We’ve seen operators worldwide suffer outages that, when they’re related to the core network, bring down services for everyone. When these large-scale outages happen, the providers with modern AI tools are best placed to identify the problems and resolve them quickly. Digital twins also enable very quick deployment of temporary solutions to maintain service in the interim while the main fault is fixed.
To support a wide range of new services, operators are increasingly taking a service-centric approach to operations alongside their historical network-centric view. As part of the reshaping of operations and maintenance, AI is providing “digital employees” to augment human teams. They help reduce end-to-end processes and enable new steps to proactively anticipate reliability issues and avoid faults, and hastening the shift to intent-based networks, which collate new data that AI tools can analyse to optimize the network.
Network Automation Levels Echo Autonomous Driving, But Telecom Is Ahead
Greater automation is now the goal of providers globally, and increasingly AI and digital twins are the main enablers of greater network automation. There are clear parallels with self-driving car models in the autonomous network levels described by TM Forum, an association for service providers and their suppliers. Self-driving cars and network automation both make use of AI.
However, leading communication service providers are arguably closer to having a fully autonomous network — classified as level 5 — than car-makers are to a fully self-driving vehicle in any conditions. Operators like China Mobile, China Telecom, China Unicom, MTN and Orange are aiming to reach level 4 by 2025, at least for some processes, and are currently deploying the necessary network tools to make that leap. Level 4 refers to a highly autonomous network where, in a complex cross-domain environment, the system enables decision-making based on predictive analytics or active closed-loop management. It uses AI modelling and continuous learning to manage service-driven and customer experience-driven networks.
Over 10 operators are members of the TM Forum’s autonomous network levels pilot study. The operator groups taking part include AIS, Antel, Elisa, Globe Telecom, Orange, MTN, Telkomsel, Telefonica, STC and Telecom Argentina. The pilot aims to assess the value of comparing autonomous network levels among operators, develop a framework for such comparisons and the requirements for autonomous network evaluation tools. Phase 2 findings will be published at Innovate Asia in November 2024.
The autonomous network level assessment goes deep into the capability at an operation task level. The goal is to help operators identify weaknesses in automation capabilities in network operations and maintenance. This assessment includes fault management scenarios and network architecture planning to understand and improve network resilience. Using a predictive and preventive maintenance approach increases the likelihood of delivering a trouble-free experience. Also, it better enables networks to cope with various challenges and changes intelligently, reduces the probability of faults and improves network stability and reliability.
We expect that many types of AI will be needed to ensure telecom services continue to operate. For example, discriminative AI will help analyse network patterns and enable fast categorization of known trends. In suitable tasks, discriminative AI avoids the risks of “hallucination” that generative AI sometimes suffers from. But generative AI can tackle more complicated tasks and work with more-flexible inputs, such as preventing signalling storms through scenario analysis. Additionally, it can offer natural language tools as an interface or to distil and communicate key parts of documentation.
New Services Like Slicing and Network APIs Increase Management Complexity
Services based on network slicing will become more common with 5G-Advanced. This increases the complexity and range of services that the core network must manage. Network slices will also be sold with more-demanding expectations of reliability and quality. This means that any instability in the core network will cause greater damage to the operator’s business than before slicing-based services.
The Open Gateway initiative and the associated network APIs will enable new services for third parties. Again, this increases the number of services an operator will offer and complicates the impact of any outages. Digital twins will help operators assess the impact of network changes for this expanded group of users — the developers working with exposed network APIs as well as traditional network users.
Service Providers Will Need Both Generative AI and Discriminative AI Tools
This year there has been enormous hype for generative AI products like large language models and image and video creation tools. Generative AI is incredibly valuable for core network information to distil specialist supplier knowledge, provide natural language interfaces and conduct simulations of the impact of outages using digital twins to improve responses to core network problems such as signalling storms. However, generative AI isn’t the only type of AI that matters to service providers.
Analysing patterns in network data and classifying them is also important, but it isn’t what generative AI is designed for. Instead, discriminative AI is the best tool for this task. One of the most widely used applications of discriminative AI is image analysis. On smartphones, this feature organizes photo libraries into categories like people, landscapes, animals, foods or types of location. In networks, discriminative AI has been used to quickly assess photos of the wiring in street cabinets to spot problems, or in data centres.
