedtest are from the Middle East. Bahrain, its number of startups in 2023. Saudi Arabia also FEB 27 2024
https://www.telecomreview.com/articles/
the UAE, Saudi Arabia, and Oman have made led the Middle East and North Africa with a record e xc l u s i v e - i n t e r v i e w s / 7 4 0 6 - a n - i n s p i r i n g -
journey-zain-ksa-and-red-sea-global- [11] Accelerating Growth: the GCC’s Start-up
significant commitments to carbon neutrality. US$1.38 billion venture capital invested last year. Ecosystem.
pioneering-world-s-first-zero-emission-5g-
Digital and intelligent transformation are also This has made the Gulf region one of the prime network Oct 09, 2023 https://communicateonline.me/category/
[11] industry-insights/post-details/accelerating-
being clearly seen in many cities and industries destinations for global startups . [5] 5G-Advanced: The Future of Wireless growth-the-gccs-start-up-ecosystemhttps://
across the region. Communications in the UAE. communicateonline.me/category/industry-
https://www.samenacouncil.org/samena_daily_ insights/post-details/accelerating-growth-the-
5.5G will likely continue to drive growth in the gccs-start-up-ecosystem Jan 31 2024
news?news=94414 March 09, 2023
At the carrier level, Middle Eastern carriers are Middle East's economy and society. This will
continuing to build high-quality communications trigger a chain reaction, supporting national
networks as they digitalize their own operations. visions like Saudi Vision 2030 and We the

This is a key milestone for du's commercial 5G-Advanced

deployment. Huawei's leading technologies will support the
realization of du's strategic vision, play an important role in
5G-Advanced service innovation and industry digitalization,
and jointly accelerate the UAE's digital transformation. [10]

— Fahad AI Hassawi, Chief Executive Officer, du

48 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 49

Network Evolution Network Evolution

Future All-optical
key technologies in the following areas: backbone networks. The acceleration of AI
• Backbone networks: adoption is further increasing requirements
Improvements in single-channel rates and for the transmission capacity of backbone

Network Architecture
system capacity to support increasing traffic networks. Backbone connectivity capacity can
between and within data centers – such be improved in three directions: improving
as that driven by China's "East Data, West single-channel rates, broadening frequency

and Key Technologies Computing" project – remain a major force

driving evolution.
bands, and evolving towards space-division
multiplexing (SDM) systems.
• Metro networks:
Changing service traffic directions are driving Direction 1: Improving single-channel rates
Tang Xiaojun transition from the traditional north-south
Chief Technology Planner, Optical Business Product Line, Huawei network architecture to a cloud-centric, low- Improving single-channel rates is a major
latency, high-quality, and T-shaped network method of increasing backbone capacity. The
architecture. evolution from coherent 100 Gbps with 50 GHz
• Access networks: channel spacing to 400 Gbps with 150 GHz

Evolving towards the 2030 optical communications As 3D applications make their way into channel spacing has improved both spectral

network system and architecture is a key issue facing homes and AI foundation models into efficiency and capacity by 33%. Requirements
the optical communications industry and requires viable devices, the focus of evolution will shift from for higher single-channel rates are often
technical options for building future-oriented and novel bandwidth improvement to high-bandwidth accompanied by expectations for lower per-
optical communications network systems. and high-quality experiences. bit costs and power consumption. To meet
such requirements, single-channel rates will
Backbone network architecture: evolve towards coherent 1.6 Tbps with 500 GHz
Evolution of single-channel rates and channel spacing. As we approach the Shannon
Optical networks form infrastructure that deliver ultra-broadband,
system capacity limit, spectrum improvement is becoming
large-capacity, and low-latency connectivity for the digital world. increasingly difficult, requiring breakthroughs in
From voice-based communications in the 1980s to modern broadband Capacit y improvement is a key driving both algorithm and component innovation. The
communications that typically support video applications, fixed force behind the generational evolution of current technical challenges are as follows:
networks have evolved through five generations. 2024 is the first
year of F5G-Advanced (F5G-A), and AI-driven applications and
the 3D display industry are now on the rise. Breakthroughs in AI
Enterprise/
foundation model technology, represented by OpenAI, are leading Campus

to the construction of numerous AI clusters and the adoption of Inner

Beijing,
Tianjin,
Yangtze
Manufacturing
River
Mongolia Hebei
many new applications. Furthermore, products based on immersive Gansu Ningxia
Delta

Guangdong,
Home
3D display technology, represented by Apple Vision Pro, are entering Guizhou Chengdu, Hong Kong,
Chongqing Macao

the market on a large scale. The enormous computing and transport

power that accompanies these developments urgently needs all-
All-optical grooming All-optical direct- Premium access Smart home network
optical infrastructure. Optical communications networks have been
backbone network access metro network network Ubiquitous beyond-10 Gbps
Integrated communications,
presented with unprecedented opportunities and challenges, which SDM, 1.6 Tbps bandwidth, All-optical access, convergence, Beyond 50G PON sensing, and computing
multi-level flexible scheduling and switching Deterministic experience
can only be addressed through breakthroughs in the architecture and

Figure 1: 2030 all-optical network architecture

50 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 51

 Network Evolution Network Evolution

• Improving spectral efficiency with higher-order delivers two-times wider spectrum than a C80 end mixers are wavelength-sensitive and must low return on investment and difficulties for long-
modulation formats and optimized algorithms: system, as well as higher single-channel rates, be re-designed. Adding these two bands will term evolution.
The spectral efficiency of 1.6 Tbps with 400 increasing system capacity from 8 Tbps, which increase stimulated Raman scattering (SRS),
GHz channel spacing (4 bit/Hz) is 50% higher was delivered in the early days of coherent a non-linear effect in the system that causes Direction 3: Evolution towards SDM systems
than that of 400 Gbps (2.67 bit/Hz). In addition, systems, to today's 32 Tbps. Beyond the C-band the transmission performance of the C-band
the modulation format is changed from QPSK and L-band, the S-band and U-band have and L-band to deteriorate, increasing system- The growing complexity of technical challenges
to CS16QAM, and long-distance backbone the potential to help broaden the spectrum. design and O&M complexity. These challenges in relation to the two aforementioned solutions
transmission must be covered. These changes However, the optoelectronic components for require further research and investment by means that backbone network evolution towards
require better modulation format designs and the S-band and U-band, such as tunable lasers, both academia and the industry. Even if the SDM systems is becoming increasingly important
optimized algorithms. photodiodes, and optical amplifiers, are not challenges are successfully overcome, system and urgent. SDM was proposed 60 years ago, and
• Components with 400G baud rates: yet mature. Moreover, modulators and receive- capacity can still only be doubled, meaning a there has been a large amount of academic research
Achieving 1.6 Tbps coherent transmission on into key SDM technologies, with solid progress
a single transceiver requires components that being made over the past dozen years. However,
support close-to-400G baud rates, such as AD/ the following issues regarding SDM productization
DA converters, modulators, and photodiodes. remain unresolved:
The latest industry developments can achieve • Academic research has proposed multiple technical
modulation and demodulation at baud rates of routes for SDM, including uncoupled multi-core
about 250G, but there is still a significant gap to fiber, coupled multi-core fiber, multi-fiber, few-
400G baud rates. mode fiber, multi-mode fiber, and orbital angular
• High-integration superchannel technology: momentum (OAM) solutions. These routes each
Limited component techniques mean baud rate have their own advantages and disadvantages,
increases are becoming increasingly difficult. An and it is too early to say which of them will
alternative way to achieve 1.6 Tbps rates per ultimately become dominant.
channel is multi-channel integration, which means • SDM can be carried by different types of optical
integrating two 800-Gbps channels or four 400- fibers, such as G.652, G.654, and hollow-core.
Gbps channels. A multi-channel solution uses • It is not yet clear whether the system architecture
lower-bandwidth components, but requires better would be WDM first followed by SDM or the
photonic integration technology that is capable of other way around.
solving issues like relative frequency control, multi- • SDM can have two, four, or more channels.
channel component yield improvement, reduced
raw-material usage, and the warping of boxes and According to Huawei's analysis, the multi-fiber
substrates. solution and uncoupled fiber solution are currently
the two most mature. The former is a mature design
Direction 2: Broadening the spectrum based on parallel single-mode fibers, while the
latter has lower crosstalk than other solutions and
System capacity is equal to the single-channel rate is made with ground-breaking drawing techniques.
multiplied by the number of channels in the system. Huawei believes that breakthroughs and consensus
Increasing the number of channels by broadening the are needed in the following four areas for SDM be
spectrum is another key direction for the evolution commercialized across the industry:
of optical system capacity. A C120+L120 system

52 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 53

Network Evolution Network Evolution

The first is compatible system architecture. SDM is adding one more dimension of multi- including relative frequency control, multi- have emerged in recent years are significantly
The WDM system was the key to the past plexing without increasing the complexity of net- channel component yield improvement, reduced expanding the application scenarios of opti-
success of optical backbone networks, meaning work management. Huawei believes that WDM- raw-material usage, and the warping of boxes cal fibers, enabling converged applications of
compatibility with the WDM system is neces- like network management on an SDM system, and substrates. Challenges to improving optical integrated sensing, communications, and en-
sary to commercialize the SDM system. SDM despite the added space dimension, is the main amplifier integration include: multi-component cryption. With multiple spatial channels, SDM
and WDM multiplex different resources, so the focus and area of breakthrough for management rare earth doping used for integrated wide- can be used for applications that integrate
design through which these two systems are and control technologies. Traditional WDM spectrum amplification, high-power and low- communication capacity expansion, integrated
made compatible must be carefully considered. systems, multi-core SDM systems, and multi- cost pumps used to support multi-channel communications and sensing, and security. For
A solution with WDM followed by SDM is fiber SDM systems all use the same network amplification, and multi-channel passive example, in a four-core system, one core is used
recommended because the compatibility in management system (NMS) to centrally manage components with high integration and low loss. for traditional optical communications, another
question can be realized through the use of OTU ports, wavelengths, and fiber cores. This The main challenge in relation to OXC is that the two cores are used for QKD-based communica-
simple-structure components, including WDM/ involves a range of technologies including additional space dimension requires high-degree tion and key negotiation, and the fourth core is
SDM multiplexers, optical amplifiers, optical network resource pooling (establishing an OXC, posing serious challenges to 128-degree used for distributed optical fiber sensing.
cross-connect (OXC) devices, and fan-in/fan- OTU port resource pool, a wavelength resource wavelength selective switches (WSS) that
out (FI/FO) devices. This solution must support pool, and a fiber core resource pool to share can only be addressed with OXC architecture Metro network architecture: Cloud-
both multi-fiber and uncoupled multi-core resources among different systems) and network innovation. In terms of optical connections for centric low-latency and high-quality
fiber technologies. The two technologies are as a service (NaaS) application programming an SDM system, another technical challenge is networks
suitable for different types of networks, with the interfaces (APIs). direct connection between multi-core fibers and
former designed for the upgrade of legacy fiber single-mode fibers, and also between multi-core AI's massive requirements for computing power
networks, and the latter designed for newly- The third is highly-integrated system architec- fibers and optical modules. will transform metropolitan area network
deployed fiber networks. It is also possible to ture. Limited equipment room spaces require (MAN) traffic from a north-south model
use the two technologies together by adding more closely integrated network equipment, The fourth is converged applications such as to a T-shaped model that goes both north-
new multi-core fibers to existing single-mode meaning that increasing system capacity communications, sensing, and security. SDM south and east-west, and cause traffic to shift
optical fibers in order to form a hybrid SDM without requiring larger equipment is critical to can support more application scenarios than from user-to-user to mainly user-to-cloud.
network. In terms of fiber types, SDM systems productizing SDM. From a system composition large-capacity communications. For example, In addition, services like AI, VR, and smart
must support both single-mode fibers and perspective, a more integrated SDM system technologies like distributed optical fiber sens- manufacturing require higher bandwidth and
hollow-core fibers. Hollow-core fibers have requires optimized component sizes including ing and quantum key distribution (QKD) that lower latency. It is essential that the metro
attracted a lot of attention in recent years due optical modules, optical amplifiers, OXC, and
to their low nonlinearity, low dispersion, and FI/FO devices. Optical module improvement
low latency, and have the potential to become a in baud rates will inevitably encounter a
disruptive fiber technology. bottleneck. Superchannel optical modules
supporting higher port rates may necessitate
AI's massive requirements for computing
The second is simplified management system. a change in technologies, which will present
power will transform metro network traffic
Another key issue to consider when productizing the same challenges as mentioned earlier,
from user-to-user to mainly user-to-cloud.

54 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 55

 Network Evolution Network Evolution

network architecture supports one-hop access Traffic flows are changed to be based on bandwidth, latency, and jitter, as well as high frequency or wavelength, providing users
to the cloud and ultra-low latency between a deterministic cloud-centric model that reliability, will be provided for diamond services with μs-level latency and jitter. This solution
AZs to meet user requirements for high-quality facilitates optical network planning and (e.g., for high-end villas), while deterministic does not require a high power budget, but
experiences. Legacy networks with an architecture O&M. Digital optical-layer technology can be bandwidth and sub-ms-level latency and jitter does require optical network units (ONUs)
that features hop-by-hop forwarding and multi- used to accurately model the optical layer, will be provided for silver services (e.g., for equipped with tunable lasers. Access
ring stacking can no longer meet the latest thereby visualizing the optical network. This apartments). On-demand bandwidth and ms- networks are highly cost-sensitive, meaning
requirements. Key technologies like all-optical supports accurate service provisioning in the level latency and jitter will then be allocated low-cost tunable lasers are a key technical
interconnection, fine-grain OTN (fgOTN), and planning phase, intelligent fault location in for copper services (e.g., for remote rural areas) challenge for this solution.
optical-layer digitalization are required to ensure the O&M phase, and quick service recovery at affordable prices.
high bandwidth and low latency for the optical on highly-reliable automatically switched • Coherent solution
metro network architecture. optical networks (ASONs). When used The high-bandwidth and low-latency require- Through the application of digital subcarrier
alongside an operation app, this will help ments of beyond 50G PON can potentially be technology, coherent networks can also
• All-optical interconnection monetize network O&M. met by the following three technical solutions: deliver high-quality P2P access and ensure
Centrally managing the wavelength resources deterministic latency with physical isolation.
of metro network access rings through OXC, Access network architecture: High • Direct modulation and detection Coherent systems can achieve high band-
with the resources in optical fibers shared bandwidth and high-quality experience This solution is based on the continued width with low-bandwidth components
among the rings, will enable the resources of evolution of current-generation technologies. and higher-order modulation, which both
each ring to be flexibly adjusted in order to Optical access network technology has evolved It provides a time division multiple access realizes higher receiver sensitivity and meets
effectively cope with traffic imbalance between along the following path: PON > GPON > 10G (TDMA) mechanism and supports low- the power budget requirements of ODNs.
rings. In addition, OXC eliminates the need for PON > 50G PON > Beyond 50G PON, providing cost, point-to-multipoint (P2MP) access, However, this requires ONUs, which are cost-
optical-to-electrical conversion, ensuring one- high-bandwidth networks for access users. but cannot provide deterministic latency. sensitive, to be equipped with intrinsic lasers,
hop access to the cloud with ultra-low latency. Emerging new services (e.g., AI, AR, VR, and The solution relies on higher-bandwidth creating a challenge regarding laser costs.
holographic display) and applications (e.g., optoelectronic components. Furthermore,
• fgOTN smart manufacturing) require access networks direct detection means a 3 dB decrease in Looking ahead to 2030, ground-breaking
OTN technology is deployed at the network that deliver not only higher bandwidth, for receiver sensitivity each time bandwidth transformations in fields like AI foundation
edge and customer edge to provide hard pipes example, 2 Gbps for 8K VR with high-quality doubles. Bandwidth improvements will also models and immersive 3D interaction will
for services. The hard pipes feature physical experience, but lower latency, lower jitter, and bring higher dispersion penalty. Reusing require an all-optical network architecture with
isolation and high security, while supporting security isolation. For example, manufacturing the optical distribution network (ODN) higher bandwidth and lower latency. Developing
the continuous evolution of single-channel requires deterministic μs-level latency and will require much higher transmit power an all-optical network architecture system
rates as well as hitless and fast bandwidth jitter. In addition, vertical industries require than that provided by a 10G PON network, will require breakthroughs in key technologies
adjustment, thus meeting service requirements multiple service networks to be carried on a posing a great challenge to high-power related to backbone networks, metro networks,
for increased and more flexible bandwidth. single physical network and be strictly and lasers. and access networks to support the connectivity
Furthermore, fgOTN hard pipes can provide logically isolated to uphold the service level required by the massive and constantly surging
deterministic latency, enabling deterministic agreements (SLAs) of each service network. • WDM/FDM & direct detection number of digital services. At Huawei, we
experience for premium industry customers. Therefore, future optical access networks will Frequency-division multiple access (FDMA) believe that all-optical network architectures
not only increase bandwidth towards beyond or wavelength-division multiple access will help make people's lives more digitalized
• Optical-layer digitalization 50G PON, but transition from a best-effort (WDMA) is implemented through multiple and intelligent and serve as the connectivity
Optical performance visualization can be model to one that ensures differentiated wavelengths, and independent high-quality foundation for an intelligent world.
extended to the edge of optical networks. services with guaranteed quality. Deterministic P2P access is implemented through a single

