This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Read our privacy policy (update in May 2018) >

AI Enables Telecom Networks: Autonomous Network and Service 2.0

At the 2018 Huawei Global Analyst Summit, Huawei launched the SoftCOM AI solution architecture. Huawei introduced AI technology based on the all-cloud network architecture, aiming to build a "never faulty" autonomous network that features automated driving and trigger a new round of network transformation.

Cross-industry competition calls for system architecture innovation

We are all saying that the world is entering a new era. Different people will endow the new era with different characteristics. "Cross-industry" may be a common characteristic.  Each industry faces the structural challenges brought by cross-industry competition, and there is no exception in the telecom industry.

First, from the perspective of revenue structure, operators' businesses are confronted with the challenges of cloud services from the IT industry and Internet industry. Telecom businesses are divided into three layers: terminals, networks, and IT infrastructure and upper-layer applications. With the rapid increase in network rates, the business model of the IT industry shifts the focus from selling products to selling services, that is, the cloud service model is becoming the main business model of IT. Cloud service providers break the boundaries of the telecom industry, IT industry, and Internet industry. Backbone networks, and even some metropolitan area networks, IT infrastructure, and IT applications become part of cloud services. Therefore, cross-industry competition has begun. If operators can do well in cloud services, they can compete with cloud service giants, such as AWS, to seize trillion-dollar market shares. Otherwise, operators may also lose many traditional telecom businesses, especially private line services between data centers.

Second, operators' efficiency and costs are also facing structural challenges. The OPEX of telecom equipment maintenance is about three times the CAPEX. Telecom networks become more and more complex, which is beyond the professional knowledge and capability of one person. As a result, human factors cause 70% of major network faults. Qing Wu, chief wireless architect of Canada's operator TELUS, said that "The telecom industry uses the 4G networks of the twenty-first century. However, the network operation and O&M are still in the eighteenth century, mechanical manufacturing moves towards automation, but the telecom industry is still in the handicraft phase."

Product innovation is far from enough to address the challenges faced by the telecom industry. The entire system architecture and business model must be innovated to improve operator competitiveness and resolve structural problems. What is system architecture innovation? Cloud computing is used as an example. System architecture innovation is not an innovation of a server or storage product. Instead, it uses a new distributed system to improve the resource utilization efficiency and is system-level innovation. Product innovation, system architecture innovation, and business model innovation support and promote each other.

To meet customer requirements in the new era, Huawei's innovation system is designed based on the three types of innovation. In the aspect of product innovation, Huawei's guiding ideology for designing network devices is "Olympic spirit", that is, large capacity and low latency. All product innovation focuses on this goal. In the aspect of system architecture innovation, Huawei aims to build an agile, automated, and intelligent network to achieve the "automated driving mode" of the network. There are two goals for business model innovation: One is that the cloud provided by Huawei+operator becomes one of the top five clouds in the world, and the other is to build an online intelligent service model in the network era.

SoftCOM AI brings brand-new value

Let us review the development path of Huawei network architecture. In 2006, we proposed the Single strategy based on the determination of the all IP technology direction. That is, the network architecture evolves from multiple networks using diverse technologies, such as TDM, ATM, FR, DDN, and IP, to a single network based on the IP system. In 2012, we introduced the concept and technology of cloud computing based on the all-IP network architecture. We proposed SoftCOM with the goal of the all cloud architecture to provide data center-centric agile networks. In 2017, with the development of AI technology, we introduced AI into telecom networks. SoftCOM AI was born to provide autonomous networks at the network architecture level and service 2.0 at the business model level.

An autonomous network with AI is introduced to construct "industry 4.0" in the network field and achieve the "automated driving mode" of the network. Industry 4.0 has three characteristics, that is, agile equipment, intelligent control, and intelligent analysis systems, to support product automation. The three characteristics are also suitable for the telecom industry. On a telecom network, the lower layer is network equipment, and the upper layer is a control layer. In terms of control and O&M of the entire network, AI technology is introduced in an end-to-end manner to implement the segmented autonomy function. Each segment of autonomy provides end-to-end autonomous capabilities based on an upper-layer operation system. In this way, the entire network can be autonomous. The biggest change brought by an autonomous network is that O&M personnel are not in the service process. The autonomous network is an automated system. We call this mode the "network automated driving mode" to provide automation, self-optimization, and self-healing of the entire network.

The goal of service 2.0 is to build an "industrial Internet" in the network field and provide online digital "intelligent services". The development of the industrial Internet has dramatically changed the business model of industrial giants. Boeing is used as an example. After selling airplanes to airlines, Boeing continues providing all-around digital services, including predictive maintenance, fuel management, and flight management, for airlines. This service concept is extended to the telecom industry. Future networks are automatically running on the operator side, and Huawei provides automated online services based on AI in the background. This service is based on the AI model with continuous iteration. "Model as a service" is built based on industry practices and is always in the Beta phase. It is constantly updated and improved.

