The Road to 5G Network Intelligence

Release Date:2018-11-30 By Wu Jiangtao Click:

 

Trends of 5G and Network Intelligence

In June 2018, 3GPP approved the freezing of the standalone (SA) 5G new radio (NR) network functions, which marks the completion of 5G Phase-I standardization task. The 5G industry has entered its last stage. 
5G brings vast industry ecosystems and innovative applications, but it also presents unprecedented challenges for network operations. Cloud-based infrastructure, full software-based application systems, and massive data are the best soil for developing artificial intelligence (AI) applications. It is commonly agreed in the industry that AI is essential to building 5G network competitiveness. In 2017, 3GPP introduced the network data analytics function (NWDAF) that was expect to become the AI engine for network functions, and ETSI established the zero-touch network and service management industry specification group (ZSM ISG) aiming at achieving automatic and intelligent network operations. Meanwhile, globally-leading network operators and equipment vendors have strengthened cooperation in the network realm by using AI. The whole industry is using the AI technology to realize 5G network intelligence.  

Progress of ZTE’s Network Intelligence 

As a globally-leading telecom equipment and service provider, ZTE has maintained high investment and played a leading role in the 5G field. It will proactively integrate AI with the cloud technology to explore the road to intelligent operations in a 5G network.  
The target of network intelligence is to implement a closed-loop self-organizing, self-healing, self-optimizing network that can significantly improve network operational efficiency (Fig. 1). The intelligent closed-loop network involves two aspects. From the service perspective, it is necessary to implement a big closed-loop self-organization of network services that cover end-to-end sub-slices, bearer network, and cloud-based infrastructure, such as slice self-configuration and self-optimization. From the local function perspective, it is necessary to implement a self-closed loop of network functions in certain scenarios requiring intelligence, such as network function self-healing. The big closed-loop has the AI-related data and computing capability that can comprehensively consider the priority according to the value sequence and technology maturity of application scenarios. The self-closed loop of network functions, which is limited by factors such as computing capability, will be gradually implemented along with the improvement of equipment performance. The application mode lays emphasis on using existing models and knowledge for reasoning. 

 

 

Network intelligence needs to be promoted and implemented by three major forces. The first is the programmable AI platform, which is the accelerator for network intelligence. The second is the DevOps automation in the cloud network, which is the cornerstone of network intelligence. The third is the specific application scenario, which is the implementation tool to generate value. ZTE has worked with cooperative partners to have a good start for network intelligence.

AI Platform: The Accelerator for Network Intelligence 

ZTE’s self-developed AI platform supports visual programming mode, traditional machine learning algorithms, and deep learning. The platform also supports GPU cluster for high-speed parallel model training. The platform can significantly reduce the threshold of AI application development, helping the developers quickly find an appropriate algorithm. As a result, the developers can focus their attention on application logic, and the development efficiency can be multiplied by several times.  
The service-based architecture (SBA) of the platform makes it easy to rapidly integrate AI applications into a product, so that the product can be intelligent in a quick manner. Based on its self-developed AI platform, ZTE has developed intelligent network products in several sectors covering wireless, bearer, and cloud core network. 

DevOps Automation: The Cornerstone of Network Intelligence   

Abundant 5G services are based on a cloud network. The process from service development (Dev) to service deployment and operations (Ops) can be iterated continuously in the cloud network. DevOps automation is the basis for intelligent networks. Only when the whole service development and operations (Dev/Ops) process is automated, the result or action from the intelligent analysis can be implemented.  
ZTE implements the whole DevOps automation process for 5G network covering network design, deployment and operations. In the design phase, a sandbox system provides the test capability that can realize automation from development to test release. In the installation and deployment phase, DevOps automation allows for automatic installation and deployment from the network function virtual infrastructure (NFVI) to upper services. The whole deployment can be orchestrated flexibly as required. In the operations phase, DevOps automation provides one-key upgrade and gray upgrade, and supports the strategy-based automatic closed-loop assurance. The automation capability is decoupled with the service scenario, and the automation capability is orchestrated as required by the scenario. 

Network Intelligence Implementation Scenario 

 
The expectations from the industry on the application scenarios based on the network intelligence are mainly reflected in fault location, slice management, performance optimization, and user experience. From the perspective of network operations, network stability is the cornerstone, and rapidly locating network faults is an optimal scenario to realize the value of network intelligence. With self-developed AI platform, ZTE can offer intelligent fault location.
A common scheme for fault location involves analyzing network performance, alarms and logs, and locating the fault root cause. With its rich experience in network operations, ZTE implements multiple intelligent fault location functions such as network performance anomaly detection, alarm root cause analysis, and intelligent log analysis. The performance anomaly detection is made in the cloud management platform and network system to find out the system anomaly in advance and eliminate it. The alarm root cause analysis can help predict rapidly the root cause of the fault, and the prediction efficiency can be promoted by 70%. The intelligent log analysis can help locate a problem precisely, find out the anomaly, and set the alarm. With the applications and feedbacks of intelligent fault location functions, the scheme for fault location is continually developed and optimized. In a 5G network, fault location will surely become more intelligent.

Prospects of Network Intelligence  

The evolution of 5G network intelligence is a long-term systematic project. As 5G speeds up its commercialization process, 5G applications are gradually booming in various vertical industries. The assistance of AI in 5G networks will be quite imaginative and creative. 
As the NWDAF specifications are completed, an intelligent closed-loop of the network function hierarchy will be realized: NWDAF can be used for smart choices of slices as well as real-time QoS management and optimization. Moreover, it is possible to implement an intelligent closed-loop of higher network function hierarchy that involves analyzing customer intent and automatically creating and optimizing a network. For example, you can just speak a few words to an interactive terminal about when, where, and how to host a sports event, the network will automatically create and activate all the eMBB slices necessary for the match. The wonderful 5G intelligence network will enjoy a promising prospect.