《GTI&中移智库：2025 5G-A×AI新技术、新服务、新模式白皮书（英文版）（65页）.pdf》由会员分享，可在线阅读，更多相关《GTI&中移智库：2025 5G-A×AI新技术、新服务、新模式白皮书（英文版）（65页）.pdf（65页珍藏版）》请在三个皮匠报告上搜索。

1、 5G-AxAI New Technology,New Case,New Model White Paper 1 5G-AxAI New Technology,New Case,New Model White Paper https:/gtigroup.org/5G-AxAI New Technology,New Case,New Model White Paper 2 5G-AxAI New Technology,New Case,New Model White Paper Version:V1.0 Deliverable Type Procedural Document Working D

2、ocument Confidential Level Open to GTI Operator Members Open to GTI Partners Open to Public Program 5G-AxAI Working Group N/A Project Project 1:Network Intelligence Project 2:Digital Twin Network Intelligence Project 3:Application Intelligence Project 4:Sustainability Intelligence Task N/A Source me

3、mbers China Mobile,DOCOMO Beijing Labs,Huawei,Ericsson,ZTE,Nokia,Intel,Qualcomm,MTK,Unisoc,Xiaomi,OPPO,vivo,HONOR Support members Editor Last Edit Date Approval Date http:/gtighttp:/gtig 5G-AxAI New Technology,New Case,New Model White Paper 3 Confidentiality:This document may contain information tha

4、t is confidential and access to this document is restricted to the persons listed in the Confidential Level.This document may not be used,disclosed or reproduced,in whole or in part,without the prior written authorization of GTI,and those so authorized may only use this document for the purpose cons

5、istent with the authorization.GTI disclaims any liability for the accuracy or completeness or timeliness of the information contained in this document.The information contained in this document may be subject to change without prior notice.Document History Date Meeting#Version#Revision Contents 5G-A

6、xAI New Technology,New Case,New Model White Paper 4 Table of Table of ContentsContents GTI 5G-AxAI New Technology,New Case,New Model White Paper.2 Document History.3 Table of Contents.4 1 Executive Summary.5 2 5G-AxAI:New Capabilities meet New Need,unleash New Value.5 2.1 5G-A Commercial progress.5

7、2.2 New AI Capabilities.6 2.3 Opportunities for the Cross-domain Convergence.7 2.4 Multiplier Effect of 5G-A and AI.11 3 5G-AxAI Breeds New Technologies.13 3.1 Network intelligence.14 3.2 Digtal Twin Network Intelligence.20 3.3 Application Intelligence.23 3.4 Sustainability Intelligence.29 4 5G-AxAI

8、 Enables New Cases.31 4.1 Differentiated Experience Assurance.31 4.2 New Calling Service.33 4.3 Industrial Deterministic Service.37 4.4 Green Energy Saving.39 4.5 High-Reliability Network.41 4.6 Multi-Modal Personal Assistant.45 4.7 Embodied Artificial Intelligence.47 4.8 Immersive Experience.48 4.9

9、 Networked Smart Driving.53 4.10 Low-Altitude Intelligent Connectivity.55 5 Industry Revolution of New Models.59 5.1 Three Innovative Business Models for ToC/BtoC.59 5.2 Three Innovative Business Models for ToB.60 6 Global Industry Collaboration Proposal.62 7 Glossary.62 8 References.64 5G-AxAI New

10、Technology,New Case,New Model White Paper 5 1 Executive Summary In the current era of technology innovation,the Artificial Intelligence(AI)technology has been advancing rapidly,bringing new opportunities for the network.The convergence of 5G-Advanced(5G-A)and AI would be an inevitable industrial tre

11、nd,unleashing the multiplier effect in telecommunications and other industries.On the one hand,5G-AxAI would meet network demand,improving network performance and efficiency.On the other hand,5G-AxAI would also provide new services for the industries,accelerating the industry intelligent revolution.

12、Innovative technology of 5G-AxAI has emerged and made effect in 4 major areas:in network intelligence field,it makes the network display exceptionally high quality;in digital twin network intelligence field,it enables the low-cost trial and high-efficiency innovation;in application intelligence file

13、d,it expands the network service scope;in sustainability intelligence,it achieves the goal of low-carbon.All of these technology provides an innovation engine for the new applications,which involve service guarantee,personal AI agent,embodied AI,autonomous driving,etc.And it also lead to new busines

14、s model,indicating the network service transformation from the connectivity to the connectivity,computing and intelligence fusion.2 5G-AxAI:New Capabilities meet New Need,unleash New Value The cutting-edge AI technology brings new capabilities for the society,enabling a novel way of human production

15、 and life1.The AI development has entered into a new phase and become light-weighted,generalized,and concrete.The optimized algorithms greatly decrease the AI cost,and reduce the barriers for the small business;the multi-modal foundation models unify the heterogeneous human data,and expand the appli

16、cation scope;the robots are integrated with understanding and thinking abilities,and make the AI objectified in real world.At the same stage,the communication network is going through the period of large-scale deployment and innovation.The number of 5G networks has reached 398 while 5G SA operators

17、reach 154 globally.And the next-phase 5G-A system begins to take shape and is implemented by more than 15 operators.A new tale happens when two advanced and fast-developing industries meet together.Along with the 5G-A network,the effect and value of AI are further enlarged.The convergence of 5G-A an

18、d AI becomes an inevitable trend and brings changes in many aspects of technology,applications,industry and so on.New need of network and industry will be satisfied based on the 5G-AxAI technology development,promoting social efficiency and creating new value.2.1 5G-A Commercial progress In the mid

19、of 2024,the 3GPP has already finished the work on Release 18,and its work on Release 19 is expected to be completed by the end of 2025.5G has officially entered the 5G-A stage,and it provides multiple values for society2-4.First,high-speed,low-latency,and almost ubiquitous connectivity capabilities.

20、Second,integration of various innovative information 5G-AxAI New Technology,New Case,New Model White Paper 6 technologies.Third,the ability base for the industry services.In a word,5G-A has achieved significant upgrades in terms of rate and intelligence capabilities.According to data,the actual 5G-A

21、 rate has jumped to 3-5Gbps and is expected to exceed 10Gbps in the future,a 10-fold increase compared with traditional 5G network.Technologies such as Extremely Large Antenna Array(ELAA)are used to improve coverage of high frequency bands,and the multiple carriers coordination technology is used to

22、 efficiently aggregate bandwidths of multiple frequency bands,thereby greatly improving spectral efficiency.On the terminal side,novel high-end chips are released in 2024 to support large bandwidth capabilities beyond 6 Component Carriers(CC).Moreover,5G-A continuous capability upgrading has also op

23、ened up a new blue ocean in the industry.For example,The 5G-A Internet of Vehicles(IoV)supports large-scale pilot projects on vehicle road clouds and accelerates the development of the intelligent connected automobile industry.From a global perspective,more than 60 operators and partners around the

24、world have announced 5G-A commercial plans.5G-A has become a hot topic in the industry.On December 10,2024,a special seminar on the theme of World First 5G-A Region Sets Sail was launched.Experts from authoritative industry organizations,regulators,leading operators and equipment vendors,such as GSM

25、A,TDRA,DU,e&,Vodafone,Ooredoo,Huawei,Ericsson,and Nokia Delegates shared the innovations of 5G-A.In China,5G-A networks have been launched in more than 300 cities,and China has set an good commercial example for the world.In other regions,lots of operators have carried out 5G-A verification,such as

26、Asia Pacific HKT,CTM,Malaysia Maxis,Europe DNA,Vodafone,Latin America VIVO,and TIM.In China,China Mobile actively promotes the implementation of 5G-A network,who takes the lead in adopting innovative technologies.For instance,it deploys 3CC aggregation networks and 5G-A intelligent control plane(NWD

27、AF),improving the transmission efficiency and providing unique assurance capabilities for key customers.China Mobile has also issued 5G-A charge plans for travel,game,and live broadcast,and are subscribed by more than 500,000 5G-A users.For example,the 5G-A business travel plan provides a high-speed

28、 experience of up to 3Gbps in the downlink and 200Mbps in the uplink.In the Middle East,operators have issued wireless broadband packages up to 300Mbps.Wireless home broadband uses the 5G-A high-speed network to provide high-quality experience similar to optical fibers,and solves the problems of dif

29、ficult fiber deployment and high construction costs.Overall,5G-A has great potential for development and monetization.It is in the stage of continuous developing and large-scale implementation,awaiting for the new driving forces to achieve further breakthroughs.2.2 New AI Capabilities AI is the tech

30、nology that enables machines to mimic human intelligence,allow them to tackle tasks like learning,comprehension,problem solving,perception,decision making,and autonomy.The latest big foundation models(e.g.,GPT-4o,Deekseek)have demonstrated their remarkable ability in understanding natural language a

31、nd thinking about the problem.AI Algorithms include supervised learning,unsupervised learning and reinforcement learning.And among them,deep learning is a kind of effective and powerful method,which has been widely used in various fields.Neural networks with multiple different kinds of layers(e.g.,M

32、LP,CNN,RNN,Transformers)have 5G-AxAI New Technology,New Case,New Model White Paper 7 demonstrated their great capabilities in perception,prediction,decision-making and optimization.These capabilities have already shown its value in healthcare,finance,transportation,governance and many other industri

33、es.1)AI Perception AI is able to perceive and understand the surrounding environment through processing the data collected from various sensors.It enables machines to simulate the way humans perceive the world,including vision,hearing,touch,and other senses.It even goes beyond human because it could

34、 also directly obtain information from the cyber-space.For example,in hospitals,AI could monitor patients vital signs and detect early signs of deterioration,in this way the medical staff would be alerted in time to provide emergency help.2)AI Prediction AI is also able to forecast future events,tre

35、nds,or outcomes based on historical data and current information.It has the ability to handle complex data and identify patterns that may not be apparent to humans.In weather forecasting,AI models can predict future weather conditions accurately by analyzing a large amount of meteorological data,inc

36、luding temperature,humidity,wind speed,and pressure.In financial field,AI could predict stock prices and market trends based on historical stock data,economic indicators,and news sentiment.3)AI Decision-making AI could analyze data,evaluate various options,and make recommendations or decisions to ac