56 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 57

Network Evolution Network Evolution

APN6 Enables Innovation Li Zhenbin

in Cloud-Network-Edge-
Chief Protocol Expert of Huawei and Former
Member of the Internet Architecture Board (IAB)
of the Internet Engineering Task Force (IETF)

Device Collaboration

As differentiated application requirements (URLLC) such as Internet of Vehicles (IoV),

With technological evolution, APN6 is reshaping the and diverse network technologies and services industrial control, smart manufacturing, and
future of network services, and is likely to initiate a emerge, a wide range of applications designed remote surgery
major transformation in Internet architecture. This article for these requirements are being developed. • As industries migrate to the cloud, cloud
explores how APN6 technology is transforming the Specific application scenarios include: applications are used over the Internet in
cloud-network-edge-device collaboration system, as well industries such as finance, manufacturing,
as its potential influence in emerging business domains. • Mobile interconnection applications oriented education, and healthcare, as well as by
towards enhanced Mobile Broadband individual users and households. They are
(eMBB) such as HD video, virtual reality, reshaping industries and people's social
cloud storage, high-speed mobile Internet activities, entertainment, and everyday lives.
In the digital age, every innovation in network technology has the access, and AI Major applications include smart cities,
potential to initiate a major industry transformation. Application-aware • Device interconnection applications oriented private networks for financial cloud, cloud-
IPv6 Networking (APN6) uses the programmable space in IPv6 packets towards massive Machine-Type Commu- based healthcare, online education, remote
to bring application information (parameters such as identifiers and/ nications (mMTC) such as environmental offices, cloud private lines for e-commerce,
or network performance requirements) into a network. This makes the monitoring, smart meter reading, and smart and cloud gaming.
network aware of applications and their requirements, and enables agriculture
refined network services and accurate network O&M, facilitating • Special applications oriented towards Ultra- Diversified application scenarios present new
innovation in cloud-network-edge-device convergence. Reliable and Low Latency Communications challenges for network O&M. Effectively

58 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 59

 Network Evolution Network Evolution

Original App
App info packet
added to SID
Original App App info added to the IPv6 H IPv6 H
Argus or SRH App info added IPv6 H IPv6 H
packet IPv6 EXH
TLV to the IPv6 EXH
SID List SID List
App info
IPv6 H IPv6 H APP Info HBH
carried in SID
SRH TLV SRH TLV Argus or SRH
HBH APP Info
IPv6 H IPv6 H
TLV for SLA
assurance APP ID User ID
APN6 means more than just enhancement and
APP ID User ID
APP Info APP Info App Req. Para.
innovation in IPv6. It provides a foundation for
App Req. Para.
Steer into a SRv6
tunnel
SLA assurance
provided for Apps
the next-generation Internet protocol.
• App-aware edge device (DL)
• App identification
• App information and
requirements added to the
• App-aware head node IPv6 EXH
• App-aware edge device (UL)
(UL) • App-aware intermediate • App-aware tail node
• App identification
• Steered into the node (UL) (UL)
• App information and
corresponding SRv6 • Service quality • SRv6 encapsulation
requirements added to the IPv6
tunnel based on App guaranteed removed Multi
EXH Leaf APP
requirements VM
DC
VNF
Leaf APP
2B
APP CPE ACC AGG ASBR DC GW Spine Leaf APP
Access Aggregation and network equipment providing services carried in APN6 packets, including application
2C
APP RG ACC AGG ASBR DC GW Spine Leaf APP
based on APN information are managed and identifier information (APN IDs) and application
controlled by the same carrier or enterprise and requirement parameter information (APN
App information directly carried App information directly carried belong to the same trusted domain. Therefore, parameters). An APN ID provides information
by IPv6 Apps in the IPv6 EXH by IPv6 Apps in the IPv6 EXH
APN6 (Network-side Solution) there are no privacy or cybersecurity issues. Its that helps the network distinguish between
disadvantage is that the application and user different application flows and different users
APN6 (Application-side Solution)
information added to packets may be inaccurate (user groups) of a certain application or
or incomplete. This is because the information is certain type of applications. Such information

Figure 1: APN6 framework

added by network border equipment, instead of may include Application Group IDs and User
applications, but the equipment cannot acquire Group IDs. APN parameters optionally carry
the information of certain applications. information that may include parameters
related to application requirements for network
realizing refined network services and accurate APN6 framework include applications, network
In the application-side APN6 solution, appli- performance, such as bandwidth, latency, jitter,
network O&M is key to meeting differentiated edge nodes, and head nodes, intermediate
cation and user information is directly added to and packet loss rate.
application requirements, ensuring SLAs and nodes, and tail nodes that provide network ser-
packets by applications. Its advantage is that it
promoting sustainable network development vices based on APN information (Figure 1).
and evolution. In-depth application-network
ensures the application and user information APN6 promotes innovation in the
integration is a feasible path to achieving this Depending on where APN information is added,
added to packets is accurate and complete. next-generation Internet protocol
The disadvantage is that it creates privacy
goal and has catalyzed the emergence of the there are two main types of APN6 solutions,
and cybersecurity challenges, as information APN6 generates significant changes to Internet
APN6 technology. application-side and network-side, each with its
must be transmitted between different trusted architecture and promotes innovation in the
own advantages and disadvantages.
domains such as terminal devices, network next-generation Internet protocol.
APN6 technology: Two solutions
infrastructure, and cloud services.
In the network-side APN6 solution, application
Network architecture consists of three key
How does application-aware IPv6 networking and user information is added to packets by
The IETF draft (draft-li-apn-framework) defines aspects: identification, forwarding, and
realize intelligent conversations between net- network border equipment. The advantage of
the application information (APN attributes) control. These aspects have different impacts
works and applications? Key components of the this solution is that network border equipment

60 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 61

Network Evolution Network Evolution

on the network architecture. Identification can space for Internet development. As a result,
IPv6 packet, and forwards it according to the three types of APN IDs: 32-bit, 64-bit, and 128-bit.
cause fundamental architecture changes as, the IP technology architecture will shift from
destination IPv6 address. This allows networks This means an APN ID used in combination with
when it changes, both forwarding and control network services based on IP addresses to
to evolve incrementally. a 128-bit IPv6 address is equivalent to a 160-bit,
must also change, which alters the entire network services based on APN IDs or on both
192-bit, or 256-bit identifier.
architecture. However, identification changes IP addresses and APN IDs.
The combination of IPv6 addresses and APN IDs
are also extremely difficult. For example,
can be regarded as another expansion of the IP The network architecture upgrade in question
changes in address identification from IPv4 The network architecture upgrade enabled
address space. The identifier, which was previously is made possible with a new identifier–APN ID–
to IPv6 require upgrade in the entire Internet by APN6 differs from the upgrade from IPv4
based on a 128-bit IPv6 address, has been in addition to the IPv6 address.
infrastructure. to IPv6. A key issue encountered during the
lengthened to include more bits — there are now
IPv4-to-IPv6 upgrade was compatibility. In-
The US launched five Internet architecture compatibility between IPv6 and IPv4 addresses
research projects through Internet2 around means that Internet infrastructure must be
2010. These projects covered Internet archi- upgraded to support IPv6. This is partly why Device-side service awareness with APN6; network-side key-service
tecture transformation, with the more well- IPv6 deployment has been slow. The APN6- assurance with slicing; E2E quality visualization with IFIT
known ones including Mobility First and NDN. enabled network architecture upgrade is based

Management
These projects provided a wealth of valuable on an identification system that consists of IPv6

layer
Visualized network and
ideas about Internet architecture, but were address and APN ID, thanks to two factors: Operation service conferencing quality in
platform (IMOC) real time
ultimately unsuccessful for three reasons:
First, the IP address system and functions are

Control
Video conferencing Network and security mgmt, control, and

layer
1. The Internet's influence is too large, meaning already well defined, and the address space issue
system (SMC) maintenance system (NCE)
it was incredibly difficult to push for trans- is resolved by the upgrade from IPv4 to IPv6.
formation. Therefore, we should shift our efforts from Video conferencing + Office e-Gov extranet (vertical IPv6+ e-Gov cloud &
e-Gov campus
devices (XC) integration), IP + Optical security
2. Transformation was incompatible with the an IP-address-based identification system to
Video slice
evolution of existing Internet technologies, an identity-like APN-ID-based identification Video slice IoT slice
Data center
and fully upgrading network infrastructure system, which involves allocation, use, and IoT slice

Physical
CloudLink Box Office slice

layer
IdeaHub
Office slice
was too costly and time-consuming. management of APN IDs, and define various
3. Existing network software and hardware protocol extensions on that basis.
Video conferencing
capabilities could not fully support new TE soft client/WebClient system (MCU)

network technologies. Second, by learning lessons from the IPv4-to-

IPv6 upgrade, incremental deployment and Conference Hop-by-hop IFIT monitoring of real-time data Conference

Service
media flow media flow

layer
APN6 ID APN6 ID
APN6 brings important changes to the network upgrade can be performed based on IPv6's SRv6 connections are automatically established, and Flex-E-
based slicing plane ensures connection quality.
identification system. APN6 can be understood backward compatibility mechanism. With an
as introducing APN IDs, in addition to IPv6 APN ID carried in the IPv6 extension header, Device Network Cloud Security

products
addresses, so that packets contain both IP an IPv6 node on the network provides services Video conferencing Network (routers &
Operation service platform
ICT devices switches)
(IMOC), network (NCE), video
e-Gov cloud
addresses and identity-like IDs. APN6 means based on the APN ID if the node identifies the (HarmonyOS) IP + Optical
conferencing system (EulerOS)
Security
Office devices (XC) (transmission)
more than just enhancement and innovation APN information in the packet. If the IPv6 node
in IPv6. It provides a foundation for the next- cannot identify the APN information in the
Figure 2: APN6-based video conferencing experience assurance solution
generation Internet protocol, creating new packet, it considers the packet to be a common

62 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 63

Network Evolution Network Evolution

Edge site 1
Latency ≈ 1.5 + 7.9 = 9.4 ms
Light load
Computing latency
Client ≈4ms Total latency ≈22.4ms
s
: 9m
n cy
te
la
The combination of IPv6 addresses and t wo
r k
Edge site 2
Ne Heavy load
APN IDs can be regarded as another Ingress Network latency: 4ms
Computing latency
Total latency ≈23.4ms
node ≈10ms
expansion of the IP address space. Ne
tw
or
k
la Edge site 3
te
nc
y: Average load
5m
s Computing latency
≈5ms Total latency ≈19.4ms

Figure 3: XR experience assurance through optimization based on both network and computing power loads

APN6 deployment use cases and at the border of the IP transport network
technical value can then steer the traffic into a network slice Next Computing
Prefix Control plane: Distributes computing power status of computing nodes
dedicated to the video service in accordance Hop Power Value
through network protocols
MEC-1 60
An APN6-based video conferencing experience with the application ID. With resource iso- Computing Data plane: Guides request forwarding based on computing service
Service ID MEC-2 20 identifiers, comprehensive computing power status, and network status
assurance solution has been deployed on an lation supported by IP network slices, other (App ID)
MEC-n 10
e-Government extranet (Figure 2). This means services will not affect traffic in the video MEC-2 MEC-n

APN6-based solutions are already commercially slice, thus guaranteeing good experience for APP 1 APP 3 4
Processing
APP 1 APP 2 Computing
at the
1 information
destination
viable for live networks. video conferencing users. Furthermore, the Computing-aware module
MEC site is
Computing-aware module
is collected
requested.
IP transport network performs in-situ flow
MEC-1
Traditional video conferencing assurance is information telemetry (IFIT) for applications 2
Edge computing
APP 1 APP 2 nodes notify
MEC-3
difficult as it is highly labor-intensive and based on application IDs to provide applica- network
Computing-aware module equipment of their
APP 3 APP 2
networks cannot support targeted assurance tion-level, service-quality visualization. This computing power
status
Computing-aware module
Network equipment
services due to a lack of integration between makes it possible to both quickly locate issues 4 generates computing
power load balancing
the video system and the network, meaning and optimize traffic when video conferencing entries
The app's routing
3
faults during video conferencing are difficult experience is poor. and computing
power status
to locate. is spread and
synchronized across
the network
APN6 can also be applied to scenarios such
With the APN6-based video conferencing as computing-network convergence (Figure
assurance solution, IPv6 packets sent by the 3). Latency is critical to good XR service Network equipment selects routes Network equipment adds an entry
1 User initiates request for App 1 2 based on the app's computing power 3 to the session table and performs
video conferencing application carry the experience. Latency assurance depends not and network status at each site. session persistence for the request.

application ID corresponding to the video only on network load, but on computing

conference that needs assurance. Equipment power load on the edge cloud. Therefore,
Figure 4: Computing-aware traffic steering solution

64 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 65

Network Evolution Network Evolution

traffic optimization must be performed with flow packets for accessing an application, based traffic steering policy. APN6 introduces Internet design principles such as the end-
consideration for both network load and the head node can select a path with a light an identity-like ID (APN ID) beyond IPv6 to-end principle and the application-network
computing power load. load, and direct the flow to the corresponding addresses, so that the ID remains unchanged separation principle.
computing service location, thus guaranteeing when the application address changes. This
A computing-aware traffic steering solution optimal application access experience. means an ID-based network policy can still Early on, APN6 caused many debates in the
implements comprehensive scheduling work without adjustment. This mechanism standardization community. But over time, a
based on the network and computing power The aforementioned two sample APN6 ap- greatly simplifies network O&M. greater consensus has been reached. APN6 can
loads (Figure 4). With this solution, the plications also show the benefit of separat- be preferentially applied to limited domains
network must be notified of the location and ing locators from IDs. Traditional Internet More agreements reached on where both the application and network are
application identifiers and load information packets have only an IP address as the iden- standardization controlled, but further development is required
of a computing service. The head node of the tifier, meaning the IP address serves both for Internet applications in which more third
network forms multiple path-forwarding rules as a locator and an ID. In mobile settings, Since early 2019, over a dozen IETF drafts parties are involved.
based on received information like different application addresses constantly change, on APN6 technology have been submitted,
computing power locations and loads of and an IP-address-based identifier system including the problem statement, framework, Regardless, APN6 has made huge progress in
the same application, points to different means the address and ID will both change, IPv6 encapsulation, YANG model, and BGP ex- terms of both theory and practical applica-
computing service locations, and identifies making it necessary to adjust the IP-address- tensions. APN6 poses challenges to traditional tions. The advancement of emerging services
different comprehensive load information like AI computing is leading to a clearer trend
relating to network load and computing in cloud-network-edge-device convergence,
power load. When receiving new service which will pave the way for the further
development of APN6.

