Case Study: China Telecom

Last updated:

China Telecom: Network AI Development Strategy and development layout

The case study “China Telecom: Network AI Development Strategy and development layout” will introduce the driving force of China Telecom’s artificial intelligence development, the future strategic vision and top-level design of China Telecom’s application and development of Artificial Intelligence technology, some specific cases, implementation measures and challenges of network artificial intelligence development.

 

1.Introduction

Founded in September 2000, China Telecom is a large state-owned communication backbone enterprise. With a registered capital of 213.1 billion yuan, an asset scale of more than 800 billion yuan and an annual income scale of more than 430 billion yuan, it has been ranked among the Fortune Global 500 for many consecutive years.

China Telecom owns the technology-leading mobile communication network. It provides global customers with comprehensive information services and customer service channel system covering all regions and services. At present, it has become the world’s largest LTE-FDD 4G operator, fibre broadband operator, IPTV operator and fixed telephone operator.

As one of the most critical technology revolution, the development and application of Artificial Intelligence technology have risen to the level of national strategy. Conforming to the tendency of an intelligent era, China Telecom officially released the transformation and upgrading strategy (i.e., Transformation 3.0) and Network Architecture Reconstruction CTNet2025 in 2016. The intelligent network, ecological business and smart operation will broaden information services, thus enhancing the network power and better serving the public.

Data, algorithms, and computing power are the three core elements for developing Artificial Intelligence. As an operator, China Telecom has an excellent IT environment and strengths for the development of artificial intelligence. China Telecom has a big data lake composed of massive network data and business data resources, which is a valuable data resource for research on artificial intelligence algorithms and application. China Telecom has many partners and also has a certain accumulation in AI algorithm research, which is an essential foundation for algorithm study and innovation. China Telecom has a nationwide and globally accessible communication and information service network, and it also has data centres throughout the country and the largest cloud resource pools among all the operator worldwide, which is the infrastructure of universal computing to support the AI development.

In addition, the telecommunication network is also the largest market segment of the AI industry. China Telecom is both a user of AI technology and a provider of AI services. Relying on the rich Data Information Communication Technology (DICT) industry service experience and channel resources, China Telecom works with a large number of government and enterprise customers to promote the application of AI projects in various fields. It then creates an AI ecosystem, and cooperate with other enterprises on research, business models and services for a win-win situation.

As the main force behind building a powerful network, smart society and a digital China, China Telecom, as the planners in cloud computing, big data, artificial intelligence and the provider of network infrastructure, will accomplish new achievements in the modern era.

2.Business Imperative

In 2016, China Telecom officially released its transformation strategy (Transformation 3.0) and network architecture reconstruction plan (CTNet2025) to adapt to the global development trend of the intelligent era. By leveraging network intelligence, service ecology, and smart operation, China Telecom has expanded comprehensive intelligent information services to help boost national cyber development, leading the digital ecosystem, promoting service industry transformation and social governance innovation.

During the steady progress of network reconstruction and the 5G pilots, the traditional manual O&M mode cannot meet the requirements of future network operation any more. In early 2017, China Telecom, together with Huawei and other partners, set up the network AI industry-standard working group, Experiential Networked Intelligence (ENI) in the European Telecommunications Standards Institute (ETSI). This working group aims to jointly promote the formulation of research and standards on AI application scenarios, requirements, and system architecture in operators’ networks.

With the development of network AI research and practice, China Telecom released China Telecom AI Development White Paper 1.0 during MWC Shanghai 2019. This white paper clearly states that China Telecom will be directed by the Strategic Transformation 3.0 and based on CTNet2025 to deeply embed AI technology capabilities and provide universal AI capability platform, applications and solutions both internally and externally. In the initial phase, cloud-network convergence is used as the breakthrough point to provide public, government, and enterprise users with quickly provisioned, customised, automated, and multi-layer intelligent services. In the future, the smart, E2E DICT solutions and services will be provided. The following figure shows the overall layout of China Telecom’s AI development.

Figure 1: Overall layout of China Telecom’s AI deployment

3. China Telecom and the Solution

In the complex market environment of ever-changing business requirements and drive of digital transformation, manually performing tasks is both time-consuming and labour-intensive. The challenge is to rapidly resolve changes in demand while achieving the goal of digital transformation.