37、hieve specific goals.It has the ability to normalize multi-modal input conditions to the unified vector space,extract the common features,and weight them to get the final decision.In autonomous driving,AI system could help analyzes data from different kinds of sensors such as camera,lidar,and radar

38、to detect obstacles,other vehicles,and pedestrians.It then makes decisions about when to accelerate,brake,or turn to avoid collisions and follow traffic rules.4)AI Generation AI is capable of creating new content by learning and analyzing vast amounts of related data.This ability could be adapted an

39、d expanded to different areas(text,audio,image,video).In creative fields,it provides creative inspiration for artists,designers,writers,musicians,helping them quickly generate creative content.In education field,it generates teaching resources such as teaching animations and personalized learning ma

40、terials,helping improve the teaching effect.All of these AI capabilities could be utilized and improved in the network field.Combined with the network capability of low latency,high-reliability,ubiquitous connectivity,etc.,communication-intelligence fusion services would emerge and drive the industr

41、y to the next phase.2.3 Opportunities for the Cross-domain Convergence 2.3.1 Meeting Evolved Network Demands The convergence of 5G-A and AI offers transformative potential in meeting the network evolution demand and enabling superior network performance.AI technique could be used to optimize multipl

42、e functional modules across an entire link or system in a data-driven way,thereby improving the entire end-to-end communication system.By combining the advanced capabilities of 5G-A and AI,networks can achieve unprecedented efficiency,adaptability,and reliability,meeting the ever-growing demands of

43、modern connectivity.Below is an analysis of how 5G-AxAI New Technology,New Case,New Model White Paper 8 AI technologies address critical challenges in network fields.1)Efficient Resource Utilization Efficient resource utilization is essential in 5G-A to meet the increasing demand for connectivity,pa

44、rticularly in dense urban areas,industrial Internet of Things(IoT)applications,and high-traffic environments.AI plays a pivotal role in optimizing key network resources such as spectrum,bandwidth,and infrastructure.Dynamic Resource Management:AI algorithms analyze real-time traffic conditions and pr

45、edict demand patterns,enabling 5G-A networks to dynamically allocate resources such as bandwidth,power,and spectrum to areas of highest need.This ensures efficient utilization of network assets and seamless adaptation to varying traffic loads.Network Slicing Optimization:Through 5G-As network slicin

46、g capabilities,virtual networks can be tailored for specific applications(e.g.,IoT,enhanced mobile broadband,and low-latency services).AI enhances this by optimizing resource allocation for each slice in real time,adapting to network performance metrics and user requirements.Predictive Traffic Analy

47、tics:AI-powered predictive models leverage historical data and usage trends to forecast network congestion or bottlenecks.This proactive approach allows networks to redistribute resources in advance,ensuring uninterrupted operation even during peak usage periods.AI-Enhanced Spectrum Efficiency:AI al

48、gorithms dynamically assign available frequencies based on real-time demand,minimizing interference and maximizing spectral efficiency.AI can also enable effective spectrum sharing between operators and technologies,boosting network capacity.Optimizing MIMO Technology:5G-As Multiple Input Multiple O

49、utput(MIMO)technology significantly enhances data throughput.AI fine-tunes MIMO parameters dynamically,optimizing system performance with minimal power consumption while maintaining high data rates.2)User-centric Network Operation User-centric operation is critical for ensuring users receive a consi

50、stent and reliable level of service,including key metrics such as bandwidth,latency,jitter,and packet loss.For 5G-A,Quality of Service(QoS)optimization is particularly important due to the wide range of services the network supports,which guarantees the user experience and balances the network load.

51、Real-Time Monitoring and Adjustment:AI enables continuous monitoring of network performance and user experience.By analyzing metrics like latency,throughput,and packet loss in real time,AI can dynamically adjust network parameters to ensure optimal QoS.For example,it can prioritize time-sensitive tr

52、affic,such as autonomous vehicles or remote surgeries,to minimize delays and packet loss while allocating sufficient resources for other services like video streaming.Intelligent Traffic Shaping:AI algorithms can classify traffic based on the type of service,such as video,voice,or IoT.By understandi

53、ng the specific requirements of each service,AI can prioritize high-priority traffic and route low-priority traffic efficiently,ensuring seamless performance and avoiding congestion.Context-Aware Traffic Prioritization:AI leverages contextual data,such as time of day,user location,and network condit

54、ions,to dynamically adjust QoS settings.For instance,during peak hours,AI can prioritize applications related to emergency services or healthcare over general browsing or streaming,ensuring critical services remain unaffected.End-to-End QoS Assurance:AI can enable end-to-end QoS control,ensuring tha

55、t the QoS standards are met not only at the network level but also at the application level.AI can analyze the 5G-AxAI New Technology,New Case,New Model White Paper 9 entire network pathfrom the edge to the coreto detect any potential issues that could affect service delivery.This holistic approach

56、ensures a better user experience,especially in critical-use cases like smart cities,autonomous driving,and industrial automation.Intelligent Traffic Offloading:AI predicts traffic patterns and proactively offloads traffic to alternative,efficient paths.This includes rerouting data to edge computing

57、nodes,neighboring cells,or other optimized network paths.Such proactive measures ensure the core network remains uncongested,improving overall network stability and performance.3)Network Automation Intelligent algorithms can significantly enhance the performance of network pipes,reduce Operations&Ma

58、intenance(O&M)costs,and increase the efficiency of new service rollouts.Furthermore,AI will drive O&M practices towards levels L4/L5,assisting telecom operators in improving service quality and increasing revenue.Agile Network:In the current O&M system,functions operate independently of each other.T

59、o achieve task goals,O&M personnel must manually orchestrate and integrate these functions,which requires frequent interactions with the system and leads to reduced O&M efficiency.AI technologies are transforming the O&M mode from being function-centric to task-centric,allowing the system to better

60、cater to human needs.By incorporating a unified knowledge model into the O&M system,O&M personnel can communicate with it using natural language.This enables the system to understand user intent and autonomously orchestrate and integrate functions to fulfill task goals seamlessly.Greener Network:The

61、re is often a contradiction between network energy saving and experience assurance.The biggest contradiction in energy saving is that we do not know when to shut down and to what extent we shut down.With AI algorithms,network could dynamically balance network resource allocation and energy saving po

62、licies to achieve smooth transition from off-peak to peak hours.Through the implementation of the foregoing intelligent strategies,efficient collaboration between experience assurance and energy saving could be achieved,significantly reducing network energy consumption while assuring the user experi

63、ence.Service Innovation Empowerment:Telecom industries are actively exploring new network capabilities and services.To meet various requirements,it is necessary to break away from the traditional working mode of service provisioning based on manual experience and operation.In the new service provisi

64、oning phase,operators could directly deliver service requirements in the form of natural language.With the help of AI methods,the network is able to understand the intent by translating the intent to the machine code,perform online simulation in the network twin environment,and automatically generat

65、e network configurations to meet new requirements.2.3.2 Meeting Diverse Industry Demands The industry demand for 5G-AxAI is growing rapidly due to the convergence of two transformative technologies:5G-A and AI.This combination is expected to enable a wide range of new applications and services acros

66、s multiple industries.1)Telemedicine and Remote Healthcare 5G-A offers low latency and high-speed data transfer capabilities,which are essential for enabling real-time video consultations and even remote surgical procedures.AI enhances these 5G-AxAI New Technology,New Case,New Model White Paper 10 c

67、apabilities by supporting diagnostic tools that analyze medical images,sensor data,and patient histories,aiding healthcare professionals in making informed treatment decisions.Telemedicine and Remote Surgery:5G-A provides the high bandwidth and low latency necessary for real-time remote consultation

68、s and surgeries.AI can assist doctors in making diagnoses or performing surgery with robotic assistance,using AI-powered imaging and data analysis to improve accuracy.Health Monitoring:AI-powered devices,enhanced by 5G-A networks,enable real-time monitoring of patient health data.These systems can p

69、redict potential health issues before they become critical by analyzing data from wearable,sensors,and other medical devices.AI-Assisted Diagnostics:AI-driven models can analyze medical scans,including X-rays and Magnetic Resonance Imaging(MRI),with remarkable accuracy.When paired with 5G-As capabil

70、ity to transfer large datasets swiftly,these tools enable faster and more precise diagnoses,particularly in remote or undeserved regions,improving access to quality healthcare.2)Autonomous Vehicles and Transportation 5G-A playing a crucial role in promoting the development of autonomous vehicles,whi

71、ch provides impressive data transmission and supports massive connectivity.Along with the computing system deployed in the network,the vehicle is able to have a more comprehensive and detailed perception of the surrounding environment and make more accurate driving decisions.Autonomous Vehicles:5G-A

72、 delivers the high-speed,ultra-low-latency connectivity required for autonomous vehicles to communicate seamlessly with each other and with infrastructure,such as traffic lights and road sensors.AI processes data from multiple sensors,including LIDAR,cameras,and radar,enabling real-time decision-mak

73、ing for safe and efficient driving.Vehicle-to-Everything(V2X):5G-A supports V2X communication,facilitating the exchange of data between vehicles,infrastructure,and even pedestrians.AI leverages this data for advanced traffic management,predictive route optimization,and real-time traffic updates,enha

74、ncing road safety and travel efficiency.Fleet Management and Logistics:5G-A enables real-time monitoring and seamless communication across large vehicle fleets.AI algorithms analyze data to predict maintenance needs,optimize delivery routes,and improve fuel efficiency,ensuring cost-effective and sus

75、tainable logistics operations.3)Manufacturing and Industry 4.0 5G-A network connects a large number of devices and sensors in factories and transmits a large amount of data(e.g.,production process parameters,equipment operation status)to the central control system in network or application server fo

76、r analysis and processing,enabling real-time monitoring and optimization of the production process.Smart Manufacturing:5G-A delivers the high bandwidth required for real-time monitoring of machines,assembly lines,and robotic systems.AI analyzes sensor data to optimize production processes,identifyin

77、g inefficiencies or potential issues before they lead to breakdowns.Predictive Maintenance:AI algorithms leverage sensor data to predict equipment failures before they occur.With 5G-As continuous data transmission capabilities,systems can send real-time alerts for maintenance,reducing downtime and e

