In the 5G era, a large amount of industrial application requirements have brought about rapid growth of network scale and service capacity, resulting in a great complexity of network architecture. Meanwhile, network users' expectations for service delivery quality and efficiency increase year by year. These pose new challenges to the transport network. The key to 5G transport network construction is to build a network efficiently and conveniently, release services quickly, perceive the network status in real time, perform service self-healing, fast fault diagnosis, traffic prediction and optimization, and make the system open and reliable. With the introduction of technologies like AI, big data, cloud computing into the telecom industry, it has been agreed that intelligentization is the inevitable road for 5G transport.
Key Technologies for Intelligent 5G Transport
Intelligent 5G transport requires many new technologies, among which the key technologies include intent-based network, intelligent control, machine learning, knowledge graph, network-based mirroring, and cloud native.
Intent-based network (IBN) matches business intent based on an awareness of the "holographic state" of the network and with the help of AI technology. Compared with the traditional networks that are managed manually, the IBN focuses on business intent. It automates network operations based on the intent, verifies in real time whether the desired intent is executed, and makes continuous adjustments to form a closed-loop O&M control system.
In the 5G era, intelligent transport network control technologies mainly include SR, centralized control, slicing and telemetry. Segment routing (SR) simplifies the control protocol and is more conducive to end-to-end network programming. The centralized control technology can uniformly control the resources of the entire network to guarantee the routing and SLA requirements for different industries. Network slicing provides multiple logical networks on top of a common physical network infrastructure to meet differentiated requirements of different industries or scenarios. Telemetry actively sends data at fine-grained levels for real-time feedback within seconds or even milliseconds, providing accurate data support for network control.
Machine learning, an important field of AI, is crucial to 5G transport. Based on the big data of the 5G transport network, it can be used for model training, and plays a key role in network traffic analysis, traffic prediction, exception analysis, network simulation, intent identification and fault diagnosis. It uses different technologies and algorithms in the above application scenarios, such as classification clustering, random forest, Bayesian network, and intensive learning.
Knowledge graph is a multi-relational graph that contains multiple types of nodes and edges. Its application in 5G transport is to use graphs to connect discrete data. The edges in the graphs have their own semantic attributes. The algorithms based on the graph theory can efficiently perform data search and reasoning, and provide valuable analysis results and decision-making support. Typical applications include network configuration check, failure analysis and diagnosis. The knowledge graph can continuously improve and optimize the knowledge logic and models through manual accumulation and machine learning, so that network intelligence can form an improvement closed-loop.
Network-based mirroring is to synchronously simulate a real network and capture comprehensive, real-time data of the network that covers topology, traffic, services and protocols. The data involves configuration data, status data, and real-time data, and includes not only current network data, but also historical and future data. It provides powerful data support for machine learning, knowledge graph, intelligent control and IBN application as well as various network services.
In the 5G era, the ability to develop new services, apply to new scenarios, and quickly respond to new requirements are becoming increasingly important. The cloud-native architecture uses PaaS to build an open service platform, adopts the app model to establish an open application model, provides open interfaces, and implements continuous integration and delivery through DevOps, thereby obtaining open capabilities that are efficient, elastic, and secure. Capability openness connects product R&D, quality assurance and network O&M processes to enable shorter service release cycles, faster time-to-market, less invasive upgrades, and higher delivery quality. The cloud native architecture provides a basic platform for intelligent 5G transport.
ZTE's Intelligent 5G Transport
Based on its great strength in intelligent technology, ZTE strives to build the highlands of intelligent 5G transport to accelerate 5G network construction and simplify O&M complexity.
Here are some typical practices that operators are most concerned about.
ZTE's Intelligent Management and Control Product
Based on a cloud-native microservice architecture, ZTE uses a self-developed ICT-PAAS platform and an app framework to support third-party app development and OpenAPI so as to develop its intelligent management and control product ZENIC ONE. The ZENIC ONE is composed of rich services/microservices. They are organized in accordance with the closed-loop lifecycle of the entire network, and are deployed or extended based on service scenarios. They also support flexible node & container expansion and unlimited network management scale. The ZENIC ONE includes an AI platform and a BigData platform to provide powerful data analysis and intelligent processing capabilities. In 2019, the number of equivalent NEs managed by ZENIC ONE reached 300,000+ as compared to 30,000 by the traditional OMC. The product has been successfully deployed on the cloud platforms in many provinces in China (e.g. China Mobile Guangdong).