66 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 67

Network Evolution Network Evolution

AI Data Lake: Breaking

Trends and challenges of large AI model corresponding shift in training data.
services Training datasets now encompass a mix of
text, images, audio, and video, rather than

Silos and Accelerating In 2023, Google released the Gemini multimodal

model, which is able to understand, operate,
text alone. This means that the volume
of data required to train today's most

Intelligence
and integrate different types of information, advanced models is a staggering 10,000
including text, code, audio, images, and video. times greater than was required for earlier
In February 2024, OpenAI released a video models (as shown in Figure 1).
model called Sora, which can create realistic • Trend 2: Computing power, algorithms,
scenes from text instructions. By combining a and data are the three core elements of
Michael Fan diffusion model with the large language model, large AI models. Through the stacking of
Sora exhibits amazing 3D consistency when computing power, aggregation of data,
Director, Data Storage Marketing Execution Dept, Huawei
learning from the physical world. and parameter scaling (from hundreds
of billions to trillions, and even up to 10
The speed with which large AI models trillion), complex behaviors emerge within
have developed has far exceeded people's the framework of deep learning algorithms.
Huawei has launched a three-layer AI data lake expectations. Witnessing the progression from For example, in a video released by Sora,
solution to break down the data silos that are ChatGPT to Gemini and now to Sora, we have we witness a stylish woman strolling down
preventing many enterprises from reaping the identified two key trends: a street. As she moves, the perspective on
benefits of AI models. the street scenes behind her seamlessly
• Trend 1: The shift from pure NLP models adjusts. As she walks by, objects and other
to multimodal models has brought a pedestrians are momentarily obscured, and
Since the release of ChatGPT in November 2022, large AI model
technology has been on a rapid upward trajectory, with model
training emerging as the driving force behind technological Sora
Training data volume: 100 million-minute videos and
progress. However, the data volume and diversity of data types 100T tokens
Data acquisition: Existing multimodal corpus insufficient,
involved in model training are increasing exponentially, meaning necessitating content synthesis and generation
GPT-4/Gemini 1.0 4D..
that data silos pose a major barrier to progress. Pangu NLP Original data volume: 50-100 PB
100GB/capsule
Hi-Res:3D
Data processing volume: ~100 TB/day 50+GB/Object
Original data volume: 380 TB Model scale: trillions
Video
Data processing volume: 3 TB/day Audio 8MB~5GB/movie

Many organizations are struggling to see the way forward. Model scale: 100 billion
Images:2D
1~5MB/song
Data processing complexity >10X
500KB~2MB/picture
Text
Numbers 500KB/record tatio n
ugmen
> 10+
A
5KB/record n
olatio
Huawei's AI data lake solution, however, does not just create Interp mpora
ry

bridges between data silos, it lights the way to a more intelligent Coding
Cutt in
g
Target Search a ll
xity, te
comple access, and
I/O
10,000x
rsion it ion data,
sm rsion
Conve Waterm n
ark recogn t conve
tio forma
tics evalua
world and a new era of AI innovation and development. Porn & Aesthe n
tio
violen
ce evalua
tion >1,000x
evalua Data scale expansion

NLP model Multimodal model Multimodal model

with 100 billion with trillions of with over 10
parameters parameters trillion parameters

Figure 1: Multimodal models for exponential growth in the volume of training data

68 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 69

Network Evolution Network Evolution

then reemerge, maintaining impeccable that AI computing power has efficient access
3D consistency and object permanence, to training materials.
faithfully replicating how the world is
perceived through human eyes. New requirements for data storage
and management The ideal AI data infrastructure should be
The transition from unimodal to multimodal
able to support the key phases of large
models will create a deluge of data, and a The ideal AI data infrastructure should be
AI model training, such as data ingestion,
solution is urgently required that can integrate able to support the key phases of large AI
scattered data resources. As a way of presenting model training, such as data ingestion, data
data preprocessing, and model training.
the real world, data serves as the cornerstone preprocessing, and model training, to provide
of training large AI models. With deep learning high-quality data services. To achieve this
algorithms, the quantity and quality of data goal, at least two layers of data infrastructure
determines how far AI training will go. However, should be considered: the storage device layer
most data owners prioritize the efficient and the data management layer.
access of data by service applications, often and access protocols may vary during the loading and checkpointing directly affects
overlooking where data is stored. Similarly, Storage device layer data ingestion phase, servers only access training efficiency. Although the volume of
most data administrators focus on whether data through the file interfaces during the pre-processed training data is small, high
data is effectively stored, and may neglect data When dealing with a large amount of heteroge- training phase. Therefore, ideal storage file access performance (OPS and IOPS) and
ownership and types. As a result, data ends up neous data from multiple sources, especially in hardware should support multi-protocol low latency are required to ensure quick
siloed in different data centers. One carrier we multimodal AI training scenarios, the ideal stor- interworking to ensure that underlying data loading and avoid wasting GPU/NPU
spoke to has seen the total amount of data age device layer should feature multi-protocol data can be accessed by different protocols computing power. Saving checkpoints during
they need to store soar to hundreds of PBs in interworking, high I/O performance, and easy or interfaces, thereby eliminating the data the training process is critical to ensure that,
recent years, and hundreds of TBs of new data scalability to address various challenges and replication caused by protocol conversion. in the event of an interruption, training can
is generated each day. However, this data is support the following pivotal phases of training • Data preprocessing: Data preprocessing be resumed rather than started from scratch.
scattered across multiple data centers. large AI models: involves the cleansing, transformation, Therefore, it requires high write bandwidth
augmentation, and standardization of for fast and frequent archiving and more
To provide as much data as possible for model • Data ingestion: In this phase, data is often diverse data to yield high-quality training stable training.
training, the carrier's technical department has scattered across different silos, stored in materials from massive volumes of raw
to migrate or replicate the data from scattered various formats, and accessed through data. In this phase, a large amount of Data management layer
data silos across regions. As a result, the data different protocols. To efficiently centralize temporary data is generated due to the
preparation process is prolonged, accounting for heterogeneous data from different sources, diversity of preprocessing tools, resulting in Based on the flexible capacity expansion and
over 50% of the entire model training duration. data storage devices need to support various data expansion. Therefore, storage devices high hybrid-load performance provided by the
data formats and access protocols, and should offer extensive shared storage storage device layer, the data management layer
Therefore, the biggest challenges and primary deliver high write bandwidth. Additionally, capacity while also delivering high read/ further provides advanced data management
considerations for constructing large AI model storage solutions need to be both flexible write bandwidth and supporting random capabilities for AI training. It helps data owners
infrastructure revolve around eliminating data and scalable, while also remaining cost- access to expedite preprocessing. and administrators maximize data value in a
silos and effectively gathering data in one effective, in order to handle new data • Large AI model training: In the model more efficient manner from three dimensions:
place, swiftly transforming collected datasets sources that may be incorporated into training phase, the performance of storage visibility, manageability, and availability.
into model training materials, and ensuring training at any time. While data formats devices in tasks such as training data

70 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 71

Network Evolution Network Evolution

• Visibility: Data asset owners and administrators storage device layer. Such a framework should both high OPS and low latency for random AI data lake solution
should have a comprehensive understanding help users streamline data preprocessing and access and high read/write bandwidth for
of their data, including data storage location, enhance processing speed to improve data sequential access. Huawei has been actively engaging in large
data volume, and data types, akin to having availability. • High capacity AI model training with customers in diverse
a data map. Considered from this dimension, The data ingestion and preprocessing industries, including carriers. Over the years,
owners and administrators can quickly Core capabilities of AI data phases involve the intensive reading, writ- we have gained extensive expertise in AI data
identify which data needs to be collected infrastructure ing, and generation of temporary data, infrastructure. Recently, we have launched
based on this map. which creates storage challenges. Data an AI data lake solution to help customers
• Manageability: After determining the data The ideal AI data infrastructure should have the infrastructure therefore needs to offer solve the problems encountered in the
to be collected, a mechanism is required core capabilities outlined in Figure 2. flexible and non-disruptive scale-out and deployment of data infrastructure for large AI
to implement policy-based data flow. For tiering capabilities. This ensures a balance model training. This enables our customers
example, a policy may be used to define the To sum up, three key features are indispensable: between cost and capacity, while deliver- to focus on model development and training.
source and target of data flow, start and end ing the necessary bandwidth, IOPS, and The architecture of the Huawei AI data lake
time window, maximum rate limitation, and • High performance OPS to handle sequential and random solution, as shown in Figure 3, consists of three
minimum rate guarantee, facilitating the High-performance data infrastructure un- data access patterns. layers: a data storage layer, a data fabric layer,
management of data. derpins and accelerates every phase of large • Exceptional ease-of-use and a data service layer.
• Availability: Raw data needs to be preprocessed AI model training, from data ingestion and Data infrastructure should feature global
and converted into training data. While there preprocessing in the early stage to training data management, efficient data mobility, Data storage layer
are numerous tools for data preprocessing, it data loading and periodic checkpointing in AI platform and toolset collaboration, and
is still important to have a data preprocessing the model training phase. High performance optimization and enhancement tailored In this layer, data is stored in different data
framework that works in synergy with the in this context is comprehensive and includes for large AI model training. This would centers.
accelerate AI training and learning by
ensuring that data is visible, manageable, In each data center, data is intelligently stored

Data Ingestion
and available. in the hot and warm tiers. The hot tier is
Model Training
Data Preprocessing
Phase Third-party collection, web Raw data +
crawling, edge aggregation, etc. Raw data Training data Training + Tuning Model
+ Evaluation

Data cleansing,
Data ingestion and import
conversion, augmentation, Training data loading Checkpointing
Centralizing and managing standardization, etc.
Descript ion data from different sources, Using training datasets and Periodically saving the
Obtaining high-quality datasets compute resources to train process data (archiving) for
in different formats, and for
from mass raw data and using large AI models resumable training
different purposes
diverse preprocessing tools

High data generation costs and

Storage
Support for multiple data
protocols (file, object, etc.)
diverse tool ecosystems Mostly files Mostly files The architecture of the Huawei AI data
High-capacity shared storage Terabyte-scale capacity and Terabyte-scale capacity and
Requirements Fast import and high
bandwidth
pools for huge amounts of data
generated during preprocessing
high OPS high bandwidth
lake solution consists of three layers: a
Data
Visible: A data map is defined and maintained to achieve visualized management across data silos. data storage layer, a data fabric layer,
Manageable: Policy-driven data mobility simplifies data ingestion.
Management Available: A data preprocessing framework works with storage devices for simplified preprocessing management and
faster preprocessing.
and a data service layer.
High-capacity resource pools (tiering available) and multi- High OPS and low latency Ultra-high bandwidth
Storage protocol interworking reduce data copies and capacity usage. shorten training data reduces GPU wait
Device High read/write bandwidth and fast random access accelerate raw loading times. times.
data ingestion and preprocessing.

Figure 2: Core capabilities of AI data infrastructure

72 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 73

Network Evolution Network Evolution

Data service layer convenience, which they can select on demand.

Intelligent tiering can be implemented between the The Huawei AI data lake solution provides The Huawei AI data lake solution is built on
commonly used service frameworks at the data years of experience and expertise in large AI
OceanStor A series and the OceanStor Pacific series.
service layer, including data preprocessing, model training. It helps enterprises eliminate
The two tiers are combined to externally present a model development, and application data silos, achieve smooth data mobility,
unified file system or object bucket. Internally, data development frameworks. and implement a data fabric between data
is intelligently and automatically tiered. applications and storage devices, making data
The data preprocessing framework performs visible, manageable, and available. As large AI
data cleansing, conversion, augmentation, models continue to evolve from unimodal to
standardization, and other preprocessing multimodal, the ever-increasing volume and
operations. Large AI model customers can diversity of data will inevitably lead to a non-
powered by the OceanStor A series, which is Data fabric layer integrate their algorithms and functions into linear increase in management complexity and
Huawei's high-performance storage specially this framework for simplified preprocessing performance requirements. An AI data lake
designed for large AI model training. It can The data fabric layer creates a seamlessly management. Furthermore, customers can solution running on a three-layer architecture
scale out to thousands of nodes. The warm interconnected network to facilitate data choose an alternative framework. can effortlessly address these challenges to
tier uses Huawei's OceanStor Pacific scale- mobility. It makes data visible, manageable, facilitate the development of large AI models,
out storage to handle mass unstructured and available, helping large AI model training Similar to data preprocessing, model development accelerate the emergence of intelligence in
data. Intelligent tiering can be implemented derive value from data. and application development are the other two model training, and push AI innovation and
between the OceanStor A series and the frameworks provided by Huawei for customers' development to new heights.
OceanStor Pacific series. That is, in the same Huawei uses a software layer powered by
storage cluster, the OceanStor A series nodes Omni-Dataverse to achieve data visibility,
Data preprocessing Model development Application development
form a high-performance storage tier, and manageability, and availability. Omni-Dat- Data service platform platform platform
the OceanStor Pacific series nodes form a averse is an important component of Huawei's layer

high-capacity storage tier. The two tiers are Data Management Engine (DME). It provides DME (Omni-Dataverse)
combined to externally present a unified file the unified management of metadata on
Visible Manageable Available
system or object bucket that supports multi- Huawei storage devices across different data
Global unified view In-depth storage-compute-
Efficient on-demand mobility
Third-party network collaboration
protocol interworking (one copy of data can be centers to form a global data asset view. heterogeneous mgmt.
Fine-grained access control
Near-compute cache acceleration

accessed through multiple protocols). Internally, Omni-Dataverse also invokes interfaces on Data fabric Global data asset view
data is intelligently and automatically tiered to storage devices to control data movement. layer DC1 DC2 DC3
achieve an optimal balance between capacity, Omni-Dataverse executes actions based on
performance, and cost. user-defined policies, and can implement GPU/
NPU-storage passthrough and intelligent file
Hot Warm
A data replication relationship can be created prefetching on demand to ensure training data
between storage clusters across data centers to is readily available, minimizing computing Intelligent Intelligent
Data storage Hot Warm DC3 Hot Warm
tiering tiering
facilitate reliable and on-demand data mobility power wait times. layer
between data centers. This ensures the data
DC1 DC2
device layer is fully prepared to support data This makes data ingestion and model training
ingestion for large AI model training. much more efficient, improving cluster utilization.
Figure 3: Architecture of Huawei AI data lake solution

74 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 75

03
Intelligent
Transformation
Intelligent Transformation Intelligent Transformation

Accelerating Intelligent Wang Su

Director, Integrated Solution Marketing Dept, Huawei

ICT with a Four-layer

Framework and Rapidly evolving AI technology is bringing carriers manage and optimize network resources

Transformation in
unprecedented opportunities and transformation better and improve O&M efficiency. At the same
potential to telecom carriers. At the business time, carriers need to explore and apply ICT
level, AI applications can help carriers increase solutions to support AI by providing stronger

Three Areas revenue by developing smart and easy-to-use

services. For telecom operations, AI can also help
data connection, storage, transmission, and
computing capabilities.

This article provides a detailed analysis of innovative AI applications in

business, telecom operations, network-cloud synergy, and infrastructure. It
Intelligence Accelerates ICT Transformation
discusses how Huawei uses a four-layer framework and transformation in
three areas to help the industry go intelligent faster and help carriers seize Business Diversified Data and
Transformation Connection Growth
new opportunities and overcome new challenges. This demonstrates Huawei's Content Interactions Connections
Digital humans, UGC AIGC
Business AIGC 3D, AR, VR
vision to work with partners to promote technological innovation. smart vehicles
Loop

Intelligence for ICT

600 GB 200 bn
Data Generation and Mobile DOU Connections
Utilization

Experience Customer-intent-based
Assurance
Transformation
UX O&M Telecom Expert-driven Model-based
Experience Assurance Efficiency Foundation Model Knowledge

Loop 20% 10x

Data Management and
Service Energy
Orchestration rollout time efficiency

Infrastructure

ICT for intelligence

Transformation Network-Cloud Synergy
Edge Center General-purpose Heterogeneous
computing Computing

Data Storage and Processing 10x 500x

Network- CPU NPU
Cloud Loop
Access DCI DCN
Connectable Task-based
resources Services
Data Connections and
Transmission 100x ms-level
Bandwidth Latency

Figure 1: Four layers and transformation in three areas

78 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 79

Intelligent Transformation Intelligent Transformation

New digital content will increase network

traffic by over 10 times and increase
the number of connections from tens of
billions to hundreds of billions, meaning
new business opportunities for carriers.