The new value of introducing AI into telecom networks is "predictability". The management and control center of a telecom network uses certain policies and rules to manage and schedule the entire network based on the southbound interface and data collection of equipment, which is based on three conditions: network reachability, SLA requirements, and resource utilization efficiency. The three conditions are the basis of network automation. However, as networks become increasingly complex, these conditions are far from enough. Online AI inference and data analysis also need to be introduced to provide traffic prediction, quality prediction, and fault prediction. Prediction is the core value of AI. Based on predicted future conditions, network resources are scheduled, faults are prevented, quality is optimized before quality is deteriorated, and traffic is adjusted before network congestion occurs. This helps provide "never faulty" automated driving networks that have features of automation, self-optimization, self-healing, and autonomy, structurally improving the O&M and operation efficiency.

AI helps improve user experience and provide three multiplications

Automated driving of networks is a long-term process and cannot be accomplished at one stroke. According to the five development steps of automated driving of vehicles, we also divide automated driving of networks into five stages. In the first stage, the AI points out "what happened". In the second stage, "why happened" needs to be determined. In the third phase, we predict "what will happen", and make decisions and take measures. In the fourth phase, AI determines "the measures to be taken", and then operations are manually performed. In the fifth phase, self-control and automated recovery of networks are provided so that the networks have the self-healing capability.

Autonomous network and service 2.0 will bring end users minute-level ROADS experience, the best network connections, and the availability of uninterrupted networks, and will bring the multiplications of O&M efficiency, resource utilization efficiency, and energy consumption efficiency for operators.

O&M efficiency multiplication: The O&M level is divided into three phases. The first phase is called Run-to-Failure (R2F). If there is a fault on a network, the O&M personnel immediately go to the site and rectify the fault. This is the lowest level; the second phase is called Preventive Maintenance (PvM). With PvM, each device is checked to prevent faults, but the efficiency is low; the third phase is called Predictable Maintenance (PdM). With PdM, we can predict the probability that a device will become faulty in the future and then perform targeted maintenance. With PdM, we hope to reduce 90% of alarms reported and faults on telecom networks, predict the failures and deterioration of 90% of key components, and further achieve network self-healing. In addition, more than 70% of the problems in network faults lie in passive equipment, such as optical fiber bending, component aging, and port loosening. In all of these situations, signals change. AI is introduced to learn the characteristics of these changes, and it is possible to predict the changes in advance and use solutions to solve passive faults.

Resource utilization efficiency multiplication: The current characteristic is that the traffic flows accordingly after the network construction is complete and the resource utilization may be unreasonable. If networks are scheduled based on traffic directions, the resource utilization efficiency will be greatly improved. However, the current networks do not have such a capability. To address this issue, AI must be introduced, and a traffic prediction model is created. This way, precise traffic prediction and the most reasonable network topology can be provided, and the network paths can be determined by the traffic directions instead of physical connections.

Energy consumption efficiency multiplication: We need to achieve the goal of "bits determining watts". That is, the network traffic volume determines energy consumption. In an equipment room or at a site, each system is configured with dozens of parameters. The heat dissipation, environment, and service load models are generated through AI training to maximize the energy consumption efficiency for sunlight, temperature, and auxiliary facilities, such as diesel generators, solar energy devices, and batteries. At the equipment layer, dynamic energy distribution is performed based on service loads. If there is no traffic, power consumption is reduced by using timeslot shutdown, RF deep sleep, and carrier shutdown. In addition, energy consumption of data center objects, such as server components, can be dynamically reduced. The third is the network system. The accurate service load prediction model is constructed to balance the traffic on the entire network and achieve the optimal energy consumption efficiency.

SoftCOM+AI is the target architecture of an autonomous network for Huawei. The specific practice is as follows: Introduce AI technology and capabilities into the three network layers: equipment and cloud infrastructure, network management and control center, and network O&M system. This is to provide end-to-end intelligent and automated network planning, deployment, running, maintenance, optimization, and operation, and enable networks to achieve optimal system performance. Huawei has built an AI training platform for operators. This platform connects the running status data of network devices to train the AI model, and continuously updates and optimizes the model to improve network system automation.

How to build SoftCOM AI? Let's take the optical network as an example to show how AI enables full-process service development. The first is a data base, that is, what data needs to be obtained. Specifically, for the optical network, the data includes optical fiber data, optical signal data, and optical route data; the next is about enabling technology, that is, AI algorithms, including data cleansing, information reorganization, machine learning modeling, and deep learning. These algorithms are irrelevant to the optical network. To provide automated driving of the optical network, a large number of models, such as the fiber model and filter model, need to be constructed; the last is about service application scenarios. The scenarios include automated optical fiber check during site deployment, service provisioning, network optimization, fault locating, and automated resource scheduling. The optimal method can be found through the model to provide fast provisioning, simplified O&M, intelligent operation, intelligent improvement of network scheduling efficiency, zero waiting, zero contact, and zero experience, letting people be not aware of the existence of the network.

The future will be an intelligent era. Operators' network intelligence is a long-term practice and cannot be accomplished at one stroke. SoftCOM AI is the implementation of Huawei's All Intelligence strategy in the telecom field. The most key AI capability of SoftCOM AI is developed based on both Huawei's long-term strategic investment and growth in All Intelligence and telecom scenarios. SoftCOM AI helps operators build "never faulty" autonomous networks and achieve digital and intelligent transformation.