The intent-based network is an application layer that expresses in simple terms, what the network wants to accomplish in a natural language; the network can then convert the intent into specific policies and automatically complete cross-domain and cross-network configurations in complex and heterogeneous environments according to the corresponding policies. Thus, the business purpose of the application layer is achieved.

According to the network practice of China Telecom, the intent-based network can be regarded as an advanced network on demand. That is, the on-demand, self-service, and elastic network services evolve to an automated, closed-loop, and intent-driven network organisation. This is one of the basic features of the evolution from CTNet2025 1.0 to 2.0.

The construction of China Telecom’s intent-based network follows the overall layout of China Telecom’s AI development. The network consists of four parts: intelligent brain, orchestration and control layer, and intelligent infrastructure of the intent-based network, and AI terminals. The intelligent brain of the intent-based network includes the Big data lake and AI enablement platform in the overall layout as well as various AI capabilities developed based on them. The orchestration and control layer of the intent-based network is the critical channel hub for implementing smart operation based on the network infrastructure. The following figure shows the overall architecture.

Figure 2: Target architecture of China Telecom’s intent-based network

China Telecom positions itself as an AI network builder, an AI industry driver, an AI technology adopter, and an AI service provider. The comprehensive introduction and development of AI technologies enable China Telecom to:

  • Accelerate the upgrade of intelligent networks
  • Form an intelligent industry ecosystem
  • Improve the level of intelligent operations
  • Build a comprehensive intelligent intent-based network by using the practical requirements and specific scenarios of telecom networks and services as the breakthrough point
  • Provide get-as-you-wish services oriented to customer requirements

China Telecom has implemented some specific cases to improve the level of intelligent operations.

Case 1: Intelligent Operation Analysis Platform for Wireless Networks

As the radio environment, network structure, user behaviour, and user distribution continuously change, the network needs to optimise and monitored continually. AI can be applied to diversified network operation analysis to quickly and precisely locate exceptions, predict network trends, and provide rectification suggestions.

An intelligent wireless network operation analysis solution is provided for operators based on analysis of their requirements of the wireless network system. This solution targets the following aspects:

  • Analysing data problems at each layer during centralised collection and aggregation of NMS data over the entire network based on existing data and functions of the NMS for 4G wireless networks and big data analysis approach.
  • Building a diagnosis model for data integrity, rationality, and stability.
  • Selecting the optimal network indicators prediction model based on diagnosis results of historical wireless performance data. This model helps predict changes in wireless KPI data to learn the KPI trends of cells to some extent.
  • Creating an intelligent warning mechanism for the integrated NMS

The intelligent operation analysis platform for wireless networks based on AI and big data was piloted in China Telecom Fujian in 2018. The application effects of its functions are as follows:

Indicator Distribution Comparison

This solution can display indicator distribution (on different network management equipment in the same period and on the same network management equipment in different periods), visualise abnormal data changes of vendors or China Telecom Fujian, and ensure proper allocation of network resources and improves network usage.

Exception Diagnosis

Exceptions of cities, provinces, cells and performance indicators are displayed, and distribution of abnormal cells are marked on the grid-based map to develop related rectification solutions. For poor-quality cells, this solution can analyse key indicators that cause poor service quality, find out root causes, and take measures accordingly. This prevents the expansion of network incidents and ensures wireless network operation quality.

Trend Prediction

Trends of specific performance indicators or performance indicators of specific cells for the next week or month can be predicted based on historical KPI data trends and time sequence deep learning algorithms. The prediction accuracy is higher than 80%. In this way, future traffic distribution can be accurately predicted, and KPIs monitored to diagnose possible incidents and take preventive measures in time, lowering the network incident rate and enhancing network connection quality.

Capacity Expansion Prediction

This solution predicts cell capacity expansion for the next six months based on wireless network performance data, perception data, and configuration data together with supervised deep learning algorithms with a prediction accuracy rate of 99%. This solution can update the number of cells to be expanded, distribution of these cells, and capacity expansion of 5H1L cells in cities and provinces in real-time each day. In this way, management personnel can adjust the cell capacity expansion solutions and perform capacity expansion promptly to improve network service quality.