Centering around data, Huawei has proposed intelligence (AGI)—which will drive transitions technologies to enable a leap from content Experience transformation: A future
a four-layer framework and transformation in to new business models. In the text-to-image creation to holographic interactions. with intelligent network O&M
three areas for the intelligent era (Figure 1). The and text-to-video fields, foundation model
four layers are service, network management, algorithms and applications are maturing, Business (or business cycle) transformation In the field of network O&M, intelligent
cloud infrastructure, and network infrastructure. with diffusion-model-based algorithms already means new ways of producing data are technologies are changing service rollout,
The three areas of transformation are business, being widely used for image generation. The increasing content production efficiency resource configuration, and O&M processes.
experience, and infrastructure. Intelligent Transformer architecture used by text-to-video (machines, not just humans, as producers), These all center on transforming user
technology has changed the way data is models like Sora follows scaling laws and content complexity (HD, 3D, and holographic experience. The intelligent transformation of
produced, managed, applied, stored, processed, realizes video generation abilities. content) and distribution speeds (faster network infrastructure will follow three trends:
and transmitted on the four layers, and carriers interactions with and ubiquitous access
need to adapt to transformed business models, AIGC is dramatically transforming content to content). This is generating massive First, from event-driven networks to intent-
O&M-based experience, and cloud-network generation and seeing fast adoption in fields amounts of data and connections. It is pre- driven networks. Traditional event-driven
infrastructure in the intelligent era. like picture generation, short video creation, and dicted that by 2030, new digital content will networks respond only to specific events, while
movie production. AIGC is also transforming increase network traffic by over 10 times intent-driven networks can understand user
Business transformation: A new connectivity from connections between people and increase the number of connections intent and adjust network configurations to
chapter for AIGC and business value to comprehensive connections between people, from tens of billions to hundreds of billions, user needs rapidly and efficiently.
objects, digital humans, smart vehicles, and meaning new business opportunities for
New ways of producing data are driving drones. These new connections involve not carriers. New ways of producing and trans- Second, from empirical decision-making to AI-
business transformation. For example, AI is only manual operations, but automation mitting data are also transforming business assisted decision-making. Traditional network
learning to create new worlds through AI- and machine-to-machine communications. models to enable innovation in personalized management decisions are made based on
generated content (AIGC). This will potentially In addition, AIGC is transforming interactive products and services and improve user ex- expert experience and analyses. With the
take AI to a new level—artificial general experiences by combining AI, 3D, VR, and AR perience and loyalty. help of AI, network management personnel

80 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 81

Intelligent Transformation Intelligent Transformation

Huawei's innovative elastic transport

networks support task-based data
express services for users.

can make decisions more efficiently using Telecom networks use the following three layers discrete data related to network resources, • Cross-domain intelligence:
AI algorithms and data analytics, which also of solutions to make network O&M intelligent: services, and status, and provide digital Collaboration can be performed across
improves network performance and reliability. twins adapted to different domains. This domains. Huawei provides intelligent
• Network element (NE) intelligence: platform's single-domain, high autonomy platforms such as NCE-Super, ADO, and
Third, problem-solving evolut ion from More real-time sensing components and capabilities, from data collection, sensing, SmartCare to support collaboration across
scenario-specific models to foundation model AI inference capabilities can be added to and analysis to simulation, decision- domains, such as fault demarcation through
generalizability. Traditional AI models are network equipment. AI-native hardware making, and control, can ensure guaranteed coordination between IP and optical
created to solve specific problems in specific will enable networks to be capable of network connection quality and timeliness. domains, and intelligent orchestration for
scenarios. Foundation models can be used to finer-grained sensing and real-time For example, Huawei provides an IP-based cross-domain services.
create more generalizable problem-solving synchronization. digital map with six visualized layers—
systems. When used alongside scenario-specific • Single-domain intelligence: physical, network, slicing, routing, service, Telecom foundation models will be key to
models, they can solve complex problems An intelligent management, control, and and application—as well as capabilities such realizing intelligent network O&M. Carriers'
through analyses that streamline different analysis platform can be used to create as congestion view, experience view, and complex O&M processes in all manner of
network scenarios and processes. digital models of networks that associate fault view. scenarios across domains will require the

82 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 83

Intelligent Transformation Intelligent Transformation

capabilities of telecom foundation models and Infrastructure transformation: The large packets, which can easily lead to load path orchestration, as well as high throughput
single-domain autonomy to implement cross- network foundation for the AI imbalance, decreasing network throughput and and high-reliability elastic 10GE private lines,
domain collaboration and the autonomous computing era reducing computing efficiency. Hyper-converged to provide users with elastic, task-based data
completion of complex processes. Huawei's Ethernet DCNs can meet data center service express services (same-hour, same-day, or next-
Telecom Foundation Model, for example, The nascent AI computing era is seeing a requirements in different phases of development day delivery).
consists of three layers: the foundation model, major leap in computing technology from and different scenarios. Ethernets that integrate
telecom industry models, and applications. The general-purpose computing to AI computing. general-purpose computing, storage, HPC, and Traditional cloud infrastructure urgently needs
foundation model layer provides multimodal, This means the number of foundation model AI improve deployment and O&M efficiency and upgrading in areas such as data center power
computer vision, natural language processing parameters and demand for computing power reduce maintenance costs. Huawei's innovative supply and cooling, hardware infrastructure
(NLP), scientific computing, and prediction will grow significantly. Trillion-parameter network scale load balancing (NSLB) algorithm, clustering, and the integrated deployment
based on Huawei's Pangu models and third- foundation models will require AI computing for example, optimizes traffic paths to achieve and collaborative O&M of computing, storage,
part y open-source foundat ion models. centers that deliver 10,000-GPU computing global load balancing, delivering 20% higher and network resources. Key requirements
The telecom industry model layer provides power and efficient data management. This is Ethernet efficiency than the industry average for network infrastructure in the cloud and
industry-specific capabilities using high-quality making coordination across computing, storage, and 10% higher than IB. intelligence era include ultra-high bandwidth
industry corpuses and efficient toolchains. networks, management, and efficiency a key and ultra-low latency, moving storage and
The application layer provides role-based trend. Networks that support AI service growth The second technology is elastic data center computing from devices to the edge, high
copilots and scenario-based agents to improve are critical to realizing ubiquitous computing interconnect (DCI) and data center access securit y and reliabilit y, and task-based
employee efficiency and customer satisfaction. power and efficient data flow. End-to-end (DCA) networks. AI computing requires fast network services.
Huawei's Telecom Foundation Model provides collaboration between devices, edge, and cloud transmission of massive amounts of data for
out-of-the-box applications and supports across all manner of scenarios will improve training and inference. Traditional private lines The four-layer framework and three-area
local deployment and incremental training by overall computing efficiency. are constrained by their lack of service elasticity transformation were proposed to systematically
carriers to deliver intelligent technologies that and high costs. Superior to the current common outline the carrier opportunities presented by
meet different user needs. Two technologies are key to achieving network- practice of shipping hard drives via express the intelligent era and help carriers build the
cloud synergy. delivery, Huawei's innovative elastic transport required capabilities to seize these opportunities.
O&M experience transformation uses intelli- networks deliver end-to-end 400GE, SRv6, The convergence of networks, cloud, and
gent technologies like foundation models, and The first is hyper-converged DCNs. AI computing and elastic high throughput. These networks intelligence is already a major trend, as new
the three-layer intelligent telecom network is characterized by small data flows with use AI-powered data flow identification and opportunities are created by intelligence
solutions to bring intelligence to network infra- powering ICT, and demand for infrastructure
structure and build AI-native, intent-driven, and upgrades is generated for ICT to boost
digital twin capabilities. These features can help intelligence. Huawei looks forward to working
carriers guarantee user experience, agilely roll with customers and partners in exploring new
out services, optimize resource configuration, architectures, applying new solutions, and jointly
and improve O&M efficiency, accelerating developing new technologies to reap the value
network evolution towards autonomy. of intelligence for data and traffic.

84 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 85

Intelligent Transformation Intelligent Transformation

EM 2.0: A Road for Chris Meng

Director, Northern Africa ICT Marketing

the Digital Intelligent & Solution Sales Dept, Huawei

Transformation of
Wang Jie
Director, Northern Africa Marketing

African Carriers
Dept, Huawei

Liu Jifan
CEO, Huawei Ethiopia

The EM 2.0 model is a partnership between Ethio

Telecom and Huawei that has accelerated digitalization
in Ethiopia and provides a valuable roadmap for
transformation in emerging markets worldwide.

The light of digitalization is now shining on Digital and intelligent

Africa, transforming the continent into one that
is brimming with dynamism and a growing According to the White Paper on Global Digital
digital economy. Economy (2023) released by China Academy of
Information and Communications Technology
The Emerging Market 2.0 (EM 2.0) model – a (CAICT), the digital economy already accounts
result of successful collaboration between ethio for 46.1% of GDP in major economies, making it
telecom and Huawei – represents the marriage an increasingly important engine for economic
of technology and innovation. EM 2.0 is more growth worldwide. The digital era is presenting
than a technology upgrade. It also exemplifies conventional telecom markets with rare
a profound change in mindsets and models. opportunities to transform, and world-leading
By optimizing digital infrastructure, building a carriers have started exploring ways to grow
digital cloud foundation, deploying digital O&M through transformation. They are employing
platforms, and enabling diverse digital services, innovative ICT solutions to transform themselves
EM 2.0 has established a new blueprint for from traditional network operators that provide
Africa's digital economy. only connectivity services into digital solution
providers that offer diversified services and that
With 5G and other innovative initiatives pushing can even serve as enablers of national digital
Africa to the forefront of the global digital transformation.
economy, EM 2.0 is a model for success that
provides valuable experience and inspiration for Africa is an emerging market with huge
other emerging markets. demographic dividends — on average, it gains

86 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 87

Intelligent Transformation Intelligent Transformation

1.5 new mobile Internet users per second. The telecom market. The emergence of new telecom oriented business model that embraces network, Strategy, collaboration, and
continent is expected to have 620 million home carriers has intensified competition in the local cloud, and intelligent transformation. From an capabilities
broadband users by 2025, nearly double that market, leading ethio telecom to carry out a architecture perspective, local industry policies
seen in 2021. At the same time, 40% of the range of transformation initiatives that ensure and the carrier's basic networks remain as the How did ethio telecom find its path forward and
African population uses mobile payment services sustainable business growth and empower it basis of the EM 2.0 model. succeed in digital and intelligent transformation?
today. According to GSMA, the total number of to seize opportunities presented by Ethiopia's
registered mobile money accounts worldwide digital economy. T h e m o d e l re s h a p e s i n f ra s t r u c t u re b y In 2021, the Ethiopian government licensed a
reached 1.75 billion in 2023, with Sub-Saharan integrating the cloud platform through multi- new telecom carrier, shaking up the once calm
Africa accounting for over 70% of total growth. The carrier has worked with Huawei to drive cloud convergence and cloud-network synergy, market. This meant that ways to maintain
As the digital economy takes the African digital intelligent transformation through the enabling robust O&M, ensuring a good robust growth in the face of intense competition
continent by storm, the most strategically EM 2.0 model by building high-quality digital user experience and higher efficiency, and moved to the top of the list of concerns for the
important issues facing African carriers involves infrastructure, creating cloud infrastructure, supporting the development and operation of CEO and other top executives of ethio telecom,
how to transform business operation models, deploying a digital O&M platform, and innovative and diversified digital services that an established carrier that mainly provided
address new market needs during digitalization, developing innovative digital services. This create a new growth curve. traditional connectivity services. At that time,
and achieve new business success. has helped ethio telecom transform from a the government was preparing a strategy to
traditional telecom network operator into a The new networks, platform, and services develop the county's digital economy. ethio
ethio telecom's successful EM 2.0 digital and intelligent solution provider that has support one another and create a ladder for telecom analyzed the local market environment
transformation achieved enormous business success (Figure 1 continuous progression. and trends, and planned to implement digital
and Figure 2).
With a 129-year history, ethio telecom is a
state-owned telecom carrier and the second Unlike EM 1.0, which focuses on connectivity,
largest carrier in Africa after MTN. It has 69.5 EM 2.0 (Figure 3) uses cloud-based digital O&M
EM 2.0 Model Concept
Digital capabilities based on a network-cloud-intelligence architecture enable carriers to
million mobile users and dominates Ethiopia's to enable digital services, forming a future- strategically transform, forge new growth paths, and support national digital transformation

One Three Five

purpose directions layers
Total Operating Revenue of ethio telecom ethio telecom revenue from new digital services
Unit: Billion ETB Unit: Million US$

800 50 48 Digital service

Digital B2C B2B B2G
provider Develop digital Industry Government

Up
services Service Mobile Money
digitalization digitalization
700 758
40
600 Digital Digital O&M platform
613 +24% 33
O&M Digital tools / Joint operation / Intelligent maintenance

transformation
500 556
30
+10%

Strategic

Middle
400 Build digital Digital Cloud platform
platforms Cloud Multi-cloud convergence / Cloud-network synergy
20
300
Access Transport Core
Digital
200 Large-capacity + Simplified + Converged +
10 Infra Open Intelligent Reliable
100 5
3 Industry policies

Down
0.3 0 Network service Optimize digital
0 0 networks Spectrum, mobile money, cloud, rural
2021 2022 2023 Mobile payment Cloud B2B provider
networks, etc.
2022 2023

Figure 1: Revenue increase over the past three years Figure 2: Growth of three new digital services Figure 3: Conceptual framework of the EM 2.0 digital transformation business model

88 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 89

Intelligent Transformation Intelligent Transformation

and intelligent transformation based on its However, these changes were accompanied by a
business development and operation model. To number of challenges.
achieve this, the carrier built two foundational
capabilities: mobile payments and a cloud The first challenge was a lack of revenue
By the beginning of 2024, telebirr had
platform. It planned to develop diversified and despite an increase in users. Within a year over 40 million users, generating a
innovative digital services and solutions for of its launch, the mobile money product's user transaction volume of 91 billion birr.
consumer and enterprise markets, expand service volume quickly increased to 22 million, but
boundaries, and compete at a higher level. service revenue remained low. The service's
actual usage rate was low due to the carrier's
In May 2021, with the support of Huawei inexperience at operating new services, a lack
products and services, ethio telecom successfully of a specialized team, and the service's limited
launched the country's first carrier mobile functions (deposits, withdrawals, and transfers). First, the company adjusted its organizational operations, and strengthened ethio telecom's
money brand "telebirr" after just five months of These factors combined led to weak profitability. structure (Figure 4). It established an independent capabilities in market development and the
development, which was key to its transformation. As a result, the product generated only Mobile Money business department headed by operation of new services. With the support of
By the beginning of 2024, telebirr had over 40 around US$3000 in revenue per month, barely the CMMO, optimized the functions of the CMO's Huawei's expert team, ethio telecom quickly
million users, generating a transaction volume of contributing to the company's overall business department, and formed teams for marketing, identified telebirr's service operation issues
91 billion birr (about US$16 billion). As a result, and putting enormous ROI pressure on ethio business operations, sales development, and and market limitations due to a lack of diverse
the service received the gold award from Future telecom. The carrier responded with measures to technical support. This new organizational offerings. To address these issues, the carrier
Digital Accreditation Institute in 2023 in the best overcome this challenge. structure aimed to better support the company's planned a product development roadmap that
mobile money supply category. operations from strategy to execution. added mobile payments to the existing mobile
money service, upgraded mobile finance,
Second, ethio telecom and Huawei planned to and expanded the telebirr product series by
develop the mobile money service, aligning on introducing value-added services to meet local
ethio telecom's new organizational structure adapted to
strategic areas such as the future transformation market demand and increase service revenue and
business transformation
roadmap and cooperation model. They agreed to profits. In August 2022, ethio telecom launched a
CEO
add mobile finance to the mobile money service mobile finance product, pulling them out of the
in order to develop a super app that would run revenue downturn within less than a year.
on a full-stack service model, and then open up
CTO CMO CIO CSO CMMO CCSO ... market opportunities by launching fintech cloud, In June 2023, the monthly revenue of telebirr
public cloud, and government cloud services increased 500-fold to US$1.5 million, restoring ethio
Information Product Sales Mobile Money Customer
System Business service in three steps. This would in turn support telecom's confidence in business transformation.
Operation
the country's digital transformation in public
services, industries, and government services. The second challenge was a financial product
DCESO ... CBO DCMO
The partners also launched the joint operation of that failed to meet market needs. A lack
B2B Solution B2B Market Massive the mobile money service. of specialized experience in financial services
Design / Product Market
Innovation / Price Design / Product coupled with an incomplete understanding of
Delivery Insight / Sales Design and
Established in
Sales past 2 years The collaboration framework leveraged Huawei's potential needs regarding financial services
global experience in mobile money services, had resulted in slow user growth and a high
increased Huawei's involvement in service bad debt rate. Huawei assigned a specialized
Figure 4: ethio telecom's new organizational structure adapted to new services