78、nhancing operational efficiency.Robotic Automation:5G-As ultra-low latency enables AI-powered robots to perform highly synchronized tasks in real-time,such as assembly and quality control,significantly improving precision and productivity in manufacturing environments.5G-AxAI New Technology,New Case

79、,New Model White Paper 11 4)Retail and E-commerce 5G-A and AI has great impact on retail and e-commerce,bringing about changes in various aspects such as shopping experience,logistics,and supply chain management.Real-time data sharing and communication make demand forecasting more accurate and custo

80、mer experience better,enhancing the overall efficiency,flexibility and attraction.Personalized Customer Experience:AI analyzes customer behavior to provide personalized recommendations and targeted promotions.5G-A supports this by enabling real-time data exchange and immersive experiences such as au

81、gmented reality(AR)in retail stores.Inventory Management and Logistics:AI can predict product demand and optimize stock levels in real-time.Combined with 5G-As ability to track shipments and warehouse operations,retailers can improve efficiency and reduce stockouts.Smart Stores:5G-A enables the inte

82、gration of IoT devices(e.g.,smart shelves,sensors)in retail environments.AI analyzes data from these devices to improve product placement,detect theft,and optimize in-store experiences.5)Smart Cities and Urban Infrastructure 5G-A and AI are the key enablers in the construction of smart cities,facili

83、tating significant advancements in smart governance,smart security,smart traffic,etc.Through the technology development,the city could monitor the real-time status,make comprehensive decisions and manage urban operations.Traffic Management:5G-A networks are ideal for smart city infrastructure,enabli

84、ng real-time data analysis from sensors across the city to optimize traffic flow,reduce accidents,and improve overall urban efficiency.AI is key in processing large-scale data to make intelligent decisions.Public Safety and Surveillance:AI-powered security systems,including facial recognition and an

85、omaly detection,benefit from the speed and connectivity of 5G-A networks.These systems can process real-time data from surveillance cameras to enhance public safety.The demand for 5G-A and AI technologies extends across a wide range of industries,each capitalizing on the combination of high-speed co

86、nnectivity,real-time data processing,and intelligent decision-making.Sectors such as healthcare,automotive,manufacturing,telecommunications,retail,and smart cities are harnessing these technologies to boost efficiency,enhance customer experiences,and unlock new applications.The synergy between 5G-A

87、and AI not only enhances current use cases but also enables the development of innovative,next-generation applications that were once unimaginable,driving digital transformation across these industries.2.4 Multiplier Effect of 5G-A and AI The convergence of 5G-A and AI creates a powerful multiplier

88、effect,amplifying the impact of each technology across multiple sectors.This convergence accelerates innovation,boosts efficiency,enhances customer experiences,and fosters the development of new business models5.5G-AxAI New Technology,New Case,New Model White Paper 12 Figure 2-1 The convergence of 5

89、G-A and AI generates tremendous value The outcome of the 5G-A and AI collision is the fusion effect rather than simple summation.Two parallel universes(5G-A domain and AI domain)collide and expand to higher dimension,and the inner particle generates interaction force with each other,releasing tremen

90、dous energy.This energy spreads to different application fields,and unfolds great value for the human society.The value is mainly reflected in three aspects:1)Boosting Network Performance and Efficiency Network Optimization:AI enhances 5G-A by facilitating autonomous network management,predictive re

91、source allocation,and fault detection.This leads to more efficient use of network resources(such as bandwidth and power),reducing operational costs for service providers while improving user experience.Networks Automation:AI enables network automation,allowing 5G-A networks to self-optimize and self

92、-heal.This reduces the need for manual interventions,speeds up network deployment,and enhances scalability to meet the increasing demands of IoT and connected devices.Low-Latency Applications:AI models help optimize traffic routing and predict network congestion,supporting ultra-low latency services

93、 such as autonomous vehicles,augmented reality,and remote surgeries.5G-As ability to provide high data rates and massive connectivity ensures smooth and uninterrupted operation of these AI-powered services.Predictive Analytics:AI-driven predictive maintenance and traffic management allow 5G-A networ

94、ks to anticipate demand shifts,detect bottlenecks,and perform proactive maintenance.This minimizes service disruptions,lowers operational costs,and enhances network reliability.2)Unlocking New Services and Applications Edge Computing and Real-Time Processing:5G-A supports edge computing,where data i

95、s processed closer to its source,such as IoT devices or sensors.This reduces latency and bandwidth usage,enabling AI applications that demand real-time processing,including autonomous vehicles,smart manufacturing,and smart cities.Massive Connectivity:5G-A can accommodate vast IoT ecosystems with mil

96、lions of devices transmitting small data packets.When combined with AIs ability to process and make decisions based on this data,industries can optimize operations,improve customer service,and develop new business models.For instance,in smart agriculture,AI can analyze sensor data to enhance irrigat

97、ion and crop management,while 5G-A ensures seamless real-time data transmission across the farm.Immersive Communication:In sectors like education,entertainment,and retail,AI-powered AR/VR applications are enhanced by 5G-A ability to deliver high-quality,low-latency immersive 5G-AxAI New Technology,N

98、ew Case,New Model White Paper 13 experiences.This opens up innovative opportunities for virtual training,remote collaboration,and virtual product try-ons.3)Accelerating the Industry Revolution Automation and Optimization:AI drives intelligent automation across industries,including manufacturing(Indu

99、stry 4.0),healthcare,transportation,and smart cities.In manufacturing,AI-powered robots and predictive maintenance systems,supported by 5G-As ultra-low latency,boost production efficiency and minimize downtime.In healthcare,AI algorithms analyze patient data in real time to create personalized treat

100、ment plans,backed by 5G-As fast data transfer.Real-Time Analytics and Personalization:With the help of 5G-As connectivity,AI enables industries to personalize services based on real-time data.For instance,in retail,AI analyzes customer behavior to offer tailored recommendations,while 5G-A ensures se

101、amless data transmission from IoT devices,sensors,and mobile apps for an enhanced shopping experience.Improved Decision-Making:AIs ability to process complex data sets provides businesses with real-time insights into market trends,customer preferences,and operational inefficiencies.This allows indus

102、tries to make quick,data-driven decisions,boosting business agility and competitiveness.Predictive and Prescriptive Analytics:In logistics,AI-powered predictive models optimize routing and inventory management,while 5G-As connectivity facilitates real-time monitoring of goods in transit.This results

103、 in cost savings,improved customer satisfaction,and greater operational efficiency.The multiplier effect of 5G-A and AI convergence is profound,creating great value of improved performance,new business opportunities,and innovation.AI drives smarter decision-making and automation,while 5G-A provides

104、the connectivity infrastructure necessary to support these advances with speed and reliability.Together,they propel the digital transformation of industries,improving operational efficiency,fostering new business models,and delivering societal benefits.The outcome is a dynamic,cross-industry ecosyst

105、em where growth and innovation are amplified,benefiting businesses,consumers,and society.3 5G-AxAI Breeds New Technologies The convergence of 5G-A and AI brings revolutionary innovations.Core innovations focus on four major areas:1)network intelligence,which aims to enhance network quality through i

106、ntelligent optimization;2)digital twin network intelligence,which aims to enable low-cost trials and high-efficiency innovation;3)application intelligence,which aims to expand network service scope through enhanced capabilities;4)sustainability intelligence,which aims to achieve important environmen

107、tal goals through intelligent energy efficiency.Together,these innovations power a new wave of applications spanning diverse areas like service assurance,personal AI agents,embodied AI,and intelligent network infrastructure lifecycle management.Through continuous advancement in these technological d

108、omains,5G-AxAI is actively reshaping network paradigms and accelerating industry transformation while driving the evolution from simple connectivity to an integrated fusion of connectivity,computing and intelligence.5G-AxAI New Technology,New Case,New Model White Paper 14 3.1 Network intelligence 3.

109、1.1 Intelligent Real-time Network Closed-loop By introducing intelligent control plane and user plane,5G-A could achieve real-time application identification,experience perception,and dynamic service control,thus supporting operators to continuously optimize network resource scheduling,and enabling

110、differentiated service assurance.1)Real-time experience perception With coordination of the Network Data Analytics Function(NWDAF)intelligent analytic capability and User Plane Function(UPF)traffic recognition capabilities,the intelligent network could provide precise and real-time user experience e

111、valuation based on massive amounts of network data.The process involves four stages:data processing,AI model structure design,model pre-training,and model fine-tuning.Figure 3-1 Framework for real-time experience perception Data processing aims to provide high-quality,readily usable corpora for mode

112、l training,laying the foundation for the accuracy and generalization capabilities of subsequent AI models.Several aspects are considered for data processing,including privacy security,feature selection,encoding,tokenization.The AI model structure design involves defining and designing basic models s

113、uitable for network operational state data compression and modeling.It designs domain models based on Transformer and Mamba structures,integrating multi-modal learning to fuse signaling,traffic,and other multi-source data.By combining knowledge distillation and model quantization techniques,complex

114、models are compressed into lightweight versions to meet the real-time data analysis and deployment requirements of UPF and NWDAF.The model pre-training is to build a basic model with general characteristics through global training on large-scale data,while optimizing the training process to improve

115、training efficiency and effectiveness.Pre-training optimizes multi-task learning performance through adaptive loss weighting algorithms,dynamically adjusting task priorities;it also enhances the efficiency of large-scale model training by combining distributed training frameworks.The model fine-tuni

116、ng provides high-precision task capabilities for models through training based on specific service scenarios.In the fine-tuning phase,transfer learning can be used to 5G-AxAI New Technology,New Case,New Model White Paper 15 quickly adapt to different scenarios,and small-sample learning can be used t

117、o improve the model performance in scenarios where data is insufficient.Based on the fine-tuning technology,application awareness and Quality of Experience(QoE)awareness tasks can share the same basic model and can be upgraded independently.2)Differentiated experience assurance Figure 3-2 Framework

118、for differentiated experience assurance Differentiated experience assurance technology facilitates operators maximizing resource utilization and enhancing profitability.Specifically,when the network runs at low-load states,the differentiated experience assurance technology could stimulate user traff

119、ic,so that users are willing to migrate high-traffic services such as live streaming to the mobile network;when the network runs at high-load states,the differentiated experience technology could provide sufficient assurance for key services of key customers,so that the network states could not degr

