ZTE's self-evolving network promotes intelligent network

As the world's leading equipment supplier, ZTE released the "5G Network Intelligence White Paper" in August 2018, and proposed the combination of AI and 5G as the entry point for 5G network deployment. In June 2019, during the MWC Shanghai Exhibition, ZTE first proposed the “Self-Evolutionary Network” solution, which enabled the AI ​​to realize the intelligentization of the 5G network and the full evolution of the enabling network.

ZTE's self-evolving network architecture

From the perspective of architecture, ZTE said that the self-evolving network solution adopts the principle of layered closed-loop, and builds an intelligent network system of network element level, single domain level and cross-domain level; modular design of AI capability, embedded in the network on demand Hierarchy, including network element layer, management layer, operation layer, self-evolutionary network with continuous evolution of construction capability and continuous value superposition, realizes network operation and maintenance through network evolution, operation and maintenance evolution, and operational evolution. Further increase revenue and reduce revenue and improve efficiency.

In terms of application scenarios, the autonomous evolution network can provide operators with a series of network AI solutions covering the whole process of network planning, construction, operation and maintenance, optimization, and operation. ZTE has practiced and applied in multiple scenarios, such as Intelligent energy-saving, AI-based Massive MIMO auto-tuning, AI-based mobile load balancing, and intelligent root cause positioning. Future self-evolving network solutions will enable the network to achieve autonomous operation and maintenance, self-recovery, and self-optimization without manual operations.

The solution's AI engine is ZTE's artificial intelligence platform uSmartInsight, which provides operators with intelligent engines that can be deployed at various levels and in different locations of the network to realize ubiquitous intelligent identification, analysis, prediction and decision-making capabilities for the network. A training platform for large-scale data processing and model training capabilities, and open to third parties.

When is network intelligence implemented?
Industry-wide promotion of hierarchical evolution

When does the network intelligence come? It is estimated that this is a problem that every communicant is very concerned about.

Zoro, the director of ZTE's system planning department, said in an interview that the current level of intelligence in the network is roughly between L1 and L2, most of the scenes are in L1, and some scenes have L2 levels. Different network architectures also restrict the level of intelligence. Compared with traditional networks, cloud networks are easier to achieve high-level intelligence. Traditional networks can only be enhanced by local intelligence, and can only reach L3 level at most. He suggested that the 5G new network can start construction with reference to the L2 level and gradually evolve to L3/L4.

What is the concept of L1/L2/L3/L4? This should refer to the grading standard of network intelligence.

Key features of each stage of network intelligent grading

As we all know, autonomous vehicles have five levels of standards, from L1 to L5 are assisted driving, partial automatic driving, conditional automatic driving, highly automatic driving and fully automatic driving. The self-driving car evolves step by step according to this level of standard, gradually achieving the liberation of the driver's hands to liberate the eyes and then to the driverless, and is now heading towards L3 and L4.

Similar to autonomous vehicles, the autonomous evolution network is also divided into five levels. From L1 to L5, it is auxiliary operation, primary intelligence, intermediate intelligence, advanced intelligence and complete intelligence to gradually liberate manpower by introducing AI.

It is very important to define a unified level standard. On the one hand, it can form a unified understanding and understanding in the industry, and on the other hand, it helps the participants in the industry to provide a reference basis for technology introduction and product planning.

After defining a uniform level standard, the next step is the scenario test.

The key to driving a self-driving car is to let the vehicle cope with environmental changes. This environment is often called a “scene”. The combination of different driving behaviors, driving conditions and environment constitutes a “scene library” for autonomous driving. The scene library is equivalent to the "question bank" of the self-driving car test evaluation. Automated driving cars evolved step by step through continuous scene testing.

Like autonomous vehicles, the autonomous evolution network will gradually be practiced and applied in multiple scenarios under different workflows such as network planning and design, deployment, operation and maintenance optimization, and business operations, while from the network element -> subnet -> Cross-domain->The whole network is gradually expanded, and finally realizes end-to-end, full-process network intelligence and automation.

Network intelligent scene-by-scene evolution

To this end, ZTE has carried out various pilot and verification work on a number of typical scenarios with many operators at home and abroad, and has achieved a series of practical results.

For example, cooperate with China Telecom to carry out 5G smart operation projects, and carry out pilot projects in Jiangsu, Fujian and Guangdong provinces around AI value scenarios such as 5G intelligent planning, 5G base station and core network root cause analysis; and China Mobile based O-RAN The architecture of AI enhanced load balancing joint verification; and China Unicom cooperated in Shandong to set up an intelligent network joint innovation laboratory to carry out 5G operation and maintenance intelligent exploration and application.

Yes, with the continuous advancement of 5G and AI technologies, network intelligence is coming to us step by step.

But the industry is equally clear-minded that an all-intelligent network cannot be achieved overnight. Like a self-driving car, the road ahead is wide but not flat.

Zuo Luo stressed that the self-evolving network mainly faces three challenges.

One is the technical challenge. There is still a lack of high-quality data sets, which leads to more model problems in training. It takes a lot of time to identify the correct model and train it. In addition, the lack of transparency in the AI ​​system, the use of AI implementation requires a large amount of network resources, and the interface specifications of the AI ​​engine and the existing network management/network element have not been standardized.

Second, personnel challenges, operators need a large number of business skills, technical skills and analytical skills.

The third is operational challenges, solving the AI ​​use cases, the complexity of deploying AI processes, and the legal and ethical issues of AI in network operations.

To this end, ZTE said that building a self-evolving network with progressive evolution and continuous value superposition requires network equipment manufacturers, chips, operators, etc. to participate, from point to line to face, gradually develop independent evolution at all levels of the network. Network practice and research. On the evolution path, it is also necessary for the standardization organization to play an active role in sharing and co-constructing models, algorithms, and platforms to realize the network goal of independent evolution at an early date.