90 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 91

Intelligent Transformation Intelligent Transformation

team of senior financial experts to join the joint station operators, causing mobile payment and other scenarios, empowering numerous and increasing user engagement. ethio telecom
operation team to help solve the problem. The transaction volumes to surge. By March 2023, industries and accelerating Ethiopia's digital has already applied FinCare across a range
team analyzed the product in terms of risk telebirr's monthly transaction volume had transformation. of scenarios, such as coffee shop promotions,
control design, tariff pricing, and user experience, reached US$3 billion, accounting for nearly Christmas promotions, and loan collections,
and provided ethio telecom with a dynamic 30% of Ethiopia's GDP. Another key factor driving ethio telecom's improving the marketing conversion rate by
pricing solution for its mobile finance product, successful transformation was xCare (Figure 5), a 60%. FinCare can display the locations of agents
optimizing over a dozen user experience metrics This meant that the carrier had successfully digital O&M platform tool that matches the EM and mobile users, allowing carriers to accurately
in the process. Quickly seeing market returns, broken the business ceiling. 2.0 architecture. xCare maximizes the potential deploy and manage agent outlets. It also
the bad debt rate dropped by 66% and profit of data using AI modeling and inference based provides location-based, real-time marketing
margins were much higher, laying the foundation While developing mobile finance services, ethio on big data from networks, users, and the capabilities that send SMS messages to users
for future growth. telecom also rolled out cloud services. In October market, providing the eyes, ears, and brain for near merchants, encouraging users to develop
2022, the carrier released the "telecloud" brand, market expansion, service operations, and O&M mobile payment habits.
The third challenge was the overall business which leverages a Huawei Cloud solution to efficiency improvements. For example, Huawei
hitting a ceiling. Through several marketing implement a digital transformation strategy has developed FinCare, the world's first mobile In multiple ways, FinCare helps quickly improve
improvement measures, telebirr's business scale based on both mobile payments and cloud. This finance operation support tool, to help with carrier capabilities to develop and operate
increased significantly, with monthly transactions service is also positioned as a 'national sovereign key issues like managing mobile money users, mobile money services.
reaching US$200 million (about 2% of Ethiopia's cloud' to help support the government's digital agents, and credit. In terms of user development,
GDP). However, growth quickly encountered transformation. FinCare identifies high-value information, such xCare provides a variety of other functions.
a bottleneck, and the company's performance as user consumption habits and transaction It offers scenario-based solutions, such
was a far cry from the government's expected Telecloud's first batch of cloud resources behaviors, to provide accurate marketing as UserCare, 2ndSIMCare, RuralCare, and
inclusive finance target. was snapped up five days after its launch, strategy suggestions for acquiring new users RoamingCare, to overcome typical problems
demonstrating great market enthusiasm.
In response, Huawei assigned a development Within ten months of launch, ethio telecom had
xCare
team of more than 30 engineers to work with encouraged multiple state organizations, such Precise Operation
ethio telecom. They scanned potential mobile as the High Court, to migrate their IT systems Service User Network
payment scenarios, such as public transport, to the cloud, and was officially awarded the operation operation operation

individual taxes, daily services, health services, national digital ID project by the government. FinCare UserCare 2ndSimCare RuralCare RoamingCare RevenueCare
Accelerating mobile Accelerating user 2nd-SIM-card Improving rural site RIO Improving roaming Precision network
and tuition fees, and developed services for the x money service
monetization
migration activation and operation efficiency revenue and
experience
investment aligned
with traffic growth
Scenarios
selected scenarios. They also launched a super In the enterprise market, the carrier promoted
2
1
app with 15 new services in two weeks, a process cloud-based industry solutions. To date, more
that would have previously taken two months. than 90 enterprises have become the carrier's
This helped the carrier shorten TTM and expand targets for market expansion.
service scope. 3 Operation
Enabling digital operation
Business
Accelerating service and user
Experience
Focus Achieving leadership in NPS
ethio telecom had integrated its fintech cloud, areas transformation development

Specialized data analytics based on the super public cloud, and government cloud services
app helped ethio telecom identify a high- based on its multi-cloud convergence strategy. Digital platforms Joint operation team Standardized process
3 SmartCare/MM BA+SA+DA Committed value
value mobile payment application scenario — With this strategy, the carrier works with Foundations Deployed for 200+ Experts with global Feasible business model
customers globally experience
gas station payments. This allowed the carrier ecosystem partners to develop innovative digital
to work with both the government and gas solutions for education, agriculture, healthcare,
Figure 5: Overview of Huawei's xCare digital platform solution

92 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 93

Intelligent Transformation Intelligent Transformation

facing the development of mobile broadband existing cloud and mobile payment capabilities Clear strategies: Carriers should formulate Suitable organizations and talent: To ensure
services in Africa. These include 4G user to create the one-stop telecom application digital transformation strategies based the robust development and operation of new
migration, improving rural coverage, activating store Tele-Gallery, forming a unified gateway on the stages of digitalization and policy services, carriers should consider adapting orga-
second SIM card slots, and international roaming for users, content, and data traffic. environments in a given country. Decisions on nizational structures to provide the specialized
operations and management. digital transformation can be driven by either talent required by services.
This would allow the carrier to monetize its national strategies or commercial factors.
In just three years, ethio telecom has achieved role as a platform and drive Ethiopia's digital In-depth joint operation: Carriers often face
its preliminary strategic goal of digital economy forward. Optimal service development paths: Carriers challenges when marketing new digital services,
transformation and has become a digital and should understand the local conditions for including insufficient expertise and a lack of
intelligent solution provider. Lessons from the EM 2.0 model developing innovative digital services and, as market experience. Joint operations, with the in-
they seek ways to transform, choose those that troduction of industry experts, agile development
However, it has continued on the transformation Twenty out of nearly 30 countries in Northern, best fit the local market (e.g., mobile money, teams, and digital O&M platforms, can make
journey. In December 2023, ethio telecom and Western, and Central Africa have developed cloud, and B2B). up for capability deficiencies and help carriers
Huawei expanded the carrier's business scope national top-level digitalization plans, with more achieve business success.
by leveraging its current digital capabilities. than 80% of these countries already licensing
For the B2C market, they planned a range mobile money services and over 20 carriers
of Internet services, such as social media, planning to deploy cloud platforms, paving the
e-commerce, streaming, and gaming. For the way for a digital Africa. As a pioneer of digital
B2H market, they planned one-stop smart transformation in the region, ethio telecom
In just three years, ethio telecom
home services with a premium network serves as a valuable reference for other carriers
has transformed into a digital and
experience (5G FWA , premium Wi-Fi, and across the continent. Important lessons have
FTTR-H). The partners also planned to add been learned in the following areas: intelligent solution provider.
digital ID and eKYC atomic capabilities to

EM2.0 service development paths

Digital transformation Platform monetization
Path 1: Step1 Step2 Step3 Mobile finance model model
Mobile Money Mobile Mobile Value-added Social
Super APP Smart city
wallet payment services networking
Smart
E-commerce
Service development paths

transportation
Government B2C
digitalization Smart taxation Gaming
Joint operation
Digital IDs Livestreaming
(Resources + Platforms +
Capabilities)
Smart education
Public service
digitalization Smart healthcare Six public-
B2B/B2G enterprise
Path 2: Step1: Step2: Cloud for large scenarios
Carrier cloud Scenario-based cloud enterprises
B2B/B2G cloud Enterprise
migration services digitalization Cloud for SMEs

Smart agriculture B2H Smart home

Industry Smart mining

digitalization Market as a Service
More paths: Smart home, 5GtoB…
Smart port

Figure 6: EM 2.0 service development paths

94 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 95

Intelligent Transformation Intelligent Transformation

Accelerating Intelligent ChatGPT rocked the world in 2023, and now is the time to take a hard look
at the current trends dominating AI development, innovation, evolution,

Transformation with
and competition. Clearer insights into feasible directions and asymmetric
competition strategies will be critical for China's AI industry as AGI draws near.

Application-driven ChatGPT 3.0's launch in December 2022 was like Breaking bottlenecks: AGI is

Industry Collaboration
the big bang for AI. Since then, development approaching a critical point
has continued at breakneck speeds, with new
foundation models, including GPT-3.5 and ChatGPT is propelling AI development forward
GPT-4, shortly following and artificial general using a model called the "data flywheel," which
intelligence (AGI) just over the horizon. AI uses statistical algorithms rooted in calculus

Li Changwei models now have more parameters than human and probability theory. This leaves ChatGPT to
brains have neurons, enabling models to make deal with two challenges: accuracy and training
Chief Strategic Marketing Expert, Huawei
inferences in similar ways as humans do. costs. ChatGPT 3.5 was trained on 1.8 trillion
parameters, costing nearly US$5 million per
AI has become the key to competitiveness training cycle, but its accuracy bottlenecked at
and leadership for the ICT industry in the the 90th percentile (Figure 1).
intelligent era.

Intelligence + Data Flywheel

Human-like AI
World-creating AI 3.0: Brain-like + Data +
ability percentile
Technical expertise
(%)
Brain-
Humanoid …
Data Flywheel Robot dog computer
robot
100% interface
World-generating AI 2.0: Data + Computing technology
Able to think and learn, preliminary understanding and generation, mainly using
the transformer architecture
AlphaGO GPT4 QoE+GPTs+Q-Star
80% xNN + yTransformer
Intelligence Flywheel
World-sensing AI 1.0: Brain-like + Computing technology Approaching and surpassing human
Object recognition and analysis, partially replacing human eyes and ears, mainly abilities , generating knowledge,
using the CNN/RNN architecture content, intelligence, and services

Deep
Cybernetics NLP CV GAN AI4S
learning
50%
AI technology
AI industry milestone
development
trend

AI
Milestones

1956 1997 2015 2020 2022 2023 … 2025 2030

Dartmouth Deep Blue beat Surpassed human Protein ChatGPT GPT-4
workshop human chess abilities in visual structure
champion recognition prediction

Figure 1: The data + intelligence flywheel approach is driving the evolution from AIGC to AGI

96 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 97

Intelligent Transformation Intelligent Transformation

AI is already changing the paradigm of Computing Index in 15 key countries correlated

economic growth. According to China's Ministry to a 3.5‰ and 1.8‰ increase in national
of Industry and Information Technology, growth economic growth and GDP, respectively. This
China's AI development has focused in AI computing scale can effectively drive new AI-driven economic development model
on strategy and business. growth in the digital and intelligent economy. will begin to impact global strategic competition
They found that every 1-point increase in the in 2024. (Figure 2)

25000 GDP(US$ B)

A3
x+
0m
To break through this bottleneck, OpenAI back-and-forth between these two components,
20000 3.
Y=
has adopted a new three-step strategy: First, initial solutions "evolve" into new knowledge,
USA
adding specialized AI, or a Mixture of Experts gradually approaching what's required for AGI. 15000

(MoE), which means linking 16 specialized A2

m x+
AI models with 110 billion parameters to Once achieved, AGI will set the stage for AI to 1.5
Y=
10000 China
the general AI foundation model which will be truly integrated into enterprise operations.
Japan
improve its accuracy. Second, providing open Microsoft's Copilot, for example, was launched x+A1 Germany
Y=m France UK
ChatGPT APIs to encourage ecosystem partners as part of its Office apps, with ChatGPT pre- 5000
Italy
South
Russia Korea
to generate their own specialized GPTs based integrated. Enterprise users have flocked to
Malaysia
Canada
on GPT and develop specialized intelligent this new service, despite its US$30 per month South India
Computing Index
Africa Brazil Australia
0
agents. Third, developing, integrating, and subscription fee, making it a new growth engine
10 20 30 40 50 60 70 80 90
absorbing new algorithms, such as Q-Star, to for Microsoft. Other current industry applications
Leading Rising Startup
get ever closer to AGI. include Tesla's humanoid robot Optimus countries countries countries
and autonomous driving system FSD V12, as
Tempted by the enormous market potential well as Huawei's Pangu Weather foundation
Infrastructure support
and the low market threshold, many major tech model, which have made breakthroughs in
Digital economy increases by 3.5‰
companies are increasing investment in AI. In critical indicators, like industrial experience and Computing capacity
2023, Google introduced its multimodal model
Gemini, which will help it move from AIGC to
accuracy, to shape new business models that
boost efficiency across industries. Computing efficiency 1-point increase in
Computing Index

AGI. The Google DeepMind team also recently GDP increases by 1.8‰
Application level
launched their FunSearch algorithm, which The turning point for AGI is almost here,
works by pairing a pre-trained large language according to many. NVIDIA CEO Jensen Huang
model (LLM) designed to output creative anticipates the advent of AGI within the next Regression analysis of Computing Index and GDP
Source: 2021–2022 Global Computing Index Assessment, Tsinghua University, IDC, etc.
solutions in the form of computer code with five years, while Tesla CEO Elon Musk believes
an automated "evaluator" that guards against that AGI will come within three years.
hallucinations and incorrect ideas. By iterating Figure 2: AI computing power is closely related to GDP

98 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 99

Intelligent Transformation Intelligent Transformation

including Facebook, Instagram, Messenger, enterprises to promote the construction of

Differentiated leadership: and WhatsApp. Apple is using AI to improve In terms of top-level strategy design, national new AI infrastructure by integrating network,
The emerging path forward for AI Siri and is expected to launch generative AI resources and those of individual enterprises t ra n s m i s s i o n , co m p u t i n g , i n t e l l i g e n ce ,
functions for its iPhones and iPads in 2024 are being organized along specified and security elements. This is helping the
AI used to be seen as an open-source and with the release of iOS 18 and iPadOS 18. development paths from the top down to nation to build an AI computing platform
non-profit technology. However, as it matures, Finally, Amazon has leveraged its advantages achieve collaboration and boost industry that will serve as critical infrastructure and
it is increasingly found at the center of closed in cloud to launch Titan AI, which includes revenue (Figure 3). promote both corporate and academic AI
and business-oriented competition. Tech text models for content generation and research and development. The aim here is
giants that used to collaborate with each embedding models that can create vector Early on, the government played a leading to build a natural momentum that will drive
other have become divided and focused on embeddings for efficient search functions. role through preferential policies, strategic development.
competition. Amazon also launched the CodeWhisperer, an investment, and data sharing. It started by
AI coding assistant that is free to users. setting digital security rules and application The government has also prioritized ecosystem
Major tech companies now either want to requirements for the government's own integration and policy collaboration at the
work with OpenAI, the biggest name currently Each of these tech giants has chosen a digitalization and intelligent application industry and enterprise levels to prevent
in the game, or follow in its footsteps. different way to develop AI. But together, scenarios. Then, it turned to state-owned excessive competition in AI foundation models
Microsoft works extensively with OpenAI, they represent the convergence of AI, security,
outperforming everyone else in LLMs. Copilot cloud, services, and devices, as all focus on
was jointly launched by the two companies. innovation in scenario-based application 1
experiences and value integration, which
Unified top-level design
Anthropic, which has a different approach ultimately enables differentiated leadership.
Government investment and data sharing
to AI safety, is now one of the biggest
competitors of OpenAI. Google, whose search Core AI players have added algorithms, like AaaS AaaS
Applications
business faces the greatest risk of being specialized AI, new Q-Star, and FunSearch, to
replaced, has launched its own AI services data statistical analysis methods to maximize 3 SaaS 4
Security
including Bard and Gemini. Google has also their strengths and move from driving
Top-down & Revenue from
taken advantage of YouTube's service data exponential growth in AI computing demand PaaS
application- Models industry
and algorithms to solidify its leadership to driving logarithmic growth. driven collaboration
PaaS
in multimodal technology. The FunSearch Leverage strengths Algorithms Leverage the advantages
in industry scenarios, of integrated resources,
algorithm recently launched by the Google China's approach: Strategy- and specialized AI, and avoid excessive
PaaS
DeepMind team now allows Google to business-driven AI development application innovation, Data competition between
and ecosystem carriers, vendors, OTT
potentially outperform other players in AGI. players, and VPs.
IaaS Computing power
Tesla launched AI systems that are closely International tech giants are not the only ones Lossless network IaaS

linked with its electric vehicles, such as FSD pursuing AI development. China is doubling
V12 and Optimus, to establish differentiated down in this field. The country's AI strategy 2
leadership. Meta AI has integrated more is seen as critical to future breakthroughs
New infrastructure enables the integration of network, transmission,
t h a n 2 0 n e w A I G C m e t h o d s , fo c u s i n g in both the ICT industry and growth of the computing, data, intelligence, and security resources
on improving user experience related to nation's digital economy. Generally, China's Use the strengths of networks to build a leading new AI computing infrastructure platform in a distributed
manner and promote ecosystem innovation and development
search, social discovery, advertising, and AI development has focused on strategy and
business communication across its platforms, business.
Figure 3: Government-led top-level design + Carrier-led new infrastructure + Application-driven + Industry collaboration

100 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 101

Intelligent Transformation Intelligent Transformation

at the PaaS layer and preserve breakthrough layer. AI computing clusters can make up for

opportunities at the SaaS and application insufficient single-point computing power and

layers. integrate scattered computing resources to

power the upgrade and iteration of developing Every 1-point increase in the Computing
This top-level design is supposed to ensure an and verifying AI foundation models. Carriers Index in 15 key countries correlated to
are prime representatives for state-owned
application-driven, top-down AI development a 3.5‰ and 1.8‰ increase in national
path. Commercial breakthroughs are expected enterprises (SOEs) here, and are expected to
economic growth and GDP, respectively.
to be made thanks to China's diversity in low- replicate NVIDIA's AI Foundations strategy

level application scenarios and AI expertise in while building new AI computing centers.

some domains. This will further encourage the They will be able to turn a profit by enabling

development and quick adoption of smaller technology companies and scientific research

models, specialized AI applications, and institutions and by leasing AI computing

carriers upgrade their primary commercial
innovative solutions (Figure 4). resources.
services through AI-based service integration, New MaaS: The AI collaboration
while achieving real growth in new market strategy for carriers
However, AI infrastructure is prioritized over By expanding and strengthening AI computing
segments.
applications in China so as to leverage the centers, they will be able to implement
For carriers, intelligent digital platforms play
country's network advantages at the IaaS "inside-out" service enablement, where
Microsoft has simplified complex service a key role in integrating telecom services with
management capabilities through AI-based AI. This is how they create new capabilities
integrated operations, which can be a big help and competitiveness in networks, operations,
The "iPhone" moment for Cloud here. For example, network automation can and O&M. As the foundation of intelligent
Applications & Solutions (Cloud-based AI video conferencing, XR, B2B…) → Intelligent
transformation (Platforms + Ecosystems)
help carriers simplify network management digital transformation for carriers, cloud
and improve customer services. SKT in South and intelligent technologies make networks,
Korea has set a benchmark for this kind of computing power, and service configurations
Wave 3: Wave 4:
Converged video
3 4 XR & AI cloud
strategic AI transformation. In November more flexible, efficient, and affordable. This

conferencing & AI 2022, SKT CEO Ryu Young-sang announced has increased their focus on a large number
cloud the company's plan to become an AI of multi-service scenarios (Figure 5).
Business-oriented AI company. It has since worked with partners
growth
to develop a Korean large language model, When it comes to AI collaboration, carriers are
Supportive cloud players
(central SOEs) which is similar to GPT-3. In May 2022, SKT now focusing their strategies on integrating
Intelligent O&M & Wave 5:
Operation cloud 2 "Eastern Data and Western Computing"
+ "Shanghai Center" 5 B2B & AI cloud also launched a Korean AI chatbot "A .", their vertical industry value chain resources
attracting more than 1 million users. At MWC (security, data, and networks) to form a
Key cloud players (industry)
AI computing centers + Open 2023, SKT showcased numerous intelligent new Model as a Service (MaaS). Under it,
ecosystems
services based on foundation models. These layer-1 XaaS technologies and capabilities are

1 services include companion AI, data AI, visual provided and supported by leading technology
AI, medical AI, and urban air mobility (UAM). vendors. Carriers, on the other hand, focus
Localization, competitiveness, and integration of network, transmission, SKT's key advantages over other carriers here on the second layer where they develop
computing, data, intelligence, and security resources is that it possesses both telecom know-how middleware to integrate XaaS at each layer.
and proprietary local data that OpenAI does This new MaaS encapsulates and integrates
Figure 4: AI foundation model services + New service planning not have.