120、ade their experiences.In the process of service assurance,the UPF continuously analyzes user service experience and reports the analysis results to the NWDAF.The NWDAF triggers the assurance policy based on poor experience.After the Policy Control Function(PCF)delivers the assurance policy,it trigge

121、rs dedicated Logo display on the User Equipment(UE).After the assurance is complete,proactively push experience reports so that users can perceive the assurance results and streamline end-to-end experience closure.3.1.2 Network Foundation Model Communication Network is the aggregation node of all tr

122、affic and could obtain massive valuable data.With the help of Large Language Model(LLM)technology,it becomes feasible to build a foundation model incorporating network knowledge,enabling precise network operations and comprehensive user analysis.The foundation model greatly improves the generality a

123、nd robustness compared to traditional AI methods,providing stronger capabilities of perception and optimization.Moreover,it makes the task adaptation more easier and avoids redundant development of network functions.5G-AxAI New Technology,New Case,New Model White Paper 16 Figure 3-3 Foundation model

124、 design for the intelligent network In general,the LLM mainly focuses on handling with speech,text,and images rather than network data.To reach full potential of LLM in network,the foundation model needs to be re-designed in the following aspects.First,it is important to customize the method of netw

125、ork data tokenization.The network data needs to be tokenized according to its characteristic,which maps the multi-dimensional network data(e.g.,traffic,user info,topology)to vector spaces and helps the model extract high-level data features.Second,the model structure and optimization objectives shou

126、ld be re-built.Appropriate setting of layer,parameter,loss function and training policy would guarantee the model convergence when it incorporates so much heterogeneous network data.Third,the cost of model adaptation and deployment should be decreased due to the current status of network,which haven

127、t been equipped with efficient computing power but requires real-time processing.This could be achieved by Parameter-Efficient Fine-Tuning(PEFT)and model compression.In the area of network traffic recognition,unlike typical methods,the foundation model doesnt depend on plain-text rules and could ext

128、ract the high-level implicit features to distinguish different kinds of packets and flows.It brings better accuracy especially in the case of encrypted traffic,private protocol and unregistered service type.Moreover,the network inherently supports to track user-related data such as location change,n

129、etwork access and traffic usage,which enables the network to perceive users preferences,lifestyles and occupations.Given the impressive semantic understanding,contextual information processing,and cross-domain adaption abilities of the foundation model,it could act as an important tool for more accu

130、rate mobile user profiling by integrating multi-source heterogeneous user-related data,which further facilitates the development of more individualized marketing strategies.3.1.3 Intelligent Network Infrastructure Life-cycle Network infrastructure involves hardware devices,cloud platforms,network fu

131、nctions,and network management systems.To better support increasingly complex upper-layer applications,network infrastructure requires for efficient and high-quality construction and delivery.Extensive and repeated verification is needed during stages such as research and development,testing,deploym

132、ent,and integration to ensure infrastructure stability.Therefore,introducing AI capabilities into network infrastructure is urgently needed to replace significant labor costs.To make the network more intelligent,AI agents could be applied and enhanced in the full lifecycle of infrastructure integrat

133、ion and verification,such as testing,integration,delivery,and operation and maintenance of 5G-A infrastructure.In the testing and validation phase,AI agents can assist in requirement management,solution formulation,acceptance validation,and pilot validation.In the construction and delivery phase,AI

134、agents can automatically complete tasks such as environment setup,delivery validation,configuration,deployment,and issue localization.In the 5G-AxAI New Technology,New Case,New Model White Paper 17 operation and maintenance evaluation phase,AI agents can perform quality assessment,problem prediction

135、,operation and maintenance monitoring,and disaster recovery drills.Accordingly,the intelligent infrastructure integration and verification technology will greatly enhance the level of test automation and management efficiency in network infrastructure.The architecture if intelligent network infrastr

136、ucture is divided into four layers,similar to the human senses,brain,nerve signals,and hands.Each layer has its corresponding functions and applications in the 5G-A network.Figure 3-4 Architecture of intelligent network infrastructure solution Interaction Layer:Serving as the senses for interaction

137、between the 5G-A AI agents and humans,this layer is responsible for perceiving and processing multi-modal information interactions with the external environment,including text,images,audio,and video.It also feeds back the results of task processing to users through the interaction layer and enables

138、iterative interaction processes.Model Layer:Serving as the brain of the AI agent,this layer is responsible for task decomposition,planning,reasoning,memory,reflection,and tool usage.For different scenarios in 5G-A infrastructure,the large model is divided into core network models,web models,cloud pl

139、atform models,etc.,for completing specific tasks in different domains.Semantic Layer:To bridge the gap between task execution and natural language,a semantic layer is established,serving as the neural signal layer for the AI agent,enabling large models to understand and execute various tasks.This se

140、mantic layer sets a unified MAML(Meta Action Markup Language)and defines sub-languages for subdomains such as network verification,web verification,cloud platform verification,deployment and configuration.Execution Layer:Serving as the hands and feet of the AI Agent,this layer is responsible for exe

141、cuting,orchestrating,and scheduling specific tasks.Leveraging intelligent pipelines,the execution layer completes tasks such as environment deployment,test case validation,quality assessment,and resource management based on automation tools or scripts,and then feeds back results to the model layer f

142、or iteration,reflection,and planning of subsequent tasks.Through this technology,5G-A network infrastructure can achieve more efficient and intelligent management and operation to meet the rapidly evolving needs of future network technologies.5G-AxAI New Technology,New Case,New Model White Paper 18

143、3.1.4 AI for Network Mobility The application of AI communication networks represents a significant breakthrough in 5G-A.One of the critical innovations is AI for network mobility.Addressing the dynamic and complex demands of networks,AI offers new solutions to optimize handovers,predict potential f

144、ailures,and reduce resource and measurement overhead.1)RRM Measurement Prediction Radio Resource Management(RRM)measurement prediction plays a pivotal role in facilitating efficient mobility management6.AI-based approach leverages historical and real-time measurement data across clusters of cells to

145、 generate predictions that improve handover decisions and reduce the need for frequent measurement and reporting.These AI models could be trained to predict measurements in the temporal,frequency,and spatial domains.The framework diagram as shown in Figure 3-5 illustrates the overall prediction stru

146、cture,where AI models can be deployed at various locations within the network.Among the proposed approaches,we emphasize that the cluster cell-based method combined with Case 3(direct prediction)demonstrates superior potential for leveraging AI prediction gains.This is achieved by incorporating rich

147、er spatial information and performing joint optimization across multiple modules,enhancing both prediction accuracy and network performance.Figure 3-5 RRM measurement prediction framework 2)Failure Prediction The prediction of Handover Failures(HOF)and Radio Link Failures(RLF)is essential for mainta

148、ining seamless connectivity during user mobility7.AI models analyze trends in signal quality over time to identify patterns that indicate potential failures,enabling proactive network adjustments to mitigate service interruptions.RLF scenarios associated with T310 expirya representative case for fai

149、lure predictioncan be addressed using both short-term and long-term approaches.In short-term scenarios,where T310 has already been triggered,AI models estimate the likelihood of failure within the remaining duration of the timer.These predictions provide immediate insights into the risk of failure,s

150、upporting real-time decision-making and recovery actions.Long-term predictions,conducted before T310 activation,focus on estimating both the probability and timing of potential failures.By incorporating extended RRM measurement trends 5G-AxAI New Technology,New Case,New Model White Paper 19 across m

151、ultiple cells,long-term approaches allow for preemptive planning and optimization of mobility strategies.Direct prediction models,which output failure probabilities directly,are particularly suited to short-term scenarios due to their ability to offer precise and actionable insights.Conversely,indir

152、ect prediction approaches,which rely on temporal RRM measurement predictions,are more effective in long-term scenarios.These methods capture the evolving conditions of both serving and neighboring cells over extended time windows,offering a comprehensive perspective for resource allocation and mobil

153、ity management.In environments with higher failure risks,such as FR2 deployments,these predictive capabilities are particularly valuable.While direct predictions address immediate risks,indirect predictions leverage multi-cell spatial and temporal data to support strategic planning,ensuring reliable

154、 network performance.Together,these methods enhance the ability to reduce HOF and RLF rates,contributing to more robust and adaptive network operations.By integrating these predictive approaches,AI models play a transformative role in ensuring seamless service delivery,even in the most dynamic and c

155、omplex mobility scenarios.3)Event Prediction By leveraging AI models,the prediction of measurement events,such as Event A3,helps optimize network operations,reducing measurement overhead and enhancing handover performance through more precise and timely event triggers8.Similar to failure prediction,

156、measurement event prediction can also be achieved through both indirect and direct approaches.Indirect prediction utilizes RRM measurement predictions as inputs,allowing the network to infer the occurrence of events based on trends in temporal and spatial data from serving and neighboring cells.This

157、 method integrates seamlessly with existing mobility frameworks,leveraging multi-cell correlations to maintain prediction accuracy while reducing operational complexity.On the other hand,direct prediction bypasses intermediate steps,with AI models directly outputting event triggers.This approach is

158、particularly well-suited for real-time applications,enabling faster response times and streamlined processing.In scenarios focused on reducing measurement overhead,such as FR1 deployments,predicted measurement events can replace actual measurements,significantly lowering signaling load while maintai

159、ning equivalent performance.For instance,temporal domain predictions allow the network to decrease the frequency of measurement reporting without sacrificing event accuracy.Meanwhile,in FR2 environments,where mobility demands are higher,these predictions enhance handover performance by proactively a

160、djusting parameters like hysteresis and time-to-trigger.This reduces the likelihood of late handovers and improves the selection of optimal target cells,ensuring smoother transitions between network nodes.Evaluating the accuracy of these predictions is crucial to their effectiveness.Metrics such as

161、missed prediction ratios,false prediction ratios,and time accuracy are essential for assessing reliability.Accurate predictions ensure that event triggers occur at the right time,enabling timely handovers and minimizing disruptions.These capabilities contribute to a more adaptive and efficient mobil

162、ity management framework,addressing the dynamic challenges of next-generation networks.4)Charting the Path Forward AI for mobility in 5G-A represents a transformative step forward,as demonstrated by advancements in measurement accuracy,failure prediction reliability,and event forecasting 5G-AxAI New

