5G Operations Get Smarter When AI and Slicing Combine

Time:2019-07-30

Trends and Challenges of Digital Operations

4G changes the life, while 5G changes the society. 5G network needs to support ultra-large bandwidth, ultra-low latency and massive connection scenarios, and can serve a variety of vertical industry applications such as automatic driving, industrial control, smart grid, big video and AR/VR. 

Diverse services, flexible deployment requirements, and complex network topology pose great challenges to 5G operations. Traditional manual or semi-automatic operations can no longer meet the requirements. 

Artificial intelligence (AI) has inherent advantages in high computation data analysis, cross-domain feature mining, and dynamic strategy generation. The introduction of AI can further improve the efficiency of network deployment and operations, increase resource utilization, and reduce Opex.

Inevitability of Smart 5G Slice Operations

Network slicing is an important feature of 5G networks. Through flexible allocation of network resources and flexible combination of capabilities, multiple logical subnets with different features can be virtualized based on a physical network to meet customized requirements in different scenarios. In essence, network slice operations provide full lifecycle management for slice instances, including design, provisioning, SLA guarantee, and termination. Network slicing not only brings great flexibility, but also increases operational complexity. It is inevitable trend to enhance automatic slice management capability based on AI.

Key Technologies of AI-Based Smart Slice Operations

When AI is introduced, the slice management system automatically executes management policies output by the AI training platform and provides smart network sensing, modeling, provisioning, analysis, judgment, and prediction capabilities to achieve a perfect balance between slicing flexibility and management complexity.

Smart Slice Provisioning

The slice management system provides smart slice provisioning that involves service customization, network planning, model design, automatic deployment, and E2E service activation.

—Service customization: The system uses data collection and machine learning to conduct deep mining of service features and provide customized, secure isolation of private slicing networks. 

—Network planning: The system comprehensively analyzes available resources of the whole network, continuously trains the optimization algorithm by using AI technology, quickly transforms service requirements into network requirements, and effectively resolves the conflict between differentiated SLA and network construction costs.
—Model design: According to the analysis result of the AI training platform, the system smartly orchestrates and schedules virtual resources and automatically outputs templates related to slice lifecycle, policy rules, and slice optimization deployment. 
—Automatic deployment: Combined with automated integrated deployment tools and slicing models, the system automatically instantiates resources at all levels, smartly matches test scenarios and use cases, and automatically performs slice testing. The deployment cycle is shortened from weeks to days. 
—E2E service activation: According to the configuration template definition, the system automatically disassembles configuration parameters to each subnet, executes automatic calculation of parameters to form a batch script, and automatically activates services through the configuration channels.

Smart Slice SLA Guarantee
 

Network slice guarantee is essential to guarantee the SLA required by users. With the smart QoS service capability, the system can smartly analyze service requirements, network capabilities and user features, make multi-standard decisions, and introduce QoS monitoring feedback to form an SLA guarantee closed loop.

—QoS guarantee: The system collects massive traffic data (such as service types and time requirements), network data (such as the number of connections, load, speed, and delay), and user data (such as user levels, communication habits, time, and locations). Through smart analysis and judgment, the system evaluates current service experience in real time, and forms one or more optimal QoS parameter sets to make the best decision and control. 
—QoS differentiated services: Based on smart judgment of time, location, access service, user communication habits, user subscription requirements and network real-time load pressure, the system forms the best matched QoS control parameters to provide users with real-time differentiated services. 
—QoS prediction and early warning: Based on massive data collection, modeling, and analysis, the system implements QoS prediction and gives QoS early warning in extreme cases, providing reference for O&M guarantee actions such as early service termination and service change. For example, based on the neural network and linear regression algorithm, the system can predict the growth rate of the same period, analyze the peak or average traffic, and predict network congestion for operations involving dynamic scheduling and traffic acceleration. 

Smart Slice Closed-loop Operations

To efficiently manage network slices and reduce the complexity and costs in operations, the slice management system must provide smart slice closed-loop operations guarantee such as network self-sensing and self- adjustment (Fig. 1). 

 

At present, the network policy is still based on manual static configuration and ignores actual network conditions. After the AI technology is introduced, the system can smartly analyze and determine the traffic, congestion level, and load status based on the time, location, and mobility characteristics. Through the dynamic slice management policies output by the AI training platform, it also implements smart scheduling.  

Moreover, real-time or historical smart analysis provides reference data such as health score, abnormal detection and prediction, and root cause analysis. Based on such data, the system performs capacity optimization, configuration optimization, resource elasticity, and fault location operations to implement closed-loop slice optimization.

Smart Slice Fault Location

The system analyzes the time, location, and event description of slice alarms and identifies the alarm clue relations based on historical frequency data, cross-NE data, intra- or cross-private network data, and service-related data. It also reasons about the relations based on the current alarms, statistics, logs, and the rules obtained through training, and obtains the matched alarm root causes. Smart slice fault location includes training, reasoning, and closed-loop optimization. 

The training process includes data extraction, data cleaning, data formatting and segmentation, algorithm execution, and results analysis. The reasoning process contains real-time monitoring of alarms, scheduled sampling of resources, and data configuration. The system uses the learned rules to comprehensively determine alarm data, resource data, service bearer relationship, and time sequence of the existing network and to find out the root cause for automatic repair or prompting the operation and maintenance personnel to repair the fault. The closed-loop optimization involves updating, modifying, and improving the rule library in accordance with the actual rule application or expert judgment.

The effect of smart fault location is measured by the number of effective alarm root cause rules and the alarm compression ratio, or evaluated indirectly through the reduction rate of the number of work orders. AI-based smart alarm location can reduce the number of work orders by over 60%. 

5G smart slicing network will experience three phases: intra-domain exploration, cross-domain integration, and high autonomy. First, sub-domains of the 5G network shall be integrated with AI to provide preliminary intelligence in network resource allocation based on big data and machine learning. Second, with the development of technologies, AI will be able to learn big data of the 5G network across domains, and integrated intelligence will emerge in some sub-domains to achieve intermediate intelligence. Finally, with the rapid development of 5G and AI technologies, network-wide coordination and high autonomy will be realized. This will greatly improve the efficiency of full-lifecycle network management and achieve advanced intelligence based on the intentions of human control networks. 

It is foreseeable that the combination of AI and 5G slicing network will produce dazzling sparks and promote the rapid development and evolution of networks.