102 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 103

Intelligent Transformation Intelligent Transformation

vertical elements throughout the value chain we must aggregate limited industry resources
leveraging operation and infrastructure through top-level design and strategic
Customer advantages. planning. This will be the only way to continue
2 H E G I T M experience
forward with the asymmetric competition
O - Online 5
applications MaaS will serve as the broad infrastructure strategy that has defined the mobile Internet

AaaS: Low-code development Application

Ecosystem-based that underpins the AI era, providing secure, era. We should start with smaller models and
Innovation - Online
platform for industry applications, ecosystem
efficient, and low-cost model usage and specialized AI, and collaborate with verticals
Open applications
general apps, etc.
cooperation
Advantages of localized 4 d e v e l o p m e n t s u p p o r t fo r d o w n s t re a m to achieve joint innovation and breakthrough.
carrier operations:
1 Middleware 5
Channels | Services
applications. Carriers will direct this usage, Iterative growth and mature AI software,
Customers | Brands
Marketing and cooperation | a s t h e y wil l prov ide t h e ke y t e ch nica l hardware, and foundation model technologies
Small
aaaS: Standardized, modularized, and Ecosystems
capabilities and industry resources needed, will be achieved through the application and
Applications & Solutions = Experience & business breakthroughs

application
Digital sovereignty | Data security |
containerized vertical apps… models Open
Privacy protection Carriers
cooperation 3 preventing excessive competition at the PaaS- business success of smaller models.
1 Middleware 4 Carriers layer, streamlining the vertical ecosystem,
Middleware 5 accelerating breakthroughs in applications These efforts will finally enable effective

MaaS platform
Specialized
SaaS: OS + Office + ERP/CRM + foundation models
Specialized foundation models and solutions, and achieving leadership in growth in new market segments and allow

Intelligent digital platform

Videoconferencing + Vertical apps... Open
AaaS: Standardized,
AI-led

cooperation
modularized, and
containerized vertical apps…
both business and scale. This will shape a countries, industries, and enterprises to build
positive cycle of technology iteration driven by their competitiveness in the new phase of
1 Middleware 3 Middleware 4

applications and businesses. vertical industry integration, innovation, and

Carriers
Specialized foundation models
SaaS: OS + Office + ERP/CRM
PaaS: O&M + Cloud + AI + Security +
GP foundation
models
+ Videoconferencing + Vertical transformation.
apps...
Data + AI computing... Accessible
The accelerated integration of AI into industry
to us Middleware 3

GP foundation models
triggered by ChatGPT has made it clear that
1 Middleware 2 PaaS: O&M + Cloud + AI
+ Security + Data + AI
computing...
Carriers
IaaS: Network + Storage + IoT + Power
supply + DC... Middleware 2
"Leading
IaaS
layer" IaaS: Network + Storage + IoT
+ Power supply + DC...
1 Middleware 1

Middleware 1
ADN: Native AI, self-optimization, self-
organization, and self-evolution Advantages of localized telecom
Carriers resources: Clouds, networks, edges,
and devices; network planning,
5G/6G/Satellite 1 Gbps/10 Gbps fiber construction, maintenance, and
networks networks optimization; and localized services
Carriers

Ecosystem-based
Innovation - Offline
Carriers
applications

HT ET GT IT TT MT Customer
experience 5
O: Offline devices

Figure 5: Transforming MaaS to build an intelligent digital platform and ecosystem leadership

104 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 105

Intelligent Transformation Intelligent Transformation

AI-ready Cloud Drives Wang Xiaobin

Intelligent Digital Chief Architect, ICT Computing Products and Solutions

(Carrier Domain), Huawei

Transformation for Carriers

To achieve business success through
digital and intelligent transformation,
carriers need to be AI-ready in terms
of corporate architecture, business
models, and infrastructure.

released the generative AI office suite tourism, NetEase's Ziyue for education, JD
Microsoft 365 Copilot, and Google DeepMind Health's Jingyi Qianxun for healthcare, and
launched Gemini. These are accelerating Ant Group's finance AI model. These are
ChatGPT: Igniting AI models globally the application of foundation models to facilitating the evolution of AI capabilities
industry and individual markets and guiding from perception to cognition and from
ChatGPT provides a complex mix of answers that can be amazing, amusing, AI development towards multimodality and identification to generation, while expanding
or utterly wrong. intelligent agents. the scope of AI applications from general-
purpose to industry-specific. AI is expected
GPTs have put AI, an industry with a history of over 70 years, under the In China, many major tech companies have to be applied to over half of core industry
spotlight again. The emergence of ChatGPT is regarded as a turning point joined the AI race by launching their own scenarios over the next two years.
for AI, transitioning it from a technology that understands the world to foundation models. These include Baidu's
one that shapes it. The integration of technologies, such as scaling laws, Ernie Bot, Alibaba's Tongyi Qianwen, Huawei's We can already see the potential of AI for
reinforcement learning, and human feedback, has given ChatGPT human- Pangu, and 360's Smart Brain, Kunlun reshaping all services in the carrier industry.
like inference capabilities. Tech's Skywork, JD's ChatRhino, iFLYTEK's Since the marginal cost of foundation
Spark, Tencent's Hunyuan, and SenseTime's models approaches 0, applying AI to internal
OpenAI went on to launch a series of innovative products, including SenseNova. In just six months, the number operations can help carriers reduce costs
GPT-4 (1.8 trillion parameters), GPT-4 Turbo, Assistants API, ChatGPT of foundation models in China increased and improve efficiency. Leveraging AI in B2C,
Enterprise, and GPTs for ChatGPT in 2023, as well as Project Q* and Sora, from around 100 to more than 200. Many B2H, and B2B services will allow carriers to
which were just disclosed. Other players on the industry value chain SMEs have also released vertical models, generate higher revenue. For example, AI-
are also accelerating innovation: Humane launched AI Pin, Microsoft including Trip.com's Xiecheng Wendao for powered CCTV solutions can create a 15%-

106 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 107

Intelligent Transformation Intelligent Transformation

settings, build nearly 7,000 AI models, and Using platforms to create AI-ready business
create over 20,000 digital employees that models, promote industry collaboration,
In the future, foundation models will provide serve Huawei organizations in more than and foster a thriving ecosystem
ubiquitous intelligent services through 170 countries and regions around the world.
tremendously diverse business models. Huawei significantly reduces operation There is no lack of companies or other players
costs with a company-wide foundation who are actively exploring applications for
model-based intelligent agent that gives all foundation models. Explorations into new
employees access to an AI assistant. AI applications for industries will likely
result in high trial-and-error costs and even
In terms of products, the AITO M9, which failure. Although carriers can participate in
plus premium over traditional solutions, and framework consists of four components: is equipped with Huawei's intelligent creating foundation models, their greater
the AI-powered 5G New Calling service can business architecture (BA), information/data driving solution, caused a sensation in the strength lies in ubiquitous cloud-network
generate more than 10% higher ARPU. AI architecture (IA), application architecture new energy vehicle market in 2023. The infrastructure that can serve as supporting
is accelerating carrier digital transformation (A A), and technology architecture (TA). voice assistant Celia, which is pre-installed and monetization platforms for foundation
with carriers around the world evolving Without an enterprise architecture that in every Huawei smartphone, improves models, as well as existing industry customers
cloud-based transformation to cloud and AI- maps strategic objectives to execution, user experience. AI-enabled energy-saving that can form a potential market as carriers
based intelligent digital transformation. digital transformation is unlikely to succeed. algorithms allow Huawei's wireless base look to commercialize foundation models.
Therefore, it is recommended that carriers stations to consume over 20% less energy Carriers' computing and O&M capabilities
Lower costs, better efficiency, and integrate AI into their enterprise architecture than competing products, in turn driving up and credibility in local communities give
higher revenue so that AI can power their strategies, carriers' revenues. In the carrier industry, them the potential to provide professional
businesses, data, and technologies. China Telecom and China Mobile also made platforms for local AI markets. With such
To seize the enormous opportunities presented "AI+" part of their group strategy in 2023. platform capabilities, carriers can engage
by AI, carriers require advancements in three At the same time, the major initiative of
areas to be AI-ready. digital transformation may encounter large
setbacks during execution. Involvement
Building a foundation model-based MaaS platform
Making AI part of the transformation and promotion by top executives will be
Working with partners to provide a one-stop foundation model platform for enterprises that covers
strategy and creating an AI-ready key to making carriers AI-ready. Huawei, the full lifecycle of foundation models, allows customers to use them out-of-the-box, and quickly
develop industry applications.
technology architecture for example, announced its AI strategy at
Industry-specific solution
HUAWEI CONNECT in 2018. In that same year,
Carriers must make AI a key part of the company's founder Ren Zhengfei issued Industry Model-invoking
Functions SDK
applications APIs
their digital transformation and overall a resolution on increasing investment in AI Dataset

transformation strategy, with specialized and using AI to improve corporate efficiency, AI development platform: training, development, inference

teams in place to implement AI strategies establishing the AI Enabling Dept headed

MaaS platform
and build AI-oriented enterprise architectures by Ren himself. Today, Huawei has already
General-purpose Telecom foundation Open-source
and capabilities. TOGAF is an enterprise made AI part of every business process in foundation model model
Partner models
models
architecture framework that bridges enterprise every department and business unit. To date,
Computing power resources
strategic planning and IT solutions, and serves Huawei has used AI to create more than 600
as the core of enterprise digitalization. This intelligent applications in over 80 business
Figure 1: Foundation model-based MaaS platform capability framework

108 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 109

Intelligent Transformation Intelligent Transformation

and aggregate foundation model providers AI computing, and supercomputing. This alli-
for business exploration and provide the ance publicly engaged influential foundation
Model as a Service (MaaS) business model model partners across the industry. At the Carriers need to be AI-ready in terms of strategy,
(Figure 1). This will enable monetization, Digital Technology Ecosystem Conference
organization, capability, and infrastructure.
encourage more partners to sell their model in November 2023, China Telecom and
products on the platform to form a long tail its partners launched the first batch of 12
market,prevent carriers from having to make industry-specific foundation models for trial
foundation models themselves that compete commercial use, which involved sectors like
with their partners, and reduce trial-and-error education, construction, finance, and mining. For individual model developers, carriers ness, cloud will be essential for AI in the
costs for uncertain models. These foundation models are all embedded should consider taking the lead to develop early stages, but AI will become a key part
on the Xingchen MaaS ecosystem service a foundation model ecosystem community of cloud further down the road. Over the
China Telecom, for example, launched a platform. By integrating networks, cloud, similar to Hugging Face to maximize the past few years, China Telecom has created
foundation model ecosystem cooperation intelligent computing, AI, and partner models, ecosystem value of foundation models a nationwide e-Cloud service. In March
alliance in July 2023 to help implement China's this business model is driving the intelligent a c ro s s s o c i e t y. R e a s o n a b l e i n d u s t r i a l 2024, China Telecom's Chairman Ke Ruiwen
cloudification and digital transformation upgrade of Telecom China's cloud services division, platform building, and ecosystem said at the company's 2023 Annual Results
strategy and accelerate the construction of ICT and is already generating commercial returns. development should be key objectives of Announcement, "Without AI, cloud does not
infrastructure centered on general computing, carriers in the AI foundation model era. have a future." He also claimed that e-Cloud
will accelerate the upgrade to intelligent
Building AI-ready infrastructure with AI- cloud to become a leading foundation model
1 2
powered clouds and data computing service provider in China.
L0 L1 L1 L2
L0 training L1 Pre-training Incremental Training Model pipeline In the future, foundation models will provide Carriers should build AI-ready intelligent cloud

App1,App2,App3,App4… ubiquitous intelligent services through solutions based on the three key elements of
L1 Telecom API tremendously diverse business models. These AI: computing power, data, and algorithms.
Foundation models
foundation models can be embedded in a An AI-ready cloud platform should feature
L2 Scenario-specific models
variety of software and hardware systems, one architecture and two capabilities:
such as intelligent vehicles and robots, to
L0 Foundation Models Anonymized industry
make their way into commercial markets in • Computing power: Distributed multi-layer
(Pangu/LLaMA.,etc) dataset L1 Telecom Private
Foundation models dataset the form of intelligent products. They can cloud architecture for ubiquitous comput-
also be deployed in the cloud to provide cloud ing power
services based on commercial foundation Distributed multi-layer cloud architecture
Incremental models and transform existing business logic. requires collaboration between the public
Full Training Training
Edge Edge
100B Parameters Model Pipeline Cloud and network infrastructure are the cloud, on-premises cloud, and edge cloud.
>1K~10K GPU/NPU
< 100 GPU/NPU
pipes through which AI can reach end users Full training is performed on the public
Huawei Cloud Stack (consumers, businesses, and homes). AI can cloud, incremental training and centralized
Public cloud On-Premises Cloud only reach where cloud and networks are inference on the on-premises cloud, and
available, so an AI-ready intelligent cloud is other inference on the edge clouds (Figure
Cost Saving Telecom Know-how Data Privacy AI Model Ownership
needed for carriers to enter the AI industry. 2). This architecture brings three benefits for
Figure 2: Customized models based on collaborative training on public and private clouds For carriers that are engaged in the AI busi- carriers to deploy foundation models.