163、 Technology,New Case,New Model White Paper 20 capabilities.The development of robust frameworks that integrate prediction outcomes into network decision-making processes will be critical to realizing these benefits.Future research will continue to enhance the adaptability of AI models to varied depl

164、oyment conditions,ensuring consistency and scalability.By embedding AI-driven insights into the heart of 5G-A,the industry moves closer to achieving its vision of a more intelligent,adaptive,and efficient network.These innovations lay a strong foundation for future capabilities in mobility managemen

165、t,paving the way for a truly connected future.3.2 Digtal Twin Network Intelligence Digital Twin Network(DTN)represents a paradigm shift in how we manage and optimize 5G-A network and beyond.By creating a virtual replica of the physical network infrastructure,DTN enables operators to simulate,analyze

166、,and optimize network operations with unprecedented precision.This digital representation serves as both a mirror of current network state and a sandbox for testing future scenarios,offering a powerful platform for innovation and network optimization.3.2.1 Network Data Mapping Technology Data mappin

167、g forms the cornerstone of network digital twin implementation,providing essential input for twin modeling.The quality and comprehensiveness of collected data directly determine how faithfully the digital twin reflects its physical counterpart.However,modern networks present significant challenges f

168、or data collection due to their vast scale,diverse device ecosystem,complex interface configurations,and dynamic traffic patterns.These challenges make high-precision data collection both resource-intensive and potentially disruptive to network performance.To address these challenges,the focus has s

169、hifted toward developing an efficient network data collection mechanism that meets the specific requirements of network twins.This mechanism adopts an on-demand approach to data collection,carefully selecting collection targets,precision levels,protocols,and transmission methods.The goal is to achie

170、ve a balanced solution that delivers comprehensive data while maintaining efficiency and energy conservation.5G-AxAI New Technology,New Case,New Model White Paper 21 Figure 3-6 The framework of AutoOPT The AutoOPT framework represents an innovative approach to this challenge by leveraging generative

171、 AI models for data generation and optimization.As illustrated in Figure 3-6,this framework operates through two distinct stages.In the data generation stage,it creates synthetic data from small-scale networks using scale-independent indicators,ensuring that DTN AI models9 can effectively generalize

172、 to larger network environments.The subsequent optimization stage automatically identifies and filters high-quality data through seed sample selection and incremental refinement,ultimately enhancing both the accuracy and generalization capabilities of DTN AI models.This approach not only addresses t

173、he immediate challenges of data collection but also creates a foundation for continuous improvement in digital twin fidelity.By using the digital twin itself as a simulated data generation entity,we can enrich machine learning training datasets and create more robust models for network analysis and

174、optimization.3.2.2 Digital Twin Modeling Technology NDT modeling technology is pivotal for realizing effective network intelligence in complex 5G-Advanced and future networks.It focuses on constructing high-fidelity virtual replicas of physical network infrastructures,balancing model accuracy with c

175、omputational efficiency.In the context of increasingly intricate and dynamic network architectures,a robust NDT model is essential for functionalities such as accurate state replication,predictive simulation,and historical event analysis,enabling proactive network management and optimization.NDT mod

176、eling is structured around three fundamental dimensions,synergistically contributing to a holistic digital representation.Firstly,network state digital twinning addresses the real-time and historical representation of network entities,their attributes,and inter-relationships.This dimension encompass

177、es both invariant network configurations and dynamic operational telemetry.Technique employed includes:Network Emulators:Utilizing platforms that simulate network environments with varying levels of abstraction,from detailed packet-level emulation to higher-level behavioral models.Data Abstraction E

178、xpressions:Employing formalized data models and schema to 5G-AxAI New Technology,New Case,New Model White Paper 22 represent network state data efficiently and facilitate semantic interoperability.This can involve utilizing data serialization formats and standardized information models.Knowledge Gra

179、phs:Constructing graph-based representations to capture network topology,entity relationships,and state dependencies,enabling knowledge-driven reasoning and inference within the digital twin.Secondly,service and environment twinning focuses on modeling the service layer and its operational context.T

180、his is crucial for understanding service performance and user experience within the network.Key methodologies include:Generative Adversarial Networks(GANs):Leveraging GAN architectures to synthesize realistic and dynamic service behavior patterns and simulate user interaction profiles.GANs are parti

181、cularly effective in capturing complex,non-linear service dynamics and generating synthetic datasets for model training and validation.Deep Learning-based Environmental Reconstruction:Utilizing deep learning algorithms,particularly CNN and point cloud processing techniques,to process sensor data(e.g

182、.,LiDAR,camera imagery,deployment blueprints)for automatic identification,extraction,and 3D reconstruction of the physical network deployment environment.This generates spatially accurate context for network simulations and visualizations.Thirdly,network behavior twinning models the functional chara

183、cteristics of network elements.This dimension employs a spectrum of modeling approaches based on the desired level of fidelity and computational cost:White-box Modeling:Implementing network protocols and device functionalities directly within the digital twin environment,often through Software-Defin

184、ed Networking(SDN)and Network Function Virtualization(NFV)principles.This provides high transparency and control over simulated network operations,suitable for detailed protocol analysis and functional verification.Black-box Modeling:Utilizing data-driven modeling techniques,such as machine learning

185、 and statistical modeling,to predict network status transitions and performance metrics based on observed historical data.This approach abstracts away from internal protocol details,focusing on input-output relationships and predictive accuracy,suitable for performance forecasting and anomaly detect

186、ion.Gray-box Modeling:Combining abstract network mechanism representations with selected actual network functions.This hybrid approach offers a pragmatic balance between modeling fidelity and computational complexity.Formal methods like Petri Nets are particularly valuable in this context,offering a

187、 mathematically rigorous framework for modeling concurrent and asynchronous events in distributed industrial IoT network digital twins,as demonstrated in prior research10.5G-AxAI New Technology,New Case,New Model White Paper 23 Figure 3-7 The NDT modeling method These three dimensions,as conceptuall

188、y illustrated in Figure 3-7,are integral to constructing a comprehensive NDT.This integrated modeling paradigm empowers network operators with enhanced capabilities for network comprehension,predictive analytics,and proactive optimization,ultimately facilitating the evolution towards intelligent,aut

189、onomous network management and supporting the advanced service requirements of modern network infrastructures.3.3 Application Intelligence Communication networks are evolving from basic connectivity services to intelligent application services.Through the deep integration of AI technology,networks n

190、ot only provide traditional voice and data services but also support new intelligent services such as multi-modal interaction,real-time translation,and digital humans.To support these innovative applications,networks have undergone technical evolution.These enhancements not only expand the applicati

191、on boundaries of networks but also create new business models for operators,shifting from traffic operations to value-based operations.3.3.1 IMS Data Channel The IP Multimedia Subsystem Data Channel(IMS DC)is an advanced feature that enables high-speed,real-time transmission of non-voice data during

192、 communication sessions,leveraging the infrastructure of IMS-based voice and video services.It is designed to meet specific requirements related to latency,bandwidth,and reliability,which are crucial for applications such as AR,real-time multimedia,and IoT interactions.This section explores the arch

193、itecture,function,and advantages of IMS DC in detail.5G-AxAI New Technology,New Case,New Model White Paper 24 Figure 3-8 IMS DC enables real-time multimedia interaction As shown in Figure 3-8,the IMS DC is integrated within the IMS framework,building on the voice and video channels to create a seaml

194、ess,multipurpose communication channel.This integration allows for synchronous transmission of data types alongside traditional voice or video calls,ensuring that data flows remain synchronized with the multimedia content.The IMS DC operates efficiently under the existing IMS architecture,utilizing

195、the inherent advantages of telecom networks,such as global connectivity via telephone numbers,unified authentication,and robust session management.The IMS DC is divided into two primary components:Bootstrap Data Channel(BDC):this channel facilitates the download and execution of IMS DC applications

196、on the terminal.These applications may include web content(e.g.,HTML5 pages),media elements,and control scripts(e.g.,JavaScript)required for real-time interactions.Application Data Channel(ADC):this is the data conduit through which applications on the terminal transmit their data to other devices o

197、r networks.The ADC ensures the real-time transmission of application-specific data in parallel with voice/video traffic,maintaining consistent user experience and minimal delay.The IMS DC is established between the terminal and the network to manage data exchange efficiently.When a communication ses

198、sion begins,the DC Server establishes a BDC to push DC applications to both calling and receiving terminals.Once these applications are active,the ADC is used to handle data transmission,ensuring real-time synchronization of all multimedia and application data.The control and media functions are sep

199、arated in the IMS DC architecture.This approach uses Data Channel Signaling Function(DCSF)to handle signaling control and Media Function(MF)to manage media resources,supporting functionalities like AR rendering and media processing,which greatly improves flexibility and scalability for future applic

200、ations.The IMS DC provides essential benefits for modern communications,including:Real-Time,Multi-Modal Interaction:By combining video,voice,and data streams,it enables immersive and interactive communication experiences,such as AR and virtual meetings,directly within the IMS framework.Enhanced QoS

201、and Security:Leveraging the robust capabilities of IMS,the IMS DC ensures QoS management and strong security protocols,protecting data integrity and minimizing latency.Flexible Application Integration:The ability to dynamically introduce new applications through the BDC streamlines the process of ex

202、tending IMS capabilities,ensuring that future innovations can be rapidly integrated into the network without major 5G-AxAI New Technology,New Case,New Model White Paper 25 infrastructure changes.In summary,the IMS DC represents a pivotal step forward in the evolution of telecommunication services,of

203、fering a unified channel for data,voice,and video communication,optimized for real-time,multimedia interactions across diverse network environments.The ongoing advancements in IMS DC,particularly with the integration of Augmented Reality(AR)and future communications services,underline its role as a

204、cornerstone for 5G-A and beyond communication systems.3.3.2 Interactive New Calling The 5G-A new call services provide users with interactive,intelligent,and immersive innovative new services and new scenarios,such as digital human,intelligent translation,intent communication etc.Figure 3-9 New AI T

205、echnologies for New Calling The new AI technologies refresh traditional call services,and promotes the expansion of business models from voice or video to multi-modal communication.With the rapid development of AI technologies,AI has become an indispensable force to promote communication technology

206、innovation.Through the AI empowerment network,continuous innovation is injected into call services.There are many new AI technology used for new calling as following:Figure 3-10 Interactive New Calling Intent Recognition:intent recognition refers to the recognition and understanding of intentions or

