The role of digital twins in autonomous networks
A key element of TM Forum’s Autonomous Networks Proect is a six step taxonomy that CSPs can use to measure their autonomous networks progress. Each AN level has a set of characteristics describing the evolutionary stage of the CSP’s journey from fully manual to fully autonomous operations.
Level 4 AN introduces decision-making based on intent-driven, predictive analysis, and the capability to perform closed-loop management of service-driven and customer experience-driven networks via AI modeling and continuous learning. Achieving these capabilities requires complex phases of technology change in CSPs’ networks.
Ultimately, CSPs need to be able to scale orchestration of zero-touch, zero-wait and zero-trouble services end to end across different network domains (for example, radio access, fixed access, core, IP, optical transport and data center/cloud networks). This requires automation of the entire service lifecycle, from ordering to fulfillment, activation, orchestration, management, assurance, optimization and billing.
But expanding or upgrading networks, and introducing new services, are costly actions for CSPs and can negatively impact network performance and customer experience. This includes introducing AI and automation. In a survey of 111 CSPs for our autonomous networks Benchmark report, we asked about the challenges CSPs face when implementing AN. The highest rated was integration across domains.
Given the complexity of networks, the lack of (or only partial) visibility of network assets and resources is a major inhibitor of cross-domain integration. As a result, digital twins are becoming essential to helping CSPs move to higher levels of network autonomy which rely on closed-loop and intent-driven automation, where accurate and reliable decision-making is critical.
Speaking at DTW24-Ignite in a panel session focused on digital twins, Dennis Abella, VP and Head of OSS, Globe Telecom, said: “Digital twins offer critical insights into the performance of autonomous networks, and addressing the challenges of assurance, given the increasing complexity of service diversity.”
Service assurance is at the heart of the business case for autonomous networks, and digital twins show CSPs how to organize and adjust to get the greatest business value from AN use cases.
AI, and increasingly generative AI (GenAI), play a vital role in creating and operating digital twins.
In turn, digital twins are critical to understanding the impact of AI-based change on networks and services: AI programs are fundamental to digital twin systems, representing the operations of the real world for complex systems that cannot be modeled using rules-based equations Digital twins provide dynamically updated data models simulating the real world that facilitate the development, deployment and operations of AI solutions.
This relationship was the focus of a TM Forum Catalyst project, RAN! Reinforcing Autonomous Networks: AIempowered digital twin for optimization. The project, which aimed to revolutionize 5G mobile network optimization, claimed to use AI in digital twin modeling in complex wireless scenarios for the first time.
“We introduced AI in digital twin modeling for mobile network scenarios,” says Ari Sondang, General Manager for Network Quality Digitalization at Indonesian operator Telkomsel and Project Lead for the Catalyst.
“We were able to analyze diverse data sources, identify problematic cells and make adjustments to optimize performance, all within an automated network environment.” The project used innovative technologies such as highfidelity digital twin simulation, a multi-modal parameter optimization agent, an intelligent beam optimization algorithm, and a realistic network simulation algorithm, to enhance wireless network performance, user experience and operational efficiency.
Frank Xie, Deputy Head, Radio Network, China Mobile Hong Kong, and Catalyst champion, outlined the challenges in a video of the project. “5G networks pose significant network optimization challenges, stemming from their inherent complexity and the vast volumes of data they handle,” he said. “By utilizing network data and employing AI-driven digital twin technology we modeled PCI – [physical cell identity, vital for network interference management] – planning…to optimize network configuration and minimize interference.” The Catalyst leverages AI-driven digital twin technology to pave the way for the deployment of AN Level 4 selfoptimization.
Among the benefits of the solution were:
Xie also noted that this solution can be applied to other use cases such as fault resolution and performance management.
Intent is the driving force of AN, communicating requirements, constraints and preferences to an autonomous system, enabling CSPs to set the goals and parameters for AI models without prescribing specific actions. This allows the network to adapt more fluidly, and enables CSPs to turn their customers’ business goals, or intents, into instructions for the network that can be fulfilled without customers needing to understand the underlying infrastructure. To be viable, intents must be grounded in data: observed, measured and controlled intents with which to drive the autonomous system, and verifiable under dynamically changeable conditions. But this is easier said than done. To manage network and service operational processes data must be accurate and easily accessible, but it is a non-trivial undertaking to integrate many sources and guarantee the integrity of the data. Inaccurate information can distort a digital twin’s reliability, even using robust data integration technologies and strong data management.
Use cases for automation, such as those identified as high-value scenarios in TM Forum’s AN Level 4 Blueprint, do not always identify domains, so cross-domain data co-operation is essential. According to Milham at TM Forum, in AN architecture CSPs need a combination of data flow, digital twins and data-as-a-service to enable this. But while CSPs have vast volumes of data, it typically resides in domain-specific silos, so understanding what data is required, where it resides, and how it can be pulled together can be challenging.
For example, Telstra is building a digital twin of its international network assets, with the intention of using AI-powered digital modeling to automate management and to solve network issues. But Roary Stasko, CEO of Telstra International, has referred to the data challenge inherent in this project.
“These networks are incredibly complex, with cable sectors and legs, cable landing stations, repeaters and every little component needing to be logged,” he said. “So, we’re systematically working through all of our inventory, all of our network data, and all of our IT data, to digitize them and put them into a clean format.”
GenAI offers capabilities that could potentially tackle these challenges and support automation by providing accurate, comprehensive data and act upon potential vulnerabilities or performance issues in the network. This is the aim of a recent Catalyst project, Bridging network data fragmentation with GenAI-enabled federation in Digital Twin.
The project, which included participation from CSP champions Telus and Vodafone, aims to help CSPs by leveraging GenAI to provide comprehensive, accurate data from fragmented network and data sources, enabling efficient operations across the network and services lifecycle. For example, the participants say it can be used to support enhanced assurance processes leading to increased end-user