Case2: Intelligent Transport Network Slice Management

In the transport network, there is a tidal effect, and slices cannot be adjusted in time, resulting in redundancy and waste of resources. Therefore, it is necessary to accurately predict traffic conditions, dynamically configure slice resources to achieve intelligent transport network slice management and ensuring service quality.

The following figure shows the intelligent transport network slicing system. The Transport Network Slice Manager (TNSM) collects the traffic throughput data of transport network slice instances and sends the data to the Intelligence Module. The Intelligence Module predicts the traffic throughput values in the next several time segments, then determines the scaling and bandwidth adjustment policies for transport network slice instances based on the prediction result and delivers the policies to the TNSM if necessary. Finally, the TNSM implements the corresponding scaling up or down procedure by reconfiguring the port bandwidths.

For traffic throughput prediction, the prediction precision reaches 91.75%. When the alarming rate is acceptable, the resource efficiency of the test set can be improved by about 30% by using the intelligent policy based on traffic prediction.

Figure 3: System structure of intelligent transport network slice management

Case3: Intelligent system of work orders

The operator’s network generates many work orders every day, the network maintenance workload is large, and the maintenance operation method of work orders is passive and inefficient. AI and big data technologies are needed to achieve active and intelligent operation and maintenance, improve operation and maintenance efficiency, and reduce labour costs.

The solution is based on AI and big data technologies such as NLP to realise the automated and intelligent processing of work orders. Firstly, there is a need to collect data of work orders, which is cleaned and preprocessed before labelling and training. Secondly, the data of work orders is labelled by experienced operation and maintenance staffs. The labels are used to describe which class every work order belongs. Then, the classification model of work orders is trained, and corresponding classification analysis can be automatically implemented. The solution can finally achieve proactive, intelligent, and efficient maintenance operation of work orders.

The intelligent system of a work order has been deployed in the Group and four provinces, which over fulfils about 50 thousand work orders this year.

4. Economic Benefits

Intelligent network operation and maintenance based on AI and big data technology can reduce OPEX. The intelligent system of work orders uses AI enablement platform for training and inference, which can realise automatic classification of several kinds of work orders and automation of the entire procedure. Through pilot deployment for these kinds of work orders, the automation rate of the whole process reaches 80%, and the intelligent system of work orders is expected to reduce labour costs by 60%.

4.1      Assumptions

OPEX savings realised from reduced costs of network management (labour, licenses) and maintenance savings. The outcomes are based on a hypothetical MNO in country profile A-D, the saving potential within network and operations management of 80% and the saving potential within the maintenance of 9%.

 

5. Implementing the Solution

To achieve China Telecom’s intent-based network, Staged Evolutionary Routes include  Early Stage (2021), Middle Stage (2025) and Middle Stage (2025).

The key implementation measures include the following aspects:

  • Build a Universal AI Enablement Platform

China Telecom will build an open AI platform that is autonomous, controllable and full-stack. The platform is a common R&D platform integrated with AI models, computing clusters and other software and hardware resources. It offers multi-layered AI capabilities as a micro-service. Users can build their models with only a few commands; the model can then be trained, evaluated and deployed. With the AI platform, it will be easier to develop AI applications and improve research efficiency.

  • Establish “Big Data Lake”

China Telecom will optimise the construction of big data platform and establish “Big Data Lake” using integrative data storage, unified platform logic, physical distribution and centralised deployment. Various open-source components carried by big data clusters are utilised to provide rich processing capabilities, aggregate system data from multiple sources and in different formats, and perform unified and standardised data storage, processing and conversion processes. Such big data capability can be leveraged to support business applications.

  • Build Infrastructure of the Intent-Based Network

China Telecom will build a new generation of information infrastructure with “one network, one cloud and one platform” as its core, and provide an intelligent service cloud with “end-to-end network changing with the cloud”.

China Telecom will develop customised servers for artificial intelligence. Considering business scenarios such as cloud training, cloud inference, edge reasoning and so on, a heterogeneous computing system (such as “CPU + GPU”, “CPU + FPGA”, etc.) consisting of computing units with different instruction sets and frameworks is developed with partners from industry. Network devices need to adapt for the Development of Artificial Intelligence. Intelligent edge computing resources in the edge data centres will become an essential carrier for artificial intelligence business. China Telecom will consider the collaborative planning of the intensive cloud data centres and the widely distributed and close to the customer’s edge data centres so that the resource superiority of both data centres can be exploited in different business scenarios.