207、 purposes expressed in human languages through natural language processing technologies like LLM,Natural Language Processing(NLP).Intent-based communication greatly simplifies the interaction procedure of new calls.The system 5G-AxAI New Technology,New Case,New Model White Paper 26 automatically ide

208、ntifies and executes the intent based on only texts and pictures.For example,if you order 1,000 pizzas,the system automatically executes the intent identification and make orders.XR+AGI Technology:the Extended Reality(XR)interactive technology refers to a real and virtual combination and man-machine

209、 interaction environment generated through computer technologies and wearable devices.It integrates virtual information and real scenarios to create a man-machine interaction virtual environment.It can be used in entertainment,education,medical care,and industrial fields to provide more abundant and

210、 immersive experience.Artificial General Intelligence(AGI)technology can be used to quickly generate XR videos and digital worlds.Digital human:digital human technology is a high-tech product that combines computer graphics,motion capture,image rendering,and artificial intelligence.It allows the cre

211、ation of virtual characters with human looks,behaviors,and characteristics.These virtual characters can exist in digital space and interact with the real world.Intelligent translation:combined with real-time communication,intelligent translation translates the voice of a subscriber into text informa

212、tion and displays the text information on the subscribers mobile phone.Intelligent translation can be used for translation between different languages or different dialects based on AI like Automatic Speech Recognition(ASR),LLM etc.It is also applicable to real-time speech/text conversion when voice

213、 calls cannot be made,for example,communication between a normal person and a hearing-impaired person.Through the application of AI technologies such as ASR,intent recognition,intelligent translation,and digital human,new calling will benefit from the following:New business model:in the future,commu

214、nications will evolve from traffic and time operations to value operations and provide value services,such as digital man,personal assistants,and avatar/ego communications.Shortening the service launch time:by integrating the AI technology,the service launch time can be simplified.Through the plug-i

215、n intelligent AI platform,new service components can be plug-and-play.3.3.3 Cloud-Edge-Terminal Collaboration Cloud-edge-terminal collaboration is central to optimizing communication networks,especially for AI-driven applications.As AI models grow in complexity,efficiently distributing tasks across

216、cloud,edge,and terminal layers is vital.Key strategies like computational offloading,resource scheduling,and data collaboration are critical to enabling seamless collaboration across these layers.1)Computational Offloading 5G-AxAI New Technology,New Case,New Model White Paper 27 Figure 3-11 On-Devic

217、e AI:energy and memory consumption With the increasing complexity of AI models,terminal devices are often not capable of handling the computational load on their own.Offloading these tasks to the cloud or edge allows for better distribution of computational workloads,ensuring real-time performance a

218、t the terminal.As illustrated in Figure 3-11,energy and memory usage on terminal devices rise significantly as the complexity of AI models increases.This highlights the importance of offloading more resource-intensive tasks to the cloud or edge,thus allowing terminal devices to focus on lighter comp

219、utations.Figure 3-12 AI model size requirements for different tasks Similarly,Figure 3-12 shows the varying computational resource needs of different AI models.Large-scale models,such as those used in deep learning,are typically offloaded to the cloud,while smaller models that need real-time inferen

220、ce are handled at the edge or terminal.This tiered approach ensures that AI applications can scale effectively,based on the specific computational requirements of the task at hand.2)Uplink Centric Broadband Communication With the popularization of AI applications such as AI video calls and AI Agents

221、,AI-human interact with each other in multi-modal,such as images and videos,and uploading of images and videos from terminal to cloud becomes necessary.According to related research,an uplink rate of 20Mbps is required to ensure interaction experience of 80%common applications and 30Mbps is required

222、 to ensure interaction experience of 60%high-experience applications like enhanced 5G-AxAI New Technology,New Case,New Model White Paper 28 Supplementary Uplink(SUL).Figure 3-13 Uplink Experience Requirements for Multi-modal Interaction Uplink Centric Broadband Communication(UCBC)improves the upstre

223、am bandwidth capability by 10 times,meeting the upload requirements of multi-modal interaction,machine vision,and massive broadband IoT in different AI application scenarios,accelerating the intelligent upgrade of thousands of industries.Currently,more spectrum and bandwidths are aggregated based on

224、 the current uplink channels to improve the uplink capability.The main technologies include flexible spectrum access,SUL enhancement,and uplink carrier aggregation.FA SUL networking could be applied to achieve large uplink bandwidth and meet multi-modal interaction requirements.In addition,the uplin

225、k and downlink slot assignment of the TDD frequency band are adjusted to increase the scheduling of uplink time-slot resources,improving the uplink capability.3)Resource Scheduling Efficient resource scheduling is crucial for ensuring that tasks are allocated optimally across cloud,edge,and terminal

226、 nodes.Dynamic allocation of resources ensures that computational power,memory,and bandwidth are used effectively,minimizing latency and maximizing throughput.AI applications often require significant uplink bandwidth,especially when processing large datasets.Meeting this demand is critical for ensu

227、ring smooth data transfer between the terminal,edge,and cloud layers.In the context of cloud-edge collaboration,the role of resource scheduling becomes even more significant.The core focus is on managing tasks based on the current network load and resource availability,ensuring that computational re

228、sources are used as efficiently as possible.This dynamic scheduling helps maintain optimal performance,reducing bottlenecks and enabling faster task execution.4)Data Collaboration Data collaboration ensures the smooth flow of information between the cloud,edge,and terminal.Data collected at the term

229、inal is first processed and cached at the edge,reducing the need to transfer large amounts of data to the cloud.This local processing minimizes latency and makes real-time decision-making possible,particularly for applications where time sensitivity is critical.5G-AxAI New Technology,New Case,New Mo

230、del White Paper 29 Figure 3-14 Cloud-Edge Data Collaboration Scheme Figure 3-14 illustrates how data is processed and managed across these layers.Data is pre-processed at the edge before being uploaded to the cloud,which then stores and further analyzes the data.This collaboration ensures that only

231、the most relevant data is transmitted to the cloud,optimizing bandwidth usage and ensuring efficient model training.It also enables secure handling of sensitive data,especially in domains like healthcare and finance.The synergy between computational offloading,resource scheduling,and data collaborat

232、ion enables cloud-edge-terminal collaboration to operate efficiently.By leveraging these strategies,the overall performance of AI applications improves,with computational power distributed appropriately,latency minimized,and data flows optimized.As AI-driven applications continue to evolve,the abili

233、ty to dynamically allocate resources and manage data across the cloud,edge,and terminal will become even more important.This collaboration not only improves the efficiency of AI processing but also allows applications to scale effectively,meeting the growing demands of next-generation services.3.4 S

234、ustainability Intelligence With the rising global electricity prices,power conservation and energy efficiency have become critical priorities in the telecommunications industry.For mobile networks,reducing energy consumption to achieve net-zero emissions has become a core goal for operators,with ove

235、r 85%of mobile network operators committed to this objective.As both the Radio Access Network(RAN)and Core Network contribute significantly to overall energy consumption,enhancing energy efficiency in both areas is crucial.The integration of AI plays a pivotal role in improving energy efficiency.By

236、analyzing real-time traffic patterns and resource utilization,AI enables optimized resource scheduling and power management,significantly reducing energy consumption while maintaining service quality.5G-AxAI New Technology,New Case,New Model White Paper 30 3.4.1 Enhancing Equipment-Level Energy Effi

237、ciency At the equipment level,the primary energy consumer is the Power Amplifier(PA)in the radio unit,which amplifies signals for transmission.Optimizing the efficiency of the PA is essential for improving overall energy savings.Existing technologies improve the PAs linearity and efficiency under lo

238、w load conditions;however,most solutions show limited optimization effects during periods of low load,which are common in network operations.Figure 3-15 Intelligent PA control To address this,AI-driven Digital Pre-distortion(DPD)techniques have been employed to improve the linearity of the PA.Tradit

239、ional DPD algorithms have been upgraded with machine learning models that dynamically adjust to different environmental and signal conditions.Deep Neural Networks(DNNs)model the nonlinear characteristics of the PA and adjust the input signal in real time to achieve better power efficiency and signal

240、 quality.AI-driven DPD systems are trained on extensive real-world signal data,taking into account various frequency bands,power levels,and environmental conditions.These systems can autonomously adjust parameters in complex network environments,reducing signal distortion and enhancing performance.3

241、.4.2 Optimizing Network-Level Energy Efficiency At the network level,AI predicts traffic load fluctuations by analyzing historical traffic patterns,weather data,and local events,allowing for dynamic adjustment of network resources to reduce unnecessary energy consumption.AI-powered shutdown mechanis

242、ms can predict low-traffic periods and automatically deactivate components such as power amplifiers and transceivers,reducing energy consumption without affecting service quality.AI models also allow for intelligent control of network equipment by analyzing real-time feedback from both terminal devi

243、ces and network components.This enables the network to dynamically adjust power distribution and transmission strategies.For instance,PA and Base-band Units(BBUs)in base stations can enter deep sleep during low-load periods and be reactivated only when traffic increases.Furthermore,AI optimization e

244、xtends beyond the RAN and includes energy efficiency improvements in the Core Network.In the Core Network,intelligent scheduling and resource management mechanisms analyze traffic patterns,user demands,and network topology to dynamically route data flows and minimize unnecessary energy consumption.5

245、G-AxAI New Technology,New Case,New Model White Paper 31 Through multi-dimensional energy-saving strategies,including time,space,frequency,and power domain optimizations,AI enables precise energy optimization without compromising performance.This capability is expected to evolve further in R18 and be

246、yond,incorporating network digital twin technology for offline energy-saving predictions,enabling more accurate energy-saving strategies for future network deployments 4 5G-AxAI Enables New Cases 4.1 Differentiated Experience Assurance As mobile internet develops,users demands for network are becomi

247、ng increasingly personalized.From 4G to 5G,operators primarily manage data services through traffic management.In the era of 5G-A,utilizing AI technology to dynamically adjust network resource allocation is crucial for meeting the differentiated experience requirements of various user levels(such as

248、 diamond,platinum,gold,silver,and standard cards),different types of services(like short videos,games,and live streaming),medium to high-speed mobile scenarios(such as high-speed rail and subway),and high-capacity scenarios(like concert venues and tourist attractions).Technologies such as network pe