110 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 111

Intelligent Transformation Intelligent Transformation

• Reduced costs: The pre-training of trained foundation models can be deployed

A powerful pipeline for end-to-end foundation model innovation
foundation models requires huge amounts on the on-premises cloud where incremental
of AI computing power (thousands of training can be performed based on a small
GPUs) within a certain period of time Super-fast
amount of private data of carriers or specific Agile development Quick training Efficient tuning
deployment
(weeks or months). For example, the industries. This exponentially reduces the 10 lines of code 20% < 2 weeks
20x < 50 ms
Accuracy &
LLaMA 2 70B model was trained using a required hardware investment (dozens of Planning Foundation model
development
Foundation model
performance tuning
Model Inference Launch
training speed compression latency
distributed supercomputing cluster with GPUs). This model featuring full training on
2,000 NVIDIA A100 GPUs, while the Falcon the public cloud and incremental training on
180B model used 4,096 NVIDIA A100 GPUs. on-premises clouds means a 100-fold saving
Most carriers, especially those outside on investment in computing power. Figure 4: A powerful pipeline for end-to-end foundation model innovation

China, do not need to purchase computing • More secure data: Laws and regulations
power in advance for this purpose. They in many countries prohibit data (such as
can instead lease public cloud computing network information and traffic information), lake that brings together scattered data lakes major open-source models. Carriers need to
power. This allows them to perform full which usually is high-quality datasets for and warehouses. This makes it possible for build independent model-related capabilities
training with the 10,000-GPU computing model training, from being transferred out one copy of data to be shared by multiple that can ensure data security and the
power already deployed on the public cloud of the local network or abroad. On-premises data analytics engines and AI engines. It also manageability and controllability of models.
and publicly available industry datasets. Pre- clouds and edge clouds can meet these enables the development of model training-
requirements. Foundation models trained on oriented data processing capabilities, such as We are stepping faster into an AI era as the
such clouds are already built on privacy and the efficient integration, cleaning, filtering, fourth industrial revolution approaches. Carriers
Converging data warehouses and
sensitive data alongside inherent knowledge. and labeling of datasets. need to be AI-ready in terms of strategy,
lakes with AI
This means they run locally and privately, organization, capability, and infrastructure.
Collaborative workbench ensuring data security. • Algorithms: Cloud-based model production They should use their cloud-network strengths,
• Better service experience: Edge clouds can line enabling lifecycle management of position themselves appropriately in the
Unified data governance & security bring AI inference closer to end users and foundation models industrial division of work, and build their
provide lower-latency experiences. The lifecycle of a foundation model includes ecosystem platform capabilities to provide
Unified data insight key phases such as development, training, Model as a Service. This will allow them to
• Data: Cloud-based data production line for fine-tuning, and deployment. Tools and bring together more industry partners and
Unified data analytics efficient data processing capabilities adapted to these phases should succeed in intelligent digital transformation.
While data is a core asset of carriers, it is be deployed on the on-premises cloud as
Unified metadata and data resources scattered in independent IT systems and does services. Simply put, the on-premises cloud
not flow. Not only are there hardware costs should enable the development, training, fine-
for storing data, its value remains untapped. tuning, and deployment of foundation models
Local cloud services require full-lifecycle one- as services to help carriers manage models
Huawei Cloud Huawei Cloud IDC data
data lake Stack data lake lake stop data governance capabilities (Figure (Figure 4). When it comes to ecosystem
3), including pool-based data storage and compatibility, cloud services should support
cross-domain collaborative data scheduling
and management. Data lake-warehouse
Figure 3: Data-AI convergence: An efficient
cloud-based data foundation for the AI era convergence can help create a logical data

112 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 113

04
Innovation
for the Future
Innovation for the Future Innovation for the Future

Jointly Defining ICT

Technologies (ICT) like 5G, cloud computing, and Driving the commercial success of 5G is top of
AI are seeing broad adoption on a global scale. mind in the industry. Carriers around the world
In addition to revitalizing user experience, these have been exploring a number of different

Architecture for 5.5G:

technologies are also bringing exciting changes business models for 5G, and along the way have
to the telecom industry, like new 5G business made huge headway in monetizing their network
models, services, and devices, new forms of home capabilities.

The Key to Unlocking connectivity, AI applications, and carrier-operated

public cloud services. These new developments Some notable examples include: The Finnish

New Opportunities
will give carriers the opportunity to maximize the carrier Elisa differentiates 5G packages based on
value of their infrastructure and stimulate new preferred downlink speed, which has driven more
demand. users to upgrade and ultimately increased Elisa's
5G ARPU by more than €3.
But to get there, carriers also need to adapt:
Dang Wenshuan
They'll have to redefine ICT architecture, evolve China Unicom Guangdong launched special 5G
Chief Strategy Architect, Huawei their ICT infrastructure, embrace autonomous livestreaming packages, which provide subscribers
networks, rethink their public cloud strategy, and with greater uplink speeds, network priority levels,
reassess their approach to data storage. These and data limits for a smoother livestreaming
efforts will put carriers in a better position to seize experience. By the end of 2023, these packages
new opportunities and raise the entire industry to brought in 500,000 new subscribers, increasing the
Fo r c a r r i e r s t o m a x i m i z e t h e va l u e o f t h e i r
new heights. carrier's ARPU by 76% compared to regular 5G
infrastructure and stimulate new demand, they
service packages.
need to redefine ICT architecture, evolve their ICT
infrastructure, embrace autonomous networks, Six megatrends in telecoms
rethink their public cloud strategy, and reassess their Thai carrier AIS now provides on-demand 5G
approach to data storage. Trend #1: New business models are paving boost packages, while Three in Hong Kong offers
the way for 5G growth 5G packages specially designed for stock market

116 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 117

Innovation for the Future Innovation for the Future

The six megatrends in telecoms will

stimulate new demand and present
industries like media, healthcare, education, ports, world provided 1-Gbps connections for roughly
new opportunities. manufacturing, oil & gas, and mining. 200 million home users. 1 Gbps has effectively
become the new standard for home broadband,
New 5G services and applications are undoubtedly driven by nonstop innovation in passive optical
creating new growth momentum in the industry. network (PON) and optical distribution network
(ODN) devices. The cost of FTTH build-out can
Trend #3: Novel smart devices are be halved by using dual-port Building PON and
tracking and gaming. Both carriers charge their Trend #2: New 5G services and applications
multiplying transparent adhesive fiber.
subscribers based on customized network priority are flourishing
levels, latency, and traffic volume.
Smart devices are the primary vehicle through We are also seeing an exponential increase in
In 2023, we saw incredible progress in new 5G
which consumers make use of network capabil- the number of homes with fiber to every room.
China Mobile launched 5G service packages services and applications that the industry has
ities, and they are also a major driver of industry Twenty-two carriers have launched Fiber to the
specifically for delivery couriers. In China, delivery been looking for.
development. A number of novel smart devices hit Room (FTTR) services as of 2023. Now, FTTR is
workers go through tons of data and minutes to
the market in 2023. So far, there are already five benefiting 12 million home users and bringing
manage logistics and contact their customers. In China Mobile Jiangsu launched its New Calling
smartphones powered by 3CC carrier aggregation better connectivity to 300,000 small- and medium-
addition to providing custom 5G packages with service in September 2023 and attracted 1.5
(CA) – a capability that's fundamental to deliver- sized businesses.
more minutes and data, China Mobile also offers million subscribers in only three months. Within
ing the 5.5G experience. The availability of these
road accident insurance and dedicated lounges the first six months of launching its cloud phone
smartphones suggests that device vendors have In November 2023, the European Telecommunica-
in their service centers, where couriers can take service, China Mobile signed up 11 million users.
outpaced network providers in the upgrade to 5.5G. tions Standards Institute (ETSI) released the F5G
a rest, get something to drink, and charge their Globally, 5G fixed wireless access (FWA) is now
Advanced Generation Definition in a well-timed
phones. These diversified offerings have attracted serving more than 160 million users, attracted
Spatial computing and AI technology are giving move to propel F5.5G forward.
a number of high-value subscribers. by 5G's enhanced experience and the growing
rise to innovative applications like glasses-free 3D,
affordability of 5G customer premises equipment
MR devices, and intelligent connected vehicles, as Trend #5: Foundation models are enabling
Moving forward, new 5G business models that (CPE). To date, there are now 155 carriers offering
well as incredible new AI devices – all of which are intelligent network O&M
monetize different combinations of service commercial FWA services around the world,
creating new possibilities for the telecom industry.
capabilities will keep emerging as carriers work double the number in 2021.
There are currently more than 300 AI applications
to better understand behavioral changes in the
Trend #4: Innovations in FTTH and FTTR are for the telecom industry. And yet, these days,
digital age. And by providing more targeted Carriers outside China are also rapidly scaling
speeding up the arrival of F5.5G carriers deploy nearly 80% of their O&M engineers
communications services, carriers will be able to their 5G business in the B2B market. Between
to work on things like network monitoring,
generate more lucrative returns and set the stage 2022 and 2023, 5G private networks for enterprise
The Fiber to the Home (FTTH) market is booming. troubleshooting, and complaint handling.
for a healthy 5G business. customers also doubled to 222, spanning
In 2023, more than 50% of carriers around the

118 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 119

Innovation for the Future Innovation for the Future

This is where an "AI for Telco" strategy can help, this platform offers more than 60 localized
harnessing the power of AI to address unique SaaS applications, exceeding the number of its
Intelligent Operation Network Cloud Public Cloud
pain points in network O&M. Role-specific AI globalized SaaS offerings.
UPF
copilots and scenario-based AI agents – built on ...

foundation models and digital twins – can help to As digitalization continues to sweep across Autonomous Cross Domain O&M Core Network
Single Domain O&M
B2B & B2C
better equip O&M engineers and further enhance the globe, localized digital services like mobile Network L4 Intelligence Plane
Foundation Digital & B2H
Control Plane
customer satisfaction. payments, online shopping, food delivery, ride Model Twin

hailing, and digital government are flourishing.

Unified Data Storage
AI for Telco applications are growing in number This presents huge opportunities for carriers with
and coming into wider use. For example, AI local service advantages to develop public cloud Ubiquitous 10Gbps
copilots are helping call centers greatly increase services. These opportunities are unfolding in Asia-
400G/800G Bearer
25/50GE
their first call resolution (FCR) rate, a metric that Pacific, Latin America, and the Middle East, where
10Gbps Mobile
represents the percentage of customer inquiries more than 20 carriers are launching or have plans
100GE 400/800GE 400/800GE
resolved in a single interaction. With the support to launch their own public clouds. The time is ripe
HTTP Internet
of AI copilots, field maintenance engineers can for carriers to rethink their public cloud strategy. 50G PON
10Gbps Home 400/800G/λ 400/800G/λ
100G/λ
also vastly cut the time it takes to perform fault
recovery. Defining ICT architecture for 5.5G
and evolving ICT infrastructure 10Gbps Campus Metro Backbone
AI agent technology is valuable for tasks like
Co
optical path maintenance: identifying faults within To seize the new opportunities brought about
one minute, locating them within three, and by these megatrends, we need to address a
recovering within five. Deploying AI agents in high- number of shared challenges in the industry,
Figure 1: Proposed ICT architecture for 5.5G
value scenarios also helps decrease the number of including defining our overall ICT architecture,
serious customer complaints. its key features, and implementation paths in
the 5.5G era. For mobile users, extremely large antenna Gpbs capacity, the industry's only dual-band 64T
With foundation models built specifically for the array (ELAA) technology and the availability MetaAAU that supports a 5 Gbps experience, FDD
telecom industry, carriers can streamline O&M in Our proposed ICT architecture for 5.5G has six of more frequency bands can help deliver a tri-band Massive MIMO sites, RRUs, mmWave
entirely new ways. defining features: ubiquitous 10-Gbps access, first-class experience with 10-Gbps downlink high-frequency AAUs, and LampSite X.
400G/800G transport networks, core networks and 1-Gbps uplink.
Trend #6: Carrier-operated public clouds are on with new user planes and intelligence planes, L4 At the same time, Wi-Fi 7, FTTR C-WAN
the rise autonomous networks, carrier-operated public Energy efficiency will also be a key metric of architecture, and 50G PON can be used to bring a
cloud services, and AI-oriented unified data future wireless network performance. To ensure 10 Gbps experience into people's homes. To help
When it comes to cloud services, over the years storage. (See Figure 1.) "0 Bit 0 Watt", where the lowest possible amount accelerate the adoption of 10G PON, Huawei has
a vast majority of carriers have gone the cloud of energy is consumed when no bits of data are launched a next-generation FTTR product, the
reseller route, while only a few have built and Feature #1: Ubiquitous 10 Gbps access transmitted, we need to constantly push the iFTTR F50 series; our brand-new MA5800T, the
operated their own public cloud platforms. But limits of our kit. Huawei has launched a series only OLT platform in the industry that supports
the tides are turning. In Africa, for example, Ethio "Ubiquitous 10-Gbps access" includes mobile, of products to help carriers build leading 5.5G GPON, 10G PON, and 50G PON on one port; and
Telecom released its own cloud computing service home, and campus scenarios. wireless networks for all scenarios, including the our scenario-specific AirPON solutions.
platform in the third quarter of 2022. Currently, industry's only TDD 128T MetaAAU with a 12

120 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 121

Innovation for the Future Innovation for the Future

For campus users and enterprises, Wi-Fi 7 E2E OXC in both metro and backbone networks, Feature #4: L4 autonomous networks Carriers have a number of advantages in local
and upgraded switches will help deliver a 10 will deliver ultra-low latency, ultra-large capacity, markets, including their infrastructure, their brand,
Gbps experience across the board. To support and optimal total cost of ownership. In this As networks grow in complexity, intelligent as well as business development and delivery
this evolution, Huawei has launched its next- regard, Huawei has launched a series of routers O&M is a must, so key technologies like telecom capabilities, so they are well-poised to find a
generation series of AirEngine Wi-Fi 7 products and optical transmission products that promise foundation models and digital twins will be needed development model that helps maximize their
and CloudEngine campus switches. an industry-leading performance and advanced to support evolution to L4 autonomous networks. opportunities in public cloud.
functionality. Building on its Autonomous Driving Network
Feature #2: 400G/800G transport networks (ADN) solution, Huawei has released a series of Feature #6: AI-oriented unified data storage
Feature # 3: Core networks with new user role-specific AI copilots and scenario-based agents,
As mobile, home, and campus networks move planes and intelligence planes including copilots for field maintenance engineers, As AI gains traction around the world, data
towards 10-Gbps connectivity, metro and backbone call centers, home broadband installation will become an even more important driver of
transport networks will need to support end-to- Core networks will need new user planes and and maintenance engineers, and marketing productivity. This will raise new requirements for
end 400G and smooth evolution to 800G. At the intelligence planes to enable 10-Gbps access, managers, as well as AI agents for optical path data storage. The size of AI training datasets has
IP layer, capabilities such as E2E SRv6 and network deliver a more intelligent and personalized maintenance, high-value scenario assurance, grown from billions of parameters to hundreds
slicing will be necessary to support multiple experience, and enable more innovative services. network risk management, and user impact event of billions of parameters, and databases need
services at the same time. These can help carriers Huawei can help with its Intelligent UDG, a management. to perform millions of reads per second, rather
satisfy transport requirements for mobile and new user plane product that supports a 10- than just tens of thousands per second. The
home scenarios, while seizing opportunities in the Gbps experience. To support new intelligence Feature #5: Carrier-operated public cloud services bar is rising for bandwidth as well. To support
enterprise market, such as multi-cloud migration. planes, Huawei has also introduced its Multi- unified data storage in the age of AI, Huawei
At the optical transmission layer, applying pooling modal Communication (MMC) and Intelligent Finding the right partner to help seize new has launched its OceanStor series of products to
architecture to metro networks, and leveraging Personalized Experience (IPE) solutions. opportunities in the public cloud market should greatly improve AI training efficiency and GPU
be a key consideration as carriers build out and utilization. OceanStor can also be used in key
operate their own cloud service platforms. Huawei application scenarios, such as billing and CRM, to
supports several different partnership models support unified data storage.
to help carriers thrive in this market, such as
deploying Huawei Cloud Stack on-premises, or A journey of a thousand miles begins with a single
running Cloud On Cloud using Huawei Cloud's step. Huawei looks forward to working with all
The six defining features of our ICT
local nodes. Building on years of experience in industry players to identify key trends and changes
architecture for 5.5G will drive the public cloud services, Huawei also provides joint in the industry, jointly define ICT architecture for
evolution of ICT infrastructure. marketing support, as well as operation and 5.5G, and drive the evolution of ICT infrastructure.
ecosystem-centric enablement to help carriers Together, we can achieve shared success and raise
make the most of their local cloud markets. the telecom industry to new heights.