  • Construction of AI Validation Platform and Evaluation System

AI-related products, devices and solutions should have reliable and accurate validation and evaluation methods. In the field of network AI infrastructure, China Telecom will establish a standardised network AI validation platform and evaluation system to test, validate and evaluate network devices using AI technology. It can also support prototype development, equipment procurement and filed test.  In the field of AI terminals, AI terminal test labs should be built, and the corresponding evaluation methods should be developed for the evaluation of various AI terminals. In the field of AI products, the assessment criteria for various AI products and basic testbeds should be developed to support the test for all kinds of smart products and business platforms.

  • Building a Team of Compound Talents for the R&D and Application of Network AI

The research and development of network AI technology require a team of composite talent, precise requirements and use cases are provided by front-end and back-end experts to AI developers along with an interpretation of data/index and application suggestions. This information is also required by the Data and AI algorithm engineers for their data annotation, feature extraction, model tuning, and related R&D and deployment work so that the network AI applications can be finally delivered to users.

6. Challenges

Implementation of network intelligence applications at least include the following challenges:

Challenges on data and algorithms

Massive data is an important foundation for the development and application of network artificial intelligence. The research of network artificial intelligence applications based on machine learning model algorithms requires a large amount of actual network data for model training and optimisation. The industry is currently focusing on applications in the areas of image recognition and natural language processing and lacks AI algorithms and models suitable for the field of communication networks.

AI Enablement Platform of China Telecom aims to provide data capability, intelligent analysis framework, a library of models and algorithms.

Low standardisation of network data

There are various types of network equipment. The data provided by more than a thousand types of equipment from different manufacturers are different in terms of data formats, calculation methods, time granularity, etc.  At present, there are expected to be tens of thousands of network data types in the field of network artificial intelligence. The lack of standard specifications for these data makes data collection, storage and feature processing very difficult.

It is necessary to establish China Telecom’s entire network data governance system (“Big Data Lake”) and to promote the development of network data industry standards for all kinds of network equipment. Improving the feasibility of introducing artificial intelligence technology into the telecommunication network from the source.

Lack of compound talents of network and artificial intelligence

The research of network artificial intelligence application model and algorithm requires compound talents who understand both the network and artificial intelligence. Pure AI talents lack an in-depth understanding of telecom network architecture and operators’ needs, while existing network engineers are mostly based on traditional equipment, lacking the ability to research AI and experience in applying AI.

China Telecom will adopt a talent cultivation system combined with internal training and external recommendation methods so that a composite talent team can be formed to support China Telecom’s AI development strategy.

Challenges on Business models

In the 5G era, it is challenging to bring new growth by relying on traditional data services. It is also a difficulty for operators to develop business models through deep participation in industry convergence. With the help of data analysis and network intelligence, operators can reasonably use resources to provide 5G services with customised SLA for a broader industry, while the industry users create commercial value in emerging markets with the help of intelligent network services. However, the value of network artificial intelligence applications is still being evaluated, and new models of industry convergence need to be explored.

7. Conclusion

With China Telecom’s Strategic Transformation 3.0, Network Architecture Reconstruction CTNet2025, 5G and the advance of future network construction, more and more problems can no longer be solved by traditional manual methods. Artificial Intelligence has become an essential approach for network intelligence, operational intelligence, and business ecology.

The intent-based network of China Telecom is a comprehensive, cross-domain, cross-network, end-to-end high-level intelligent network architecture. Does it evolve from the on-demand, self-catering and elastic network to a closed-loop automated network driven by the business intent? which is one of the essential features of CTNet2025 from 1.0 to 2.0, and also the long-term goal and good vision of China Telecom’s future intelligent network.

The construction of the intent-based network is a long-term evolution under the guidance of a unified framework. China Telecom is eager to join upstream, and downstream partners from the industry to jointly promote the formulation of network artificial intelligence standards, to promote the development of AI industry, accelerate the upgrade of intelligent networks, enhance the level of intelligent operation, expand comprehensive intelligent information services, and to improve people’s livelihood. It will help build the country into a cyber power and give a substantial boost to the extensive digital transformation of the economy and society.