249、rception,experience assurance,and experience evaluation can significantly enhance user satisfaction.Network perception:Currently,the intelligent network perception module supports the identification of tens of thousands of service categories,including mainstream domestic and international service tr

250、affic.It also finely identifies various sub-services within super apps,such as voice calls,video calls,live streaming,video conferencing,and cloud gaming,with an overall sensing rate exceeding 95%.Experience Assurance:Based on the results of network perception,targeted assurance strategies are provi

251、ded for corresponding services.Key technologies for experience assurance adopted on the wireless side include precise service assurance prediction,multi-objective optimization service assurance strategies,and intelligent multi-frequency coordination.On the core network side,network intelligence base

252、d on NWDAF is introduced to provide dynamic Guaranteed Bit Rate(GBR)assurance according to user needs.Experience Evaluation:Experience evaluation involves developing a comprehensive set of QoE metrics based on the characteristics of different services.For video services,this includes assessing clari

253、ty,smoothness,and timely interaction through multiple dimensions such as content quality,transmission quality,and interaction quality,which are then integrated into a unified standard score 1,5.For instant messaging services,the focus is on factors like service type and response timeliness,with metr

254、ics including text,voice,image,and video transmission types,along with service volume,duration,speed,and performance indicators under TCP/UDP protocols(such as packet loss rate,delay,jitter,and TCP retransmission rate),all fitted to calculate the Mean Opinion Score(MOS)for instant messaging services

255、.This comprehensive approach ensures that the evaluation accurately reflects the quality experienced by users across various types of services.Test results on the existing network are exhibited as follows:On the wireless network side:After deploying the precise service assurance strategies,the laten

256、cy for various types of services,such as short video,QR code payment,and web 5G-AxAI New Technology,New Case,New Model White Paper 32 browsing,was reduced by approximately 20%.Figure 4-1 Reduction in latency for short video,QR code payment,and web browsing services.On the Core Network side:After int

257、roducing NWDAF-based intelligence,when a decline in user experience is detected,the GBR assurance is then employed to enhance users experience.Figure 4-2 For VIP users,upstream speed was 2Mbps before congestion,dropped to 255Kbps after congestion,and restored to 2Mbps after establishing GBR.Addition

258、ally,during the service assurance process,the logo on the users mobile device can dynamically display the text VIP Assurance Active in real-time,matching the enhanced experience and elevating the users perception.After the assurance process ends,users receive an experience report providing real-time

259、 feedback on the measurable assurance effects.This completes the closed-loop experience management,achieving a higher level of care for the customer.5G-AxAI New Technology,New Case,New Model White Paper 33 Figure 4-3 Zhejiang Mobile Test Results 4.2 New Calling Service In recent years,the developmen

260、t of multimedia large models,AI Agent,and DC technologies is driving the communication industry from traditional audio and video to multi-modal,interactive communication.This provides users with a series of rich and colorful services such as intelligent video calls,interactive video calls,and call i

261、ntelligent assistants,significantly enhancing the communication experience and efficiency for individuals and enterprises.It promotes the prosperous development of the communication industry and opens up new growth space for operator businesses.In 2023,China Mobile,in collaboration with partners,lau

262、nched the 5G New Calling service and completed the construction of the first phase of the New Calling network covering 31 provinces nationwide by the end of the year.Relevant New Calling network elements were deployed in the resource pools of the eight major regions,involving the upgrade of existing

263、 IMS network elements and the deployment of new calling capability elements.This supported the release of six New Calling service scenarios:lighting up screen,fun calling,celebrity calls,AI shorthand,speech to text,and real-time translation.1)Lighting up the screen:from voice to video to content,ope

264、ning up a new space for content operation In 2023,China Mobile Jiangsu worked with Huawei and other partners to develop a new call lighting up screen service to meet consumers emotional expression requirements.The service was released in Q4 of that year.Figure 4-4 New call lighting up screen service

265、 Lighting up the screen is a product centered on user communication social attributes.Users can use the product to set personal virtual images.During a voice call,users can transmit and 5G-AxAI New Technology,New Case,New Model White Paper 34 display preset pictures,videos or virtual digital human i

266、mages to the calling party without switching video calls,so as to convey emotions and express individuality.Enterprise users can customize corporate images,spread corporate value,or conduct product marketing.During calls,they can trigger introduction videos or pictures based on different keywords to

267、 improve communication efficiency and effect.Carriers can publicize anti-fraud and security knowledge to assume more social responsibilities.In less than a year,Jiangsu Mobile has more than 5 million users of the screen-lighting service,and more than 100 million content displays per month.It has rec

268、eived good market feedback and has become a new call service and is popular with consumers.China Mobile Guangdong enables the ToB application for all installation and maintenance personnel.During communication with customers,keywords are used to trigger Fiber to The Room(FTTR)brand promotion and pac

269、kage promotion videos,facilitating service promotion and efficient communication.In the second half of 2024,China Mobile upgraded its screen-lighting business,implanting Artificial Intelligence Generated Content(AIGC)self-creation capabilities,allowing users to create their own stylized images and s

270、how them to each other during calls.Jiangsu Mobile also explores the Another Me function of digital human calls.Based on the users appearance characteristics,the stylized digital human image is generated by AI.During a call,the digital human is driven by the network AI based on the users voice.Real-

271、time synchronization of my voice and digital persons expression and lip shape,infusing more fresh experience into the call.By the end of 2024,more than 15 million users had subscribed to the light-up screen service.Turning on the screen enables carriers to upgrade the call duration operation to an a

272、verage of 90 seconds,which brings greater business and social value to calls and becomes a new mobile media.2)Real-time translation&Speech to text:Technology for Good,Building a Bridge for Real-Time Communication In September 2023,at the 19th Asian Games in Hangzhou,China Mobile launched the Smart T

273、ranslation service for the Asian Games.When cross-language communication is required,the Smart Translation service can automatically identify and translate voice content during native video calls without relying on translation software.Bilingual subtitles are displayed.During the Asian Games,real-ti

274、me translation between English,Korean,Japanese,Arabic and other languages can be supported,allowing communication to cross the language gap.The Call Caption service can identify the voice content of calls and display the voice content in large fonts on the mobile phone screen to reduce communication

275、 obstacles and greatly improve the call experience of the hearing impaired.Figure 4-5 Real-time translation and call captioning service In 2024,the real-time translation and call captioning service was launched in all provinces of China Mobile.In less than one year,more than 5 million users subscrib

276、ed to the call captioning or 5G-AxAI New Technology,New Case,New Model White Paper 35 call captioning service.The new call truly fulfills the tenet of technology for good.Help users to move towards the goal of barrier-free communication.By the end of 2024,the number of new 5G call users on the live

277、network has reached 40 million,greatly improving user call experience and communication efficiency,and bringing corresponding business and social values to operators.In addition to the original voice and video channels,DCs are added to transmit data during audio and video calls.Interactive calls are

278、 introduced to implement multimedia content such as image and video sharing,message communication,screen sharing,AR annotation,and file transfer during calls.Improve the experience and efficiency of remote communication and expand the industry value boundary.In the fourth quarter of 2024,China Mobil

279、e launched the trial commercial use of interactive video users on the live network.It launched six DC applications,namely,content sharing,message box,fun call+,real-time translation+,screen sharing,and digital person,to implement quick setting,sharing,and convenient interaction during calls.Figure 4

280、-6 Interactive video users With the collaboration of new communication industry partners,GSMA released TS.66 in June 2024,which defines the DC API standard on the device side.Mainstream chips from chip vendors such as Qualcomm,MediaTek,and UNISOC also support DC capabilities.Some mainstream terminal

281、 models from Vivo,OPPO,Xiaomi,Samsung,and Huawei support the DC function.The DC interactive call industry ecosystem is preliminarily established.It is estimated that 2025 will be the first year of large-scale commercial use of DC series applications.3)AI Agent Enable New Calling,Helping Carriers Bui

282、ld AI Service Entrances AI comprehensively upgrades the native call experience.As an intelligent call assistant,AI provides services such as intelligent pick-up and intelligent chat,and consolidates the personal call entrance.Upgrade the enterprise customer service hotline and build the enterprise i

283、ntelligent application portal.In the second half of 2024,China Mobile began to gradually build calling agent capabilities on the network.In Zhejiang,Guangdong,and Jiangsu provinces,China Mobile conducted service tests and trials in service scenarios such as intelligent pickup,AI-assisted chat,and st

284、ar call.The goal is to build a digital assistant for each user.Build a personal AI service entry.5G-AxAI New Technology,New Case,New Model White Paper 36 Figure 4-7 Intelligent call assistant In scenarios such as unreachable power-off,express delivery,and promotion,the call intelligence helps users

285、intelligently answer calls.Based on call content analysis,the interception conditions can be flexibly set to reduce misjudgment.The interception content is also notified to users and can be called back.To achieve the purpose of anti-harassment,anti-leakage and anti-fraud;It can also help users sort

286、out massive information during calls,remind important information in real time based on call scenarios,reduce users memory burden,and implement intelligent chat.The call agent can also invoke external vertical domain agents and search tools to help users complete meal booking and booking,improve cal

287、l experience and efficiency,and make calls a unified entry for personal AI services.In June 2024,China Mobile Jiangsu streamlined the business processes of vehicle insurance claims and loan customer service scenarios of Company As customer service hotline.AI was used to help enterprise customer serv

288、ice upgrade intelligently,evolving from traditional voice interaction to multi-modal agents of natural language,video,and data channels.The enterprise digital human customer service personnel obtain key information at a time through natural language interaction.Before the large-scale commercial use

289、of native terminals of the Data Channel,the multi-modal agent capability solves the problems of complex Dual Tone Multi Frequency(DTMF)interaction,multi-layer nesting,and time-consuming.Enterprises can quickly close transactions during calls,reduce transaction costs,achieve win-win among users,enter

290、prises,and carriers,and help carriers expand the enterprise market space.Figure 4-8 Intelligent customer service Users can invoke third-party vertical domain agents through code numbers based on actual requirements.Carriers can use code numbers to build a unified entry for call applications,implemen