In the network core, discriminative AI is needed to correlate network settings and performance indicators across time and location to better understand how and why faults occur. This spatio-temporal correlation analysis can tackle complicated tasks that no human could address. Also, discriminative AI can detect issues in telecom networks, such as bottlenecks, that could make a network less resilient in the event of a signalling storm.
AI Tools Are Already in Use at Leading Operators
At TM Forum’s Ignite 2024 event in June, several operators discussed the use of AI to move ahead with greater automation in their networks. Hong Kong Telecom has moved to an integrated service operations centre, instead of a network-centric approach, with a closed-loop mechanism. The operator is using AI to improve deep packet inspection to understand problems with home broadband customers’ Wi-Fi.
Orange has built a digital twin network to assess the impact of potential configuration changes on future performance. It also helps predict how types of event could affect performance such as signalling storms or outages. Orange is already using generative AI for a natural language interface to document and assess network cluster logs, evaluating how closely problems correlate with previous fixes.
China Mobile uses digital twins and AI tools to handle complaints manage faults. Industry customers have strict requirements in their service level agreements that can be challenging for complaint teams to meet. To improve efficiency and quality, China Mobile Zhejiang has used a complaint agent powered by a telecom large language model to help understand customer complaints accurately. Then a dialogue-based question-and-answer interface speeds analysis and resolution of core network signalling. To resolve problems the model connects with the work order system to issue orders, providing information on the causes and any related cases for reference.
As a result, China Mobile says it has seen the efficiency of troubleshooting improve by one-third, which is equivalent to adding 30 or more employees to the network operation and maintenance team. The company reports that in multiple scenarios the model delivered advice in seconds based on searches of unstructured knowledge, calculating indicators and assessing quality. In essence, it provided an interactive operations and maintenance assurance service to support teams to make them more effective.
In June 2024, Globe Telecom’s network was somewhere between level 2 and level 3 according to James Lim, VP Core Planning, Engineering and Implementation at DTW Ignite in 2024. Having a stable core network is a critical foundation on which to build AI-based automated operations to achieve level 4. AI provides new opportunities to improve customer experience, service delivery and operations to which Globe is committed alongside responsible AI usage.
Indonesia’s XL Axiata is also using a digital twin alongside TM Forum’s Expected Demand Not Served AI-based framework to estimate the impact of network anomalies and then act. By quickly identifying affected users, XL Axiata can take steps to mitigate problems, for example by adjusting cell settings to improve their reach and compensate for outages in nearby cells. Also, it enables prioritization of issues. This is especially important because when a major core network outage happens, it causes a flood of tickets and alarms that can mask the root cause and make it harder to act effectively to minimize the time and extent of the impact on customers.
In Malaysia, Digital Nasional is testing AI intent-based operations to manage usage of a 5G wholesale network by customers of six mobile network operators. The approach uses network slicing to provide concurrent connectivity tied to a shared multi-operator core network. Using AI tools to manage such a complex core network set-up that is used by six operators highlights the importance of AI to modern networks. With 5G-Advanced now arriving, the need for more-intelligent operations and management to manage multiple slices is becoming more important because 5G-Advanced enables more complicated configurations that digital twin simulations and AI can help to optimize.
Reducing energy usage and maintaining key operational network experience metrics for users is another clear application of AI tools by service providers. For example, Far EasTone reported it has achieved 25% savings in daily power consumption using AI tools. Given AI is often criticized for increasing data centre usage, it’s important to understand AI tools can also be used to reduce the energy costs of core networks and access networks too. Again, operations teams can use digital twins to simulate the impact of reduced energy usage on network services and users, ensuring that key performance metrics are maintained, as well as improving network sustainability.
In Pakistan, Jazz has been using AI algorithms to automatically collate and summarize live network data — replacing manual processes — to improve operations and management efficiency. It reports reducing the time to visualize network performance and identify network problems from four hours to just 90 minutes.
AI-Based Automation Enables Scaling beyond Headcount Limitations
In all regions, communication service providers are increasing the automation of their networks as they seek to launch new services. Scaling up manual processes and headcounts linearly can’t keep up with the exponential rise in network complexity from newer 5G releases and the wider range of customer types and needs.
By using AI tools to streamline operations and management, service providers can proactively anticipate problems and how to resolve them through generative AI and digital twins. They can analyse past events to correlate network settings, faults and the impact on customers using discriminative AI, and they can better manage live network operations and management by using modern AI tools to respond to current events more quickly. By using AI, service providers can enhance the stability of the core network. AI is a key enabler to help companies achieve level 4 of autonomous networks.