Automatic Intent-Based Service Provisioning
The ZENIC ONE includes an intent engine, an automation engine, and a perception engine. The intent engine consists of three components: intent translation, intent perception, and intent assurance. Intent-based service provisioning mainly involves intent translation. ZTE's intent-based automatic service provisioning is simple and fast. The user only needs to select a service scenario, and the system will perform an intelligent analysis and display only the information that must be entered for this scenario. Other default recommended data is provided by the intent engine based on the scenario and self-learning. Then the system automatically provides multiple service solutions that comply with the user's intent. After the user selects a solution, the system converts it into the device configuration information according to the internal processing flow and sends it to the relevant device through the automation engine, thereby completing the service provisioning. The intent-based service provisioning have been applied to A1's existing network in Belarus, increasing service provisioning efficiency by 80%.
Optical Layer Adaptive Control
The optical layer adaptive control is mainly to enable the system to automatically deal with optical network emergencies without the need for human intervention so as to maintain the stability of the customer's service. ZTE's optical layer adaptive control function can optimize optical power, compensate for optical damage and obtain better OSNR performance for the optical link through machine learning algorithms without changing the rate, the spectral interval, and the modulation mode, so that the optical link can obtain better OSNR performance. In the case of limited resources, machine learning algorithms can be used for flexible spectrum modulation conversion with rate control to increase the room for path selection. In the event of a failure, the function can make full use of current physical resources to ensure customer service connectivity, providing the most reliable service guarantee. It has been tested and verified by multiple operators.
Network simulation can identify potential hidden dangers or bottlenecks in a timely manner without affecting the operation of the existing network, improving the quality and efficiency of network planning and O&M. ZTE's network simulation function relies on network-based mirroring to simulate the internal and external environment changes that may occur in a real network and trigger the execution of specific network simulation behaviors (e.g. fault simulation, traffic simulation, quality simulation, and protocol simulation). In addition, the system can simulate one or two faults to evaluate the overall anti-attack capability of the network, study the network robustness and provide quantitative results to guide subsequent network planning, adjustments, and hidden trouble elimination. Take traffic simulation for example. Capitalizing on network- based mirroring service that provides the network topology, protocols, and historical data and change information of traffic, the traffic simulation and prediction algorithms will be invoked according to the user's expected traffic increase. The traffic simulation results are presented to the user in the form of a traffic topology diagram that identifies network bottlenecks and offers suggestions for network optimization or expansion. These simulation behaviors have been tested and verified by multiple operators.
Configuration check is to identify network configuration anomalies and potential risks quickly and automatically, improving the efficiency of network O&M. ZTE's configuration check function uses the knowledge graph technology to extract the configuration structure feature of the device from the existing network to generate the device's role fingerprint. Meanwhile, it completes network sub-graph mining, learns semantic rules through statistical/connection approaches, performs analysis in view of graph neural networks, and finally scans device configuration based on device roles and NLP technology to identify abnormal configurations and potential risks. This function won the Best Network Intelligence Award at the Broadband World Forum in 2019. It has now been deployed on the entire IPRAN network of China Unicom Guangdong, and played an important role in network assurance during the Spring Festival 2020.
Fault diagnosis is a high-frequency operation of network O&M that determines the network quality to a certain extent. ZTE's fault diagnosis function uses the fault relational dependency graph based on the knowledge graph technology to automatically diagnose faults of various networks and service objects, and uses the Bayesian network-based fault propagation graph to improve the probability analysis of suspected root causes, making fault location more accurate. The function has been verified on existing networks of China Mobile Shenzhen with diagnostic efficiency increased by 70%.
With intelligent 5G transport gradually becoming a reality, ZTE will invest heavily in intelligentization and work closely with operators and industry partners to apply the ZENIC ONE to more commercial networks and address the needs of industries and niche markets more flexibly and quickly.