291、t One Number,One Agent,and build an enterprise application ecosystem.The calling Agent technology will develop rapidly in 2025,gradually reshaping the call industry ecosystem.As the saying goes,Calling make AI everywhere,and AI makes calling omnipotent.5G-AxAI New Technology,New Case,New Model White

292、 Paper 37 4.3 Industrial Deterministic Service With the rapid development of 5G technology and the gradual improvement of network infrastructure,significant progress has been made in leveraging 5G to empower the digital transformation of industries.Information technologies such as 5G and artificial

293、intelligence have been widely applied in production management and auxiliary production processes.Solutions like data collection,video surveillance,and Automated Guided Vehicle(AGV)have achieved large-scale deployment.However,for production control scenarios,which require extremely high levels of de

294、terminism due to short packet transmission cycles(in milliseconds),stringent packet-level requirements(three consecutive packet anomalies can cause system downtime),and a variety of protocols with significant parameter configuration differences,current 5G technology has not yet been fully applied in

295、 this domain.Two main challenges remain:Traditional 5G QoS Identifier(5QI)mechanisms only guarantee average network performance and cannot provide the required packet-level determinism.This makes it difficult to meet the long-term stable operation needs of industrial control systems.Even with advanc

296、ed networking techniques such as network slicing,differentiated service strategies(such as DS frame structures and PDCP out-of-order delivery),and link redundancy(dual transmission with selective reception),some highly demanding scenarios,especially those requiring extremely low network jitter,still

297、 find it challenging to meet service requirements.The existing network solutions are insufficient to fulfill these stringent demands.To ensure deterministic service experiences,an 1+2+3 deterministic assurance solution is introduced by leveraging AI technology,enhancing the networks deterministic ca

298、pabilities.1 refers to an AI-based service sensing algorithm,focusing on industrial scenarios where AI algorithms are used for self-learning of service characteristics.This continuous iterative optimization accurately identifies packet-level service types,message lengths,transmission intervals,arriv

299、al times,and key words in industrial control packets.2 refers to two intelligent computing infrastructure solutions,including industrial smart cards for base stations and edge intelligent UPF.1)Industrial Smart Cards for Base Stations:These cards can quickly upgrade traditional base stations into in

300、telligent industrial base stations through plug-in methods,providing a foundational computing platform for AI capabilities such as intelligent service sensing.With forward-compatible design,they flexibly adapt to existing outdoor and indoor mainstream station models without requiring complete statio

301、n replacement,effectively reducing costs.2)Edge Intelligent UPF:Based on this infrastructure,an intelligent sensing-inference-orchestration closed-loop assurance technology system is built for deterministic services.It enables intelligent inspection of deterministic protocol packets,intelligent prom

302、otion of deterministic service scenarios,and intelligent management of deterministic orchestration strategies.This addresses challenges such as difficult inter-system coordination,high orchestration complexity,and long deployment cycles.Additionally,it integrates 5G computing power with industrial c

303、ontrol and industrial AI capabilities,achieving hybrid real-time virtualized operating systems and base computing power orchestration,promoting deep integration of 5G with industrial 5G-AxAI New Technology,New Case,New Model White Paper 38 application ecosystems,thus providing more efficient,intelli

304、gent,and deterministic production and management methods for enterprises.3 refers to three service-based scheduling enhancement solutions,optimizing packet-level scheduling based on characteristics by service sensing to achieve precise matching of network resources with service needs.This enhances d

305、eterministic performance while increasing the capacity of deterministic users,ensuring long-term stable operation of services.The three levels of assurance include:1)Fixed Configuration Assurance Based on Service Type:Adjust user-level protocol parameters,such as disabling power-saving features,inac

306、tivity timers,BWP switching,and UM mode,to provide differentiated assurance between industrial control terminals and ordinary terminals.2)Precision Scheduling Assurance Based on Service Characteristics:Optimize resource allocation strategies for service flows.On one hand,reserve resources based on p

307、acket periods and sizes,making resource allocation more precise and reducing transmission waiting delays.On the other hand,coordinate multiple industrial control flows for staggered transmissions to avoid conflicts,significantly enhancing the ability to ensure bounded low latency under multi-user co

308、ncurrency.3)Key Packet Scheduling Assurance Based on Service Logic:Optimize scheduler queuing strategies and reliability assurance strategies for key packets within service flows to ensure timely and accurate transmission,preventing watchdog resets and system downtime,thus achieving long-term stable

309、 operation of industrial control services.Figure 4-9 Example of intelligent industry 5G-AxAI New Technology,New Case,New Model White Paper 39 Figure 4-10 End-to end latency distribution Currently,the deterministic assurance technology solutions and products have been validated in over 40 pilot proje

310、cts at factories such as Lynk&Co Automotive,Ansteel,and Panda Electronics.Network performance:the deterministic network latency has reached 12ms99.99%,and the deterministic latency jitter has reached 8ms99.99%.The effectiveness of deterministic assurance has improved by more than 30%,significantly e

311、nhancing latency and jitter performance as well as the reliability of one-time successful transmission.Practical implementation:this technology is widely applied in various scenarios including full-directional wireless Programmable Logic Controller(PLC),remote control of overhead cranes,and collabor

312、ative control of AGVs.By enabling wireless industrial control,these solutions facilitate intelligent production in factories,leading to increased production efficiency,improved production quality,reduced labor costs,lower maintenance costs,and accelerating the digital and intelligent transformation

313、of industries.4.4 Green Energy Saving Driven by global green development trends and sustainable development goals,the deployment of low-carbon mobile communication networks has become an industry consensus.With the rapid expansion of 5G base stations,electricity costs now constitute a significant po

314、rtion of operators overall operational expenditures,making energy conservation and efficiency improvements crucial for enhancing operational efficiency.Digital technologies play a vital role in achieving carbon neutrality and are essential in helping the world address climate change.Actively explori

315、ng innovations and applications in energy-saving technologies,and utilizing digital and intelligent methods to enhance network energy efficiency,is not only an effective way to reduce operational costs but also a critical step in promoting sustainable development within the industry.Energy saving te

316、chnologies for base stations have already achieved comprehensive multi-dimensional energy consumption management across time domains,frequency domains,power domains,and spatial domains.However,given the complexities of multi-standard and multi-frequency networking,diverse service scenarios,and varyi

317、ng user perception needs,traditional broad-brush energy saving strategies struggle to meet these varied demands.The key challenge lies in precisely predicting service needs,dynamically triggering energy saving mechanisms,and differentiating energy saving strategies under the complex network conditio

318、ns.Achieving this precision and efficiency is essential for the successful application of energy saving.5G-AxAI New Technology,New Case,New Model White Paper 40 To address the limitations of single-target base station energy saving strategies that cannot adapt to complex network structures,diverse s

319、cenarios,and service needs,big data analysis and AI algorithms are introduced.These technologies enhance the real-time performance and precision of energy saving strategies across time,frequency,spatial,and power domains,thereby expanding energy saving potential.The schemes are stated as follows.1)P

320、recise Identification of Energy saving Cells By conducting joint regional analysis to identify co-covered areas,automatically configuring energy saving compensatory cells and discovering more energy saving opportunities,we can improve network energy efficiency.To achieve this,AI algorithms analyze l

321、arge amounts of base station data based on differentiated characteristics of different types of base stations.This allows for automatic identification and classification of base stations.Using a data relationship model between base station energy efficiency and traffic,problematic base stations are

322、identified,and energy saving technologies are employed to enhance overall network energy efficiency.Figure 4-9 Energy-saving cell identification 2)Accurate Prediction of Energy saving Durations Analyze historical network carrier/cell group data using regularity analysis and train optimal load predic

323、tion algorithms with time series models.Identify periods where low load conditions meet criteria for triggering carrier shutdown or deep sleep functions.Utilize AIs self-learning capabilities to iteratively adjust threshold parameters while monitoring fluctuations in performance metrics(basic KPIs,s

324、ervice KPIs,sensing KPIs).This helps find the best balance between energy savings and system performance by optimizing energy saving thresholds.Figure 4-10 Energy-saving duration prediction model 3)Intelligent Optimization of Energy saving Strategies For multi-scenario,multi-standard,multi-frequency

325、,and multi-energy saving technologies,develop coordinated energy saving strategies.Collect network status data such as Measurement Reports(MR),Performance Measurement(PM),and engineering parameters.Use Gaussian 5G-AxAI New Technology,New Case,New Model White Paper 41 Mixture Density-Based Spatial Cl

326、ustering of Applications with Noise(DBSCN)clustering algorithms to determine cell energy saving states.Train models for different cells and optimize multi-dimensional network performance indicators under various parameter configurations.Through joint self-learning within regions,continuously iterate

327、 and optimize to generate the best combination of energy saving technologies and network parameter configurations that match current coverage scenarios and service requirements.This achieves intelligent dynamic energy saving strategies tailored to“one site-one time-one strategy”,aiming for maximum e

328、nergy savings.4)Network Energy Saving Based on Service Sensing Introduce service sensing capabilities to optimize energy saving strategies based on different service types and requirements.For example,for latency-sensitive services,develop differentiated centralized scheduling strategies to balance

329、the impact of energy saving strategies on network performance.This enables the network to activate energy saving features without compromising service quality and extend the duration of these features.Additionally,dynamically adapt and optimize energy saving strategies based on changes in service ch

330、aracteristics and network monitoring indicators,ensuring network performance and meeting the needs of latency-sensitive services.Figure 4-11 Differentiated centralized scheduling based on service sensing The intelligent schemes such as service forecasting,energy saving cell identification,and energy

331、 saving optimization strategy have been deployed across more than 25 provinces in China Mobiles networks,serving approximately 4.5 million base stations.Service-differentiated energy saving schemes have also been piloted nationwide.On top of conventional energy saving features,intelligent energy sav

332、ing strategies can achieve an average daily energy saving gain of over 5%,ensuring a balance between energy savings and performance.This has significantly promoted the green and sustainable development of wireless networks.4.5 High-Reliability Network Network stability involves multiple aspects,incl

333、uding site-level and data center-level operations.This white paper discusses the challenges,solutions,and practical outcomes of network operation assurance from both site and data center perspectives.1)Site Operation Stability Assurance With the large-scale deployment and development of 5G,network structures are becoming increasingly complex,and application scenarios and service requirements are d