122 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 123

Innovation for the Future Innovation for the Future

AI: The Bridge to 6G Tong Wen

Huawei Fellow, CTO
Zhu Peiying
Huawei Fellow, Senior
of Huawei Wireless Vice President of Wireless
Research, Huawei

In the age of AI, the core carrier of 6G services and Ma Jianglei Chen Yan
applications will shift from the mobile Internet and
Technical Vice Senior Expert of Wireless
smartphone apps to AI agents across various sectors.
President of Wireless Research, Huawei
This means AI will serve as a bridge to 6G.
Research, Huawei

exceed human limitations and create digital for 6G deployment. Technological innovation

Progress in foundation models and AI is promoting digital transformation twins in the virtual world. The 6G vision is an in 4G LTE brought us into the mobile Internet

across industries, while laying a foundation for the future of communications embodiment of global consensus and a key age, in which smartphone apps were the main

technology and playing a vital role in the shift towards 6G. milestone on the path towards a globally- carriers of applications and services. In the
unified 6G standard. age of AI, the core carrier of applications and

In June 2023, the International Telecommunication Union (ITU) completed services will shift from mobile apps to AI agents.

the recommended framework for the 6G vision, which answers the question Within the three scenarios beyond communi-
[1]
"What is 6G?" from two aspects . First, 6G will continue to evolve mobile cations, integrating AI with communications AI agents can sense and proactively take action.

communications and expand in three usage scenarios, i.e., enhanced Mobile mainly focuses on how 6G can be designed to Capable of sensing, learning, and acquiring

Broadband (eMBB), Ultra-Reliable and Low-Latency Communication (URLLC), natively support massive AI services and ap- knowledge, they can set action objectives based

and massive Machine Type Communication (mMTC), to provide immersive plications in the future. Over the next five to on the environment and constantly improve their

and deterministic communication experiences and support massive ten years, 99% of all development, design, and capabilities. The recent success of foundation

connections. Second, 6G will go beyond the scope of mobile communications administrative tasks is expected to be done by models has taken AI agent capabilities to a

to achieve integrated sensing and communication, the integration of AI with AI. In the near future, foundation models will new level, going beyond just generative AI,

communications, and ubiquitous space-air-ground integrated connectivity. even replace manual architecture design chip. to creating interactive AI capable of complex

These advances will allow us to "observe" the physical world in ways that This future trend will overlap with the window dialogues and decision-making. Therefore, in the

124 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 125

Innovation for the Future Innovation for the Future

6G era, networks will power not only AI agents, AGI for 6G To effectively support an AI-agent-powered to-human communications, for example, are
but artificial general intelligence (AGI). Huawei's 6G communication framework, 3GPP standard grounded in two core concepts:
vision of Connected Intelligence (Figure 1), In the 6G era, the basic model of communica- design must consider uplink channels that
proposed in 2019, assumed support for native- tion for AGI will be based on effectiveness com- support both sensing and learning, as well • First, everyone can use a foundation model,
[3]
AI capabilities and involves two aspects: AGI for munication , as proposed by Warren Weaver, or as downlink channels that support inference, which is a generative pre-trained transformer
[2]
6G and 6G for AGI . intent-based communication. Such a framework low latency, and metaverse applications. The (GPT), as an agent.
would go beyond Claude Shannon's model of remainder of this section will focus on the first • Second, each person's foundation model can
This article covers both of these aspects, with communication which involves the transmission two types of 6G communication. The other two use a GPT and spatial computing to
a particular focus on 6G for AGI. As shown in of only bits. Bits do not represent understanding types are similar to the first two in terms of generate a multimodal virtual real-time
Figure 2, 6G for AGI looks to explore areas like and are not intelligent, which essentially outlines how they use the AI-agent-based framework, response, representing a proxy response of
how to design communications capabilities how AGI-enabled 6G communication differs but differ in terms of specific communication the person's deterministic behavior. In the
like eMBB+, URLLC+, and mMTC+, and how from traditional communication. requirements, depending on scenarios. digital world, a communication system for
to use networks' sensing capabilities to better such GPT-based agents is known as system
support AI and make 6G into the neural center AI agent-powered 6G communications can be AI-agent-assisted human-to-human 1. It is worth noting that foundation
that connects future AI agents and a key part broken down into four types: communication models cannot accurately learn and model
of AI learning, training, and inference. If 6G non- deterministic behaviors such as
is to succeed in these areas, the 6G system 1. Human-to-human system-1 and In the future, foundation-model-based emotions. A communication system that
needs to be designed with an architecture that system-2 communication communication will organically integrate transmits such information in the physical
expands beyond connectivity and must integrate 2. Machine-to-machine intent-driven communications in both the physical and digital world in real time is known as system 2,
the four basic functions of AI agents: sensing, communication worlds. This will give rise to a post-Shannon- which can also be designed based on GPT
cognition, decision-making, and action. This 3. Human-to-machine ultra-reliable model AGI communication architecture. Human- foundation models.
architecture should use efficient, intent-based low-latency communication
communications to closely integrate the physical 4. Machine-to-human spatial-computing-
and digital worlds and thus influence how the based metaverse communication
physical world operates. CO
NN
EC
TI
narios of I MT
Y NG
s ce
L IT
BI - 20
TH
Sensing
N
A
a ge 30
E

Us

U
AI

N
ST

CO
SU
AGI eMBB

NN
Immersive

EC
Communication devices mMTC

T
ED
Integrated
Sensing Integrated Al
and
AGI
and
eMBB Communication
Communication
IMT-2020
Action Understanding

AI Connected Connected mMTC URLLC

LLM Agent AGI AGI Intelligence Massive Hyper Reliable
Communication and Low-Latency
Communication URLLC A-RAN
Ubiquitous

CE
Connectivity
AGI for 6G

EN
UB
Decision-

IL
IQ
IT

ES
/R

U
O
US

6G for AGI
IN
TE IV
AC
Y
making
LLI
GE / PR
Y
NCE RIT
SECU

Figure 1: Connected Intelligence = AGI for 6G + 6G for AGI

Figure 2: 6G's AI-native capabilities and the 6G for AGI framework

126 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 127

Innovation for the Future Innovation for the Future

loop communication between the AI agents of communication mechanism allows system-2

Each person can release their own foundation to enhance and optimize the communication
both ends without occupying communication communication to reduce data traffic by 100-
model, allowing people to access each other's process. Within this framework, everyone will
resources (shown in Figure 4). System 2 is fold or even 1,000-fold compared with direct
models before communicating. This facilitates have two major GPT foundation models: system
slow communication and can use wireless video communication. Furthermore, the AI agent
system-1 communication between foundation 1 is used for local intelligent processing, and
channels (shown in Figure 5). It should be noted of both system 1 and system 2 can constantly
models, which is essentially local communication system 2 for physical communications.
that the use of an AGI-based intent-driven update the general-purpose foundation model.
that does not use wireless communication
resources. Wireless channels will be used for First, everyone will need to train the GPT
communication in cases where foundation foundation model used by their system-1 AI
Fast and slow agent communication model
models fail to generate what's necessary for agent, as well as the GPT foundation model
Agents based on GPT foundation models Spatial Computing
system-2 communication. used for system-2 physical communication. Such
training will primarily involve supervised offline Fast solver Fast solver Fast solver Slow solver
The Fast Path 1

6G communications between AGIs (shown learning. Training based on a general-purpose Local, Low latency, high throughput, ensured 1
coverage
i n F i g u re 3 ) i n c l u d e i n t e r n a l - c h a n n e l foundation model in a broader sense can enable
The Slow Path 2 System-1 System-1

communications based on the Shannon model continuous updates and ensure humans are User-1 GPT User-2 GPT Emergent System-2
Very low speed wireless communications,
reliable achieved with low rate coding Agent
and external-channel communications between kept in the loop, making the resulting GPT
Slow solver
neural networks, between foundation models, model more accurate and powerful. Second, an
and between agents. It should be noted that emergence detector will be required to detect More advanced GPT model for System-1
• Reduce Communications Bandwidth
radio air interfaces are increasingly being whether system 1 is working properly. Parts • Increase the computing need for GPT model
• Increase the Emergent Effect to increase the “Existential Risk”
powered by GPUs, at the cost of higher power that system 1 cannot learn or model will be World Model GPT (LLM)

consumption without higher performance. distributed to system 2 for learning.

Figure 4: System 1 based on a GPT foundation model – Fast local communication

Human-to-human communications assisted These will create an architecture that combines

by AI agents are an advanced method of both fast and slow communications. System 1 is System-2 with wireless-channel transmission
interaction that uses powerful AI capabilities fast communication that facilitates local closed- Agents based on GPT foundation models

The Fast Path 1

System-2 Local, Low latency, high throughput, ensured coverage

Communications
The Slow Path 2

Channel Channel Very low speed wireless communications, reliable achieved with low rate coding
Encode Decode

Physical Noise
Fast solver Fast solver Fast solver Slow solver
System-1 System-1 System-1 System-1
world Supervised Learning AIGC Inferencing AIGC Inferencing Supervised Learning 2
1/100 ~ 1/1,000
Bandwidth
Compression
Digital System-1 System-1

User-1 GPT User-2 GPT Emergent

world Agent
System-2

Slow solver

Trained System-1 Models Trained System-1 Models

AGI AGI

World Model GPT

Figure 3: Post-Shannon-model AGI communication architecture based on GPT models Figure 5: System 2 based on a GPT foundation model – Slow physical communication

128 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 129

Innovation for the Future Innovation for the Future

Uploading
raw video AI

video
eMBB
5G
URLLC

• 3 GHz band with large uplink

6G is essentially about integrating
communication, AI, and sensing to create a GPT
Cloud-network-based
neural center for numerous AI agents. self-driving vehicle 1,000X ↓
6G
URLLC +

• Sub-3 GHz activated

Figure 6: 6G-powered, intent-driven machine-to-machine communication

AI-agent-assisted machine-to-machine • Step 1: Use an AI agent on devices to 6G for AGI can also help significantly reduce transmission
communication perform primary preprocessing and analysis power consumption while enhancing wireless
of the scenario, which is also known as goal- 6G sensing provides big data sources for AI transmission performance.
In terms of machine-to-machine communi- oriented filtering, in order to clean sensing learning
cation, uploading visual-sensing results (e.g., data in real time. Data extracted from 6G sensing can be used
complete videos and point clouds) to support • Step 2: Perform embedding in the transformer 6G's integrated sensing and communication for modeling the physical world in areas the
foundation-model computing on the edge or foundation model on the extracted objects (ISAC) is another unique advantage of applying network can reach, as well as providing a source
cloud will result in a huge amount of uplink to obtain the simplified mathematical AI agent services on 6G networks. ISAC brings of big data for AI learning (shown in Figure 7).
traffic, limiting the number of machines that descriptions (tokens) of intents. new opportunities to wireless communication People, machines, vehicles, buildings, materials,
can be supported. To combat this, a primary AI • Step 3: Transfer data back to the edge or systems—providing wireless sensing services and even weather can be objects of 6G sensing.
agent can be used on devices for the purpose cloud through wireless channels to align while supporting communications. Native Wireless sensing can provide big data on the
of real-time token alignment with foundation intents (represented by tokens) on both convergence of sensing and communication environment through parameter estimation,
models on the cloud through wireless channels ends in real time, thereby realizing efficient enables mobile base stations and devices imaging, and even mass spectrometry, all
to facilitate AI agent collaboration across machine-to-machine communication with to obtain a larger sensing scope and higher of which are transmitted over radio waves.
devices, pipes, and cloud, thus realizing massive device-pipe-cloud synergy. sensing accuracy through collaborative Attention and study are both essential for
machine-to-machine communications (shown sensing, without additional spectrum or sensing across the entire communications
in Figure 6). Compared with direct video transmission, this increased equipment costs. With shorter radio spectrum, including centimeter wave, millimeter
transmission mechanism can reduce data traffic wavelengths, broader spectrum resources, and wave, and sub-THz bands. THz technology has
Specifically, an AI-agent-based post-Shan- by 100-fold or even 1,000-fold, increasing the larger antenna apertures, 6G can support the the potential to see wide adoption in high-
non-model communication framework uses number of communicating users the system can highly-accurate, real-time reproduction of the precision sensing.
intent-driven communication according to the support by an order of magnitude. physical environment as a service. This capability
following steps:

130 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 131

Innovation for the Future Innovation for the Future

6G-based intelligent and inclusive A-RAN Shannon-model communication architecture behavior and performance, in what we run a small-scale AI model. These devices
and A-Core that supports AGI (i.e., the C+A+S radio access call digital twins. This section covers two can sense the physical world through wireless
network) is called A-RAN (shown in Figure 8). technologies related to digital twins. The signals. By processing the sensing signals, the
Foundation models for natural language first is building digital twins that can mirror devices can generate a point cloud that both
processing, represented by ChatGPT, will 6G networks can also be built with AI agents, the physical world in real time through depicts and describes the physical world.
match and exceed human capabilities in with each serving as a logical net work intent-driven communication. The second is
the near future. However, such human- element (NE). Huawei's 2012 Laboratories performing accurate spatiotemporal inference Then, with a given intent , the physical
like intelligence will be possible only when proposed the concept of application-driven about the physical world based on the real- world is selectively reproduced through AGI
supported by the computing power provided networks in 2015. The core idea behind time twin world. Both technologies have communications between an AI foundation
by supercomputing clusters. For example, 500 this is automatically generating customized the potential for wide application across model in the network and the small AI model
billion neural network parameters require net works for customers' application numerous scenarios like self-driving vehicles, running on the 6G devices. The twin world
over 10 million watts of power supply to run. requirements, and cancelling such networks robotics, smart industrial production, and in question does not have to completely
In the next 10 years, it is unlikely that such when the applications are no longer telemedicine. describe the physical world, but is capable
foundation models will be able to run on needed. The infrastructure is like a unified of identifying things relevant to the intent.
mobile devices. computing platform that stretches across the 6G devices can be placed in a small physical This reduces communication overhead, while
entire network. It is a task-based network space, such as a small room, to support protecting personal privacy and keeping
6G networks must be able to deliver architecture and the prototype of 6G A-Core integrated sensing and communication information about the location confidential.
inclusive intelligent services for all people (as shown in Figure 9). provide the AI computing power required to
and all things, anytime and anywhere.
T h i s re q u i re s 6 G n e t w o r k s t o a d o p t a Creating a real-time twin world with 6G AGI
C+A+S (communication + AI + sensing) Agent B

architecture powered by foundation models. In the digital world, virtual copies of physical
AGI
The convergence of communication, AI, and entities are built based on application intent
System-1 Model
sensing is a key feature of 6G. The post- in order to simulate and analyze real-world for B

LLM
AI and ML Integrated sensing,
communications, and AI

6G Point Cloud Radio Rendering Ground Truth

Processing

Agent A

AGI
6G-sensed point cloud AI reproduction algorithm Physical world System-1 Model
for A

Feeding zettascale-volume of data for exascale AI computing·

Neuron Center (1T) Neuron Edge (100B) Neuron Agent (10B)

Figure 7: 6G sensing is a major data source for future AI foundation models Figure 8: A-RAN architecture: Communication-AI-sensing converged access network

132 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 133

Innovation for the Future Innovation for the Future

The sensing system senses point-cloud in- ChatGPT, such a system can sense the physical will be essential in the future foundation
formation, which is lattice representations of world in real time, perform semantic, temporal, model era:
References :
objects in the three-dimensional space. To and spatial inference, and then present the • Connected Intelligence = AGI for 6G +
reduce the amount of data to be transmitted results in the form of natural language. For 6G for AGI
[1] Recommendation ITU-R M.2160-0, Framework
and protect user privacy, we upload semantic example, the positions of an object and • AGI for 6G: Effectiveness communication and overall objectives of the future develop-
ment of IMT for 2030 and beyond.
spatiotemporal information to the cloud person can be displayed in the digital twin in powered by the post-Shannon-model
for fusion only when the semantic target real time in the form of a rendered animated communication architecture [2] Wen Tong and Peiying Zhu, "6G: The Next
Horizon – From connected people and things to
matches the original sensing information. object and figure. A 6G AGI system can use • 6G for AGI: An inclusive intelligent neural
connected intelligence," Cambridge University
That is why we have developed a unique the sensed point cloud information to identify center that integrates AI learning, Press, May 2021.

semantic spatiotemporal fusion and prediction actions relevant to preset intents (e.g., reading training, and inference
[3] C. E. Shannon and W. Weaver, The Mathemati-
algorithm that can fuse semantic information a book), and locate the spatial positions of • 6G A-RAN which integrates communication, cal Theory of Communication. The University of
Illinois Press, 1949.
to form a digital twin. The algorithm enables such occurrences in real time. Continuous sensing, and AI
a digital twin to mirror the information of improvements to the sensing system will • Task-based 6G A-Core that is built
concern in almost real time, while reducing further increase inference accuracy. with agent NEs
the required uplink transmission bandwidth by
orders of magnitude. In conclusion, we believe that 6G is essentially
about integrating communication, AI, and
Object positioning and human-pose tracking sensing to create a neural center for numerous 6G networks must be able to deliver
can already be driven by natural language AI agents. We also believe that the following inclusive intelligent services for all people
through a real-time twin world. Unlike elements of 6G native-AGI communication and all things, anytime and anywhere.

APP Slice-Agent
CN-Agent
External Agents AutoGPT/Generative Networks Load-Agent

In-network Agents Scheduling OAM-Agent

NRT adaption (mobility…) Execution LLM native

Infrastructure Provisioning Intelligence Provisioning

Connectivity Compute Data Model Parameter

Networks O&M

Figure 9: A-Core: Task-based network architecture

134 HuaweiTech 2024 Issue1 HuaweiTech 2024 Issue1 135